(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF

NEW HORIZONS IN THE ANALYSIS OF CONTROL AND RAISING
Studies in Natural Language and Linguistic Theory
VOLUME 71
Managing Editors
Marcel den Dikken, City University of New York

Liliane Haegeman, University of Lille
Joan Maling, Brandeis University
Editorial Board
Guglielmo Cinque, University of Venice

Carol Georgopoulos, University of Utah
Jane Grimshaw, Rutgers University
Michael Kenstowicz, Massachusetts Institute of Technology
Hilda Koopman, University of California, Los Angeles
Howard Lasnik, University of Maryland
Alec Marantz, Massachusetts Institute of Technology
John J. McCarthy, University of Massachusetts, Amherst
Ian Roberts, University of Cambridge
The titles published in this series are listed at the end of this volume.
NEW HORIZONS IN THE
ANALYSIS OF CONTROL
AND RAISING
Edited by
William D. Davies
University of Iowa, USA
and
Stanley Dubinsky
University of South Carolina, USA
A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-1-4020-6175-2 (HB)

ISBN 978-1-4020-6176-9 (e-book)
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
Printed on acid-free paper
All Rights Reserved

© 2007 Springer
No part of this work may be reproduced, stored in a retrieval system, or transmitted
in any form or by any means, electronic, mechanical, photocopying, microfilming, recording
or otherwise, without written permission from the Publisher, with the exception
of any material supplied specifically for the purpose of being entered
and executed on a computer system, for exclusive use by the purchaser of the work.
ACKNOWLEDGEMENTS
Any scholarly endeavor such as this one requires the goodwill and
support of many people. This book is no exception. We would like to thank the
anonymous referee for Springer who carefully read the manuscript and offered
suggestions that strengthened the volume.
Thanks as well to the students of the Spring 2007 Syntactic Theory class
at the University of South Carolina who read and commented on a prepublication
draft of this book: Minta Elsman, Carlos Gelormini, Analía Gutiérrez, Eun
Hee Lee, Stephen Mann, Linnea Minich, Eun Young Shin, Stacey Warnick, and
Henry Yum.
Thanks to Helen van der Stelt and Jolanda Voogd for encouraging us to
pursue the project and for their editorial and logistical help. Thanks also to Linnea
Minich for her assistance in preparing and proofreading the manuscript.
We would also like to thank those who helped to bring off the 2005
LSA Institute workshop, New Horizons in the Grammar of Raising and Control,
which gave rise to this book. First of all, our thanks go out to the some seventy-
five workshop participants, who contributed their presentations, posters, and
commentary. Critical to the successful organization of the workshop were
Cheryl Murphy, of the Harvard University Linguistics Department, and our
own graduate students, Craig Dresser and Mila Tasseva-Kurktchieva, who
provided assistance before, during, and afterwards. Major financial support for
the workshop was provided by the National Science Foundation, under Grant
No. 0417880. Additional funding was provided by the University of South
Carolina and the University of Iowa. Sponsoring units at the University of
Iowa were the Office of the Vice President for Research, the College of Liberal
Arts and Sciences, and the Linguistics Department. At the University of South
Carolina, funding was provided by the Office of Research and Health Sciences,
the College of Arts and Sciences, the Department of English Language and
Literature, the Department of Languages, Literatures and Cultures, and the
Linguistics Program.
As usual, our families provided more support and forbearance than we
merit, through our many disruptive visits and absences, always providing the
right balance of inspiration and exasperation. We continue to be grateful for
their love and support through it all. So our very special thanks to Melissa,
Elijah, & Isaac, and to Patty, Billy, & Kate.
v
CONTENTS
Contributors ix
Part I: New Horizons in the Analysis of Control and Raising 1
Looking Out Over the Horizon 3

WILLIAM D. DAVIES AND STANLEY DUBINSKY
Part II: New Views of Raising 13
Raising in DP Revisited 15
IVY SICHEL
The Late Development of Raising: What Children Seem

to Think about Seem 35
CHRISTOPHER HIRSCH AND KEN WEXLER
Raising of Major Arguments in Korean (and Japanese) 71

JAMES H. YOON
Part III: Raising or Control in Greek 109
Not Really ECM, not Exactly Control: The ‘Quasi-ECM’

Construction in Greek 111
GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
Control in Modern Greek: It’s Another Good Move 133

KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
Finiteness and Control in Greek 159

VASSILIOS SPYROPOULOS
Part IV: Control in Romance 185
Moving Forward with Romanian Backward Control and Raising 187

GABRIELA ALBOIU
vii
viii CONTENTS
Agreement and Flotation in Partial and Inverse Partial

Control Configurations 213
CILENE RODRIGUES
Null Subjects in Brazilian Portuguese and Finnish:

They are not Derived by Movement 231
MARCELLO MODESTO
Part V: Extensions and Alternatives to the MTC 249
On (Non-)Obligatory Control 251

CEDRIC BOECKX AND NORBERT HORNSTEIN
Control and Wh-infinitivals 263

MICHAEL BARRIE
Control via Selection 281

JOHAN ROORYCK
Movement-Resistant Aspects of Control 293

IDAN LANDAU
Bibliography 327
Name Index 341
Subject Index 345

CONTRIBUTORS
Gabriela Alboiu Christopher Hirsch

York University Department of Brain and Cognitive
4700 Keele St. Sciences, MIT
Toronto, ON M3J 1P3 46-3027C
Canada Cambridge, MA 02139
galboiu@yorku.ca USA
ckh@mit.edu
Michael Barrie
Department of Linguistics Norbert Hornstein
1866 Main Mall, Buchanan E270 Department of Linguistics
University of British Columbia University of Maryland
Vancouver BC V6T 1Z1 College Park, MD 20742
Canada USA
mibarrie@interchange.ubc.ca nhornste@umd.edu
Cedric A. Boeckx Konstantia Kapetangianni

Harvard University University of Michigan
Boylston Hall 313 1120 Undergraduate Science Building
Cambridge, MA 02138 204 Washtenaw Avenue
USA Ann Arbor, MI 48109-2215
cboeckx@fas.harvard.edu USA
kapetang@umich.edu
William D. Davies
Department of Linguistics George Kotzoglοu
University of Iowa University of the Aegean
Iowa City, IA 52242 1 Demokratias Av.
USA Rhodes 85100
william-davies@uiowa.edu Greece
kotzoglou@gmail.com
Stanley Dubinsky
Linguistics Program
University of South Carolina
Columbia, SC 29208
USA
dubinsk@gwm.sc.edu
ix
x CONTRIBUTORS
Idan Landau T. Daniel Seely

Department of Foreign Literatures Eastern Michigan University
& Linguistics 612D Pray-Harrold
Ben Gurion University Department of English
P.O. Box 653 Ypsilanti, MI 48197
Beer Sheva 84105 USA
Israel tseely@emich.edu
idanl@bgu.ac.il
Ivy Sichel
Marcello Modesto Department of English
Dept. de Lingüística - FFLCH The Hebrew University of Jerusalem
Universidade de São Paulo Mount Scopus
Av. Professor Luciano Jerusalem 91905
Gualberto, 403 Israel
São Paulo, SP 05508-900 isichel@mscc.hji.ac.il
Brazil
modesto@usp.br Vassilios Spyropoulos
University of Aegean
Dimitra Papangeli Department of Mediterranean
AGG Studies
Academy of Athens 1 Demokratias Av.
Chadzimchali 8 Rhodes 85100
Athens 10558 Greece
Greece spiropoulos@rhodes.aegean.gr
dpapangeli@academyofathens.gr
Ken Wexler
Cilene Rodrigues Department of Brain and Cognitive
Instituto da Linguagem/Unicamp Sciences, MIT
Cidade Universitaria Zeferemo Vas 46-3029
Sln. Barão Geraldo, Ca pinas Cambridge, MA 02139
São Paulo USA
Brazil wexler@mit.edu
cilene@iel.unicamp.br
James Hye Suk Yoon
Johan Rooryck Department of Linguistics
ULCL, Leiden University 4080A Foreign Languages Building
P.O. Box 9515 University of Illinois
2300 RA Leiden Urbana, IL 61801
The Netherlands USA
j.e.c.v.rooryck@let.leidenuniv.nl jyoon@uiuc.edu
I
NEW HORIZONS IN THE ANALYSIS

OF CONTROL AND RAISING
WILLIAM D. DAVIES AND STANLEY DUBINSKY
LOOKING OUT OVER THE HORIZON
1. THE GENESIS OF THIS BOOK
Raising and control have been central concerns of generative syntax since the
1960s and continue to be an empirical focus of every comprehensive model that
has come along since. The analysis of these constructions in each framework has
typically relied crucially on the most fundamental assumptions underlying that
framework. Thus, raising and control continue to provide an excellent window
into generative models of syntax, and a useful tool for measuring the validity
of their claims. In the 40 years since the publication of Rosenbaum (1967) and
the 33 years since the publication of Postal (1974), attention to these construc-
tions has persevered through each significant paradigm shift in generative syn-
tax. Interest in these constructions has also broadened (from an initial focus
on English and French) to include analyses of similar (or apparently similar)
grammatical phenomena in a wide range of languages. Most recently, interest in
raising and control has once again surged with the rise of the Minimalist Pro-
gram. At the same time, some of the most recent analyses venture into relatively
underexplored languages and/or grammatical phenomena. Concerned as we are
with empirical results informing theoretical paradigms, we think that renewed
attention to these two constructions, combined with an expanding empirical
basis for analysis, makes this a particularly appropriate time to produce a book
that gathers in one place some of the more interesting work being done on the
topic at this time.
The chapters in this book represent, for the most part, a selection of the papers
and posters presented at a workshop titled ‘New Horizons in the Grammar of
Raising and Control’, which was supported in part by a National Science Founda-
tion grant and which took place at Harvard University as part of the 2005 LSA
Linguistic Institute. The agenda for this workshop was a deeper exploration into
the analysis of raising and control, and was set in part by the presentations and
public discussion held at a January 2005 symposium at the LSA annual meeting.
The aim of that January 2005 symposium was to articulate a set of research ques-
tions to be addressed at the July 2005 workshop. Issues arising from the January
panel included the following:
1. What are the empirical properties of raising and control? How can each be clearly
identified, or has the question become irrelevant? With the movement theory of
control proposed by Hornstein (1999), and adopted in subsequent work (e.g.
Boeckx and Hornstein 2003, 2004, 2005; Polinsky and Potsdam 2002, 2003),
3
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 3–12.
© 2007 Springer.
4 WILLIAM D. DAVIES AND STANLEY DUBINSKY
the ‘base-generated’ analysis of copy raising (Potsdam and Runner 2001) and
others, for some (but by no means all), the line between raising and control
has become less and less prominent. Is the distinction empirically motivated
or simply an artifact of terminology inherited from a rich history of work in
generative linguistics?
2. What constructions (besides the canonical ones) might be subject to a rais-
ing or control analysis? What constructions that have been treated as raising
or control might turn out not to be so? Raising or control have been posited
for cases (such as Japanese) in which the complement is finite and has an
overt complementizer. Backward control (in which the controllee rather than
the controller is overt) has been posited for Tsez and Malagasy (Polinsky
and Potsdam 2002, 2003). It has been observed that control, but not raising,
is possible in nominalizations. Possessor–possessee relations expressed out-
side of the NP have been characterized as possessor raising (and sometimes
possessor control).
3. Besides the core class of obligatory control (OC), what classes of control must
be recognized? What is the relation of partial control, arbitrary control, and
more generally non-obligatory control (NOC) to the canonical cases (Landau
2000; Jackendoff and Culicover 2003)? In some cases, the control and raising
label has also been applied to constructions in which the controlled nominal
or target of raising is overt (i.e. copy raising). In many instances copy raising
combines with issues of finiteness or possessor raising. In other cases, the rela-
tion between the controller and controllee is not local (i.e. super-equi or long-
distance control).
4. What are the syntactic attributes of raising and control? What part does tense,
or finiteness, or clausal completeness play in restricting their distribution? How
are restrictions on the controllee and raisee (e.g. the fact that they must be
complement subjects) determined? And what is the role of semantics in these
determinations?
Workshop proposals were encouraged to address these and other relevant issues,
and proposers were encouraged to bring new empirical data, especially from
understudied languages, into focus.
2. AN OVERVIEW OF THE CONTENTS
This volume presents 13 out of the 22 papers and posters that were part of
the workshop, plus one invited chapter that was not presented there. The book
opens with a section of three papers on raising. Sichel examines the grammatical
characteristics of nominals in Hebrew and provides evidence that some nominals
may best be analyzed as instances of raising, dramatically contradicting the con-
ventional wisdom that DPs are not a domain in which raising can occur. Hirsch
and Wexler report on a series of experiments designed to reveal children’s knowledge
LOOKING OUT OVER THE HORIZON 5
of raising. They argue that the results indicate that raising is not acquired until
around age 7, well after the acquisition of control, and discuss the possible ramifi-
cations for a theory of control involving movement. Yoon examines the uncertain
status of raising in Korean (and by implication Japanese), a structure which has
received both raising and control analyses for almost three decades. Despite much
evidence seemingly to the contrary, he argues for a raising analysis, but one in
which not a subject but a Major Subject raises.
The next three papers examine the status of raising and control in Greek, in
which both structures take finite complements. In examining the so-called raising
structure, Kotzoglou and Papangeli conclude that there is a wealth of data that
resist a raising analysis. They propose instead a prolepsis analysis in which the
raised nominal is never a dependent of the sentential complement. Kapetangianni
and Seely propose a movement account for control in Greek clauses, arguing that
the crucial difference between na complements with control and those without is
that in the control structures na is phi-defective, which provides an environment
for the controlled element to move out of the complement. Examining much of
the same data on control, Spyropoulos argues that subjunctive na clauses are fully
finite and that it is not the case that the controlled element can always be PRO
or open to a movement analysis. Rather, he concludes, control into finite clauses
must be enriched by additional syntactic devices as well as pragmatic and seman-
tic considerations.
Section 3 considers phenomena in two Romance languages: Romanian and
Brazilian Portuguese. Alboiu tackles the so-called backward control structure
in Romanian, where the structure evinces all the hallmarks of control, but the
overt nominal surfaces in the complement clause rather than the matrix clause.
Examining a variety of data points, she argues for the superiority of a movement
analysis of these control phenomena. Rodrigues, too, argues for a movement anal-
ysis of control in Brazilian Portuguese, illustrating how such an analysis provides
an insightful account of some novel agreement and quantifier float facts. Modesto
adopts a different perspective. Focusing on the distribution of pro in embedded
finite complements, which have the properties of OC, he argues that a movement
analysis is unable to account for all of the data and that pro is actually A’-bound
by a topic.
The final four chapters represent four distinct perspectives on the appropriate
treatment of control. In their chapter, Boeckx and Hornstein attempt to bring
NOC under the generalizations of the movement theory of control (MTC),
appealing to processing constraints in part to account for some of the other-
wise intractable features of this construction. For his part, Barrie proposes
two distinct syntactic analyses for the domains of OC and NOC, focusing his
attention on the distribution of obligatorily and non-obligatorily controlled
wh-infinitivals. Rooryck’s chapter presents evidence against the MTC and pro-
poses instead that control be analyzed in terms of semantic selection, rather
than movement or syntactic devices. Finally, Landau’s chapter lays out a distinct
battery of empirical arguments against the MTC, arguing that his Agree-based
approach provides a superior account for split control, partial control, case
percolation, and others.
3. SURVEYING THE RESULTS OF THE RESEARCH
The proposal that OC is best analyzed as movement (Hornstein 1999) and the
ensuing debate has blurred the distinction between control and raising somewhat.
Regardless of one’s position in the debate, this has brought a welcome focus on
the border of control and raising. The chapters by Sichel; Hirsch and Wexler;
Yoon; and Kotzoglou and Papangeli examine issues on this border.
One area where there has been assumed to be a bright line between control and
raising is the distribution of these two constructions in nominals. It has long been
held that while control is possible in nominalizations, raising is not. Evidence of
this is seen in the paradigms in (1) and (2).
(1) a. The regents attempted to undermine the faculty.

b. the regents’ attempt to undermine the faculty
(2) a. The candidates appear to be late.

b. *the candidates’ appearance to be late
Facts such as these have been cited as evidence against the movement theory of con-
trol by Culicover and Jackendoff (2001) and led Boeckx and Horstein to conclude
that the syntax of verbal phrases and nominal phrases is distinct, particularly as
regards this issue.
Sichel’s paper reopens this debate with some startling evidence that certain
DPs in Hebrew are most insightfully analyzed as involving raising. Marshaling
evidence from selectional restrictions, expletives, and idiom chunks, Sichel argues
that Rina in the DP in (3) has raised from the CP complement of the nominal
sikuyim ‘chances’ into its position in the DP.
(3) ha-sikuyim Sel rina [le-hagi’a ba-zman]

the-chances of rina to-arrive on-time
‘the chances of Rina arriving on time’
Of course, granting that the raising analysis is the superior solution does not close the
debate. Indeed, the Hebrew facts raise a fistful of new questions. Why is raising possi-
ble in the Hebrew DP but seems not to occur in English and other languages for which
raising from a clausal complement is not a marginal structure? What are the particu-
lar features of Hebrew (or any other language) that sanction raising in DP? What
instances of raising in DP might there be that have gone unnoticed or misanalyzed?
In order to answer these and other questions, data from sources other than
synchronic syntactic paradigms may provide avenues to new insights. One such area
is language development, and in their chapter Hirsch and Wexler discuss new data
on the raising in first language acquisition. They report results from a series of
experiments that indicate a strong correlation between raising (Bart seems to Lisa
to be playing an instrument) and passives of the psychological predicates remember,
hate, love, and see. They account for this strong correlation and the fact that both
structures are acquired late using phase theory and are dependent on whether the
children recognized a defective v (one that does not assign an external theta-role),
a feature the structures share. Hirsch and Wexler also touch on the relationship
between raising and control. The literature fairly uniformly indicates that OC (with
verbs like try) is acquired by age 3, whereas their studies indicate that raising is
acquired later, at age 7. (For a dissenting view on the age of acquisition of raising,
see Becker 2005, 2006.) Taken at face value, one might see this as evidence against
the movement theory of control. Nevertheless, Hirsch and Wexler hold out the pos-
sibility that, under the phase theory analysis they propose, such a radical conclusion
might be unnecessary. This could, of course, provide impetus for devising experi-
ments that would more directly bear on this particular issue.
Another area in which the line between control and raising has always been blurry
is that of raising out of finite complements. Over the last three decades a number of
cases have been cited, most notably in Japanese (Kuno 1976) and Greek (Joseph 1976;
Ingria 1981). In early generative models, raising out of a finite complement, while rare,
posed no particular theoretical quandaries, since in those analyses the nonfiniteness
of the complement was (optionally) triggered by the raising rule. However, this state
of affairs changed with the rise of Government and Binding Theory and the notion
of triggered (Case-motivated) movement. Both raising and control type analyses have
been proposed for the Japanese and Greek constructions, and the debate has contin-
ued for over 30 years since the issue was first raised. In his chapter, Yoon examines this
issue in Korean (and to an extent in Japanese) and Kotzoglou and Papangeli do the
same for Greek. Interestingly, the authors arrive at different answers regarding raising
or control type analyses for the respective languages.
Yoon proposes that raising is involved in the derivation of the sentence in (4).
(4) Cheli-nun Yenghi-lul yenglihay-ss-ta-ko mitnun-ta.

C-TOP Y-ACC smart-PST-DECL-COMP believe-DECL
‘Cheli believes Yenghi to have been smart.’
In his analysis, Yenghi raises from a position in the complement clause to its posi-
tion in the matrix clause. While there are many advocates for a raising analysis,
there are others who cite troublesome data that seem to indicate a raising analysis
is untenable and that some type of control analysis is superior. But in Yoon’s
analysis it is not the subject of the embedded clause that raises but the Major
Subject, a term that traditionally refers to the first nominative element in a mul-
tiple nominative construction. Yoon not only shows that his analysis can account
for the thorny data but also presents data that are problematic for both standard
raising and control analyses.
For their part, Kotzoglou and Papangeli reach a very different conclusion for raising
from a finite complement in Greek. They propose a prolepsis analysis in which an
accusatively marked NP, Maria in (5), is never a dependent of the complement clause.
(5) o petros ithele ti maria na traghudhai oli mera.

the Peter.NOM wanted.3SG the Maria.ACC SBJ sing.3SG all day
‘Peter wanted Maria to sing all day.’
Under their analysis, Maria is generated as an object of the matrix clause and is
coindexed with a null pronoun in the complement. Citing a variety of evidence
including data from adverb placement, negative polarity items, and clitic dou-
bling, they demonstrate that a raising analysis consistently fails to account for
data that the prolepsis handles naturally.
Yoon’s Major Subject proposal adds to the partial typology of raising pro-
posed in Davies 2005, where common properties of raising and copying raising are
contrasted with those in the prolepsis analysis proposed for Madurese, and which
extend quite naturally to Kotzoglou and Papangeli’s analysis for Greek (which
arose simultaneously and independently). Among the properties that the Korean
structure shares with the prolepsis structure are: the raised NP not being limited
to the embedded subject, the interpretation of idioms, immunity to islands, and
various interpretive differences. However, as Yoon shows, while the raised NP need
not be limited to embedded subjects, it is nonetheless constrained to Major Sub-
jects, and in addition to sharing a number of properties with prolepsis structures,
there are important differences as well. In addition to these structures are the kind
of backward raising structures described by Polinsky and Potsdam (2006), and
which are touched on in passing in Alboiu’s chapter on Romanian, where there is
evidence of an embedded dependent in the matrix clause despite the fact that that
NP is spelled out in the sentential complement.
Thus, a comprehensive typology of raising must include a variety of structures
with detailed accounts of their characteristic properties. However, with regard to
the contributions in this volume, it remains to be seen whether the Major Subject
analysis or the prolepsis analysis might not be extended to and even more appro-
priate for other reported cases of raising out of finite complements.
In some ways, the chapters on control can be viewed largely as a referendum
on the MTC. While true in part, the authors actually do much more. They add
new observations and paradigms to the empirical storehouse, at times with very
fine-grained and detailed data, that speak to some of the issues swirling around
control. As noted above, the distinctiveness of control and raising has often been
blurred. To the extent that data contraindicate the MTC, they potentially pro-
vide evidence for distinguishing the two constructions. To the degree that data are
consistent with the MTC, they provide potential evidence for not distinguishing
between control and raising in terms of movement (and perhaps not at all, syn-
tactically).
Many of the empirically focused papers deal with or touch on control into finite
clauses. This may be unsurprising: one of the key issues that has emerged within
the past decade is the status of what seems to be control into finite clauses (which
complements in many ways the study of raising out of finite clauses, such as that
discussed in Yoon). The chapters here are all concerned with how ‘complete’ the
finite complement is.
Rodrigues and Modesto, both analyzing Brazilian Portuguese, come to
different conclusions about the MTC. Rodrigues examines partial control, which
has been argued by some to present severe challenges to the movement theory.
Rodrigues sheds new light on partial control in her analysis of sentences like (6),
with its unexpected agreement pattern.
(6) A vítima quer se encontrar bêbada/* bêbadas.

the victim.FEM.SG wants.3SG se meet.INF drunk.FEM.SG/drunk.FEM.PL
‘The victim wants to meet drunk.’
Here the participle bêbada in the embedded clause agrees with the matrix subject,
which is singular, rather than the embedded PRO, which is presumably plural.
This is predicted by neither the MTC nor the Agree-based analysis. Rodrigues’
novel proposal of stranding a pro initially adjoined to the embedded DP provides
an intriguing solution to this tough nut of partial control.
Modesto, while not directly addressing the issue of control in Brazilian Portu-
guese, analyzes the null subject of embedded finite clauses, a DP that has all or most
of the characteristics of the obligatory-controlled position in nonfinite clauses. He
argues against a movement analysis of pro, citing a number of problems. Perhaps
the most empirically interesting of these is the shifting interpretation of the embed-
ded null subject. In most instances the embedded null subject must be interpreted as
coreferent with the matrix subject. However, this interpretation shifts if the object is
questioned (or relativized).
(7) Quem2 que a Dani1 convenceu t2 que e*1/2 pode se eleger?

who that the Dani convinced that can self elect.INF
‘Who did Dani convince that s/he can get elected?’
Modesto proposes an analysis in which pro must be A’-bound by an element in a

functional projection in the complementizer layer. Modesto’s account for the vari-
able interpretation of a controlled (or bound?) null subject in a finite clause may
prove illuminating in analyzing other recalcitrant data.
Alboiu’s chapter on backward control in Romanian not only argues for the MTC
but also touches on the issue of what constructions might ultimately be included in
the domain of control or raising. The DP that is involved in the control relation can
be realized in multiple positions, some of which are in the complement clause, a fact
which leads Alboiu to argue that PRO-based analyses are unsuited to Romanian,
as they do not include an embedded lexical DP. Conversely, the copies of movement
are available in the complement for spell out in the MTC analysis. As is true of
the Greek structure discussed in the Kotzoglou and Papangeli, and Spyro-
poulos chapters, the complement in the Romanian control construction is often
subjunctive, and thus finite. In Alboiu’s solution, the permeability of the complement
owes to the fact that it is not a phase. The choice of the site to spell out the DP is linked
to semantic and pragmatic considerations. Alboiu’s attempt to explain the factors
underlying this choice represent an important step in understanding and delimiting
backward control.
In looking at OC vs. NOC in Greek subjunctive complements, Kapetangianni
and Seely reach conclusions similar to but not matching those of Alboiu: the
subjunctive complement of some predicates is incomplete. For Kapetangianni
and Seely this incompleteness is manifested in what they call ‘phi-defective Agr’.
Operationally, matrix predicates in obligatory-control structures select comple-
ments with a phi-defective Agr. Building on Chomsky’s (2001a) proposal that only
phi-complete Agr can value Case, they assert that the phi-defective Agr is unable
to value Case and therefore the embedded subject DP must look to the matrix
clause for Case checking. Thus, Kapetangianni and Seely propose an analysis con-
sistent with the MTC (and explicitly argue against some PRO-based proposals).
In examining some of the same Greek data, Spyropoulos comes to a very dif-
ferent conclusion. Spyropoulos argues that the controlled element in Greek OC
structures is pro. He bases this principally on the fact that a lexical DP is pos-
sible in this position and that case agreement facts indicate that the controlled
element has nominative case. Spyropoulos thus rejects both MTC and PRO-based
analyses. Instead he proposes a solution that relies on the nature of the [Tense]
feature on the embedded C. Its absence or dependent nature induces matching
of the matrix AGR and the complement AGR, which licenses pro and assures its
phi-features. In this regard, Spyropoulos’ analysis is reminiscent of some of the
mechanisms introduced in Landau’s work absent the presence of PRO.
The final four chapters of the book (by Boeckx and Hornstein; Barrie;
Rooryck; and Landau) all contend in some manner with the syntactic unity of
control phenomena, and the ability of the MTC to account for the full range of
these.
Boeckx and Hornstein’s chapter, while obviously well-disposed toward the
MTC, notes that certain aspects of NOC are troublesome for that account (and
that – as they acknowledge – the MTC literature has mostly ignored NOC). At the
same time though, they note that adopting a distinct, nonmovement analysis of
NOC (e.g a null pro account) does not solve the problem. B&H would rather not
categorize sentence types as OC or NOC, since these are more properly relations
between nominal expressions than they are types of clausal complements. Instead,
B&H propose utilizing the MTC to handle all cases of control, while also allowing
null pro to occur freely in non-case positions.
Parser preferences are such, according to them, that trace is always preferred
to pro when possible, meaning that the empty category in (8) will be trace, and
coindexed with Mary rather than John. If pro could be dropped into this position,
then (8) might have an interpretation analogous to (9).
(8) John1 persuaded Mary2 [[empty category]2, 1 to leave].

*
(9) John1 persuaded Mary2 [that he1 should leave].
In cases where movement (hence trace) is not possible, the insertion of pro leads to
the relevant interpretations. In this way, by both allowing the insertion of pro and
by assuming that the parser is transparent to grammatical principles, the distribu-
tion of pro (vs. movement) in NOC constructions is accounted for.
For his part, Barrie explores a different approach to the OC/NOC distinction,
focusing on the distribution of wh-infinitivals. Contrary to B&H’s assertion that
the terms OC and NOC merely characterize distinct types of relations between
nominal expressions, Barrie presents the distinction as one that types complement
clauses, and one that is determined in part by semantic selection and in part by
pragmatic meaning. One of the (several) interesting syntactic contrasts noted in
Barrie’s chapter is that in (10), where extraction out of a wh-island is permitted

under generic (or arbitrary) control, but not under OC.
(10) a. What kind of bee does Mary know how to defend oneself against?
b. *Where did John wonder who to introduce himself to?
Barrie’s categorization of control in wh-infinitivals is distinct both from tradi-

tional analyses (which typically posit PROarb for all the relevant cases) and from
Landau (2003) (in which partial control and NOC are analyzed as one category).
Under his analysis, OC (which includes both exhaustive and partial control) is
derived through movement, while NOC involves a null generic pro.
Rooryck’s chapter, if anything, is a further step away from the perspective
of B&H, contending as he does that the distribution of control and its subtypes
can be accounted for under a fine-grained theory of lexically governed comple-
ment selection. The empirical motivation for his proposal (supported by Dutch,
English, French, and German data) comes from a consideration of four gener-
alizations that are inexplicable under the MTC: (1) the case of variable but local
control (VLC) verbs; (2) the semantic coherence of variable control verbs; (3) the
semantic coherence of each subtype of variable control verb; and (4) the case
of split control and progressive Aktionsart. Invoking the semantics of the event
structure of VLC verbs, Rooryck sets forth a semantic account for control meant
to supplant the MTC.
The final chapter of the book, by Landau, continues in the same vein as that of
Rooryck, presenting further challenges to the MTC and generally arguing for the
approach adopted in Landau (2002, 2003, 2004a). Here Landau’s empirical objec-
tions to the MTC are founded on the claim that it fails to account for Minimal Dis-
tance Principle (MDP) violations, partial control, OC into finite complements and
wh-infinitives, case-marking of PRO, PRO-gate effects, passive of subject-control
verbs and sideward movement from OC complements. In addition to using English
data, Landau’s arguments consider data from a variety of languages, including
Brazilian Portuguese.
4. ASSESSING THE RESULTS
A fair amount of the discussion in this book tests the boundaries of the raising/
control distinction, or challenges the assertion that there is a distinction. As one
will readily see from the content of many of the chapters, the MTC lurks as a
major protagonist, on- or just off-stage. In some chapters, such as Sichel’s, evi-
dence that had proved troublesome to the MTC is shown to be less of (or no longer)
a factor. In others, such as Hirsch and Wexler’s, new evidence that troubles the MTC
(or any other unificational account, such as that of Culicover and Jackendoff 2001)
is brought out onstage. Proposals brought forward here range from those which sup-
port the contention that control (like raising) involves movement, to those which
claim that some apparent raising constructions (like control) do not. There is
clearly not yet a complete consensus on whether raising and control are the same
construction, nor on the precise syntactic (or semantic) mechanisms that should
account for either (or both).
Another area of contention showing up in these pages concerns the unity of
control itself. As is clear from reading Boeckx and Hornstein, the integration of
NOC into OC under a unified MTC is troublesome at best. Barrie’s chapter, in
contrast, suggests that there may be good empirical reasons for dividing them
from one another syntactically (a position with which we have some sympathy).
One the one hand, it is good to have more empirical arguments put on the
table, as so many of these chapters do. On the other hand, it is clear that the
empirical evidence is still contradictory in certain data domains and in certain lan-
guages, and in these circumstances the decision of whether to separate or merge
one’s account of raising and control, and whether to attribute their properties to
syntactic or semantic components of the grammar still winds up being, to some
extent, a matter of theoretical predisposition.
It is clear from the contributions in this volume that we now know much, much
more about the phenomena than was known previously (even as recently as a
1999, when the MTC made its debut). The MTC proposal itself has stimulated
a host of new research on this topic, and has made an enormous contribution
to the field (regardless of whether it should ultimately turn out to be the correct
proposal or not). We should be grateful for this. This new research on raising and
control, a respectable sampling of which appears in this volume, has delved into
underexplored languages, new syntactic data paradigms, and other subfields of
linguistics. The Hirsch and Wexler chapter, for one, shows the advances that might
be gained from applying first language-acquisition research methods to the prob-
lem. Other work of this sort has been presented in Becker (2005, 2006), among
others. We are of the mind that empirical considerations must play a greater role
than theory-internal ones for any lasting progress to be made in the understanding
of these phenomena, and believe that research contributions from other quarters
(e.g. psycholinguistics, historical linguistics) will ultimately be of great importance
to the analysis of raising and control (or raising-control).
II
NEW VIEWS OF RAISING

IVY SICHEL
RAISING IN DP REVISITED
1. ON THE ABSENCE OF RAISING
The absence of raising in DP has figured prominently in syntactic theory ever

since Chomsky (1970). As is well known, English has nominal counterparts to
control configurations (1) though apparently no (non-gerund) nominal counterparts
to raising and exceptional case marking (ECM) with infinitives (2):
(1) a. John1 refused/promised [PRO1 to leave]
b. John1’s refusal/promise [PRO1 to leave]
(2) a. John1 appeared/was certain [ t1 to be on time]
b. *John1’s appearance/certainty [ t1 to be on time]
c. We wanted [Mary to arrive on time]
d. *Our desire [of Mary to arrive on time]
e. Our desire [for Mary to arrive on time]
The restriction against raising in DP has contributed to theories about various

subcomponents in the grammar. Beginning with the lexicalist hypothesis, the
development of GB has often proceeded comparatively, clauses vs. noun phrases,
or verbs vs. nouns, with the categorical contrast with respect to raising and ECM
playing a role in the development of government, the ECP, and ECM (Kayne
1984); inherent vs. structural case marking (Chomsky 1986a); and theories of
argument–structure, complementation, and C-selection (Abney 1987; Grimshaw
1990). The claim made in Chomsky (1970) that the input to nominalization can-
not be derived by transformation left open the possibility that a post-nominaliza-
tion A-movement operation produces (2b). Expanding the empirical paradigm
to include the restriction on ECM (2d), Kayne (1984) argues that both (2b) and
(2d) derive from the deficiency of N° (vs. V°) with respect to exceptional govern-
ment of the subject of the embedded IP, resulting in a Case Filter violation in
the latter and an ECP violation in the former. Focusing on predicted similari-
ties between noun phrases and clauses, and broadening the empirical context to
include raising in gerund DPs (John’s being certain/likely to be late), Abney (1987)
argues that raising in gerunds attests to a VP within DP and that N° (unlike V°)
15
© 2007 Springer.
16 IVY SICHEL
does not select reduced clausal complements (a prerequisite for both raising and
ECM) or impose a subject requirement (a prerequisite for A-movement). In the
spirit of Kayne’s proposal, Abney’s treatment invokes a single factor (absence of
V°), but implies a double violation, raising now being excluded due to both the
unavailability of nominal IP complements and inactive EPP. Similarly, Chomsky
(1986) treats the ungrammaticality of raising as stemming from two independ-
ent violations: the restriction of Case-assignment by N° to its arguments (‘inher-
ent Case’) and an independent semantic requirement that A-moved DP must be
‘affected’.
Judging from the heterogeneity of accounts given for (2), and persistent lack
of clarity regarding the ultimate source of violation – the source position within
IP, the target position spec DP, or both – it appears that the restriction was never,
in fact, fully understood. And despite its centrality to syntactic theory, the empiri-
cal claim has gone virtually unchallenged since Postal (1974). Postal (1974) argued
that alternations such as those in (3), with infinitive and gerund complements
to N°, support Raising-to-Subject in noun phrases.1 Yet unlike raising in IP, the
construction in noun phrases fails to produce grammatical results with standard
diagnostics such as expletive and idiom chunk movement, in (4):
(3) a. the tendency [for John to leave]
b. John1’s tendency [t1 to leave]
c. the likelihood [of Nixon’s winning]
d. Nixon1’s likelihood [of t1 winning]
(4) a. *there’s tendency to be a problem
b. *its tendency to rain in June
c. *its tendency to annoy me that Jane is late
d. *its likelihood of raining/of annoying me that Jane is late
e. *the shit’s tendency to hit the fan in these situations
f. *the shit’s likelihood of hitting the fan in these situations
While it is no doubt conceivable that the theoretical tools provided by GB were not
sufficiently restrictive to produce a conclusive understanding of the restriction, it
is equally possible that facts regarding its ungrammaticality in English DPs with
infinitives are insufficient, on their own, to fully determine its analysis. I argue
here that in fact, contrary to the expectation for a universal restriction raised by
the proposals mentioned above, raising in DP from infinitives does exist. Hebrew
DPs headed by nonderived nouns such as ‘chances’, ‘tendency’, ‘opportunity’,
denoting, roughly, modality or degrees of certainty with respect to the eventuality
denoted by the embedded infinitive, exhibit the range of effects typically found
in clausal raising constructions.2 As shown below, they differ systematically from
uncontroversial nominal counterparts to control predicates in (5) and (6)3:
RAISING IN DP REVISITED 17
(5) a. ha-nisayon Sel rina1 [PRO1 le-hagi’a ba-zman] control

the-attempt of rina to-arrive on-time
‘Rina’s attempt to arrive on time’
b. ha-havtaxa Sel rina1 [PRO1 le-hagi’a ba-zman]
the-promise of rina to-arrive on-time
‘Rina’s promise to arrive on time’
(6) a. ha-sikuyim Sel rina1 [ t1 le-hagi’a ba-zman] raising

the-chances of rina to-arrive on-time
‘Rina’s chances of arriving on time’
b. ha-netiya Sel rina1 [ t1 le-hagi’a ba-zman]
the-tendency of rina to-arrive on time
‘Rina’s tendency to arrive on time’
DPs of the sort in (6) show typical Case/theta splits, such that a possessor DP
may be theta-marked as an embedded subject, yet Case-marked genitive in the
DP domain. It is shown, in section 2, that both expletives and idiom chunks are
licensed in the genitive position. Section 3 goes on to motivate a movement analy-
sis of the theta-Case split, as schematized in (6), based on the distribution of focus
particles, agreement, negative concord, and extraposition.
The implications of raising for the analysis of control in DP and preliminary sup-
port for obligatory control (OC) are discussed in section 4. With the abandonment
of government within the Minimalist Program (Chomsky 1995), new and more fine-
grained approaches to infinitives neutralize the significance of the ever mysterious IP/
CP raising/control infinitive divide (Martin 2001; Bošković 1997; Wurmbrand 2001),
and reopen empirical questions surrounding the similarities and differences in the inter-
pretation and distribution of raising and control and how they are to be encoded in the
grammar. Recent studies of control lead to firmer conclusions regarding the insepara-
bility of the distribution and the interpretive properties of the null embedded subject,
i.e. OC vs. NOC, and assimilate OC to anaphors, from which the distribution of OC is
derived in various ways (Wurmbrand 2001; Landau 2001; Hornstein 1999; Hornstein
2001). On Hornstein’s approach, the anaphoricity of OC is directly related to its status
as an NP-trace, leading to the expectation that OC should be licensed in exactly those
configurations which license A-movement, including the sidewards variety (Nunes
2004). DP may serve therefore as an important testing ground for the A-movement
hypothesis such that if OC is A-movement and control is attested within DP, so is rais-
ing expected (Culicover and Jackendoff 2001). Given earlier conclusions regarding the
unavailability of raising in DP, its existence in Hebrew, and possibly English as well,
will shape predictions regarding control; while Culicover and Jackendoff (2001) con-
sider the absence of raising in DP as straightforward evidence against the A-movement
hypothesis, its presence makes control in DP much less surprising. However, whether
or not the availability of raising within Hebrew DP (and possibly also in English DP)
provides new evidence directly supporting an A-movement analysis depends ultimately
on the nature of control in DP and on the complement or adjunct status of the infini-
tive – questions which have not yet been conclusively settled on independent grounds
(Stowell 1981; Grimshaw 1990; Hornstein 2001; Boeckx and Hornstein 2003). As a step
toward resolving these questions, I show that, allowing for pervasive implicit Agents of
nominals, control in DP is most probably of the OC variety.
18 IVY SICHEL
2. EVIDENCE FOR THE NON-THEMATIC NATURE

OF THE GENITIVE DP
The empirical basis for a distinction between raising and control in clauses is
grounded in the relation between thematic licensing and Case-assignment, such
that in raising the matrix subject is theta-licensed by the embedded predicate and
Case-marked in its surface position in the matrix clause. Such theta-Case splits are
observed most clearly with non-referential subjects such as expletives and idiom
chunks, diagnostics which in English do not produce the expected grammatical-
ity if Postal was correct in claiming that nouns like ‘tendency’ allow raising (see
(4) above). While for English the case for raising had to be made on independent,
perhaps less convincing grounds, leaving open the possibility that non-referential
possessors were excluded for independent reasons, Hebrew does show evidence of
the sort familiar from IP. IP-raising diagnostics are shown in (7) for English and
in (8) for the Hebrew-raising adjectives xayav ‘certain’ and alul ‘possible’, which
select infinitival complements. These include the absence of selectional restric-
tions imposed on the subject and the possibility of having an expletive or an idiom
chunk raised from the embedded subject position where licensed:
(7) a. The theory1 seems [t1 to be correct]
b. There1 seems [ t1 to be a problem with the theory]
c. The shit1 seems [ t1 to have hit the fan]
(8) a. ha-te’oria1 xayevet / alula [t1 lihiyot nexona]

the-theory.F.S certain.F.S / possible.F.S to-be correct.F.S
‘The theory is certain / should (to) be correct’
b. ze1 xayav/ alul [ t1 lihiyot naxon [Se-dina nicxa]

it certain / possible to-be true that-dina won
‘It is certain / could (to) be true that Dina won’
c. ha-kerax1 xayav / alul [t1le-hiSaver be-macav kaze]

the-ice certain / possible to-break in-situation like-this
‘the ice is certain / could (to) break in this kind of situation’
Within DPs, a consistent difference is observable between nouns such as ‘chances’,

‘tendency’, and ‘opportunity’ on the one hand, and nominal counterparts to
control verbs on the other. The former appear to impose no selectional restric-
tions on the genitive DP and allow expletives and idiom chunks associated with
the embedded subject position. Yet in both types of DP, the DP is marked genitive
by Sel, which on standard assumptions is associated with the nominal portion
of the DP, N° or its functional projections. In other words, there exists a class
of nouns to which I refer as ‘raising’, which Case-marks their possessors with-
out bearing a thematic relation to them, on a par with the syntactic subjects or
clausal raising predicates. Examples (9) and (10) show a contrast between raising
and control nouns with respect to selectional restrictions, (11) and (12) show the
difference with genitive expletives, and (13) and (14) show it for idiom chunks as
genitive DPs. For convenience, the underlying structure is given under the Roman
numeral heading each diagnostic4:
I. Selectional Restrictions: N° [the theory to be correct]
(9) a. [ha-sikuyim Sel ha-te’oria lihiyot nexona] kluSim le-maday ‘raising’

the-chances of the-theory to-be correct.F.S slim quite
‘The chances of the theory being correct are pretty slim’
b. [ha-netiya Sel te’oriot lihiyot nexonot] yedu’a le-kulam

the-tendency of theories to-be correct known to-all
‘The tendency for theories to be correct is known to all’
c. [ha-histabrut Sel ha-te’oria lihiyot nexona] krova le-efes

the-probability of the-theory to-be correct.F.S close to-zero
‘The probability of the theory being correct is close to zero’
(10) a. *[ha-havtaxa Sel ha-te’ora lihiyot nexona] hirgiza otanu ‘control’

the-promise of the-theory to-be correct.F.S annoyed us
b. *[ha-nisayon Sel ha-te’oria lihiyot nexonot] hirgiz otanu

the-attempt of the-theory to-be correct annoyed us
c. *[ha-seruv Sel ha-te’oriot lihiyot nexonot] hirgiz otanu

the-refusal of the-theory to-be correct annoyed us
II. Expletives: N° [ it to-happen [CP that . . . ]]
(11) a. [ha-sikuyim Sel ze likrot [Se-bibi yibaxer]] tovim ‘raising’

the-chances of it to-happen that-Bibi will.be.elected good
‘The chances of it happening that Bibi is elected are good’
b. [ha-netiya Sel ze likrot [Se-metunim ne’enaSim]] lo yexola lihiyot mikrit

the-tendency of it to-happen that-moderates get.punished can’t be accidental
‘The tendency for it to happen that moderates are punished cannot be accidental’
c. [ha-histabrut Sel ze likrot [Se-bibi yipasel]] krova le-efes

the-probability of it to-happen that-bibi will.get.disqualified close to-zero
‘The probability of it happening that Bibi gets disqualified is close to zero’
(12) a. *[ha-nisayon Se ze likrot [Se-bibi yibaxer]] hifti’a otanu ‘control’

the-attempt of it to-happen that-bibi will.be.elected surprised us
b. *[ha-seruv Sel ze likrot [Se-bibi yipasel]] hifti’a otanu

the-refusal of it to-happen that-bibi will.get.disqualified surprised us
c. *[ha-havtaxa Sel ze likrot [Se-bibi yibaxer]] hid’iga otanu

the-promise of it to-happen that-bibi will.get.elected worried us
III. Idiom Chunks: N° [ SUBJ-idiom PRED-idiom]
(13) a. [ha-sikuyim Sel ha-kerax le-hiSaver be-macav ka-ze] kluSim ‘raising’

the-chances of the-ice to-break in-situation like-this slim
‘The chances of the ice breaking in this kind of situation are slim’
20 IVY SICHEL
b. [ha-netiya Sel ha-kerax le-hiSaver be-macav ka-ze] yedu’a le-kulam

the-tendency of the-ice to-break in-situation like-this known.F to- all
‘The tendency for the ice to break in this kind of situation is know to all’
c. [ha-efSarut Sel ha-kerax le-hiSaver be-macav ka-ze] mutelet besafek

the-possibility of the-ice to-break in-situation like-this held.F in-doubt
‘The possibility of the ice breaking in this kind of situation is doubtful’
(14) a. *[ha-nisayon Sel ha-kerax le-hiSaver be-macav kaze] hu tipSi ‘control’

the-attempt of the-ice to-break in-situation like-this is silly
b. *[ha-seruv Sel ha-kerax le-hiSaver be-macav kaze] hu tipSi

the-refusal of the-ice to-break in-situation like-this is silly
c. *[ha-havtaxa Sel ha-kerax le-hiSaver be-macav kaze] hi tipSit

the-attempt of the-ice to-break in-situation like-this is silly
The facts in (9), (11), and (13) consistently point to the existence of a class of
nouns which bears no thematic relation to the genitive Sel-DP: it imposes no
selectional restrictions on it and allows expletives and idiom chunks which are
clearly licensed only in the embedded clause. Control nouns, in contrast, impose
selectional requirements and do not allow non-referential DPs such as expletives
and idiom chunks, a consequence of the theta-relation they bear to Sel-DP.
Yet while (9)–(14) show that there is no necessary thematic relation between
raising N° and the genitive DP, and that a genitive DP may be non-referential, it
is still conceivable that when Sel-DP is referential and is semantically compatible
with the requirements for possession, it is then thematically licensed by the posses-
sion relation itself, mediated possibly by functional material in DP. If it is, it will
bear a theta-role and a thematic relation with whichever portion of DP licenses
possession.5 On that scenario theta-Case splits in DP would be limited to non-ref-
erential genitive DP, and the theta-criterion would impose control by a referential
genitive of the embedded infinitive subject, as depicted in (15):
(15) a. ha-netiya Sel rina1 [ PRO1 le-hacliax taxat laxac]

the-tendency of rina to-succeed under pressure
‘Rina’s tendency to succeed under pressure’
b. ha-sikuyim Sel rina1 [PRO1 le-hacliax taxat laxac]

the-chances of rina to-succeed under pressure
‘Rina’s chances of succeeding under pressure’
A control analysis, in contrast to raising, would imply that the denotations of nouns
such as ‘chances’ or ‘tendency’ may in some sense be possessed by individuals, and
that this relation is independent of and in addition to the relation of the noun to
the eventuality denoted by the embedded infinitive. On a raising analysis, chances
and tendencies are monadic, and are associated only with eventualities, on a par with
raising predicates and epistemic modals.6 Whether or not it is feasible to consider the
relation of the genitive to N° in examples like (15) as a possession relation turns out,
however, to be tangential to the syntax of these constructions since it can be shown
that possession, in general, is not necessarily mediated by theta-role. Clausal posses-

sion, at least in Hebrew, has the possessor expressed as a dative (16). The dative posi-
tion allows non-referential DPs such as expletives and idiom chunks, on a par with
the DP-internal genitives discussed above, provided that these are licensed from below,
within the possessed DP. The expletive in (17) has its source within a possessed DP of
the sort seen in (11); the expletive may surface as a dative ‘possessor’, in (17b)7:
(16) yeS le-dina be’aya

be to-dina problem
‘Dina has a problem’
(17) a. ha-sikuyim Sel ze1 [t1 likrot [Se-bibi Suv yibaxer ]]

the-chances of it to-happen that-bibi again will.get.elected
‘the chances of it happening that Bibi will be reelected’
b. yeS le-ze1 [sikuyim [t1 likrot [Se-bibi Suv yibaxer ]]]

is to-it chances to-happen that-Bibi again will.get.elected
‘There are chances of it happening that Bibi will get reelected’
The possibility of having a non-referential DP as clausal possessor provides

independent evidence for the Raising-to-Have analysis proposed in (Szabolcsi 1983,
1994 and Kayne 1993). It also suggests that quite generally possession is not medi-
ated by a thematic relation. Since non-referential DPs clearly cannot possess, a
possessor interpretation of a dative or genitive DP, unlike the kinds of interpreta-
tions regulated by the Theta-Criterion, turns out not to be enforced. The possibility
for the possession interpretation to be withheld from a position to which it can in
principle be assigned is compatible with its neutralization in derived nominals. The
requirements on the possessive interpretation appear to be relatively weak, weaker
even than the requirement on theta-role assignment to subjects of process nominali-
zations, which, as is well known, can be withheld when no genitive is projected, i.e.
the demolition of the house took an hour. Optionality of the possessor interpretation
with overt genitives implies therefore that the genitive position in DP, or the dative
position in clauses with ‘have’, is not restricted in its local domain. More generally,
the possession interpretation, when present, as in (16), is probably not to be associ-
ated with a contentful predicate or head which assigns a theta-role to a DP argu-
ment. If possession is not thematic, the Theta-Criterion does not stand in the way
of a raising analysis of referential Sel-DP: the DP gets its theta-role as subject of
the embedded predicate, with the possibility of having possession configurationally
‘overlayed’ in its derived position in the DP domain.8
Independent motivation for excluding thematic possession is provided by the pres-
ervation of truth-value with an embedded passive. With control nouns, which assign a
theta-role to the genitive DP, passive in the embedded clause identifies the genitive as
an embedded Theme, affecting the denotation of the DP and ultimately truth condi-
tions (18). No such effect is apparent with ‘chances’ or ‘tendency’ (19)/(20):
(18) a. ha-racon Sel ha-eved lir’ot et ha-malka

the-desire of the-slave to-see O.M the-queen
‘the slave’s desire to see the queen’
22 IVY SICHEL
b. ha-racon Sel ha-malka lehera’ot al-yedey ha-eved

the-desire of the-queen to-be.seen by the-slave
‘the queen’s desire to be seen by the slave’
(19) a. ha-sikuyim Sel ha-cava lifgo’a be-mafginim

the-chances of the-military to-hurt in-demonstrators
‘the chances of the military hurting demonstrators’
b. ha-sikuyim Sel mafginim le-hipaga al-yedey ha-cava

the-chances of demonstrators to-be.hurt by the-military
‘the chances of demonstrators being hurt by the military’
(20) a. ha-netiya Sel ha-cava lifgo’a be-mafginim

the-tendency of the-military to-hurt in-demonstrators
‘the tendency for the military to hurt demonstrators’
b. ha-netiya Sel mafginim le-hipaga al-yedey ha-cava

the-tendency of demonstrators to-be.hurt by the-military
‘the tendency for demonstrators to be hurt by the military’
To the extent that there is no change in ‘chances’, from the chances of the military
hurting demonstrators to the chances of demonstrators being hurt by the mili-
tary, and similarly for ‘tendency’, it appears that the only argument these nouns
are associated with is the embedded infinitive. With control nouns, which bear a
thematic relation to the genitive, it matters which of the participants in the embed-
ded eventuality holds a desire (or refusal, attempt, etc.) for that eventuality. With
the nouns in (19) and (20) it does not, suggesting that there is no possession of a
tendency or chances by an embedded event participant and more generally that
the genitive is thematically licensed only from below. If so, the theta-Case splits in
DP observed with non-referential genitives apply to referential genitives as well,
paving the way for a raising analysis.
3. EVIDENCE FOR MOVEMENT
Genitive Sel-phrases occur in Hebrew after the noun they are associated with.
Therefore, the conclusions reached in the section above are not sufficient in
and of themselves to conclusively determine a movement analysis. In addition
to a string vacuous movement analysis, which may be assimilated either to
Raising-to-Subject or Raising-to-Object, schematized in (21a), the post-nominal
position of Sel-DP and the theta-Case split it exhibits are also compatible with
an ACC-ing analysis (21b). The genitive marker Sel is external to an opaque
clause (21b) on a par with the potential ACC-ing complements to English
nouns given in (22):
(21) a. N°. . . Sel DP1 … [t1 infinitive . . . raising (to ‘subject’/‘object’)
b. N° . . . [Sel [DP infinitive . . . ACC-ing (off an N°)

(22) a. the chances of [there being a violent retaliation]
b. the possibility of [it raining]
c. the chances of [the shit hitting the fan]
According to an ACC-ing analysis of (22a), the genitive marker is associ-

ated with the gerund complement, and an opaque, non-ECM-like clausal
boundary intervenes between ‘of ’ and the embedded subject (Reuland 1983);
in other words, the DP following ‘of ’ is not necessarily itself associated with
the genitive Case assigned in DP, the theta-Case split is only apparent, and
no movement bringing the DP close to its Case assigner is involved. By exten-
sion, it is perfectly feasible that Hebrew infinitives are nominal like gerunds in
English and require genitive licensing, in which case the embedded ‘genitive’
DP is not directly associated with genitive and no movement operation would
be involved. Therefore, in order for the argument for raising to be complete,
an ACC-ing type of analysis must be excluded for the Hebrew nouns with the
infinitive complements discussed so far.
The raised status of Sel-DP, as in (21a), is motivated by comparison of
the internal syntax of its containing DP with the internal syntax of a similar
construction which does exhibit the opacity effects expected on an ACC-ing
analysis. Nouns such as ‘phenomenon’ and ‘result’ (23) take nonfinite comple-
ments. They contrast, as shown immediately below, with raising nouns on a
variety of effects diagnosing clausal opacity, including the distribution of focus
particles, clitic doubling, negative concord, and extraposition. They are easily
distinguished from raising nouns since the embedded clause is headed by a par-
ticiple, while in raising constructions the embedded clause is infinitival9:
(23) a. [ha-tofa’a Sel [xayalot metaxkerot palestina’im]] hitrida et batya

the-phenomenon of soldiers.F interrogating Palestinians disturbed batya
‘The phenomenon of soldiers interrogating Palestinians disturbed Batya’
b. [ha-toca’a Sel [yeladim lo yoc’im me-ha-bayit]] hayta cfuya

the-outcome of children not leaving from.the-house was expected
‘The outcome of children not leaving home was expected’
The following set of contrasts is sensitive to the permeability of the clausal bound-
ary within DP. They all show that the embedded clause in DPs of the sort in (23) is
delineated by an opaque boundary. Like the English ACC-ing constructions stud-
ied in Reuland (1983), they do not exhibit ECM-like domain-extension effects,
i.e. it never looks as if the embedded subject may be external to its clause raising
constructions, on the other hand, consistently show ECM-effects. Their presence
attests to a ‘real’ theta-Case split since the embedded subject, even if it were in
situ in the embedded clause, would be structurally Case-marked by the nominal
genitive Sel. Focus particles and negative concord facts show, in addition, that
domain-extension effects are derived by the movement of the embedded subject to
a clause-external position in the DP.
24 IVY SICHEL
Consider first the distribution of focus particles such as also/only, which seem
to be directly dominated by IP material. They cannot normally intervene between
a preposition and its DP complement, unless there is also an IP clausal boundary
separating the preposition qua C° and the DP. This is seen in the following English
ECM construction (24):
(24) a. I baked the cake (only) for (*only) John
b. I prefer for [only John to be there]
A focus particle in (24b) can intervene between P° and DP because here, unlike
(24a), a clausal boundary between P° and DP provides the IP material necessary to
host the particle. The restriction against having a focus particle embedded within a
PP or DP is seen also in Hebrew, including genitive phrases headed by Sel (25b):
(25) a. dibarti (rak/gam) im (*rak/gam) dani

spoke.I only/also with only/also dani
‘I spoke only/also to Dani’
b. ha-sefer hu (rak/gam) Sel (*rak/gam) dani

the-book COP only/also of only/also dani
‘The book is only/also Dani’s’
Given that focus particles cannot intervene between P and DP unless there is IP
material to host them, they are a good diagnostic for sentential boundaries. They
show that a clausal boundary falls between Sel and DP with nouns like ‘phe-
nomenon’, but not with raising nouns of the ‘chances’ type. A focus particle may
immediately follow Sel in (26), but not in (27):
(26) [ha-tofa’a Sel [rak/gam xayalot metaxkerot palestina’im ba-rexov]]

the-phenomenon of only/also soldiers.F interrogating Pals in.the-street
hitrida otanu
disturbed us
‘The phenomenon of only/also soldiers interrogating Palestinians on the street disturbed us’
(27) rak/gam [ha-sikuyim Sel (*rak/gam) xayalot le-nace’ax] tovim

only/also the-chances of only/also soldiers to-win good
‘Only/also the chances of Rina winning are good’
The possibility of having a focus particle between Sel and DP in (26) implies IP mate-
rial and an embedded clause following Sel and preceding DP. Therefore, DP must
occupy an embedded subject position. The ungrammaticality of a focus particle
between Sel and DP in (27) suggests the absence of IP material and that DP is a direct
complement of Sel.
The contrast between (26) and (27) suggests an ECM-like configuration for (27),
in which the embedded subject is directly Case-marked by genitive Sel. Nevertheless,
it does not track movement per se since Sel and its complement could be internal
to the embedded clause (similar to the DP following ‘want’ in English) or exter-
nal to the embedded clause (the result of movement). Similarly, pronominalization
contrasts attest to an opaque clausal boundary with participial clauses following

‘phenomenon’, and a permeable, possibly ECM-like, boundary with the infinitive
under ‘chances’. There are two strategies for pronominalizing the genitive DP in
Hebrew; the pronominal may be cliticized to Sel (28b) or to N° (28c), the latter strat-
egy optionally allowing clitic doubling in the presence of Sel and a doubled DP:
(28) a. ha-simla Sel dina

the-dress of dina
‘Dina’s dress’
b. ha-simla Sela
the-dress of-her
‘her dress’
c. simlata (Sel dina)

dress-her of dina
‘her dress’ or ‘Dina’s dress’ (with clitic doubling)
Raising nouns like ‘chances’ allow both types of pronominalization. The

pronominal clitic may attach to Sel (29b) and to N° (29c), with the option to
double, exactly as in (28c). The class of nouns claimed to take opaque clauses
excludes both. The embedded DP, as a pronominal, cannot cliticize to Sel (30b)
or directly to N° (30c):
(29) a. [ha-sikuyim Sel dina le-nace’ax] tovim

the-chances of dina to-win good
‘The chances of Dina winning are good’
b. [ha-sikuyim Sela le-nace’ax] tovim

the-chances of.her to-win good
‘The chances of her winning are good’
c. [sikuyeha (Sel dina) le-nace’ax] tovim

chances.her of dina to-win good
‘The chances of her(/Dina) winning are good’
(30) a. [ha-tofa’a Sel [xayalot metaxkerot palestina’im ]] hitrida et batya

the-phenomenon of soldierS.F interrogating Palestinians disturbed batya
‘The phenomenon of soldiers interrogating Palestinians disturbed Batya’
b. *[ha-tofa’a Selahen metaxkerot kSiSim palestina’im] hitrida et batya

the-phenomenon of-them interrogating about-60 Palestinians disturbed batya
c. *[tofa’atan Sel [(xayalot) metaxkerot palestina’im]] hitrida et batya

phenomenon.their of soldiers interrogating Palestinians disturbed batya
Assuming that pronominalization is subject to a syntactic locality constraint,

the presence of an opaque clausal boundary between Sel and DP will prevent
pronominalization. On this analysis, the DP is not a complement of Sel, not
even in the most liberal ECM-like sense. The possibility of pronominalization
26 IVY SICHEL
under ‘chances’, on the other hand, implies that the embedded subject DP is
close enough to Sel and N° for cliticization to proceed. As with focus particles,
this could be due to movement of DP to a clause-external, but it may also mean
that DP is in situ with a permeable clausal boundary producing an ECM-like
cliticization effect.
The distribution of negative concord and extraposition point conclusively to
movement of the DP to a clause external genitive position. Negative DPs, includ-
ing subjects and objects, are licensed in Hebrew by clausemate negation, as seen
in (31a) and (31b). Negation a clause up from the negative DP, or a negative DP a
clause up from embedded negation are impossible, as in (31d) and (31e):
(31) a. *(lo) baxarti be-af talmid

NEG choosed.I in-no student
‘I didn’t chose any student’
b. af talmid *(lo) nice’ax

no student NEG won
‘No student won’
c. zaxarti [Se-af talmid lo hitkonen]

remembered.I that no student NEG prepared
‘I remembered that no student prepared’
d. *lo zaxarti [Se-af talmid hitkonen]

NEG remembered.I that-no student prepared
e. *af talmid zaxar [Se-hu lo hitkonen]

no student remembered that-he NEG prepared
With negation outside of the DP containing the embedded clause, and a negative
DP within the nonfinite clause embedded in DP, the result is grammatical for rais-
ing nouns (32b) and ungrammatical for ‘phenomenon’ (32c). (32c) is expected,
given the clausemate requirement, on a par with (31d). The grammaticality of
(32b) is similar to the situation with the ECM complement under ‘remember’
(32a), with high negation licensing an embedded negative subject. It could imply
movement of ‘no student’ to a position outside of the embedded clause, in which
it is close enough to negation, but it could also suggest domain extension, as in the
examples considered so far.
(32) a. lo zaxarti [af talmid mitkonen] ‘ECM’

NEG remembered.I no student preparing
‘I didn’t remember any student preparing’
b. lo he’emanti [ba-sikuyim / netiya Sel af talmid le-hitkonen]

NEG believed.I in. the-chances / tendency of no student to-prepare
‘I didn’t believe in the chances / tendency of any student preparing’
c. *lo saladeti me-[ha- tofa’a Sel [af talmid mitkonen]]

NEG disapproved.I from-the-phenomenon of no student preparing
The pattern of grammaticality with the reverse relative order of negative DP and
negation, however, clearly favors a movement analysis. In (33) negation is within
the clause embedded within DP. Here, a negative DP subject under ‘phenomenon’
is grammatical, as expected, if negation and the negative DP are within the same
clause. (33b), with raising nouns, produces ungrammaticality:
(33) a. *zaxarti af talmid lo mitkonen

remembered.I no student NEG preparing
b. *he’emanti ba-sikuyim / netiya Sel af talmid lo le-hitkonen

believed.I in.the-chances / tendency of no student NEG to-prepare
c. saladeti me- ha- tofa’a Sel [af talmid lo mitkonen]
disapproved.I from-the-phenomenon of no student NEG preparing
‘I disapproved of the phenomenon of no student preparing’’
The ungrammaticality of (33b) can be understood as the result of movement,

which places the negative DP outside of the clause which contains negation, in
which case a clausal boundary will intervene, as in (31e). The combination of
(32b), in which a lower negative DP appears to be high enough to be licensed by
high negation, together with (33b), in which the negative DP appears to be too
high to be licensed by a lower negation, strongly suggests that the genitive DP is
in fact raised from its base position within the embedded clause, and that overt
movement underlies the domain extension facts observed above with focus parti-
cles and pronominalization.
The conclusion that the genitive DP raises, albeit string vacuously, from its
embedded position to a genitive position within DP, is further supported by con-
trasts in extraposition.10 A finite clause embedded within DP may be extraposed
to a clause final position (34). If the genitive DP has raised from its embedded
position, it is expected not to undergo clausal extraposition. Raising nouns with
extraposed infinitives are indeed impossible (35):
(34) a. ha-sikuyim adayn kayamim [Se-dina tibaxer]

the-chances still exist that-dina will.be.elected
‘The chances still exist that Dina will get elected’
b. ha-hizdamnut adayn lo huxmeca [Se-dina tenace’ax]

the-opportunity still not missed that-dina will.win
‘The opportunity hasn’t yet been missed that Dina will win’
(35) a. *ha-sikuyim adayn kayamim Sel dina le-hibaxer

the-chances still exist of dina to-be.elected
b. *ha-hizdamnut adayin lo huxmeca Sel dina le-nace’ax

the-opportunity still not missed of dina to-win
The problem with extraposition in (35) is not likely to be related to the nonfinite
status of the complement, or the presence of the genitive Sel, since extraposition
28 IVY SICHEL
is allowed under ‘phenomenon’ and ‘outcome’, which similarly take nonfinite

embedded clauses preceded by Sel:
(36) a. ha-tofa’a adayn kayemet [Sel anaSim medabrim ba-telefon ba-rexov]

the-phenomenon still exists of people speaking on.the-cell.phone in.the-street
b. ha-toca’a lo exara le-hagi’a [Sel yeladim lo yoc’im me-ha-bayit]

the-result not late to-arrive of kids not leaving from-the-house
The difference in extraposition between (35) and (36) suggests therefore that in
(36) the genitive is part of the clausal constituent, while in (35) it is not, the result
of movement into the DP domain.
4. CONTROL AS MOVEMENT WITHIN DP
The possibility of raising in DP may carry implications for the analysis of control,
especially if control is A-movement, as argued in Hornstein (1999, 2001). With the
abandonment of the government relation within the Minimalist Program, a shared
conclusion emerging from recent studies of control is that its interpretive hetero-
geneity, OC vs. NOC, correlates with structural heterogeneity, either the size of
the infinitive (Wurmbrand 2001) or its position within the clause it is embedded
in (Hornstein 1999, 2001; Landau 2001). According to both Landau and Horn-
stein, OC is anaphoric. For Landau it is derived by Agree, limited to complements,
while for Hornstein it is derived by A-movement, leading to the expectation that
OC should be licensed in exactly those configurations which license A-movement,
including the sideward variety (Nunes 2004). DP therefore serves as an important
testing ground for the A-movement generalization such that if OC is A-movement
and control is attested within DP, so is raising expected (Culicover and Jackendoff
2001). The objection raised by Culicover and Jackendoff (2001) based on the absence
of raising may be neutralized by the facts presented above. Still, whether or not the
availability of raising in DP directly supports an A-movement analysis depends on
a number of factors discussed briefly below: (i) whether the differences observed
between raising and control in DP reduce to the theta-checking procedure devel-
oped in Hornstein (1999); and (ii) given the analogy between A-movement and OC,
whether control in DP is of the OC or NOC variety, an empirical question not yet
fully resolved (see Hornstein 2001, 2003, and Boeckx and Hornstein 2003 for recent
discussion). Preliminary evidence based on the comparison of nominalizations and
gerunds suggests OC, granting pervasive control by an implicit argument.
Many of the differences between raising and control observed above reduce to
the thematic properties associated with control and the genitive DP. Selectional
restrictions and the unavailability of non-referential genitives such as expletives
and idiom chunks will follow directly from the theta-feature assigned by the
head noun and checked by the genitive DP. Similarly, the fact that truth-value
or denotation of a control noun is not preserved with an embedded passive (18),
but is with raising nouns (19)/(20), reduces to the thematic relation between the
control noun and genitive DP and the absence thereof with raising nouns.
The distribution of possessive datives, however, has been taken to track A-movement
(Borer and Grodzinsky 1986; Landau 1998) yet shows a difference between raising
and control. A possessive link is possible when the dative possessor c-commands the
possessed or its trace (37). A possessive link is therefore impossible between the dative
and the subject in (37a), but good in the unaccusative structure in (37b):
(37) a. ha-yeladim pihaku le-rina ba-hafgana

the-children yawned to-rina at.the-demonstration
→ no possessive link between ‘rina’ and ‘the children’
b. ha-yeladim1 ne’elmu le-rina t1 ba-hafgana

the-children disappeared to-rina at.the-demonstration
‘Rina’s children disappeared at the demonstration’
→ possessive link bet rina and subj
Similar to the effects discussed in Burzio (1986) with impersonal SI, possessive datives
distinguish raising from control. They are incompatible with unaccusative PRO and good
with unnaccusative NP-trace. In keeping with the basic requirement that the dative c-com-
mand the trace of the possessed, both the control examples (38) and the raising examples
(39) contain an embedded unaccusative. When the matrix verb is a control verb, the rela-
tion is impossible, and when it is a raising predicate, a possessive reading is possible11:
(38) a. ha-yeladim1 nisu [PRO1 le-he’alem le-rina t1 ba-hafgana]

the-children tried to-disappear to-rina at.the-demonstration
‘The children tried to disappear on Rina at the demonstration.’
b. ha-yeladim1 tixnenu [PRO1 le-he’alem le-rina t1 ba-hafgana]

the-children planned to-disappear to-rina at.the-demonstration
‘The children planned to disappear on Rina at the demonstration.’
(39) a. ha-yeladim1 alulim [t1 le-he’alem le-rina t1 ba-hafgana]

the-children likely to-disappear to-rina at.the-demonstration
‘Rina’s children are likely to disappear at the demonstration’
b. ha-yeladim1 amurim [t1 le-he’alem le-rina t1 ba-hafgana]

the-children supposed to-disappear to-rina at.the-demonstration
‘Rina’s children are supposed to disappear at the demonstration’
The contrast between raising and control surfaces also in DP, control DPs in (40),
and raising DPs in (41):
(40) a. [ha-nisayon Sel ha-yeladim1 [PRO1 le-he’alem le-rina t1 ba-hafgana]] nixSal

the-attempt of the-children to-disappear to-rina at.the-demo failed
‘The children’s attempt to disappear on Rina at the demonstration failed.’
→ no possessive link between ‘to-rina’ and ‘the children’
b. [ha-racon Sel ha-yeladim1 [PRO1 le-he’alem le-rina t1 ba-hafgana]] muvan lanu

the-desire of the-children to-disappear to-rina at.the-demo clear to-us
‘The children’s desire to disappear on Rina at the demonstration was clear to us.’
→ no possessive link between ‘to-rina’ and ‘the children’
30 IVY SICHEL
(41) a. [ha-sikuyim Sel ha-yeladim1 le-he’alem le-rina t1 ba-hafgana] kluSim le-maday

the-chances of the-children to-disappear to-rina at.the-demo meager rather
‘Rina’s children’s chances of disappearing at the demonstration are pretty slim’
b. [ha-netiya Sel ha-maftexot1 le-he’alem le-rina t1] yedu’a le-kulam

the-tendency of the-keys to-disappear to-rina known to-all
‘The tendency for Rina’s keys to disappear is known to all’
c. [ha-hizdamnut Sel ha-yeladim1 le-he’alem le-rina t1] avra kvar

the-opportunity of the-children to-disappear to-rina passed already
‘The opportunity for Rina’s children to disappear at the demonstration has passed’
The difference between raising and control DPs with possessive datives might
be more challenging for an A-movement analysis of control since it seems to
be sensitive to the source constituent at the tail of the chain. Assuming that the
dative binds into a position within the possessed DP (Borer and Grodzinsky 1986;
Landau 1997), it could be argued that PRO exhausts a full DP and so cannot be
possessed by an external dative, while a lexical noun phrase leaves enough struc-
tural space in DP for a variable or trace bound by the dative. If control is exactly
like A-movement, except for the theta-role the moved DP receives in its landing
site, a simple tail analysis of the sort just sketched will not be available.
Finer implications of raising for the analysis of control depend on the nature of
control in DP, whether it follows the obligatory or non-obligatory pattern familiar
from IP. Though often mentioned in the literature, control in DP has been scarcely
studied and its status still remains unclear. Part of the challenge may be related to
the poorly understood position of infinitives within DP, as complements (Stowell
1981, among others) or adjuncts (Grimshaw 1990), leaving little ground for solid
predictions. Another puzzle has its source in the pervasive optionality of an overt
controller (42), compared with the stricter requirement familiar from clauses:
(42) a. John tried to leave
b. *It was tried to leave
c. John’s attempt to leave
d. the attempt to leave
The obligatory presence of the controller in (42a) has been considered (beginning
with Williams 1980) to be a hallmark of OC. Similarly, Hornstein (2001, 2003)
and Boeckx and Hornstein (2003) treat the presence of the controller in DP to
be central among the properties of control. Observing that DPs exhibit OC with
overt controllers and an NOC pattern when the controller is absent, including lax-
ing of the c-command and locality requirements, the assumption that absence of
an overt controller implies NOC leads to indeterminate conclusions. Either DPs
exhibit NOC generally since controllers are never obligatory (Hornstein 2003;
Boeckx and Hornstein 2003) or DPs exhibit OC with overt controllers and NOC
when the controller is absent (Hornstein 2001). The former solution is coher-
ent and is consistent with the supposed absence of raising in DP, yet it leaves
unresolved, the OC effects found with an overt controller located within the DP.
The requirement for strict c-command and the available interpretations (bound
variable and de se readings, sloppy identity), absent in clausal NOC, remain unac-
counted for. Furthermore, given that an overt controller always produces OC,
long-distance control across a potential controller is never attested. In clauses, on
the other hand, the NOC pattern, as observed with verbs such as ‘help’, allows the
dative to control or be skipped and exhibits optionality:
(43) Mary knows that it would help Bill [PRO to behave himself/herself] in public
While full discussion of control in DP is beyond the scope of this article, the fol-
lowing facts suggest that the problems stemming from the treatment of ‘obliga-
tory presence of controller’ as central to OC may disappear if controllers may be
covert and pronominal.12 Once implicit argument control is granted, DP appears
to exhibit OC generally, though sometimes by a null pronominal which may be
identified from afar. The differences, from this perspective, between control in
clauses and in nominals fall neatly into place. Null controllers are allowed more
liberally than in clauses because independently, Agents of nominalizations are
not required by an EPP. The laxing of c-command and locality follow from the
properties of the covert controller, which may be identified by a nonlocal and
non-c-commanding antecedent, not from the properties of the controlled subject.
This is seen by comparing nominalizations with covert Agents with their gerund
counterparts. Gerunds, unlike nominalizations, do not allow long-distance con-
trol without c-command:
(44) a. John1’s mother2 supported the decision [PRO to vindicate himself1/herself2/oneself]
b. John1’s mother2 supported deciding [PRO to vindicate herself/*himself/*oneself]
(45) a. John’s mother was in favor of the refusal [PRO to vindicate himself/oneself/herself]
b. John’s mother was in favor of refusing [PRO to vindicate herself/ *himself]
The greater flexibility of nominalizations with respect to absence of c-command

can be shown to be independent of control of the embedded subject. Covert
subjects of gerunds generally require c-command and a local antecedent, and in
this respect may be assimilated to PRO, as seen in (46):
(46) a. People who know John often discuss [his / *ec working too hard]
b. Sam thought that Mary discussed [his / *ec shaving himself]
The hypothesis that c-command variability with nominalizations (44a) and (45a)
is related to the pronominal nature of the implicit Agent rather than to the sub-
ject of the infinitive can also explain the contrast in locality variability between
gerunds and nominalizations. On a par with (46b), null subjects of gerunds are
32 IVY SICHEL
incompatible with long-distance controllers, and nominalizations allow optional-

ity when the antecedent is external to DP:
(47) a. John2 regretted that Mary1 put down the decision [PRO to love himself2 / her1 /
herself1]
b. John2 regretted that Mary1 put down deciding [PRO to love herself1 / *himself2 / *her1]
c. John2 regretted that Mary put down his2 deciding [PRO to love himself2]
(48) a. John2 was aware that Mary1 counted on the promise [PRO to love her1/himself2/
herself2]
b. John was aware that Mary counted on promising [PRO to love herself2/*her2/
*himself1/*oneself1]
The comparison of nominalizations with gerunds suggests therefore that NOC

effects detected in the absence of a controller internal to DP may be reducible to
an implicit Agent argument which controls the embedded subject but which itself
is pronominal. The pronominal implicit argument may have as its antecedent a
remote or non-c-commanding DP, producing the appearance of NOC; yet the
relation between the implicit argument and the embedded subject patterns with
OC, as if it were overt. Sichel (2006) shows that the interpretation of control by
implicit arguments also patterns with OC, allowing de se, bound variable, and
sloppy identity interpretations, bringing the source position closer to the NP-trace
observed in raising constructions. Whether and how the implicit argument is rep-
resented syntactically remain open to further study.
5. CONCLUSIONS
Contrary to expectations based in Government and Binding for a universal restric-

tion against raising in DP, Hebrew appears to allow theta-Case splits in which a DP
is marked genitive yet thematically licensed exclusively in an embedded infinitive
clause. Non-referential DPs, such as expletives and idiom chunks, may instantiate a
genitive position in DPs headed by a class of nonderived nouns, roughly denoting
notions of modality. It was furthermore shown that these theta-Case splits attest
to a movement operation which relocates the embedded subject to a position in the
DP domain in which it receives genitive Case. In light of raising in DP, objections to
control as movement based on its absence should also be reevaluated. While a fuller
understanding of control in DP awaits further study, the comparison of nominali-
zations with gerunds may suggest that control in DP is consistently of the obliga-
tory variety and in this respect is compatible with the distribution of raising.
ACKNOWLEDGMENTS
For valuable and insightful comments I thank the audiences to which parts of
this paper have been presented: at GLOW (2005) at the University of Geneva, the
University of Southern California, UCLA, The Hebrew University of Jerusalem,

the ‘New Horizons in the Grammar of Raising and Control’ workshop at Harvard
University, Ben Gurion University, CUNY, and the University of Maryland.
Research for this paper was supported by The Israel Science Foundation, Grant
#0322358.
1
This is with the exception of a structural-case analysis of genitive Case in Hebrew construct DPs,
supported by the existence of construct-state ECM (Siloni 1997).
2
Discussion of the source of the English restriction, and the precise difference between English and
Hebrew are beyond the scope of this paper: the range of possibilities remains fairly broad, even with
the elimination of some of the older Case-based and government-based approaches. First, it is pos-
sible that DPs with gerund complements, of the sort in (3d), do involve raising, despite the absence
of expletives and idiom chunks. If so, it is possible that the parametric difference is related to the
overt expression of nonfinite tense, such that Hebrew infinitives are underlyingly similar to English
gerunds. Alternatively, the difference may be related to the typology of A-movement and its para-
metric availability within DP. Richard Kayne (personal communication) gives the following contrast
between standard raising nominals and ECM-raising nominals, when embedded under ‘despite’:
(i) a. ?Despite its tendency to snow around here a lot, it’s been relatively mild this year
b. *Despite his belief to be a genius, the rest of us were not as convinced
Setting aside the improvement under ‘despite’, the contrast may suggest a fairly deep dif-
ference between (1a) and (1b), such that only ECM-passive is derived by true A-movement
(see Jacobson 1990 for a non-movement analysis of English clausal raising). Similarly, the
parametric difference between Hebrew and English may be tied to the typology of passive and
A-movement, such that in Hebrew DP-internal A-movement is tolerated more readily than in
English or that DP never allows true A-movement, and Hebrew raising constructions are not
derived by true A-movement.
3
Clausal control in Hebrew, the counterpart of (1a), is perfectly grammatical, and bears the obliga-
tory/non-obligatory distinction familiar from English. See Landau 2000 and Sichel 2006 for further
details.
4
The English glossing of the good cases should not be taken to imply identical structure. English
‘chances’, ‘probability’, etc. with a gerund complement may well be ACC-ing constructions with
the embedded subject in situ bearing no morpho-syntactic relation to the embedding DP, an analysis
argued against for Hebrew in section 4.
5
See for example the discussion of possessor adjuncts in Partee and Borschev 2003, where it
is claimed that (non-inherent/alienable) possessors are not direct arguments of nouns, and are
licensed semantically by a relation R represented higher in the structure.
6
Part of the difficulty in determining argument structure on independent semantic grounds hinges
on the intensionality of these nouns in conjunction with the extensional/intensional status of pos-
session. Given that chances and probabilities may be negative and that tendencies do not imply the
truth of their complement, it appears that no extensional object need be possessed in cases such as
Mary’s chances of winning (are less than zero) or John’s tendency to be late.
7
See Sichel (2005) for more detailed discussion, argumentation for the expletive status of ze, and
similar examples with idiom chunks.
8
See Sichel (2005) for further details.
9
There is also an interpretive difference, possibly related to the way in which the nonfinite clause com-
bines with the head noun. In raising constructions, the clause restricts the denotation of the head
noun, on a par with control infinitives in nominals and complements in VP generally. The nonfinite
clauses in (23), on the other hand, specify the content of the head noun, like an appositive:
(i) ha-tofa’a hi Se-xayalot metaxkerot palestina’im ba-rexov

the-phenomenon is that-soldiers.F interrogate Palestinians on.the-street
‘The phenomenon is that soldiers interrogate Palestinians on the street’
34 IVY SICHEL
10
See Sichel (2003) for genitive Case assignment as a spec-head configuration between genitive DP
and the head Sel and further elaboration of the functional portion of DP consistent with Structural
genitive case and movement.
11
In the absence of a possessive link, (38a) and the examples with control allow an ethical dative
reading, in which the dative is in some sense affected by the event.
12
See Sichel (2006) for more detailed discussion of control in DP.
CHRISTOPHER HIRSCH AND KEN WEXLER
THE LATE DEVELOPMENT OF RAISING: WHAT

CHILDREN SEEM TO THINK ABOUT SEEM
1. WHY THE ACQUISITION OF RAISING SEEMS TO MATTER
While it is clear that the study of language development contributes to linguistic

theory, it is perhaps less widely recognized that by examining the time course of
language development we can integrate the study of language into the broader
study of biological development. The hope persists that this type of analysis will
play a role in the genetic underpinnings of language, as it has already done in
some areas of grammar.1
One area of grammar notorious for demonstrating late development involves
various kinds of long-distance dependencies. In their comparative analysis of the
development of different linguistic structures, Borer and Wexler (1987) argued
that structures containing A-chains develop late.2 We will base our study on more
up-to-date and empirically correct versions of Borer and Wexler’s A-Chain
Deficit Hypothesis (ACDH), but the logic is the same: certain grammatical rep-
resentations allowed by Universal Grammar (UG) are ungrammatical for young
children because of constraints imposed by their particular biology as opposed to
adult biology.
The basic argument for such maturation is Borer and Wexler’s (1987) ‘Triggering
Problem’ (Babyonyshev et al. (2001) ‘Argument from the Abundance of the Stimu-
lus’ (AOS)) that parallels Chomsky’s argument from the Poverty of the Stimulus
(POS; following Descartes). Since evidence for a particular structure is abundant
in the input, why should it take so long for the structure to develop? Both the AOS
and the POS provide evidence for biological (genetic) underpinnings of linguistic
representations. In this paper, we present evidence for the late development of one
structure in particular: subject-to-subject raising, thus providing evidence for the
role of biology in the development of linguistic structure.
At the same time, developmental evidence can play a role in helping to determine
the correct linguistic analysis of structures. Given good evidence for the develop-
mental delay of a certain grammatical process G, then if a structure S is found not
to be delayed, this provides evidence that S in fact does not make use of G. In this
regard, subject-to-subject raising provides an important test case. According to the
theory of development of long-distance structures that we take to be most empiri-
cally adequate, raising should show delayed development. Control structures, on the
other hand, are not subject to the developmental constraint that we assume, thus
they should not be delayed at anywhere near the level of raising structures.
35
© 2007 Springer.
36 CHRISTOPHER HIRSCH AND KEN WEXLER
Certain recent analyses, however, propose that control structures do not arise
from a separate control module, but are actually a type of raising structure. Given
the theory of development that we present, and depending on the precise syntactic
analysis, these ‘control’ structures perhaps should be delayed, possibly pattern-
ing with subject-to-subject raising structures. Closer inspection suggests that this
is not so obvious, as we will see. Thus, evidence concerning how raising vs. control
structures develop can play a role in determining which analysis of this structure
is correct but much depends on the theory of development. As we might expect,
the role of development is Janus-faced; it looks out on and contributes both to
linguistic theory and biological theory.
2. RAISING ISSUES FOR THE THEORY OF LINGUISTIC

DEVELOPMENT
Raising constructions involve movement or some other form of long-distance

relation. The classic theory of the delay of long-distance relations in children is
Borer and Wexler’s (1987) ACDH.
(1) ACDH: A-chains are ungrammatical for children until a certain age. As children age, their
brains mature such that A-chains become grammatical.
A great deal of evidence has accumulated that verbal passives and unaccusa-
tives are very much delayed in young children.3 On the other hand, ever since
Borer and Wexler (1992) it has been known that the VP-internal subject hypoth-
esis poses a problem for ACDH. If subjects are generated internal to the VP,
then their movement to [Spec, IP] forms an A-chain. Yet empirical acquisition
evidence shows that children are not delayed in placing the subject correctly
outside the VP (Stromswold 1996). The field for the most part concentrated
on demonstrating late development for ‘object-to-subject’ A-chains, leaving the
problem of VP-internal subjects moving to [Spec, IP] to be solved. To address
this problem, Wexler (2004) proposed the Universal Phase Requirement (UPR),
in place of ACDH.
(2) UPR: (holds of premature children, until around age 7) v defines a phase, whether or
not v is defective.
The theory is couched in the Minimalist framework. Chomsky (1998, 2001a) derives
on minimalist considerations a very strong cyclic theory of syntax. Essentially
Merge proceeds from the bottom to the top of a derivational tree with most of the
derivation closed off to further analysis or change as it proceeds. He proposed the
Phase Impenetrability Condition (PIC) in (3):
(3) PIC: When working at a phase, the edge (the head and any specs) of the next lower
phase is available for analysis, but nothing lower than the edge. In particular the
complement is not available.
THE LATE DEVELOPMENT OF RAISING 37
C (Complementizer) and v (the light-verb head of a phrase that takes VP as its

complement and selects the external argument as its spec) are substantively defined
as phasal heads (on an argument from completeness of semantic properties). Sub-
jects of vP can move to [Spec, IP] because the subject, although inside a phase vP,
can be probed at the next phase CP because the subject is at the edge (specifier
position) of v. Passives and unaccusatives, however, create a special problem for
the theory, as the object of V must move up to T (INFL) in the higher phase.
Chomsky proposed that the v of passives and unaccusatives is ‘defective’, that is,
it does not assign an external argument and it does not operate as a phase. Thus
the object of V is available at the higher C above it.
UPR states that children take the v of passives and unaccusatives to be phasal
(although they know that this v does not assign an external argument). As such,
verbal passive and unaccusative representations are ungrammatical for the prema-
ture child. This is the correct result empirically. On the other hand, the ‘VP-internal
subject’ of a transitive (i.e. one with an external argument) clause is generated in
[Spec, vP] is at the edge of VP. At the next higher phase, C, the subject is available
for analysis according to PIC. Thus T can ‘see’ the subject, meaning that Agree and
Move can take place; there is no need for non-phasal v. The child subject to UPR is
also unhindered since there is no non-phasal v involved. UPR predicts that the child
can raise the subject from the edge of vP with no problem. This solves the problem
of VP-internal subjects that dogged ACDH.
The upshot of UPR is that there is no problem with any particular kind of
chain, and no special assumptions about objects or subjects moving, or even
Agreeing. The problematic constructions are (some of) those that demand a
defective phase (e.g. non-phasal v). Consider the subject-to-subject raising
structure in (4a).
(4) a. Bert seems to Ernie [t to be wearing a hat].
b. It seems to Ernie that Bert is wearing a hat.
The traditional analysis of (4a) considers the surface subject of the sentence to
start out in the subject position of the lower predicate, from where it is moved
to [Spec, IP] of the matrix clause. Unraised versions of the raised sentence exist,
with an expletive in subject position, as in (4b). The movement that derives (4a)
creates an A-chain. The subject moves to an argument (A) position, [Spec, IP]
of the matrix clause. As such, Borer and Wexler (1987) predicted that subject-
to-subject raising structures would be delayed.
On UPR, even some structures with defective phases can be unproblematic for
premature children. Consider an unraised structure like (4b). Presumably seem with
an expletive subject is defective, since it does not assign an external argument. Thus
a child will take this defective v to be phasal given UPR. There is no reason, how-
ever, that the child cannot make the derivation converge. No relation holds between
matrix T and anything in the lower clause, so that even if the child takes v to be
phasal, nothing in the computation is interrupted. UPR predicts that unraised seem
in sentences like (4b) should converge for the immature child, even if the derivation
they use (with phasal v) is slightly different than the adult representation. On the
other hand, consider the analysis of (4a), the raised construction:
(5) Berti T vdef seems to Ernie [ti Tdef to be ti v* wearing a hat]
Wexler (2004:165) writes:

The embedded clause in raising constructions like [5] does not contain C, and it
contains defective Tense, Tdef. The embedded verb has normal v*, with its directly
merged external argument Bert. See the arguments in Chomsky (1998). The external
argument raises to [Spec, Tdef], checking the EPP feature of Tdef. Since Tdef doesn’t con-
tain a full complement of phi-features, the phi-feature set of the subject Bert remains,
and Bert is active. All this has taken place cyclically, in the phase determined by root
C. v of the root verb seems does not define a phase, since it is defective, as we saw
earlier. Since Bert is still active it can raise to the matrix T, deleting T’s uninterpretable
phi-features and EPP feature, resulting in convergence of [5].
This is Chomsky’s analysis of raising. It demands a non-phasal v. By UPR, the

child takes v in (5) to be phasal. T then cannot probe Ernie (which is in the com-
plement of v), and T will end up with uninterpretable features unchecked. This is
Wexler’s derivation from UPR of the prediction that raised sentences like (4a) will
be ungrammatical for children subject to UPR.
It thus becomes crucial to test the early status of raised sentences like (4a) in
children who are at the premature age. UPR makes the interesting prediction that,
although raising involves movement of a lower subject, it patterns (is late in devel-
opment) with the movement of objects (passives, unaccusatives) and not with the
movement of another kind of subject (the VP-internal subject of a transitive clause).
If the predictions are confirmed, it will be evidence that UPR, and not a constraint
against some form of chain or movement, is constraining children’s early grammar.
An alternative proposal for accounting for delays in passives and unaccusa-
tives, but allowing VP-internal subjects, was Babyonyshev et al. (2001) Exter-
nal Argument Requirement Hypothesis (EARH), according to which children
demand external arguments, which they formalize in Minimalist terms as in (6):
(6) EARH: Young children consider structures with defective v to be ungrammatical.
EARH predicts that verbal passives and unaccusatives will be ungrammatical for
young children. Furthermore, it predicts that raised structures like (4a) will also
be ungrammatical, as both of these structures lack external arguments. On the
other hand, EARH predicts that VP-internal subjects can raise to [Spec, IP] with
no problem, since these structures have an external argument.
As Babyonyshev et al. recognize, EARH also predicts that unraised struc-
tures like (4b) will be ungrammatical for young children, since such structures
contain no external argument. The comparison between raised and unraised
sentences (4a vs. 4b) carried out in this paper provides an empirical test to
distinguish EARH from UPR.
Hyams et al. (2006:30) look for another explanation for why passives are
delayed in acquisition. They write, ‘Descriptively speaking, children’s difficulty
seems to be restricted to those A-chains that derive a misalignment of
thematic and grammatical hierarchies. The argument structure associated with

transitive . . . and unergative . . . verbs . . . specifies an actor-like external argument,
which is not represented in the passive, which involves a promotion of the theme
to the external argument position. Unaccusative verbs have no external argument
hence no violation of canonical alignment, and the alignment is also respected
in subject raising.’ In essence, they take the problem with passives to be that
a ‘canonical alignment’ is not respected, which premature children find
ungrammatical. As we have noted, children have problems with unaccusative
structure, which would speak against this hypothesis. More to the point for the
present study is that the hypothesis also predicts that there will be no problem
with subject-to-subject raising.4
3. ACQUISITION OF RAISING: PREVIOUS RESEARCH
Few studies to date have investigated children’s comprehension of raising struc-

tures. To our knowledge, the first detailed comprehension study of raising in Eng-
lish was carried out by Froud et al. (in preparation). They conducted a two-choice
sentence-picture matching experiment testing both raised and unraised sentences,
using thought-bubbles to represent the thinking aspect of seem.5 Thirty-three par-
ticipants were run, in the 4- to 5-year age range. On unraised, expletive sentences
with seem and an experiencer (e.g. It seems to Homer that Lisa is eating a sand-
wich), the mean correct response rate was around 80%. On raised sentences (e.g.
Lisa seems to Homer to be eating a sandwich), the mean response rate was around
45%. That is, on raised sentences, children performed near chance level, whereas
on unraised sentences the children performed significantly above chance.6
Recent experimental work by Becker (2005, 2006) purports to demonstrate
that young children not only comprehend raising, but that they actually ana-
lyze control structures as raising structures. Becker offers a pair of experiments
to support these two claims. In her first experiment, she finds that 64% of
3-year-olds and 47% of 4-year-olds accept as grammatical control-verb sentences
with inanimate subjects and ‘compatible’ complements (e.g. The flower wants to
be pink). The same children also accept raising verbs with ‘compatible’ predicates
(e.g. The flower seems to be pink).
There are many strategies that children could be employing to derive this
result, but Becker first attempts to rule out the possibility that children sim-
ply ignore the matrix verb, producing some form of copula structure (e.g. The
flower is pink), since then children’s judgments might reflect nothing more than
whether the subject can occur with the embedded predicate. She tested chil-
dren on a second experiment, this one a truth-value judgment task in which a
character either ‘seemed’ or ‘wanted’ to do something, but in fact, did not. For
example, to test the raising structures, one scenario involved a white dog who
stood under a black light, and thus appeared (‘seemed to be’) purple. The child
was then asked to judge the truth of the sentence The dog seemed to be pur-
ple. Becker (2006:448) writes, ‘A child parsing only the dog . . . be purple should
respond “false”, since the dog was not in fact purple; but a child parsing the dog
seemed to be purple should respond “true” since the dog did seem to be purple
when standing under the lamp.’ Similar scenarios were constructed for the con-
trol conditions. For example, in one, a pig wants to eat a donut, but actually ends
up eating a banana. The child is then asked to judge the sentence The pig wanted
to eat the donut. Ignoring the matrix verb should lead the child to respond that
the sentence is false, since the pig in fact did not eat the donut, whereas the child
should respond correctly if he does parse the matrix control verb.
Given children’s above-chance performance on this test, Becker deduces that
children must be paying attention to the raising and control verbs in the first experi-
ment, and concludes (1) that young children do in fact understand raising construc-
tions, and that (2), it is control verbs that they cannot handle, treating them instead
as raising verbs.
While we acknowledge the ingenuity of Becker’s experiments, there are a number
of important problems for her two conclusions, which ultimately cast into serious
doubt her claims about raising and control. Most doubtful is her claim concerning
control verbs being non-thematic raising verbs for children. For one thing, there is
no shortage of experimental evidence (which we review in detail later in the paper)
and commonsense/anecdotal evidence that children do correctly understand control
verbs, not least of which are strikingly contradictory data from Becker’s own experi-
ments.7 To wit: Becker reports that her subjects performed fine on the control verbs
in her second experiment, but on her hypothesis children should be interpreting
these verbs as raising verbs, which should have produced wrong answers. According
to Becker’s hypothesis, children interpret want as seem (or some other semantically
simple raising verb),8 so if they are indeed parsing the main verb, they should inter-
pret the sentence The pig wanted to eat the donut as roughly The pig seemed to eat
the donut, which is false, and should prompt the children to respond as such. On the
hypothesis that children interpret control verbs as raising verbs, the results of Beck-
er’s second experiment are unexplained. On the commonsensical hypothesis that
children readily comprehend control verbs, these results are unproblematic.
Furthermore, Becker’s hypothesis that children treat control verbs as raising verbs
makes (at least) two syntactic predictions against which production data speak. First,
if control verbs like want are actually raising verbs, then they should not allow bare
DP complements. While such structures are allowed in the adult grammar (e.g. The
man wants an apple), they should be ruled out for the child since bare DPs are not
allowed with raising verbs (e.g. *The man seems an apple). Yet a brief glance at data
on the CHILDES corpus turns up thousands of examples of control verbs with bare
DPs. Second, if control verbs are raising verbs, then a control verb like want might
be expected to have an ‘unraised’ counterpart (e.g. It wants that the flower is pink).
Yet, there is no evidence from production data that children ever use control verbs
in such a manner. Children’s use of bare DPs with control verbs, children’s lack
of ‘unraised’ forms with control verbs, and Becker’s own second experiment,
and 20 years of research on the acquisition of control strongly speak against Becker’s
claim that children provide a raising analysis to sentences containing control verbs.
A plausible alternative explanation for the findings from Becker’s first experi-
ment is that children who accept sentences like The flower wants to be pink are simply
those children operating under the assumption that the sort of cartoon inanimate
objects used in the experiment can be agents/experiencers (either for the purposes of
the experiment or more generally). This is particularly likely in a story-based, game-
like experimental setting, where children are often willing (and even encouraged) to
suspend normal judgments and anthropomorphize pictured objects.9 In any case, if
a slightly modified version of Becker’s first experiment (one that discourages chil-
dren from imputing animacy to inanimate objects) significantly reduced the propor-
tion of children who judge the relevant sentences to be acceptable, then Becker’s
conclusion would further be called into serious question.
While Becker’s second experiment suggests that children comprehend raising
structures, an alternative explanation, in the spirit of the very idea that Becker was
attempting to control for, presents itself. Becker’s reasoning for conducting the second
experiment was to test whether children could simply be ignoring raising and con-
trol verbs (giving rise to the findings from her first experiment), presumably because
children might ignore them if they could not understand them in the relevant structure,
as predicted, for example, by UPR. If the children ignored the verb, Becker assumes
that children’s representation for The dog seemed to be purple would be The dog . . .
to be purple. This is only true if children failed to notice the past tense morphology
on the matrix verb. If children did recognize seemed as a past tense form, but did not
know its meaning, then their likely parse would be The dog was purple. Such a parse,
however, would lead children to correctly answer the raising sentences, since during
the scenario, the dog was purple (when he stood under the black light). This analysis,
where children can parse the past tense morphology of a verb they do not under-
stand in the relevant linguistic frame, extends to two of the other raised sentences
Becker tested: the horse used to be small (where he is now big) and the rhino happened
to be under the tree (where he is now somewhere else). Substituting the past tense
form of the copula for the raising verb produces correct responses, since during the
scenarios the horse was small and the rhino was under the tree (they just no longer
are). This account, however, fails on the final raised sentence Becker tested: the horse
tends to eat hay (where he ate something else). In this case, the matrix verb tends is not
a past tense form, and substituting the present tense copula form yields the ungram-
matical string the horse is eat the hay. Interestingly, though, while children did quite
well on the first three raised sentences, as predicted by substitution of the past tense
form was, children had such great difficulty with this last tends sentence that Becker
excluded it from all subsequent analyses. That only this sentence would be problem-
atic is predicted if children parse nothing more than the past tense morphology of
the raising verb. These two experiments, therefore, offer little concrete evidence that
children can comprehend raising structures.
While a few studies have examined children’s knowledge of raising, further
experimentation is very much needed.
4. ADULT AND CHILD PRODUCTIONS CONTAINING SEEM
Previous acquisition studies of raising have failed to investigate the degree to

which children hear raising structures and raising verbs, as well as the extent to
which children produce such structures and verbs. Before asking whether children
comprehend raising structures, it is desirable to know whether children are even

exposed to such structures and whether they produce them. In order to determine
the degree to which children hear and produce raising structures, we examined the
child-directed and child-produced speech for all 1,051 English-speaking children
on the CHILDES corpus containing the raising verb seem (MacWhinney 2000).10
Detailed analyses were limited to utterances containing the verb seem after initial
searches made clear that seem was by far the most frequent raising verb in the
corpus and would thus serve as the sole verb in the comprehension experiments
to follow.11
From 552 child-directed utterances containing seem, 448 analyzable, non-
repetitive utterances were extracted for further investigation. For the sample
considered, this implies that a child hears a unique sentence containing seem
every 1,700 utterances. While this might seem rather scarce, an average Ameri-
can child hears about 7,000 utterances per day (Cameron-Faulkner et al. 2003);
by a child’s third birthday, he will have heard 4,500 sentences containing seem.
Further, the number of input examples containing seem exceeds the number of
input utterances containing verbs most researchers (and parents!) would judge
young children to know, including crawl, feed, hug, lift, pass, and rub. Thus, it is
unlikely that any difficulties children might have comprehending raising struc-
tures involving seem could be directly attributed to the rarity of the verb in the
input, as the verb simply is not that rare. Also, difficulties could not be attrib-
uted directly to children failing to hear the verb used in raised constructions.
The vast majority of parental use (87%) is in the raised, non-expletive form.
Furthermore, it is not the case that adults in some way modulate their use of
unraised and raised structures, such that the raised sentences are only used with
older children. The type of structure used (raised or unraised) is not a function
of age; raised forms are used at all ages, even with children who have not yet
reached their first birthday.
While these input analyses make clear that children do indeed hear raising
structures with the verb seem, analysis of children’s use of such structures shows
children rarely produce them. In all, only 67 child-produced utterances con-
taining seem appear in the corpus, of which only 33 constitute non-repetitive,
analyzable examples. There is thus an almost 14:1 ratio of adult-use to child-use
of sentences containing seem, while the overall ratio of adult-produced to child-
produced unique utterances in the corpus is less than 2:1. This input–output
discrepancy is even more striking given that for verbs with similar input frequen-
cies to seem, such severe discrepancies do not exist (e.g. 1.9:1 for carry, 2.1:1
for climb, 3.4:1 for crawl, 2.3:1 for dance, 0.8:1 for feed, 0.7:1 for hug, 4.0:1 for
lift, 2.1:1 for pass, and 4.1:1 for rub). Of the analyzable child-produced exam-
ples, only 11 appear in children younger than 5. Furthermore, the percentage of
raised uses is less in the children than the adults (64% vs. 87%). The few children
who do, however, produce raised structures with seem are for the most part
5 years of age or older.
A closer examination of children’s raised sentences reveals a striking asym-
metry between the type produced by children younger than 6 years of age
and those 6 and older. All the examples produced by the younger children
involve small-clause adjectival complements (e.g. that seems fun), while eight
of the nine older children’s utterances have verbal complements (e.g. they seem
to be following the same direction). It is thus possible that younger children’s
grammar allows raising with adjectival complements, but not with verbal com-
plements, which might be due to seem having different syntactic entailments
depending on which type of complement it appears with.12
Given that natural production data demonstrate that children hear many
sentences involving the raising verb seem, but rarely produce raising structures,
we now turn to experimental work investigating children’s comprehension of
such structures. Later in the paper, we return briefly to further analyses of chil-
dren’s productions of raising structures.
5. COMPREHENSION OF RAISING: AN EXPERIMENTAL

INVESTIGATION
To investigate the acquisition of raising, four sentence structures were tested:

transitive-active sentences (7), sentences with the verb think and finite embedded
clauses (8), unraised, expletive-it sentences with seem (9), and raised sentences
with seem (10).
(7) Homer is eating a sandwich.
(8) Lisa thinks that Bart is playing an instrument.
(9) It seems to Homer that Marge is pushing a cart.
(10) Homer seems to Maggie to be bowling a ball.
To assess children’s comprehension of these sentence types, we conducted a two-

choice sentence-picture matching task in which children were shown two pictures
side by side on a laptop computer screen and were asked to choose the picture
best matching the sentence they were read. Answers were logged on the computer
before proceeding to the next item. All sentences were read aloud twice before
children were allowed to respond. Item presentation was randomized on an indi-
vidual subject basis. In order to minimize task demands, only four characters
(from The Simpsons television cartoon), with whom the children were familiarized
during the introduction, were used throughout the experiment. Thought-bubbles
were used to convey the notion of ‘thinking’ for the think condition and both seem
conditions. The notion of thought-bubbles was familiarized in the introduction.
Previous research has shown that children comprehend such pictorial depictions
of thinking (Wellman et al. 1996).
The active-transitive condition involved pictures in which one character
interacts with an object. The foil picture for this condition had a different, non-
mentioned character interacting with the same object. For the think condition
and seem conditions, the correct pictures involved one character thinking about
Figure 1. Sample picture for picture-matching task
another character performing some action. Thus the picture in Figure 1 would
constitute the correct picture for the following three sentences: Lisa thinks that Bart
is playing an instrument (think-condition), It seems to Lisa that Bart is playing an
instrument (unraised-condition), and Bart seems to Lisa to be playing an instru-
ment (raised-condition). For these three conditions, three different foil types were
constructed. Matrix-reversal (MR) foils involved switching the character who
does the thinking. Thus the MR foil for Figure 1 would involve Bart playing the
saxophone thinking about Lisa. Embedded-reversal (ER) foils involved switch-
ing the character who performs the action denoted by the embedded predicate.
With respect to the picture, this would involve Lisa playing the saxophone, think-
ing about Bart. Finally, double-reversal (DR) foils involved switching both who
is doing the thinking and who is performing the relevant action. The DR foil
to the picture would therefore have Bart thinking about Lisa playing the saxo-
phone. The use of these three foil types allows for the pinpointing of any difficul-
ties in comprehension, whether it be with determining who is doing the thinking
(MR foils), with who is performing the action mentioned in the embedded clause
(ER foils), or both (DR foils). On any given trial, the child was always presented
with the correct picture and one of the three foil types. Each of these foil types was
tested six times per condition. Each child thus saw 18 items for the think-condition,
unraised-condition, and raised-condition (only 12 items were used for the active-control
condition). Location of the correct picture (left side or right side of the screen)
was balanced across conditions and the entire experiment.
At this point, we should address the decision to use an experiencer to-phrase
with the seem sentences. According to UPR, in no way is the presence of the
to-phrase required to elicit poor performance from the children. This theory
straightforwardly predicts that children are incapable of grammatically representing

sentences involving subject-to-subject raising, whether or not this raising takes
place over an experiencer. The decision to include the to-phrase was based
exclusively on experimental considerations. It was felt that even if children cannot
grammatically represent raising structures, they would still be able to roughly infer
the meaning of a sentence of the form A seems to be Z by directly associating
A and Z. How to design a plausible foil for such a sentence using a sentence-
picture matching task is not at all clear. The problem is not even specific to this
particular methodology, as the same concern was raised earlier for the items used
in Becker (2005, 2006), where alternative explanations exist for her findings with
experiencer-less raising structures. By including the experiencer, we are able to eas-
ily construct an experiment that is simple for children and affords the possibility
of detecting difficulties with raising.
There are many reasons, however, why one would prefer an experimental
design that did not involve the experiencer phrase. Many languages that have rais-
ing structures, nonetheless do not allow raising over (non-clitic) experiencers (e.g.
Icelandic, Italian, and for many speakers, French). There is thus the possibility
that any problems children might have with the raising sentences tested in this
experiment could simply be due to children having a problem with raising over
experiencers, whether it be for grammatical reasons or processing reasons, and
not a general problem with raising. We return to this possibility in detail later in
the paper. Regardless, given that English does allow raising over experiencers, this
experiment examines, at a very minimum, whether children have acquired this
property of their language.
The transitive-active sentences were meant to serve as control items to ensure
that children were paying attention to the experimental stimuli, given that the
correct item could be determined simply by attending to the subject of the
sentence. There is voluminous evidence that even very young children have no
difficulty with such sentences. Any difficulties with this condition would be a
reflection of attentional problems and not core grammar. Thus, those children
who did experience difficulties would be subject to elimination from the experi-
ment. The think sentences were meant to serve as cognitive controls for the sen-
tences involving seem. If children are able to comprehend the think sentences,
then there is no reason why any difficulties with seem are due to either problems
comprehending thought-bubbles or a general deficit in theory of mind.
Data were gathered from 70 children (34 girls, 36 boys), with ten children in
every 1-year interval from 3 to 9 years of age, with participant details in Table 1.
The experimental results, collapsing momentarily across foil type, are sum-
marized in Table 2.
Overall, children performed extremely well on the transitive-active controls. All
age groups were 99% accurate for these control trials. No child made more than a
single mistake on this condition, and only 4 of 70 children made even one mistake.
Thus, no children were omitted from subsequent analyses due to inattention. Chil-
dren likewise performed quite well on the think trials, with all age groups scoring
above 88% correct. This indicates that children generally had no difficulties compre-
hending thought-bubbles. Similarly, children performed quite well on the unraised
Table 1. Participant data
Group Number Age range Mean age

3-year-olds 10 3.04–3.87 3.51
4-year-olds 10 4.18–4.95 4.52
5-year-olds 10 5.13–5.83 5.49
6-year-olds 10 6.03–6.98 6.45
7-year-olds 10 7.05–7.79 7.52
8-year-olds 10 8.10–8.77 8.45
9-year-olds 10 9.05–9.96 9.49
Total 70 3.04–9.96 6.49
Table 2. Combined results
Group Actives Think Unraised Raised

3-year-olds 100.0% 88.3% 85.6% 43.9%
4-year-olds 99.2% 92.8% 88.9% 45.6%
5-year-olds 99.2% 95.6% 92.8% 44.4%
6-year-olds 99.2% 95.6% 91.7% 51.7%
7-year-olds 100.0% 96.1% 96.7% 71.1%
8-year-olds 99.2% 98.3% 98.9% 75.6%
9-year-olds 100.0% 100.0% 98.9% 92.2%
Total 99.5% 95.2% 93.3% 60.6%
condition, with no groups scoring below 85% accuracy. Children generally had no
difficulty comprehending the verb seem, at least in its unraised form.
As predicted by UPR, however, children had great difficulty with the raised
sentences. Across the 40 youngest children, accuracy did not differ from chance
level (t(39) = −0.978, p = 0.334). No group scores noticeably better than chance level
until the 7-year-olds. Across the first four groups (3- to 6-year-olds), development
is flat, with only a 6.7% increase in performance over these 3 years. In the follow-
ing year alone, however, performance rockets up an impressive 22.8%. This type
of rapid growth following years of level stagnation is exactly what is expected on
a maturation account, where prior to some genetic event, children lack the neces-
sary grammatical representation to derive the correct sentence meaning, but after
maturation, such analyses are possible. This sudden increase in raising performance
is further noted in individual subject analyses, counting the number of children in
each age group who score above chance (minimum 14 of 18 items correct). As seen in
Table 3, before the age of 7, only eight children scored at above-chance level on the
raised condition. In the subsequent 7-year-old group, there are already six children
scoring above chance. Of the 41 children who fail to score above chance on raising,
78% of them are less than 7 years of age. Meanwhile, 70% of the children 7-years-old
and up score above chance on this condition.
Table 3. Performance on raising
No. of children scoring above

Grouping chance on raising
3-year-olds 2
4-year-olds 1
5-year-olds 2
6-year-olds 3
7-year-olds 6
8-year-olds 6
9-year-olds 9
A preliminary examination of the data thus supports the class of grammatical

acquisition theories (including UPR) predicting that structures involving subject-
to-subject raising are delayed. Children comprehend unraised structures involving
seem, but cannot comprehend their semantically equivalent raised counterparts
until around the age of 7. Before this age, very few children (only 20%) compre-
hend raising structures, whereas most children older than this do comprehend
raising. While the above data certainly demonstrate a delay for raised sentences
with seem, certain children did have difficulties with think and unraised seem trials,
as made clear by an examination of performance as a function of foil type.
Children do rather well (>75% correct) on all foil types with think at all age
ranges. Yet for these think sentences, it is also clear that children have the great-
est difficulties with the MR foils. Children are overall 5% worse with MR foils
compared to the average of the other two foil types. The fact that the younger
children score significantly better on think trials with ER and DR foils com-
pared to trials with MR foils suggests that ER and DR trials are somehow easier
for children. This might be because when given a sentence of the form X thinks
that Y is doing Z, even children who did not know the meaning of think, but
who nonetheless correctly parsed the embedded clause, would still be able to
correctly reject the ER and DR foils, since both (incorrectly) involve pictures in
which the subject of the embedded clause is not performing the action denoted
by the embedded predicate. When presented with the MR foil, however, children
cannot simply look to the embedded clause to determine which picture to choose
since both the correct picture and the MR foil have the subject of the embedded
clause performing the action denoted by the embedded predicate. In order to
consistently choose the correct picture over the MR foil, children must under-
stand that the matrix subject in the think sentences denotes the experiencer.
Thus a test of children’s knowledge of think is whether or not they score well on
those think trials involving MR foils. We take ‘above chance’ performance on think
sentences with MR foils to be 83% accuracy and greater.13 As indicated in Table 4,
14% of all the children fail to meet this level of proficiency. Five of ten children fail-
ing on the MR think trials are 3-year-olds, and nine of ten are younger than 6. These
children either do not comprehend the verb think, do not comprehend the pictures
used to depict characters thinking (i.e. thought-bubbles), or fail at theory of mind
Table 4. Performance on ‘think’ with MR foils
No. of children % Children Think-

Group Think-MR <83% MR <83%
3-year-olds 5 50%
4-year-olds 2 20%
5-year-olds 2 20%
6-year-olds 0 0%
7-year-olds 1 10%
8-year-olds 0 0%
9-year-olds 0 0%
Total 10 14%
such that they cannot represent one character thinking about another. Numerous
studies suggest that theory of mind develops around the age of 4 (for a review see
Wellman et al. 2001), and this could account for the difficulties that some children
had on the think trials. Children who do not comprehend these think items fail to
offer interesting data as pertains to knowledge of raising, and are excluded from
many of the subsequent analyses.
Turning to children’s comprehension of the unraised sentences as a function
of foil type, by 4 years of age all foil conditions are answered above 75% correct.
Again, however, children have the greatest difficulty with MR foils, scoring 10%
worse on the MR foils with unraised seem compared to the average of the DR and
ER foils. Just as children can answer the think sentences with ER or DR foils by
merely parsing the embedded clause so too can children comprehend the unraised
sentences with ER and DR foils by doing nothing more than correctly parsing
the embedded clause. In a sentence of the form It seems to X that Y is doing Z,
children cannot just look to the embedded clause with MR foils since both the
correct picture and MR foil have the subject of the embedded clause performing
the action denoted by the embedded predicate. In order to correctly reject the MR
foil with unraised seem, children must also comprehend who is doing the thinking
(i.e. correctly understand the relationship between seem and the to-phrase experi-
encer). In order to understand this relationship, children must comprehend seem.
Thus, accurate performance on unraised sentences with MR foils serves as a test
of whether or not the children comprehend the verb seem. Again, taking ‘above
chance’ performance to be 83% accuracy (minimum 5 of 6 items correct), 19% of
children have difficulty with unraised seem. As shown in Table 5, all of the chil-
dren who have trouble with unraised sentences are younger than 7 years of age.
Fully 60% of the 3-year-olds fail on unraised seem, constituting nearly half of all
those who fail.
These data make it clear that while the good majority of children 4 years of
age and older comprehend the verb seem (at least in its unraised form), most
3-year-olds do not. This suggests that raising cannot be studied in children younger
than 4, as they do not even know the meaning of the raising verbs used in the
Table 5. Performance on unraised ‘seem’
No. of children % Children

Group Unraised-MR <83% Unraised-MR <83%
3-year-olds 6 60%
4-year-olds 3 30%
5-year-olds 2 20%
6-year-olds 2 20%
7-year-olds 0 0%
8-year-olds 0 0%
9-year-olds 0 0%
Total 13 19%
125%
100%
Accuracy
75% MR-foil
ER-foil
50% DR-foil
25%
0%
3 4 5 6 7 8 9
Age Group (Years Old)
Figure 2. Raised Comprehension by Foil Type
unraised form, which brings up many challenging questions for Becker’s research.
She tested 3-year-olds’ knowledge of raising constructions, but had no independ-
ent assessment of whether the children actually knew the raising verbs used in
the experiment (e.g. she never tested the verbs in their unraised form). Once again,
the results of those children who do not comprehend seem in its unraised form
cannot speak to the question of children’s comprehension of raising and will be
omitted from many subsequent analyses.
Children’s comprehension of the raising sentences appears quite different than
their comprehension of the other sentence forms. All foil types are answered at
or below-chance level until 7 years of age, as seen in Figure 2 (below-chance level
is indicated by the line at 38%).14 While MR and ER foils are answered at chance
level before age 7, DR foils are consistently answered at below-chance level. That
is, children actually prefer the DR foil to the correct picture, and are not randomly
guessing when a DR foil is paired with a raised sentence, though they do appear
to be guessing when given MR and ER foils. Systematic preference for DR foils
and chance performance with MR and ER foils is further reflected in individual
subject analyses, where of the 31 children younger than 7 who score at chance
level on the average of ER and MR trials (33–66% accuracy), 84% score below
chance on DR trials (<17% accuracy). Children are not selecting the DR pictures,
though, simply because such pictures are inherently more attractive. As noted ear-
lier, children actually prefer the MR pictures with think sentences and unraised
seem sentences.15
According to UPR, children lack the syntactic means necessary to compute
raising. This inability to mediate the dependency between the matrix subject posi-
tion and embedded subject position appears quite compatible with children’s poor
performance on raising trials. What then to make of the much worse and below-
chance performance for the DR foils compared to the other foils? The effect of
foil type suggests that children are applying different strategies when interpreting
raised sentences depending on which type of foil is presented along with the cor-
rect picture. The data rule out several strategies.
First, children are not blindly guessing when presented with a raised sentence.
Guessing would result in chance performance across all foil types, but children
consistently choose the incorrect picture when presented with DR foils. Second,
children are not simply analyzing the nearest noun to the embedded predicate as the
subject of the embedded clause. This would predict below-chance performance for
both ER and DR foils, but children only demonstrate below-chance performance
for the DR foils. Third, they are not simply ignoring the matrix subject plus seem
and just parsing the embedded clause, as this would predict below-chance perform-
ance for ER and DR foils.
Instead, children appear to be providing a ‘think analysis’ to the interpretation
of raising sentences. The sentence Bart seems to Lisa to be playing an instrument
leads children to interpret it as meaning Bart thinks Lisa is playing an instrument. This
interpretation maps directly to the DR foil and straightforwardly accounts for
why children choose the DR foil over the correct picture. Since for ER and MR
foils neither the correct picture nor the foil matches such an interpretation,
children simply guess, which accounts for chance performance with ER and MR
foil types.
This think analysis might work syntactically in one of many ways. First, children
might replace seem with think and ignore the fact that the embedded clause is non-
finite, resulting in the forced parse X thinks to Y to be Z. Such an analysis requires
children to ignore the preposition to and to ignore the fact that think elsewhere in
the grammar requires a finite embedded clause. An alternative analysis would be
that children actually take think to be a possible object-control verb. While this is
not licit for adult speakers in English, such an analysis is fine for certain semanti-
cally related verbs (11).
(11) John believes/imagines/understands Mary to wear a hat (every Sunday).
As for the preposition to, children might ignore it, or even take it as a marker
for think that takes a nonfinite complement. Regardless of the exact represen-
tational details, it is clear from the acquisition data that some structure with
these semantic entailments holds for children’s analysis of raising structures
containing seem and an experiencer.
Many things can be said about why children invoke such an analysis. First, and
most importantly, children are unable to provide the (correct) adult representa-
tion for these raising sentences. Nonetheless, children have heard seem used many
times, most often in contexts where it is clear to them that something like ‘think-
ing’ is being referenced. At a level of general semantic conception, think and seem
share many properties. As the experimental evidence also makes clear, children
comprehend quite well sentences with think, probably as soon as they develop a
theory of mind. While children do not comprehend raising sentences, they none-
theless attempt to find some reasonable interpretation for such sentences, most
likely unconsciously. Returning to the earlier CHILDES data, it is worthy of
note that the utterances children hear containing seem might lead them toward a
think analysis for raised seem. The majority of raised sentences that children hear
contain animate subjects, where animacy is a prerequisite for sentience, and only
sentient entities may be subjects of think.
It must be asked whether this strategy is particular to the context of the experi-
ment, or whether it reflects core knowledge on the part of the child. That is, do
children merely substitute think for seem given the demands of the experimental
task, or do they actually come to the task with a lexical entry for seem along the
lines of an object-control version of think? Also, should we expect this analysis to
hold for all raising verbs? While such an analysis can easily be extended to appear,
it is unclear how it could apply to certain other raising verbs (e.g. used (to)). Also,
it is important to note that children cannot be extending this analysis to sentences
with unraised seem since this would most likely lead to comprehension difficulties,
while children performed quite well on unraised structures. That there are so few
examples of seem in children’s productions suggests that children do not actively
maintain such an analysis of seem, since if they did have such a representation we
might expect them to produce more utterances with raised seem. Recent experi-
mental work by Hirsch, Orfitelli, and Wexler (2006), discussed below, does suggest,
however, that the think analysis for raised seem is rather pervasive.
We conclude that premature children have no problem with unraised structures
but demonstrate a very significant delay on subject-to-subject raising structures.
In natural production, young children do not produce raised structures, although
they produce unraised adjectival complement structures with the same verb seem.
Even more strongly, in our experimental comprehension study, children younger
than 7 years of age performed extremely poorly on raised structures although they
performed very well on unraised structures. Good performance on think trials and
unraised seem rules out ‘cognitive complexity’ explanations for difficulties with
raising.
These are exactly the predictions that the UPR makes. Since raising structures
demand a defective v, and UPR says that children will replace this by a fully phasal
v, raising structures will be ungrammatical for them. As such, they will either
guess at the answer or use an interpretive strategy, assimilating the structure to
another structure that is grammatical for them. Evidence was reviewed that many
children treat the raised structures as if they are a non-raised structure with a verb
meaning roughly think. In many ways, this is the use of a syntactic-homophone
(Babyonyshev et al.’s term), along the lines of Borer and Wexler’s (1987) analysis
in which children use the adjectival passive structure (grammatical for them) as a
substitute, when it works, for the verbal passive (ungrammatical for them).
EARH, on the other hand, does not fare well. EARH, which says that all
structures must have an external argument, correctly predicts that raised structures
will be delayed for premature children since they lack an external argument. On the
other hand, unraised structures using expletives and seem are also predicted to be
problematic by EARH since they too lack an external argument. Yet children
do quite fine on these structures thus EARH cannot be maintained. Hyams,
Ntelitheos, and Manorohanta’s hypothesis about canonical structure also fails to
predict the data since it expects children to have no problem with raised structures.
Hyams and Snyder’s maturation theory of strong freezing does predict that these
structures will be delayed, although it also predicts that raising without the
experiencer should be understood, which we argue against later in the paper.
Thus our data select between a constraint against non-phasal v and a constraint
against non-theta-role assigning v (no external arguments). The former turns out to
be empirically correct. It looks as if children are not biologically prepared to handle
categories that should be phasal (on the simplest minimalist terms, as Wexler (2004)
argues) but are not. On the other hand, children have no trouble with structures in
which v does not assign an external argument so long as v does not have to be non-
phasal. As for ACDH, it fares fine on the raising data, but it already has trouble
with VP-internal subjects. In the following sections, we consider further evidence in
support of UPR.
6. CHILDREN’S COMPREHENSION OF RAISING AND PASSIVES:

AN EXPERIMENTAL INVESTIGATION
Maturation theories such as UPR and ACDH predict that the acquisition of
raising should match that of passives, both across age groups, and more impor-
tantly, within individual children. In order to examine whether raising and passive
acquisition do indeed mirror one another, a few words on the acquisition of pas-
sives are required. With respect to English, the acquisition literature is quite clear
that passives are delayed (Slobin 1966; Turner and Rommetveit 1967; Bever 1970;
Maratsos and Ambramovitch 1975; Maratsos et al. 1985; Gordon and Chafetz
1990; Fox and Grodzinsky 1998; Hirsch and Wexler 2004a).
Not only do English-speaking children show a delay for passives, they have
significantly more difficulty with passives with ‘psychological’ verbs (e.g. see,
love, hear; subject-experiencer verbs) compared to passives with actional verbs
(e.g. push, kick, wash), but importantly not actives. This interaction is confirmed
by every study that has crossed voice and verb type (Maratsos et al. 1979; Maratsos
et al. 1985; Sudhalter and Braine 1985; Gordon and Chafetz 1990; Fox and
Grodzinsky 1998; Hirsch and Wexler 2004a, 2006; Hirsch and Hartman 2006).
Coupled with Horgan’s (1978) findings that children’s early passives describe
states and not events, Borer and Wexler (1987) hypothesized that children lack the
syntactic means to represent verbal passives (their ACDH), while better perform-
ance on actional passives was due to an adjectival strategy. Since verbal passives
and adjectival passives are homophonous in English, and given that children
otherwise lack a syntactic parse of verbal passives, children treat actional verbal
passives as adjectival passives, for which their grammar does provide an analysis.
As (subject-experiencer) psychological verbs universally tend not to form adjectival
passives, children are unable to comprehend psychological passives, resulting in
guessing and chance performance.
There is much evidence supporting the hypothesis that young children’s
(actional) passives are adjectival. Babyonyshev and Brun (2003) present relevant
evidence from Russian, a language with different verb forms for imperfective
and perfective aspect. They studied the passives children hear and produce with
respect to the aspectual form used. In terms of what children hear, for active voice
sentences there was no significant difference between the use of perfective and
imperfective aspect. In the passive voice there was also little difference, with a
slight minority of perfective forms used by parents. Yet, 91% of the passives
produced by the children were perfective. Babyonyshev and Brun suggest that
this striking asymmetry is understandable in terms of an adjectival strategy, since
perfective passives, but not imperfective passives, are homophonous in Russian
with the adjectival passive form. Terzi and Wexler (2002) asked how children would
comprehend actional passives in a language in which verbal passives and adjectival
passives are not homophonous (Greek). They found that unlike English-speaking
children, who master actional passives quite early, Greek children had great
difficulty with actional passives, even at later ages. In a recent study, Hirsch and
Hartman (2006) present experimental evidence that the earliest passives children
comprehend are not the paradigmatic actional passives used in previous experiments
(e.g. with hit), but those with object-experiencer verbs (e.g. with scare). In part,
their explanation centers on the fact that achievement verbs make even better
adjectival passives than activity verbs.
In order to investigate the relationship between the acquisition of raising and
passives, the same 70 children who participated in the previous raising study
were administered a passive test within 2 weeks of having taken the raising test.
This experiment tested four conditions, crossing voice (active vs. passive) and
verb type (actional vs. psychological).16 Eight verbs were used, consisting of four
actional verbs (push, kiss, kick, hold) and four psychological verbs (remember,
love, hate, see). Eight items were constructed for each active condition, and 16
items for each passive condition. All sentences were semantically reversible.
In order to minimize task demands, once again only four Simpsons cartoon
characters were used throughout the experiment. To assess children’s compre-
hension of these four sentence types, we employed a two-choice sentence-picture
matching task wherein children were shown on a laptop screen two pictures side
by side depicting opposite events. Children were told to choose the picture best
matching the sentence they were read, after which their answers were logged on
the computer before continuing to the next item. All sentences were read twice
to the child before he was allowed to respond. The location of the correct pic-
ture (left or right side of the screen) was balanced across the individual verbs,
conditions, and the entire experiment. Items were presented in a randomized
order to each child.
Table 6. Results for passives test
Group Actional actives Psych actives Actional passives Psych passives

3-year-olds 96.3% 98.8% 65.6% 38.1%
4-year-olds 95.0% 98.8% 86.3% 50.0%
5-year-olds 97.5% 98.8% 92.5% 58.8%
6-year-olds 97.5% 98.8% 89.4% 45.6%
7-year-olds 97.5% 97.5% 95.6% 75.6%
8-year-olds 100.0% 98.8% 92.5% 82.5%
9-year-olds 100.0% 100.0% 95.0% 90.6%
Total 97.7% 98.8% 88.1% 63.0%
The results of this experiment appear in Table 6. Children performed extremely

well on both active conditions. Excellent performance on the psychological actives,
which were comprehended slightly better than their actional counterparts, indicates
that children have no general problem comprehending psychological verbs, nor any
difficulty with the experimental methods for assessing such knowledge. We replicate
all past experiments crossing voice and verb type, and find much worse performance
for psychological passives as compared to actional passives. Furthermore, the study
replicates previous findings that psychological passives are not comprehended until
around 7 years of age (Maratsos et al. 1985; Hirsch and Wexler 2006).
For the purpose of examining in detail the relationship between the acquisi-
tion of raising and that of passives, it is vital to minimize the influence of any
compensatory strategies that children might be employing in an attempt to com-
prehend structures for which their grammar is unable to provide a representation.
Since these strategies are hypothesized to be dependent on environmental factors
and mechanisms of general cognition and not on pure syntactic competence, they
must be isolated for the purpose of exploring deeper syntactic knowledge.17 For
raising, this means examining only the data from MR and ER trials since chil-
dren employ a consistent strategy when given DR foils (i.e. the think analysis). As
for passives, only psychological passives should be included, as actional passives
are ‘understood’ using an adjectival strategy. Thus, in all subsequent analyses,
performance on raising will be examined only with respect to comprehension of
raising sentences paired with MR and ER foils. Similarly, for passives, only scores
on psychological passives will be examined further.
Plotting children’s scores on raising and passives by age group reveals very
similar development curves (Figure 3). Before 7 years of age, children show gen-
erally no improvement in comprehending either raising or passives, with sudden
and dramatic improvement at age 7 across both structures. It is only with the
7-year-old group that there is a noticeable deviation from chance level for both
structures.18
The similarity in the patterns of acquisition is further demonstrated by consid-
ering the number of subjects in each age group who score above chance on each
structure.19 As reflected in Figure 4, before the age of 7, no more than three of ten
children score above chance in any age group. The 7-year-old group, however, has
100%
75%
Accuracy
Raising
Passive
50%
25%
3 4 5 6 7 8 9
Figure 3. Comprehension of Raising & Passives
10
# Individuals AC (of 10)
6 Raising
4 Passive
0
3 4 5 6 7 8 9
Figure 4. Above Chance Comprehension: Raising & Passives
six children who score above chance on raising and passives. This sudden increase
in above-chance comprehension accounts for the sudden increase in group accu-
racy, as opposed to all children doing just slightly better.
While such data strongly suggest that the acquisition of raising and passives
are fundamentally linked, providing exciting evidence in support of certain matu-
ration theories, what such theories predict is not simply that the average age of
acquisition for both structures should match (in this case somewhere between age
6 and 7), but rather that acquisition of both structures should correlate within
individual children.
It is important to note, however, that the maturation theories do not predict
that a strictly linear correlation should hold between children’s scores on raising
and passives. Rather, these theories predict that two groups of children should be
observed. The first group would consist of children who have not undergone the
relevant maturation, and who should thus score poorly on both raising and pas-
sives. The second group of children would be made up of those whose grammar has
matured such that these children are expected to comprehend both raising struc-
tures and passives. Before grammatical maturation takes place, comprehension of
raising and passives is guided by relatively idiosyncratic and independent strategies
(i.e. the think analysis for raising and the adjectival analysis for passives). Thus, there
is no great expectation of a strong correlation between the actual scores on raising
and passives, especially for the younger children. Rather, the strong prediction of
UPR is that there will not be children who comprehend raising but not passives,
and vice versa. That is, the maturation theories predict very strong correlations of
above-chance performance on raising and passives.
Before examining such correlations, it is necessary to remove from considera-
tion those children who failed to comprehend the think and unraised seem trials
during the raising experiment, as measured by performance on MR trials. Prob-
lems with theory of mind or simply in understanding seem guarantee problems with
raising, even if the relevant linguistic maturation has taken place to make passives
grammatical. Removing such cases leaves data from 53 children. The data for these
remaining children is summarized in the scatter plot below (Figure 5), where lines at
75% indicate above-chance level for raising and passives. Significant, and very high
correlations obtain when either exact scores (r(51) = 0.799, p < 0.0001) or above-
chance performance (r(51) = 0.851, p < 0.0001) are examined.
The scatter plot shows that in general, older children tend to cluster in the
upper right quadrant (above chance on both structures), while children younger
than 7 tend to populate the lower left quadrant (not above chance for either
structure). As predicted by UPR, very few children seem to fall outside these two
100%
Psych Passive Accuracy
3 year-olds
75%
4 year-olds
5 year-olds
50% 6 year-olds
7 year-olds
8 year-olds
25%
9 year-olds
0%
0% 25% 50% 75% 100%
Raising Accuracy
Figure 5. Raising vs. Passive Comprehension

quadrants. There are only four apparent contradictions to the predictions of this
maturation account.
Upon further review, three of these turn out to be very marginal contradic-
tions. Had one subject missed just one more passive item he would not be an
exception (not above chance for either structure). Had another subject done one
item better on passives, he would not be an exception (above chance for both
structures). Had the third child done one item worse on raising, he would also not
be an exception (not above chance for either structure). That is, had these three
children scored just one point differently they could not be considered exceptions.
That leaves only one true exception, a 4-year-old child who got all of the raising
items correct but was at chance for the psychological passives.20 In fact, it is
amazing that the 3-year-old who does well on raising also does well on passives,
and likewise, that the 9-year-old who fails on raising also fails on passives.21 While
maturation theories predict that there should be few such children, it is telling that
these apparent age ‘exceptions’ are not exceptions to the correlation predictions
of UPR. Despite the one counterexample, the maturation theories receive tremen-
dous support from the near-perfect correlation between raising and passives.
7. RAISING OVER EXPERIENCERS, (ADULT) PROCESSING,

AND MORE ACQUISITION FINDINGS
In this section, we review how the experiencer to-phrase in the raising sentences
might shape children’s poor comprehension of such structures. Two general lines
of inquiry will be pursued, one addressing processing costs incurred by raising
over an experiencer, and another addressing grammatical implications of such
movement. Upon review, it is our conclusion that neither processing nor a gram-
matical ban on raising over experiencers accounts for children’s delay in raising.
Many theories of (adult) processing predict increased costs associated with
forming long-distance dependencies across an intervening DP, as occurs in raising
sentences with an experiencer (e.g. Gibson 1998; Gordon et al. 2001). Could such
processing costs coupled with an assumption about children having more limited
processing resources (for which we know of no evidence) explain why children fail
to comprehend the raising structures tested in our experiment? This appears to be
unlikely for several reasons. First, even if processing limitations could account for
children’s general difficulty with raising, which will be addressed momentarily, it is
utterly unclear how any such processing account could explain the specific problems
children have with raising (i.e. chance performance for ER and MR foils and below-
chance performance for DR foils).
To address the possibility of a processing explanation though, we had 24 native
English-speaking adults, half men and half women (18–24 years, mean age 19.0
years), complete the same raising test as administered to the children, with two
small changes. First, the sentences appeared at the bottom of the screen for the
adults to read. Subjects were free to respond at any time once the pictures and
sentence were presented. Second, in addition to accuracy, subjects’ reaction times
(RTs) were also recorded from the time that the pictures were first presented to
Table 7. Results of raising test administered to adults
Condition Accuracy Reaction time (ms)

Actives 99.0% 2150
Think-DR 98.6% 3056
Think-ER 98.6% 2979
Think-MR 99.3% 2911
Mean Think 98.8% 2982
Unraised-DR 98.6% 3164
Unraised-ER 97.9% 3216
Unraised-MR 99.3% 3235
Mean Unraised 98.6% 3205
Raised-DR 94.9% 4028
Raised-ER 96.2% 4490
Raised-MR 95.7% 4344
Mean Raised 95.6% 4284
when the subject pressed the key to choose the picture matching the sentence. The
task was therefore completely self-paced, other than the introduction, which was
identical for the children and adults. In order to minimize the influence of outly-
ing RTs, we eliminated all responses falling three SDs above the mean response
time (calculated with respect to each subject’s mean) and those faster than 900
ms, resulting in a loss of only 1.9% of the total data. We also eliminated these
responses from accuracy counts since it is unclear what to make of any sentences
answered extremely quickly or slowly. The adult data is summarized in Table 7.
As is clear from these data, compared to their semantically equivalent counter-
parts, subjects answered the raising sentences an average of 3.0% less correctly and
1,079 ms more slowly. This is in line with the predictions of the processing accounts.
What is also clear, however, is that there is no evidence of any foil effects, which were
so important in explaining children’s difficulties. While children had more difficulty
with MR foils for think and unraised seem, no foil effect is found for either sentences
structures in either accuracy (Fthink(2,429) = 0.201, p = 0.818; Funraised(2,429) = 0.505,
p = 0.604) or RT (Fthink(2,429) = 0.387, p = 0.679; Funraised(2,429) = 0.088, p = 0.916).
Furthermore, while children had much greater difficulty with DR foils for raised
sentences, again no difference is found by foil type in the adults for either accuracy
(Fraised(2,408) = 0.132, p = 0.876) or RT (Fraised(2,408) = 1.57, p = 0.208). The lack of
even minute RT effects by foil type for the adults in light of children’s striking com-
prehension differences according to foil type is powerful evidence against processing
explanations for the children’s data.
A further argument against processing explanations is the observation that
children have no difficulties mediating long-distance dependencies across inter-
vening DPs when A-bar movement, as opposed to raising’s A-movement, is
involved. Children as young as 3 have no difficulties producing or comprehend-
ing object-extracted wh-questions (Stromswold 1995; Hirsch and Hartman 2005).
Also, children are known to comprehend object-extracted relative clauses, which
involve forming a dependency between an object gap and a filler which crosses
the intervening subject, as long as relevant pragmatic conditions are taken into
consideration (Hamburger and Crain 1982).
Finally, processing theories working over surface-structure relations predict
children should perform much better if the experiencer to-phrase is fronted (e.g.
To X, Y seems to Z). While we have not yet fully completed an experiment examin-
ing this issue, we do have pilot work from 30 subjects (4–6 years) testing just such
structures, with the experiencer fronted for both unraised and raised sentences with
seem. Children have no difficulties comprehending the unraised sentences with a
fronted experiencer. Even with the experiencer fronted, children continue to have
difficulties with the raised sentences. Difficulty is greatest with MR foils, which is to
be expected, since with DR and ER foils, children need only look past the fronted
PP to determine the correct picture.22
Having considered and found inadequate processing explanations for these
data, we turn now to a possible grammatical ban in children’s early grammar on
raising over experiencers. It is well established that many languages that allow
raising nonetheless prohibit raising over experiencers (e.g. Icelandic, Spanish,
French, Italian, amongst many others). This ban has often been taken to reflect
a strong locality requirement on A-movement (McGinnis 1998; Torrego 2002;
Collins 2005). Perhaps English-speaking children assume they are speaking
Icelandic, at least with respect to rules about raising over experiencers. That is,
perhaps children’s early grammar simply rules out raising over experiencers, and
not raising generally as predicted by UPR.23
What would such an account look like? Presumably the idea is that raising over
experiencers is marked, that children start out with the unmarked value of the
parameter (no raising over experiencers) and ‘learn’ or ‘reset’ the parameter with
experience. Such an account goes up against Borer and Wexler’s (1987) Trigger-
ing Problem. Why should it take 7 years for the child to reset the parameter, given
that the child has the relevant experience? Why do most children suddenly set it at
exactly the same age (but not sooner)? In other words, even if this were the correct
account, we would still need a maturational theory to explain the slow development.
This would be even more surprising since learning language-specific parameter
settings is accomplished very early (Wexler’s (1998) Very-Early Parameter-Setting).
More difficult for the idea that children’s only problem is with raising over
experiencers is that on such an account the incredibly strong correlations between
the acquisition of raising and passives is completely unexpected. On UPR, how-
ever, not only is the correlation not unexpected, it is strongly predicted. Ultimately,
though, the best evidence that children’s difficulties are not determined by raising
over an experiencer comes from new experimental evidence that children have
difficulties with raising even when the experiencer is absent.24
Comprehension of raising structures without experiencers was examined in
an experiment by Hirsch, Modyanova, and Wexler (2006). The crucial scenario
involves a character (Mary) looking for something (e.g. her hat), but unable to
find it. Unbeknownst to Mary, however, she is already wearing the hat! Mean-
while, a second character (John) who is some distance away, is watching all of
this. He cannot quite make out what is on Mary’s head, but he is pretty sure it is
her hat. Mary meanwhile recognizes that her hat is at least not on John’s head.
An impartial puppet who has himself been watching all of this is then asked to
comment on what he has seen. The puppet may answer using one of four sentence
types: sentences with think, unraised sentences with an experiencer, raised sen-
tences with an experiencer, and raised sentences without an experiencer. Children
must then judge the truth of the puppet’s statement.
We will discuss data from ten 4-year-old children who have completed the
experiment (average age 4.7 years).25 These children perform brilliantly on the
think sentences (96.3% correct) and unraised sentences (97.5% correct). The task
was designed such that children could not have performed well on either condi-
tion by simply parsing the embedded predicate, which in the unraised condition
requires children to comprehend the verb seem. Every child performed well on
these two conditions, such that any subsequent problems with raising, either with
or without the experiencer-phrase, cannot be attributed to a problem of not know-
ing the raising verb, since children consistently comprehended all of the unraised
sentences. When presented with a raised sentence with an experiencer (e.g. John
seems to Mary to be wearing a hat), children always get such sentences wrong
(5.0% correct). This is further evidence for the think analysis, since with respect to
the scenario discussed, while it is not true that John seems to Mary to be wearing a
hat, it is true that John thinks that Mary is wearing a hat.
What then to expect for the raised sentences without an experiencer
(e.g. Mary seems to be wearing a hat)? At first blush, a think analysis seems
impossible:
(12) *Mary thinks to be wearing a hat.
Yet structures like (12) are quite grammatical in other languages, where think can
take a nonfinite complement:
(13) Jean pense PRO porter un chapeau. [French]

Jean thinks to wear a hat.
Jeani thinks hei is wearing a hat.
(14) Franz denkt, PRO einen Hut zu tragen. [German]

Franz thinks a hat to wear.
Franzi thinks hei is wearing a hat.
If English-speaking children extend their think analysis to seem sentences with-

out an experiencer, then they should consistently get such sentences wrong.
Given this same scenario involving Mary searching for the hat she happens
already to be wearing, it is true that Mary seems to be wearing a hat, but false
that Mary thinks she is wearing a hat. On such sentences children show mixed
performance. As a group, accuracy is 40.0%. Already this demonstrates that
children have difficulties with raising independent of the experiencer, in support
of UPR. Individual subject analyses, however, reveal two groups of children:
those who get all of the raising without experiencer items correct and another
group which gets them all wrong. From the children’s explanations, it is clear
that the latter group is getting the wrong answer by means of a think analysis.26
The other group of children, who all failed on raising with an experiencer, then
might just be ignoring the verb in the cases without an experiencer, and treating
them as copular constructions, which derives good performance when the expe-
riencer is absent but would not work when the experiencer is present, thus their
use of the think analysis in those cases.27
This study demonstrates that children’s difficulties with raising structures are not
tied to a grammatical ban on raising over experiencers. Even when no experiencer
is present, at least half of the children misunderstand such raising sentences. The
particular pattern of comprehension further suggests children treat seem as think.
In the next two sections we discuss the relevance of other types of long-distance
structures to theories of linguistic development, in particular the structure of control.
8. ACQUISITION OF CONTROL: PREVIOUS RESEARCH
UPR (as well as ACDH and EARH) predicts that raised structures will be quite
delayed for children. In the spirit of the original ACDH, an important considera-
tion is that A-bar relations are not delayed in nearly the same way; this fact under-
lay Borer and Wexler’s (1987) proposal. There is another class of long-distance
structures, though, that are important to consider because they seem to involve
relations between argument positions. This is the class of control structures. In a
sentence like (15), PRO is controlled by the subject John. Let us call these cases of
obligatory control (OC).
(15) John tried [PRO to leave]
The standard analysis of control does not take the relation between John and PRO in
(15) to be an Agree relation (much less a Move relation). Rather John and PRO
are coindexed or made coreferential (or PRO is made to be referentially dependent
on John in some other way, perhaps in the semantics). Thus UPR does not predict
any difficulties with OC cases like (15).28 Nevertheless, in some structural respects,
the relation between the controller and PRO in a sentence like (15) seems similar
to an Agree or Move relation; the controller c-commands and is fairly local to the
controlled element. Similarly a moved element c-commands and is fairly local to
the position from which it moved. The same holds generally for Agree. The major
difference seems to be that the control relation is not sensitive to phases in the
way that Agree and Move are. There is a non-defective, phasal v in (15), the v that
selects the VP tried PRO to leave. PRO is in the complement of this phasal v, yet
there is no problem in relating it to its controller. Phases do not seem to play the
same type of role in control as they do in Agree or Move. Thus, UPR does not
predict a problem for control. But we can ask: is this prediction correct? Does
control develop earlier than structures that depend on non-phasal (weak) v? If so,
this would be further evidence for UPR. Alternatively, if OC develops as late as
the structures with non-phasal v, then we would have evidence against UPR, and
in favor of a problem with all relations that appear to involve local c-command.29
Reviews of the development of control can be found in Wexler 1992 and

Guasti 2002. McDaniel and Cairns (1990:316) argue on the basis of their experi-
ments that ‘there is a stage, previously unattested as far as we know, during
which children lack control.’ Wexler (1992:260), surveying the data, concludes:
‘[T]here is an early stage in which children don’t know that the empty subject in
complements and adjuncts must be controlled.’ This holds though only at very
young ages. Wexler notes that of 20 children from 3;9 to 5;4 years old, only one
of the four youngest children (3;9–3;10) lacked complement control (OC) and
only one of the older 16 children (3;11–5;4) lacked OC.30
Sherman (1983) conducted a comprehension study using an act-out task with
sentences like Mary told John PRO to leave. The group from 5;0 to 5;11 had a
mean accuracy of 81%. These numbers are much better than what we see on rais-
ing sentences, where in our experiment the group at this age answered at chance
level. Guasti (2002:371) reviews many studies and agrees that initially children
lack control in certain complement constructions, but, ‘by 3 years children know
that PRO is distinct from lexical pronouns.’
The consensus seems to be that OC is in place at about 3 years of age. Before
this, external control is possible for children. This is a much earlier age than the
7-year-old age range in which raising appears to become mastered. Wexler (1992)
proposed that until about age 3, children overextend the case filter due to a matu-
rational process, requiring that all NPs have case. This would outlaw PRO. What-
ever the explanation, control of complements by objects is mastered much earlier
than raising (or verbal passives of psychological verbs in English).31
Of course, children also use control verbs like try quite often. Pinker (1984)
argues that in natural production the use of an external controller for PRO is
almost nonexistent in the data that he analyzed.32 It would be somewhat difficult
to determine the correct interpretation from transcripts of natural production,
so elicited production studies would be desirable. Nevertheless, we know of no
evidence that past around the age of 3 children comprehend or produce PRO with
an external controller when the sentence involves complement control.
Given this strong asymmetry between the ages at which good performance on
control and on raising develop, UPR is supported, as it predicts this asymmetry.
Thus we can take the development of control as further support for UPR.
9. CONTROL AS RAISING: HORNSTEIN’S PROPOSAL

AND IMPLICATIONS FOR ACQUISITION
Hornstein (1999) argues that Minimalist considerations lead us to eliminate the

control module, and that OC is best understood as raising. There is controversy
in the syntactic literature on this point: Culicover and Jackendoff (2001) and
Landau (2003) argue against Hornstein’s proposal, and Boeckx and Hornstein
(2003) reply. Obviously we cannot get into the details of this argument. We will
simply attempt to see how Hornstein’s proposal meshes with the developmental
results. What does control as raising predict about development and how do these
predictions fit the facts?
Initially, one would think that if control is raising, then the prediction is that
control and raising should develop at about the same time. Since this is greatly at
odds with the acquisition facts (see the previous sections for the developmental
comparison of OC and raising), we could conclude that the developmental data
do not support the control as raising hypothesis.
But let us do a bit more justice to the control as raising hypothesis. Let us look
at the proposal in detail to see if we can be more explicit about the relation between
the proposal and the developmental principles (like UPR) that are known.
Hornstein’s (1999) analysis is illustrated in (16) (his (19)):
(16) a. John hopes to leave
b. [IP John [VP John [hopes [IP John to [VP John leave]]]]]
We follow Hornstein in his explanation of the derivation (16b). John merges with
leave. John then raises to the embedded [Spec, IP]. John raises again to [Spec, VP]
of hope. By principles that Hornstein adumbrates, the chain that John heads has
‘two theta-roles, the leaver role and the hoper role.’ John then raises to [Spec, IP]
of the matrix clause.
Clearly there are A-chains here, but ACDH is not under discussion, UPR
is. Is UPR violated by this analysis? It is hard to tell because Hornstein does
not discuss an analysis incorporating phasal considerations, that is, any kind of
strong cyclicity. There is no vP, only VP. But let us see what seems reasonable
if we attempted to understand the derivation in phasal terms. Suppose there is
a phasal head v between the embedded IP and the lower VP. The first raising,
from [Spec, VP] to [Spec, IP] would be the movement of a phrase in the comple-
ment of v (after all, it is in the VP that v selects) to the next higher phase. If John
is actually merged into [Spec, vP], then raising to [Spec, IP] is licit given PIC
because John comes from the edge of the next lower phase. So far no phasal vio-
lations are incurred on the adult analysis, similarly for the child analysis, since
no defective phases are involved.
Now John raises to [Spec, VP] of hope. Let us assume again that hope is intro-
duced in a VP that is a complement of a v. This v is phasal, since hope assigns
an external argument. Since the raising of John allows it to check the external
theta-role feature of hope, let us assume that the raising goes to [Spec, vP].33 Then
at phase CP, T can attract John, in the edge of the lower phase, and it can raise
for the last time. Under this analysis it looks as if no defective phase is needed
in the derivation. The strict cycle can be followed, with all material except edges
shipped off to interpretation and not available at the next stage, except the edge of
the lower phase. Thus for the child, UPR is not violated. The child will be able to
compute the (raising) derivation for control.
If there is any reason that the first raising of John must go to [Spec, VP] of
hope rather then to [Spec, vP] then we are in a different situation. Since John is
then in the complement of the highest v, T cannot attract it, and the highest v will
have to be non-phasal. This seems to be against the spirit of phase theory, since
hope assigns an external argument feature. At any rate, if this were the case in the
derivation, then there would be a UPR situation for the child; the child would take
the highest v as phasal and the derivation would crash, predicting difficulties for
control structures like (16).
But it seems most reasonable, till further considerations come in, to take the
former analysis, with movement to [Spec, vP]. If this is indeed correct, then UPR
predicts no trouble for control, even with a raising analysis of control. In such a
case, the development of control cannot distinguish the control module vs. raising
analysis, at least not in these terms. Raising, presumably, would still look the way
it traditionally does, a defective v will be needed, and raising will be predicted to
be late.
It is quite interesting then that UPR, a theory of acquisition that relates
to phases rather than to chains, seems to predict no problems for children on
control, even with a raising analysis. The chains are not what matter; only the
phases and their conditions (PIC) matter. This should be a familiar lesson: labels
(‘raising’) are not as important as analyses.34 On deeper inspection, even with a
raising analysis, control might not demand extraordinary conditions on move-
ment; raising does, passives do, and unaccusatives do.
We make these observations quite tentatively; deeper syntactic analysis might
contradict us, but it will take that. For the moment, we predict good control and
poor raising until UPR matures, whether control is a control module or a raising
analysis of the sort that Hornstein provides. These predictions are in strong con-
cordance with the developmental data.35
10. CROSSLINGUISTIC EVIDENCE: A NOTE ON CONTROL

INTO FINITE CLAUSES IN GREEK
In Modern Greek, it is well known that OC occurs into certain finite (in particular
subjunctive) complements (that is, with particular matrix verbs). Varlakosta (1994)
argues that these structures are in fact control structures, offering an analysis in
terms of PRO. To the extent that the correct analysis of these structures does not
demand any defective phases, we expect that children will not be strongly delayed in
their development.
Goodluck et al. (2001:171) study the development of these structures and
conclude that they are not delayed.36 They write that ‘there is evidence that four
to five-year-olds have a grasp of the patterns particular to their language. Such
crosslinguistic contrasts support the view that children aged four and older have a
category PRO available for the subject of sentential complements.’ Thus we might
take the evidence on this variational possibility to confirm our prediction that
raising (i.e. defective phase type raising), but not control, is late in development.
This fits neatly with the data confirming that passives (both actional and psycho-
logical, as discussed earlier) are delayed for Greek-speaking children.
Kapetangianni and Seely (this volume) offer an account of the Greek finite
control facts in terms of a raising analysis in the spirit of Hornstein (1999). To
the extent that this analysis does not demand defective phases (which seems to be
the case), there would be no more reason to predict that Greek control in finite
clauses is delayed than under Hornstein’s analysis for control (as raising) in Eng-
lish, which we discussed in the previous section.
On the other hand, Kotzoglou and Papangeli (this volume) provide an analysis
of Greek that accounts for some control into finite clauses phenomena by positing
an extra- thematic role. They ask whether the control facts are ECM-like and con-
clude that they are somewhat different. Wexler (2004) argues that ECM construc-
tions should not be delayed in children because the UPR will not apply, so if these
phenomena are ECM we predict that they should not be delayed in acquisition.
On the other hand, if they involve an extra-thematic role, there is still no reason
to suppose that there is a defective phase, so again we predict that there should be
no delay. Either way, the data on development is consistent.
Spyropoulos (this volume), argues that control in Greek does not involve
PRO, nor does it involve raising the subject from a lower position. He argues
that the subjunctive is finite and that even pronouns can be controlled. In the
spirit of Varlakosta (1994) and Landau (2000), he argues that ‘it is the licensing
of the temporal [properties of the subjunctive clause] that regulates the control
pattern.’ One of the advantages of this analysis, he points out, is that ‘As a con-
sequence, we maintain and strengthen the assumption that [NOM] case is the
by-product of agreement valuation . . .’.
Under Spyropoulos’ analysis there does not appear to be the need for a
defective phase in order to account for the finite control facts in Greek. Thus
UPR does not predict a delay in control into finite clauses, and the Goodluck et
al. results thus are consistent and expected. In addition, we know fully well that
children at an even younger age know that nominative case is the by-product of
agreement valuation. As Schütze and Wexler (1996) argue, nonfinite root clauses
produced by children (‘Optional Infinitives’ in the sense of Wexler (1993)) can be
missing either agreement or tense. When they are missing tense and agreement is
present, nominative case is used on the subject. When they are missing agreement
and tense is present, the default case for English (non-nominative) case is used.
Children almost never produce sentences of the form ‘him goes’, that is, agree-
ment on the verb and a non-nominative subject although they do produce many
examples of ‘him go’. Thus we can conclude that the ingredients (at least with
respect to agreement and case) are present for children’s analysis of finite control
in Greek. The fact that the developmental data imply that children do well on
these forms suggests that, to the extent that this is the correct analysis, children are
aware of the temporal licensing properties of the subjunctive and how this relates
to control. Needless to say, the topic deserves further study.37
11. WHAT IT ALL SEEMS TO MEAN
This paper represents a collaboration of linguistic theory (UG), the theory of

linguistic development, and the experimental study of acquisition. We tested a
crucial case – the development of raising, comparing raised to unraised sentences.
Until now, the literature has been fairly quiet on the empirical facts relating to
this issue. We demonstrated that raising is delayed until about 7 years of age,
whereas similar sentences without raising were acquired much earlier. The result
follows from the UPR and the theory of UG. None of the other developmental
theories we considered can capture this new result, except for ACDH, which has
other empirical problems. Thus we confirm UPR in a new domain, raising. Fur-
thermore, we showed that there is a strong correlation between the development
of passive and raising; any theory of development will have to derive this fact.
UPR is the only candidate on the horizon; moreover, UPR is a very natural theory
given a Minimalist approach. Raising constructions should provide another tool
with which to probe the genetic basis of language. We can look forward to studies
integrating genetics and phasal computation.
* We would like to thank Misha Becker, Karen Froud, Jeremy Hartman, Karen Law, Alec
Marantz, Gregory Marton, Nadya Modyanova, Robyn Orfitelli, Alexandra Perovic, David Pesetsky,
Vina Tsakali, the entire Wexler ab/Normal Language Lab, as well as all the children and day cares that
participated in these studies. The preparation of this article was supported in part by an NSF Gradu-
ate Fellowship awarded to the first author, and by a Marcus Fund awarded to the Department of Brain
and Cognitive Sciences at MIT.
1
In particular, the development of obligatory finiteness is genetically determined, as argued in
Wexler (2002). The latest and most systematic behavioral genetic evidence strongly confirms
that the development of finiteness is controlled by genetics and that the genetic source of
finiteness is independent of the genetic source of phonological working memory (Bishop et al.
2006). There is also behavioral genetic evidence for the biological maturation of verbal passives
(Ganger et al. 2004).
2
Borer and Wexler (1987) considered maturation to take place around age 5. We have argued on the
basis of much more detailed evidence that the age of development is closer to 7.
3
Babyonyshev et al. (2001) showed that native Russian-speaking children at age 5 could not system-
atically provide the correct analysis of unaccusatives, as tested by the Genitive of Negation in
Russian. Miyamoto et al. (1999) demonstrated that children acquiring Japanese omitted nomina-
tive case for subjects of unaccusative verbs, but for no other verbs. They interpreted this to mean
that the children had difficulty in producing the A-chain between the object of the unaccusative
and [Spec, IP]. Without this chain, nominative case is not assignable, thus resulting in the lack
of nominative case. Lee and Wexler (2001) for Korean provided further evidence for this position.
Hirsch and Hartman (2006) demonstrated experimentally that children do not comprehend non-
agentive object-experiencer verbs used in their ‘active’ form (e.g. The shadows scared Mary),
while also showing that children have no difficulty with their agentive counterparts (e.g. The
witch scared Mary). Belletti and Rizzi (1988) argue that non-agentive object-experiencer verbs
(their preoccupare-class) are unaccusative, involving derived subjects, while their agentive counterparts
are not unaccusative.
4
Hyams and Snyder (2005) suggest alternatively that premature children accept a very strong ver-
sion of Wexler and Culicover’s (1980) Freezing Principle, which prevents smuggling. Following
Collins (2005), raising past experiencers requires smuggling, so young children are predicted not
to comprehend such structures. The smuggling approach predicts that there is no problem with
raising without an experiencer. As we argue later in the paper, raising even without experiencers
is delayed for children.
5
Since the basic methodology of our experiment follows theirs, the reader can get an idea of what
they did in the section that describes our experiment.
6
The presence of the experiencer is discussed in detail later in the paper.
7
A cursory, search of the CHILDES corpus of children’s utterances turns up thousands of want
control constructions such as the following:
[I want to read this paper]
[I want to shave too]
[I want to have some espresso]
[um # I want to be a gypsy]
[I want to drive]
[I want to ride on a panda]
[I want to hold a lamb # I didn’t hold it]
Needless to say, it is not very plausible that the 3-year-old speaker of the third utterance means
to convey that he actually seems to be having an espresso (the full transcript makes it clear that
he does not). By the same token, the other utterances are odd indeed on the assumption that the
children are using want to mean something like seem or any other raising verb.
Corpus-based evidence for children’s correct interpretation of control verbs is not limited to pro-
duction. Even when responding to a parent, young children show an unambiguous understanding
of verbs like want, as in the following example and many hundreds of others:
MOT: do you want to do that again?
CHI: (o)k.
(bates/free20/hank20.cha:240)
It is unclear what to make of this exchange if we are operating under the assumption that children
interpret control verbs as raising verbs.
8
For Becker, if all control verbs, regardless of their particular adult meaning, map to the syntax of raising
verbs, it would stand to reason that whatever raising verbs the control verbs map onto should them-
selves be relatively free of particular semantic meaning in order to accommodate such a large class of
interpretations. Semantically vague raising verbs like ‘seem’ and ‘appear’ make more likely targets for
such interpretation than raising verbs with more inherent meaning like ‘tend’ and ‘happen (to)’.
9
That the results of the first experiment are due to younger children taking the inanimate subjects
to be sentient in this experiment is strongly suggested by children’s justifications for accepting the
control sentences with inanimate subjects (Becker 2004):
Test item: # The bucket wants to be in the sandbox
Child: I think the bucket should be in the sandbox.
Inv: But do you think the bucket could want to be in the sandbox?
Child: I think so. (age 3;11)
Test item: # The flower wants to be pink
Child: And the bees want to eat them!
Inv: Do you think the flower could want to be pink?
Child: Yes, and green too! (age 3;1)
10
At the time the searches were conducted (5/2004), this constituted every English-speaking child
available in the CHILDES database.
11
While these analyses focus exclusively on seem, it should be noted that other raising verbs do
appear in the input to these children, including dozens of examples of appear, tend (to), used
(to), and happen (to).
12
Why should there be a difference in the age at which seem with VP complements (verbal-seem)
develops vs. seem with small-clause (adjectival) complements (adjectival-seem)? Note that there
are many considerations suggesting that adjectival-seem does not involve raising of the subject.
Adjectival-seem tends not to allow an experiencer to-phrase between itself and its complement:
(i) *John seems to Bill sad.
(ii) John seems to Bill to be sad.

The experiencer, though, is allowed with adjectival-seem if it is fronted, but a fronted experiencer
is also allowed in copular constructions:
(iii) To Bill, John seems sad.
(iv) To Bill, John is sad.
When taking a verbal complement with a stage-level predicate like available, verbal-seem allows
both existential and generic readings:
(v) Firemen seem to be available.
a. There exist some firemen x such that x seem to be available (existential reading)
b. For all firemen x, x seem to be available (generic reading)
With adjectival-seem, however, only the generic reading is present:
(vi) Firemen seem available.
a. *There exist some firemen x such that x seem available (existential reading)
b. For all firemen x, x seem available (generic reading)

In Diesing’s (1992) framework, the existential reading is derived via lowering the subject from [Spec, IP]
back into [Spec, VP]. This, of course, is only possible when the subject is initially raised from [Spec, VP] to
[Spec, IP]. For the generic reading, on the other hand, the subject is directly generated in [Spec, IP], form-
ing a subject–predicate relationship with a control analysis. The lexical NP in [Spec, IP] controls a PRO
subject in [Spec, VP], which is assigned a theta-role by the verb. This suggests an analysis like (vii):
(vii) John seems sad
[IP DP John [I′ INFL seems [VP PRO sad]]]
Consequently, the subject cannot lower and only the generic interpretation is licensed. Thus, adjectival-
seem is syntactically very similar to the copula. Crucially, there is no defective v involved, in fact no rais-
ing of the argument at all, thus no violation on UPR. As such, young children’s early productions
with adjectival-seem are no challenge to UPR, which actually predicts their early use, alongside the
much later development of verbal-seem.
13
Strictly speaking, statistically significant above-chance performance for six items, before even com-
pensating for multiple comparisons, requires getting all six items correct. To compensate for chil-
dren’s distractibility, a slightly more liberal cutoff has been used (5 of 6 correct).
14
Below-chance performance was calculated for six items per (foil) condition across an age group of
ten children, not correcting for multiple comparisons, which yields a mildly liberal cutoff.
15
Note that this also indicates that children’s preference for MR foils with think and unraised seem is not
due to any inherent preference for the MR pictures since with raised seem children prefer DR foils.
16
The passives conditions were further subdivided according to whether or not they contained a by-
phrase, half of the passives being full passives (with a by-phrase) and half being truncated passives
(without a by-phrase). It has been claimed by Fox and Grodzinsky (1998) that children compre-
hend truncated psychological passives, which they take as evidence that children do not have a
general deficit in passive comprehension. Their claim that truncated psychological passives are not
delayed, based on only eight children, has since been shown to be false. Not only has their finding
not been replicated, but numerous studies using both more children and more items demonstrate
that truncated psychological passives are just as delayed as their full counterparts (Gordon and
Chafetz 1990; Hirsch and Wexler 2004a; 2005; 2006). Of the 140 subjects examined by Hirsch and
Wexler (2006), only two children have scores matching the Fox and Grodzinsky pattern of good
performance on truncated psychological passives and poor performance on full psychological pas-
sives. Fox and Grodzinsky’s result appears to be due to methodological problems, including too few
subjects, too few experimental items, a failure to randomize conditions, and delays between when
conditions were tested.
17
That children’s analysis of actional passives is at least in part determined by environmental factors
is demonstrated in recent work by Hirsch, Modyanova, and Wexler (2006). They administered a
survey to the parents of the 140 children tested in the Hirsch and Wexler (2006) study, querying
parents about numerous biographical details. It was found that many environmental factors (e.g.
parents’ education, child’s age at enrollment in day care, number of hours child is read to, etc.)
predict acquisition of actional passives. None of the more than 20 factors surveyed, however, pre-
dicted the acquisition of psychological passives. The child’s age (younger or older than 7) was the
only predictor of psychological passive acquisition.
18
An anonymous reviewer asks about the slight drop in passive comprehension in the 6-year-olds in
Figures 3 and 4. This appears to be just coincidental, as the 6-year-old comprehension level is not
significantly different from that of the 5-year-olds in either analysis, nor does the accuracy of the
6-year-olds differ from chance level.
19
Note that above-chance performance here is different than in Table 3, as responses to DR items are
not being considered. Here, above-chance performance for raising is at least 75% correct (9 of 12
items correct). For passives, above-chance performance is defined as at least 75% correct on both
truncated and full psychological passives (6 of 8 items correct on both subconditions), which is
relatively conservative.
20
Perhaps this child has undergone the relevant linguistic maturation but for some unknown reason
fails to recognize the morphological markers of passives. If this were true, he would not constitute
an exception.
21
It is even more impressive that poor raising and passive scores correlate in the 9-year-old child since
it is known from other experiments in which the child participated that she has for her age above
average IQ as measured by KBIT, above average vocabulary as measured by PPVT, and above
average grammatical competency (at least for non-raising and nonpassive structures) as measured
by TROG. In light of these facts, poor performance on raising and passives thus demonstrates the
acquisition of these structures to be independent of other aspects of both general cognition and
linguistic development, while nonetheless being dependent on one another.
22
Perhaps most striking is the finding that many of the children even have difficulties with DR and ER
foils in the raised condition. This means, for example, that when presented the sentence To Lisa,
Bart seems to be playing an instrument many of the children are willing to (incorrectly) choose the
ER foil in which Lisa is playing an instrument while thinking about Bart. These same children never
make such a mistake with the unraised sentence To Lisa, it seems that Bart is playing an instrument.
23
Notice that the ‘simplest’ processing explanation, namely, that the mere presence of an experiencer-
phrase (i.e. extralinguistic material) accounts for children’s particular comprehension difficulties is
surely not correct. First, this processing notion makes no predictive distinction between the raised
and unraised conditions studied since both contain an experiencer, yet children’s difficulties are
confined to the former. Second, with respect to studying comprehension differences between rais-
ing sentences with and without an experiencer-phrase, it is not the case that merely adding more
material ceteris paribus results in detrimental processing overload. As noted by an anonymous
reviewer, the addition of a by-phrase to actional passives does not result in poorer comprehension
(e.g. Fox and Grodzinsky 1998; Hirsch and Wexler 2004a, 2006). The relevant possibilities explored
here are that raising over an experiencer causes either (i) processing difficulties in establishing a
long-distance dependency across the experiencer or (ii) ungrammaticality (in the standard repre-
sentational sense).
24
In addition to the experiment by Hirsch, Orfitelli, and Wexler (2006), which argues against prob-
lems with raising being due to the presence of an experiencer, Hirsch and Wexler (2004b) found
that while children have difficulties with raising (over an experiencer) in declarative sentences (e.g.
Bart seems to Lisa to be playing an instrument), the same exact children have no difficulties with
raising (again, over an experiencer) when wh-movement is also involved (e.g. Who seems to Lisa to
be playing an instrument?). With certain assumptions about Improper Movement, this is expected
on UPR (see Wexler 2004).
25
In all, 40 children were tested, with ten children in each 1-year age range from 4 years of age to
7 years of age. Similar results obtain for the 5- and 6-year-olds, while 7-year-olds perform much
better on the raised items.
26
One child, when directly asked ‘When I say “Barbie seems to be carrying a pineapple” what does it
mean?’ answered: ‘Barbie thinks she has a pineapple’. This provides direct insight into children’s
analysis of seem sentences as think sentences.
27
In the original raising study by Froud et al. (in preparation), they too included raising sentences
without experiencers (Bart seems to be wearing a hat). They found children performed well on
such structures, but noted that this is to be expected if children merely ignored the verb seem. Of
relevance, none of the foils used in this experiment were felicitous with the nonfinite think reading
suggested by Hirsch, Modyanova, and Wexler (2006). That is, there were no foils in which the sub-
ject thought he was doing the relevant action (e.g. Bart thinking that he was wearing a hat).
28
EARH also does not predict any difficulties for (15) since even the embedded sentence has an exter-
nal argument, PRO, which has referential content (it is not an expletive).
29
Of course, such local c-command structures would include Principle A of the binding theory, and
we know that reflexive binding develops much earlier than passive structures, roughly around age
3, depending on the quantitative standards and types of experiment used (Wexler and Chien 1985;
Chien and Wexler 1990; amongst many others). Such phenomena already suggest that UPR is
more on the right track than difficulties with local c-command relationships.
30
We are ignoring adjunct control, which complicates the picture somewhat, but for different reasons.
See Wexler 1992.
31
We return to the question of the very small set of subject control verbs like promise in the next section.
32
Pinker’s observations are written in terms of Equi, but the observation is equivalent to what we note.
33
Chomsky’s system does not allow movement to the first [Spec, vP], but presumably this could be
allowed in another system. Details would have to be worked out.
34
Confusion over labeling versus analyses might be at the heart of claims for early passive acquisition
in some languages (e.g. Sesotho, Inuktitut). Crawford (2005), in reviewing the literature and in her
own novel contributions, notes that the ‘passives’ in such languages might not be counterexamples to
UPR. Regardless, all claims of early passive acquisition are based solely on natural production stud-
ies. To date, there is no experimental evidence for the early acquisition of passives in any language.
35
Boeckx and Hornstein (2003) argue that the late development of subject control with verbs like
promise (when taking an ‘object’; e.g. A promised B to do Z) supports the movement analysis of
control because object control is expected given the Minimal Link Condition and the movement
analysis. Thus subject control is marked and late development is expected. This analysis, however,
does not account for the extremely late development of subject-to-subject raising. Why should it
take such a huge amount of time for subject-to-subject raising to develop given that it is possible
and is in the adult input? Further, as Boeckx and Hornstein themselves note based on informal
observations, promise with an object and nonfinite complement is ungrammatical for lots of
English-speakers. If this is true, which we think it is based on some of our own recent pilot studies,
then poor comprehension by many of the children cannot be taken as evidence for late acquisition,
but merely as a reflection that such structures are ungrammatical for many speakers. Also, it bears
noting that the relevant structure (A promised B to do Z) never appears in the child-directed speech
for any (of 1,051) English-speaking children on the CHILDES corpus. That is, children are never
presented evidence in the form of adult speech that promise is a possible control verb when an
object is also present. Wexler (1992, 2004) suggests that Larson’s (1991) account, which involves
an A-chain (and presumably defective v) in the analysis of promise subject control, together with
ACDH (or UPR), predicts the late development of such structures. A reasonable research strategy
would be to see if the correct analysis of promise involves a non-phasal v so that UPR would
predict that the computation would not converge for the premature child.
36
We would like to thank an anonymous reviewer for pointing out the relevance of this paper.
37
Note that Avrutin and Wexler (1999/2000), in an experiment, find that Russian-speaking children
make errors on obviation of the subject in subjunctive embedded clauses. They argue that children
know the syntactic properties of the subjunctive, but make discourse errors. This whole issue of
obviation and control in subjunctive clauses would make an intriguing topic for further study.
JAMES H. YOON
RAISING OF MAJOR ARGUMENTS IN KOREAN

(AND JAPANESE)*
1. INTRODUCTION – THE STATUS OF SUBJECT-TO-OBJECT

RAISING IN JAPANESE AND KOREAN
Since Kuno (1976), the consensus in generative inquiries of Japanese (and Korean)
syntax has been that the case alternation on the embedded subject shown in (1a)
and (1b) is the counterpart in these languages of the English constructions in (2a)
and (2b), respectively. That is, Japanese and Korean possess the Subject-to-Object
raising construction (and/or the Exceptional Case-Marking/Long-Distance Agree
construction). Some recent examples of the dominant view include Hiraiwa 2002
and Tanaka 2002 for Japanese and S.-M. Hong 2005 for Korean.
(1) a. Cheli-nun Yenghi-ka yenglihay-ss-ta-ko mitnun-ta (Korean)

C-TOP Y-NOM smart-PST-DECL-COMP believe-DECL
‘Cheli believes that Yenghi was smart.’
b. Cheli-nun Yenghi-lul yenglihay-ss-ta-ko mitnun-ta

C-TOP Y-ACC smart-PST-DECL-COMP believe-DECL
‘Cheli believes Yenghi to have been smart.’
(2) a. John believes (that) he is innocent
b. John believes him to be innocent
However, Hoji (1991, 2005) (see also Saito 1983; Oka 1988; Sells 1990; Takano
2003) for Japanese and K.-S. Hong (1990, 1997) (see also P.-Y. Lee 1992) for
Korean have challenged the conventional wisdom. These researchers take the
construction in (1b) in Japanese/Korean not to be an SOR/ECM construction,
but one where the accusative NP is base-generated in the matrix VP. In this view,
(1a) and (1b) are not related syntactically by movement and/or optional case-
assignment. In their recent book on raising and control, Davies and Dubinsky
(2004: Chap. 10) appear to side cautiously with the latter regarding Japanese.
Independently of the fate of Japanese/Korean SOR, they show that a number of
constructions in different languages previously analyzed as involving SOR do not
seem to be raising/ECM constructions, but something else. Davies (2005) argues
that the construction in Madurese previously taken to exemplify SOR involves
a base-generated object in the upstairs clause, on a par with the similar English
(Prolepsis) construction shown in (3) below.
71
© 2007 Springer.
72 JAMES H. YOON
(3) I believe about/regarding Bill that he is responsible for the failure
In this paper, I show that while there is apparent, but often ignored, evidence
indicating that (1b) may not be a typical raising construction, the evidence can be
reinterpreted to support a raising analysis – if we posit that what undergoes raising is
not the embedded subject, but the embedded Major Subject. A Major Subject is the
traditional designation for the ‘extra’ subject-like nominal of Multiple Nominative
Constructions (MNCs). However, as argued by Heycock and Doron (2003), Major
Subjects can be equated with subjects of categorical judgment sentences, or Categorical
Subjects. Such subjects are not restricted to sentences with two subject-like nominals,
as is well known (Diesing 1992). The hypothesis that the construction in (1b) involves
raising of the embedded Major and/or Categorical Subject accounts for many of the
apparent problems for the raising analysis. In turn, it is supported by evidence show-
ing that the accusative-marked nominal in (1b) displays properties that could only
have been determined in the lower clause. The argument for this position is made on
the basis of detailed investigation of Korean. It is applicable to Japanese as well,
insofar as the properties of the two languages parallel each other.
The organization of this chapter is as follows. In section 2, we introduce
a number of properties of SOR in Korean and Japanese that appear to be
problematic for the assumption that the languages possess a genuine SOR
construction. In section 3, we propose that SOR in Korean (and Japanese)
involves the raising of the embedded Major Subject. We demonstrate first that
there is a correlation between the possibility of an embedded nominal – sub-
ject or nonsubject – to undergo SOR and its ability to be expressed as a Major
Subject. We then show how this analysis accounts for the apparent problems
for the raising analysis of SOR. Section 4 compares the proposed analysis with
an alternative, base-generation analysis. We show that while the two are largely
equivalent in terms of descriptive coverage, the base-generation analysis suf-
fers from a number of drawbacks which favor the raising analysis defended in
the paper. Section 5 concludes the paper.
2. APPARENT PROBLEMS FOR THE RAISING ANALYSIS OF SOR
If we take the properties of the English construction in (2b) to be representative

of SOR crosslinguistically, a number of properties of the Korean construction in
(1b) appear to indicate that it may not be a canonical SOR construction.1
First, the complement clauses of SOR verbs are unreduced, finite clauses,
unlike English. This is a possible problem for movement and/or case-marking,
since it is commonly assumed that a finite clause boundary acts as a barrier to these
dependencies. However, the problem is internal to some theoretical assumptions
and may not be considered critical.
A second apparent problem stems from the fact that unlike the English
construction in (2a) and (2b), accusative-marking alternates with Nominative-
marking on the complement subject. In other words, SOR is optional. These
properties do not bode well for some theoretical assumptions either. They seem to
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 73
indicate that the complement subject has multiple cases and/or that the movement
in question lacks a Case-theoretic motivation. These too are problems for certain
theoretical assumptions.
A third apparent problem is not so theory-centered. Unlike what is assumed
in many previous investigations of SOR in Korean, we find that what raises in the
SOR construction is not just the complement subject. Nonsubjects of complement
clauses can raise, as pointed out already in Yoon 1987 (see also K.-S. Hong 1990,
1997, J.-M. Yoon 1991). A range of nonsubjects can be raised as shown below,
where we annotate the sentences with the role of the raised nominal in the embed-
ded clause. The complement clause without raising is given along with the SOR
sentences for comparison.
(4) a. Initial DP in Possession-type MNC:

Na-nun Cheli-lul meli-ka coh-ta-ko mit-nun-ta
I-TOP C-ACC head-NOM good-DECL-COMP believe-PRS-DECL
‘I consider Cheli to be smart.’
cf.
Cheli-ka meli-ka coh-ta
C-NOM head-NOM good-DECL
b. Initial scene-setting adverbial NPs:

Na-nun LA-lul (mikwuk-eyse) hankwuksalam-i kacang manhi santa-ko
I-TOP LA-ACC US-LOC Koreans-NOM most many live-COMP
mit-nun-ta
believe-PRS-DECL
‘I believe LA has the greatest number of Koreans in the US.’
cf.
LA-ka/ey mikwuk-eyse hankwuksalam-i kacang manhi sa-n-ta
LA-NOM/LOC US-LOC Koreans-NOM most many live-PRS-DECL
c. Raised (Non-accusative) Object:

Na-nun Pwukhansan-ul mwul-i manhi nanta-ko sayngkakhanta
I-TOP Mt. Pwukhan-ACC water-NOM a.lot flow-COMP think
‘I believe that there are a lot of springs flowing from Mt. Pwukhan.’
cf.
Mwul-i Pwukhansan-eyse//*ul manhi nanta
Water-NOM Mt. Pwukhan-LOC/*ACC a.lot flows
Though it is known that raising need not be restricted to complement subjects in

some languages (e.g. Choctaw, Kipsigis), the range of embedded constituents that
can apparently be raised in Korean seems larger than those in other languages.
For movement/advancement analyses of SOR, nonsubject raising would be prob-
lematic, as nonsubject raising violates proposed constraints on such processes.2
While the problem of nonsubject raising by itself does not seal the case against
SOR, a number of additional properties, when taken together, appear to spell
doom for the analysis. We turn to these below.
A fourth problem for the raising analysis of (1b) comes from the fact that
SOR can target an embedded constituent within an island. Resumptive pronouns
are acceptable for most speakers when the raised nominal is related to a position
74 JAMES H. YOON
within an island (5), but even in the absence of island boundaries, as long as there
is sufficient distance between the nominal and the constituent targeted for raising
in the embedded clause (6).
(5) Na-nun Yenghi-luli [[ ei/kunye-ka e ha-nun] il]-i

I-TOP Y-ACC she-NOM do-ADNOM work-NOM
mopemcek-ila-ko sayngkakhanta
exemplary-COP-COMP think
‘I think of Yenghi that the things she does are exemplary.’
(6) Na-nun Cheli-lul hangsang kunyesek-i taytanhan malssengkkwuleki-la-ko

I-TOP C-ACC always the.guy-NOM extreme troublemaker-COP-COMP
sayngkakhayssessta
had.thought
‘I had always considered Cheli to be a real troublemaker.’
Though the possibility of resumption by itself does not necessarily constitute

evidence against the existence of a movement-like dependency, since there are
copy-raising constructions showing movement locality (Moore 1998; Potsdam
and Runner 2001), resumption combined with lack of locality strongly suggests
that movement might not be implicated in Korean/Japanese SOR.
A final category of apparent problems for the raising analysis of (1b) comes
from the fact that non-raised (1a) and raised (1b) constructions differ in terms of
a number of interpretive properties.
Idiomatic readings are not retained when subject idiom chunks undergo
raising, as shown below (J.-S. Lee 1992).3
(7) Hankwuksalam-un cakun cochwu-ka maypta-ko sayngkakhanta→ idiomatic, literal

Koreans-TOP small pepper-NOM hot-COMP thinks-DECL
vs.
…..cakun kochwu-lul maypta-ko sayngkakhan-ta → *?idiomatic, literal
small pepper-ACC hot-COMP thinks-DECL
Literal reading: Small peppers are hot.

Idiomatic reading: Size/height is not a measure of toughness.
Persistence of idiomatic readings on the raised subject has been used as a key
diagnostic of SOR, which distinguishes it from object control. As such, the fact
that idiomatic readings are lost in apparent SOR in Korean (and possibly Japa-
nese) appears to militate against a raising analysis.4
J.-M. Yoon (1989) pointed out another difference between (1a) and (1b).
A raised indefinite subject differs interpretively from a non-raised one, as shown
in the following pair of sentences modeled on Takano 2003.
(8) a. Kyengchal-i myes-myeng-uy namca-lul peminila-ko tancenghayssni?

police-NOM how.many-CL-GEN man-ACC culprit-COMP conclude.INT
‘How many men of the men do the police consider to be culprits?’
b. Kyengchal-i myes-myeng-uy namca-ka peminila-ko tancenghayssni?

police-NOM how.many-CL-GEN man-NOM culprit-COMP conclude.INT?
‘How many men do the police consider to be culprits?’
The raised nominal in (8a) is interpreted primarily in the specific (that is, partitive
or presuppositional) sense, while the same nominal in a sentence without raising
in (8b) is interpreted in the nonspecific (that is, cardinal) sense.
O’Grady (1991) (see also J.-G. Song 1994) points out another interpretive
difference between raised and unraised structures. He notes that (9a) with rais-
ing can describe a situation where, say, John wakes up at night upon hearing a
noise and thinks that an intruder has broken in, but does not realize that it is his
wife. (9b), by contrast, implies that John is aware that the one making the noise
was his wife.
(9) a. John-un caki anay-lul totwuk-i-la-ko sayngkakhay-ss-ta

J-TOP self wife-ACC thief-COP-DECL-COMP think-PST-DECL
‘John thought that his wife was the thief.’
b. John-un caki anay-ka totwuk-ila-ko sayngkakhay-ss-ta

J-TOP self wife-NOM thief-COP-COMP think-PST-DECL
‘John thought that his wife was the thief.’
The first reading is the de re reading. The second can be thought of as a de se read-
ing, where the anaphor is under the scope of the higher intensional verb.
P.-Y. Lee’s (1992) data shown below also involve the availability of de re
readings. A ‘mistaken identity’ reading where John mistakenly thinks that the
individual named Cheli is Tongswu (due to obstructed vision, for example)
is felicitous with the raised version in (10a), but not (10b). The latter can be
uttered only if John believes that the individual named Cheli also goes by
another name, Tongswu. That is, the de re reading is possible in (10a), but not
in (10b).
(10) a. John-un Cheli-lul Tongswu-la-ko sayngkakhanta/hayssta

J-TOP C-ACC T-COP-COMP thinks/said
‘John thought that Cheli was Tongswu.’
b. John-un Cheli-ka Tongswu-la-ko sayngkakhanta/hayssta

J-TOP C-NOM T-COP-COMP thinks/said
‘John thought that Cheli was Tongswu.’
Relative scope of quantifiers differs in raised and unraised structures. The following
examples from Japanese (Oka 1988 via Takano 2003) show that while a non-raised
complement subject with a passivized embedded predicate commutes in terms of
scope with a nonsubject of the embedded clause (11a), a raised subject cannot. In
other words, a raised nominal does not reconstruct to the embedded clause for the
purposes of scope.
76 JAMES H. YOON
(11) a. Mary-wa sannin-no gakusei-ga subete-no sensei-ni syookais-are-ru

M-TOP three-GEN student-NOM all-GEN teacher-to introduce-PASS
bekida-to omotteiru
should-COMP thinks
‘Mary thinks that three students should be introduced to all the teachers.’
three > every, every > three
b. Mary-wa sannin-no gakusei-o subete-no sensei-ni syookais-are-ru
M-TOP three-GEN student-ACC all-GEN teacher-DAT introduce-PASS
bekida-to omotteiru
should-COMP thinks
‘Mary believes three students to have been introduced to all the teachers.’
three > every, *every > three
Reconstruction for bound variable readings is similarly barred in raising, as the

following pair of Korean sentences illustrate. The bound reading of the anaphor
caki is (somewhat marginally) possible in (12a), but absolutely ruled out in (12b),
where raising has taken place.
(12) a. ? Na-nun caki sensayng-uy chwuchense-ka citohaksayngtul-eykey

I-TOP self teacher-GEN letter-NOM advisees-DAT
kakkak kongkay-toy-eyahanta-ko sayngkakhanta
each release-PASS-must-COMP think
‘I believe that their teacher’s letters of recommendation should be released to
each advisee.’
b. *Na-nun caki sensayng-uy chwuchense-lul citohaksayngtul-eykey
I-TOP self teacher-GEN letter-ACC advisees-DAT
kakkak kongkay-toy-eyahanta-ko sayngkakhanta
each release-PASS-must-COMP think
In sum, a number of properties of Korean (Japanese) SOR seem surprising if

indeed raising is what relates (1a) and (1b).
3. SOR IN KOREAN-JAPANESE IS RAISING OF EMBEDDED

MAJOR SUBJECT
We shall argue in this section that the generalizations noted as problematic for the
raising analysis of Korean (Japanese) SOR are only apparent problems. Our argu-
ment for this conclusion rests on the claim that SOR does not raise an embedded
subject directly, but an embedded Major Subject. The term Major Subject is the
designation for the initial nominative-marked DP in an MNC, shown below.
(13) a. Cheli-ka(MS) apeci-ka(GS) pwuca-i-si-ta

C-NOM father-NOM rich-COP-SUBJ.HON-DECL
‘Cheli’s father is rich.’
b. I hakkyo-ka(MS) enehakkwa-ka(GS) coh-ta
this school-NOM linguistics-NOM good-DECL
‘The linguistics department at this school is very good.’
MNCs are characterized by the fact there is more than one subject-like constituent,
the Major Subject (=MS) and the Grammatical Subject (=GS). A grammatical
subject is the subject of the VP, an unsaturated predicate. The role of the
grammatical subject is often borne by the external argument of the verb.
A Major Subject is a subject on which the sentence consisting of the grammatical
subject and VP are predicated. While sentences are thematically closed, in MNCs,
sentences can be turned into predicates and that is why there can be two (or more)
subjects. When sentences function as predicates, we call them Sentential Predicates.
The choice of these terms (except for the term ‘grammatical subject’ for what
is traditionally called ‘Minor Subject’) is strictly intentional. They are used to
highlight the insights in traditional studies of MNCs that there is more than one
subject-like constituent in these constructions.
As long recognized in both traditional and generative approaches to MNCs,
the process of Sentential Predicate formation can be recursive, yielding more than
one Major Subject and nested Sentential Predicates. For simplicity, we deal with
MNCs with only one Major Subject. The discussion is not affected by this
simplification. When there are multiple Major Subjects, only the first, highest,
one undergoes SOR.5
A Major Subject occupies a position higher than the grammatical subject.
We shall argue that all instances of SOR, even those that seem to target
embedded subjects, target the Major Subject position. Thus, SOR in K/J should
be analyzed as the raising of a subject, albeit that of a Major Subject. Once this
is recognized, most of the unexpected properties of K/J SOR can be naturally
accounted for.
The key components of the analysis can be summed up as follows:
(14) (i) Verbs that govern SOR in Korean select complement clauses with a Major Subject
(that is in construction with a Sentential Predicate) when SOR takes place.6
(ii) The Major Subject of the embedded clause, and not the grammatical subject,
undergoes SOR, which is an instance of A-movement (J.-M. Yoon 1989; Yoon
2004a, b).
(iii) The Major Subject of the embedded clause may be co-indexed with a null or
overt pronoun within the Sentential Predicate. The co-indexation, however, is not
movement.7
(iv) The GR of the gap/pronoun co-indexed with the Major Subject is not restricted
to that of grammatical subject, though often the Major Subject and the
grammatical subject (or a constituent within the grammatical subject) are
co-indexed.
(v) The Major Subject and the Sentential Predicate that is in construction with it must
satisfy certain semantic conditions in order to be felicitous (Kuno 1973; J.-M.
Yoon 1989; K.-S. Hong 1997; Yoon 2004a, b).
(vi) The position occupied by the raised Major Subject is a derived/non-thematic

Major Object position in the matrix clause.
78 JAMES H. YOON
The following is a schematic representation of the derivation of Korean (Japanese)

SOR, where we indicate Predicate Abstraction yielding a Sentential Predicate by a null
Operator:
(15) ….
DP’i V’/VP
XP V
: movement
DPi ZP : coindexing
: predication/theta-role assignment
Opi ….. pron /ei ……
DP’: The surface/derived Object position of moved Major Subject

DP: Base position of the Major Subject
ZP: Sentential Predicate
XP: Sentential constituent containing a Major Subject and Sentential Predicate
pron/e: Constituent within Sentential Predicate co-indexed with Major Subject
In what follows, we demonstrate that the proposed analysis of SOR in K/J

can insightfully explain a number of unexpected properties of SOR introduced in
section 2.
3.1 Restrictions on objects that undergo SOR
Recall that a prima facie problem for the raising analysis of K/J SOR is that
nonsubjects, including objects, seem to undergo raising (4). Since we know that
in other languages SOR is restricted to complement subjects (and perhaps objects
– see Davies 2005 for discussion), the fact that a variety of constituents that
do not function as embedded subjects can seemingly undergo SOR appears to
jeopardize the movement analysis of K/J SOR. We shall argue that the problem
can be resolved under the analysis proposed above. This is because nonsubjects
that appear to raise do not raise directly from within the complement clause.
Instead, it is the Major Subject that is co-indexed with the embedded nonsubject
constituent that undergoes raising. Thus, contrary to appearance, only embedded
subjects (Major Subjects) raise in SOR.
In this section we chart our path through a specific prediction that this analy-
sis makes about nonsubject raising. An embedded nonsubject should be allowed to
raise in SOR if and only if it can be expressed as an embedded Major Subject. We
show that this prediction is confirmed. We will develop this argument on the basis of
embedded objects that appear to undergo raising.
Regarding the raising of objects in SOR, many researchers who have cursorily
examined the facts have assumed that embedded objects cannot raise at all. The
assessment cannot be correct, since we have seen felicitous examples of objects raised
in SOR earlier (4).8 However, it remains a fact that compared to the raising of embed-
ded subjects, nonsubject raising is much more difficult. The question that arises is why
this should be so. We propose that the reason it is more difficult to raise embedded
objects is because MNCs possessing Major Subjects that are co-indexed with an object
gap (or a gap within an object constituent) are licensed under more stringent condi-
tions than those where Major Subjects are co-indexed with a grammatical subject or a
constituent embedded within one. If so, and if SOR raises embedded Major Subjects,
we can see why SOR of an embedded object will be restricted. This is the argument
we develop below.
In the vast literature on MNCs, it is commonly assumed that the Major
Subject must be related to a grammatical subject. Examination of garden-variety
MNCs appears to bear out this assessment. For example, (16) below is a
representative MNC where the Major Subject binds a constituent within the
grammatical subject:
(16) Johni-i(MS) [ ei khi-ka](GS) khu-ta

J-NOM height-NOM great-DECL
‘As for John, his/the height is great (=John is tall).’
It is this type of sentence that figures prominently in the literature on MNCs.

However, further examination reveals that Major Subjects need not always be
related to a grammatical subject. There are MNCs – albeit not common – where
the Major Subject does not bind a variable in (or within) the grammatical sub-
ject. And even when the Major Subject does bind a variable within the Sentential
Predicate, the variable need not be in (or within) the grammatical subject position.
(17a) and (17b) below are MNCs that contain Major Subjects that are not related
to a position within the grammatical subject. The Major Subject binds the object
in these sentences and yet the MNCs are acceptable (Yoon 2004a).9
(17) a. Ilen chayk-ii(MS) [SP salamtul-i(GS) ei culkye ilknunta ]

this.kind book-NOM people-NOM enjoying read
‘As for/it is this kind of book (that) people enjoy reading.’
b. Mikwuk-yenghwa-kai(MS) [SP salamtul-i(GS)) enu kukcang-eyse-na

American-movie-NOM people-NOM which theater-LOC-ever
yocum swipkey ei po-l swu iss-ta/po-n-ta ]
these.days easily see-COMP can be-DECL/see-PRS-DECL
‘American movies can be seen by people in any movie theater.’
By contrast, the following MNC with a similar structural profile (i.e. where the
Major Subject binds an object) is judged to be quite degraded by native speakers:
(18) *Mikwuk-yenghwa-kai(MS) [SP John-i(GS) cikum ce kukcang-eyse

American-movie-NOM John-NOM now that theater-LOC
ei po-ko iss-ta]
see-COMP be-DECL
‘It is an American movie that John is watching in that theater now.’
80 JAMES H. YOON
What is responsible for the difference between acceptable and unacceptable

MNCs where the Major Subject binds an object observed above? Following Yoon
2004a, we propose that the difference boils down to whether or not the Sentential
Predicates in the MNCs in question satisfy the interpretive condition called the
‘characteristic property’ condition (aka ‘Aboutness Condition’) identified by
Kuno (1973). That is, while the Sentential Predicates of MNCs in (17a) and (17b)
predicate a characteristic, or distinguishing, property of the Major Subject, that
in (18) does not. The property of John watching the movie right now in a theater
is not a pragmatically plausible property that characterizes American movies,
compared to the property of being able to seen in just about any theater in town.
Yoon 2004a unpacks the factors involved in the ‘characteristic property’
condition into the following components:
(a) Preference for generic/habitual vs. episodic interpretation of the Sentential
Predicate
(b) Preference for the lexical predicate within the Sentential Predicate to be an
individual-level predicate
(c) Preference for the Major Subject to be more salient than the grammatical subject
As we can see, the MNCs in (17a, b) satisfy these preferences, while that in (18)
does not.
We have just seen that MNCs where the Major Subject binds (into) an object
are possible. But why should they be more restricted than other types of MNCs
where the Major Subject binds (into) the grammatical subject? We would like to
offer the following, somewhat speculative, line of reasoning as a possible answer.
In an MNC like (16), the VP is predicated of the grammatical subject, the
most salient argument of the verb, while the Major Subject is co-indexed with
a constituent within the subject. Note that the Sentential Predicate forma-
tion in MNC does not change the salience of the grammatical subject/external
argument relative to the other arguments of the verb. In contrast, in the MNCs
in (17), while the VP is predicated of the grammatical subject, the larger
predication by the Sentential Predicate targets the object, as the Major Subject
is co-indexed with it. This creates a potential conflict since predication by VP
treats the Subject as salient while that by the Sentential Predicate treats the
object as salient.
If we understand the MNCs in (17) this way, we can see how the factors noted
earlier help in the identification of a Sentential Predicate in such MNCs. All three
factors mentioned above have the function of demoting the grammatical subject
in salience relative to the Major Subject, thus facilitating the identification of the
Sentential Predicate.10
Let us take stock of what we have done so far. First, we saw that the Major
Subject in an MNC need not be restricted to binding (into) the grammatical
subject. The variable within the Sentential Predicate, if there is one, can be in
(or within) the object position. Second, MNCs where the Major Subject binds
(into) an object are rare compared to those where it binds (into) the grammatical
subject. We attributed this to the difficulty of constructing a felicitous Sentential
Predicate that satisfies the ‘characteristic property’ condition in the former type
of MNCs.
We now have the necessary ingredients to explain (i) how an embedded object
can appear to raise in SOR, and (ii) why such raising is more restricted when
compared to the raising of embedded subjects. The answer to both is grounded
in the hypothesis that what undergoes raising in SOR is the embedded Major
Subject. Major Subjects can be co-indexed with objects, but the co-indexation is
quite constrained.
Now that we have explained the possibility of embedded object raising as
well as the reason for its restricted distribution, let us turn to the evaluation of
the specific prediction of the Major Subject raising hypothesis – all and only
embedded objects that can be expressed as a Major Subject of the embedded
clause will undergo SOR.
This prediction is not directly testable when we embed the MNCs in (17) and
(18) introduced above under verbs that govern SOR. This is because while (19a)
(the putative raised version of 17) is well-formed, so is (19b) (the putative raised
version of 18). The prediction that only an object that can be expressed as the
Major Subject should undergo SOR seems to be falsified.
(19) a. Na-nun ilen chayk-ul salamtul-i culkye ilknunta-ko saynkakhanta

I-TOP this book-ACC people-NOM enjoy read-COMP think
‘I think that people like to read this kind of book.’
b. Na-nun mikwuk-yenghwa-lul Cheli-ka cikum ce kukcang-eyse poko

I-TOP US-movies-ACC C-NOM now that theater-LOC see
issta-ko sayngkakhanta
is-COMP think
‘I think that Cheli is watching an American movie in the theater right now.’
Fortunately, we need not come to this conclusion. There is no reason to think that
(19b) or (19a), for that matter, is derived by SOR. There is another parse of the
sentences where they are derived by scrambling of the embedded object. The
two derivations (SOR derivation and Scrambling derivation) cannot be teased
apart for these sentences. Therefore, what we need to test the prediction is the
following. We need to find embedded clauses containing predicates that do not
govern accusative case on their objects. If such objects can show up accusative-
marked in a position preceding embedded clause constituents, the only way in
which they could have gotten there is by SOR since Scrambling does not yield
case alternations. Our analysis predicts that the raised non-accusative object in
such sentences should be expressible as a Nom-marked embedded Major Sub-
ject. We show below that this prediction is borne out.
(20a) is a sentence with the requisite properties. Availability of Acc-mark-
ing (which is unavailable without raising, as shown in 20b) and word order
relative to the embedded clause show that the Acc-marked DP that is inter-
preted as the embedded object in (20a) has undergone SOR. We predict the
raised object to be expressible as the Major Subject of the embedded clause.
82 JAMES H. YOON
The prediction is borne out, as we see in (20c). We take (20a) to have the analy-
sis sketched in (20d).
(20) a. Na-nun Pwukhansan-ul mwul-i manhi nanta-ko sayngkakhanta

I-TOP Mt. Pwukhan-ACC water-NOM a.lot flow-COMP think
‘I believe that there are a lot of springs flowing from Mt. Pwukhan.’ (=4e)
b. Mwul-i Pwukhansan-eyse/*lul manhi nanta

water-NOM Mt. Pwukhan-LOC/*ACC a.lot flows
‘Many springs flow from Mt. Pwukhan.’
c. Pwukhansani-i(MS) mwul-i ei manhi nanta

Mt. P-NOM water-NOM a.lot flows
‘As for/it is Mt. Pwukhan (from which) a lot of springs flow.’
d. Na-nun Pwukhansani-ul [ei(MS) mwul-i ei manhi nanta-ko] ….
We have seen thus far that the constraints on the apparent raising of embedded
objects correlate with the constraints on the ability of a Major Subject to be
co-indexed with an embedded object. We take this to be evidence that the raised
nonsubject in SOR is the embedded Major Subject that is co-indexed with the
nonsubject constituent. In fact, the correlation between raising and Major
Subjecthood extends beyond objects, though we have focused on objects in
this section. As we saw earlier in (4), any embedded constituent that is not
the grammatical subject and appears to undergo SOR is expressible as the Major
Subject of the embedded Clause. Therefore, the following correlation holds:
Major Subject-SOR Correlation
A nominal raised in SOR must be felicitous as the Major Subject of the complement
clause.
3.2 Governedness of embedded predicates in embedded subject raising
In the previous section, we showed that nonsubjects only appear to undergo SOR.
It is the Major Subject co-indexed with the nonsubject constituent that is raised.
In making this argument, we capitalized on the interpretive restrictions on the
embedded clause in nonsubject raising and the ability of the raised nonsubject to
be expressed as a Major Subject. A consequence of our analysis is that embedded
clauses in nonsubject raising always possess two (or more) overt subjects – the
Major Subject and the grammatical subject. And it is the former, structurally
higher, subject that undergoes SOR.
The question that arises at this juncture is the following: what about clauses
where embedded subjects are raised? Do such clauses also have two subject posi-
tions and is SOR restricted to the higher, Major Subject, position? We will defend
an affirmative answer to this question in what follows.
A cursory examination of standard SOR sentences like (1b) may lead us to suppose
that subject raising patterns differently from nonsubject raising. That is, any embedded
grammatical subject can be raised, perhaps on the strength of its subjecthood

alone. However, a careful investigation of constraints on subject raising will reveal
that this is not the case. In the previous section, we saw that there are restrictions
on both the raised nonsubject (that is, the Major Subject co-indexed with it)
and on the embedded clause in the case of nonsubject raising. What leads us to
hypothesize that raised subjects are no different from raised nonsubjects is the
following: even when an embedded subject undergoes raising, we find the same
restrictions. We will interpret this to mean that both subjects and nonsubjects
raise from a higher, Major Subject, position and not from a lower, grammatical
subject, position.
J.-S. Lee (1992) pointed out that embedded predicates in SOR are restricted.
He claimed that only non-case-assigning predicates are allowed as the predicate of
the embedded clause when SOR takes place (which for him targets only embedded
subjects). That is, he claimed that unaccusative intransitives are possible as the
predicate of the embedded clause, but not unergatives (if we assume they are
potential case-assigners) or transitive verbs. However, when we examine the data
systematically, we find that the actual generalizations are different.
Among intransitive verbs, individual-level predicates are optimal, (21a) and
(21b), while stage-level intransitives are not, (22a) and (22b). Correlated with
the restriction on the embedded clause predicate when SOR takes place is the
interpretation of bare plural indefinite subjects (Diesing 1992). In (21), the subject
is interpreted generically while in (22) it is interpreted existentially. This restriction
is in force only when SOR takes place, which is shown by the fact that (22a, b)
without SOR (nominative-marked embedded subjects) are acceptable.
(21) a. Cheli-nun tolkolay-lul/ka phoyutongmwul-ila-ko sayngkakha-n-ta

C-TOP dolphins-ACC/NOM mammal-COP-COMP think-PRS-DECL
‘Cheli considers dolphins to be mammals.’
b. Cheli-nun tolkolay-lul/ka yenglihata-ko sayngkakahanta

C-TOP dolphins-ACC/NOM intelligent-COMP thinks
‘Cheli considers dolphins to be intelligent.’
(22) a. Cheli-nun tolkolay-*?lul/ka pointa-ko sayngkakhanta

C-TOP dolphins-*?ACC/NOM visible-COMP think-PST-DECL
‘Cheli considers dolphins to be visible.’
b. Cheli-nun tolkolay-*?lul/ka mwul-eyse ttwie ollassta-ko sayngkakhayssta

C-TOP dolphins-ACC/NOM water-from jump up.PAST-COMP thought
‘Cheli considered dolphins to have jumped from the water.’
However, it is not that stage-level predicates are ruled out altogether when SOR
takes place. Even when the predicate is a stage-level intransitive verb SOR is possible
as long as the entire VP of the embedded clause can be interpreted as predicating a
characteristic property of the raised nominal. Thus, (23a) and (23b) are acceptable.
As expected, the raised bare plural subject is interpreted generically, rather than
existentially.
84 JAMES H. YOON
(23) a. Cheli-nun tolkolay-lul yeki-se cal pointa-ko saynghakhanta

C-TOP dolphins-ACC here-from easily visible-COMP thinks
‘Cheli believes dolphins can be easily seen from here.’
b. Cheli-nun tolkolay-lul mwul-eyse nophi ttwieollul swu issta-ko sayngkakhanta

C-TOP dolphins-ACC water-from high jump.up can be-COMP thinks
‘Cheli considers dolphins to be capable of jumping up high.’
VPs headed by transitive verbs are also possible in embedded clauses so long
as they can be interpreted as denoting a characteristic property of the raised
nominal, as the contrast between (24a) and (24b) demonstrate.11 Again, when
SOR does not take place, there are no restrictions on the embedded predicate (the
nominative-marked embedded subjects in (24a) and (24b)).
(24) a. Cheli-nun wonswungi-*?lul/ka banana-lul cikum meknunta-ko sayngkakhanta

C-TOP monkey-ACC/NOM banana-ACC now eat-COMP thinks
‘Cheli considers a/the monkey to be eating a banana right now.’
b. Cheli-nun wonswungi-lul/ka banana-lul cal meknunta-ko sayngkakhanta

C-TOP monkey-ACC/NOM banana-ACC well eat-COMP thinks
‘Cheli thinks monkeys love to eat banana.’
We claim that the restriction on embedded predicates when SOR takes place is none
other than the ‘characteristic property’ condition that holds for Sentential Predi-
cates in MNCs discussed in section 3.1. There we noted that while individual-level
predicates are preferred as the lexical predicate of MNCs, stage-level predicates are
allowed as long as the entire Sentential Predicate can be construed as denoting a
characteristic property of the Major Subject. What we have seen above is that even
when there is a single subject in the complement clause of SOR verbs, the predicate
in construction with that subject must satisfy the ‘characteristic property’ condition
if SOR is to take place.
This invites the following question. Why must the embedded predicate of clauses
satisfy the ‘characteristic property’ condition when SOR takes place, regardless of
whether an embedded subject or a Major Subject co-indexed with a nonsubject
undergoes raising? We propose that the answer lies in the fact that SOR targets
Categorical Subjects.
A Categorical Subject is the subject of sentences expressing a categorical
judgment, in the sense of Kuroda 1972 and Ladusaw 1994. Ladusaw (1994)
reinterprets the stage-individual distinction in terms of the theory of judgment
forms. Individual-level predicates occur in sentences expressing a categorical judg-
ment, while stage-level predicates typically occur in those expressing a thetic judg-
ment. In this sense, the categorical-thetic distinction is similar to the individual-stage
distinction. However, the two cannot be equated. As Ladusaw 1994 shows, while
lexically individual-level predicates are found in sentences expressing categorical
judgments, sentences containing lexically stage-level predicates can express cat-
egorical judgments when the VP they head is construed as denoting a property.
As we have seen, the restriction on embedded predicates in K/J SOR cannot

be phrased in terms of the stage-individual distinction, since lexically stage-level
predicates are possible in the embedded clause. However, we have also seen that
when the embedded predicate is lexically stage-level, the entire embedded clause must
be construable as stating a property of the raised nominal. The property-denoting
restriction is in force when nonsubjects appear to raise. It is also present when
embedded subjects undergo SOR.
A logical interpretation of these observations is that verbs that govern SOR
select embedded clauses construable as expressing a categorical judgment. In
other words, SOR targets embedded Categorical Subjects. Heycock and Doron
(2003) observe that the Major Subject in an MNC has the interpretive properties
of Categorical Subjects. Thus, it is not surprising that the Major Subject avails
itself of SOR, regardless of whether it is co-indexed with (a constituent within)
the embedded subject or with a nonsubject constituent. In embedded sentences
without multiple subjects, the complement subject can undergo SOR as long as it
is construable as a Categorical Subject.
However, this cannot be the end of the story. What needs to be explicated
is what prevents embedded Thetic Subjects from undergoing SOR. We know
that verbs governing SOR can take embedded clauses construable as expressing
a thetic judgment when SOR does not take place (as in (22) and (24a) with
nominative-marked embedded subjects). The fact that embedded Thetic Subjects
cannot undergo SOR appears to undermine the movement analysis of SOR, since
an embedded Thetic Subject is the highest A-specifier of the embedded domain
in such cases.
We argue that the reason Thetic Subjects do not undergo SOR is because the
positions of two types of subjects are different. Categorical Subjects – whether
in MNCs or sentences with single subjects – are Major Subjects occupying the
highest A-specifier position of the embedded cluase. Thetic Subjects, by contrast,
occupy a low subject position. We also hypothesize that there is a higher subject
position in sentences expressing thetic judgment. It is the presence of the higher
subject that prevents embedded Thetic Subjects from undergoing SOR. The
different positions of two types of subjects are illustrated schematically below:12
(25) a. e(MS) tolkolay-ka(GS) yeki-se cikum pointa (thetic subject)

dolphins-NOM here-LOC now visible
‘I can see some dolphins from here.’
b. Tolkolay-kai(MS) ei(GS) yeki-se cal pointa (categorical subject)

dolphins-NOM here-LOC often visible
‘Dolphins can be easily seen from here.’
We begin our argument for the analysis sketched above by noting that the inability
of movement to target embedded Thetic Subjects is not restricted to K/J SOR.
There is another case where movement discriminates between Categorical and
Thetic Subjects – Small Clauses in English (Basilico 2003).
86 JAMES H. YOON
As is well known, there are two types of SC’s in English – verbal and
adjectival SCs. One difference between the two types of SCs is that only the
subject of adjectival SCs can undergo A-movement. Another difference is that
adjectival SCs often express a categorical judgment, while verbal SCs seem to
always express a thetic judgment.
(26) a. John was considered [SC t unreliable] (categorical/adjectival SC)
a´. We considered [SC John unreliable]
b. *John was seen [SC t leave the room] (thetic/verbal SC)
b´. We saw [SC John leave the room]
It may seem that what is at stake is that subjects of embedded verbal SCs cannot
undergo movement, while those of adjectival SCs can. However, this is not the
case. There are certain adjectival SCs that express a thetic judgment (as can be
ascertained by the generic vs. existential interpretation of bare plural subjects).
In such cases, the embedded subject cannot move, as the following illustrates (Bill
Davies and Stan Dubinsky, personal communucation):
(27) a. We considered [firemen generally available] (categorical/adjectival SC)
a´. Firemen were considered [t generally available]
b. We considered [firemen late this morning] (thetic/adjectival SC)
b´ *Firemen were considered [t late this morning]
The correct generalization is the following: an embedded subject of categorical

SCs can be A-moved while a subject of a thetic SC cannot. Note that this contrast
is exactly what we observed with SOR in Korean. Predicates that govern SOR
select both categorical and thetic embedded clauses and yet only the subject of a
categorical clausal complement can undergo raising.13
Basilico (2003) proposes to make sense of the English paradigm introduced
above in the following way. First, he assumes that Categorical and Thetic Sub-
jects occupy different positions, the former higher than the latter. He then makes
the assumption that even in thetic sentences, there is a higher subject (Topic, for
him) position. It is this subject which prevents the lower, Thetic, Subject from
moving over it, as the movement in question would constitute a violation of
Relativized Minimality (Rizzi 1990).14
The proposal that a higher subject position exists even in sentences expressing
a thetic predication is neither novel nor far-fetched. As Basilico (2003) notes, a
number of researchers have defended such a position. In the terminology of this
paper, a sentence expressing a thetic judgment can also be thought of as a Sen-
tential Predicate, which is predicated not of a DP subject but a spatio-temporal
event argument (Kratzer 1995) or a Stage Topic in the sense of Erteschik-Shir
1997 which occupies the higher, Major, subject position.
Basilico’s (2003) analysis is straightforwardly applicable to Korean SOR. Let

us assume with Basilico and other researchers that there are always two subject
positions in a sentence – a higher, Major Subject, position, and a lower, grammatical
subject, position. In a sentence expressing a categorical judgment, the higher subject
position is filled by the Categorical Subject, while the lower subject position is filled
by the grammatical subject (which may or may not be co-indexed with the higher
subject). In a sentence expressing a thetic judgment, by contrast, the high subject
position is filled by the spatio-temporal event argument, while the lower subject
position is filled by the Grammatical, Thetic, Subject. Because of the presence of
the event argument in the higher subject position, raising of Thetic Subjects to the
matrix VP is barred. The analysis is illustrated schematically below:
(28) …… v’
VP v
WP V’
ZP V
Z MS: Major Subject = Categorical Subject

or Event Argument
MS GS: Grammatical Subject = Thetic
GS
Subject
WP: Landing site of SOR
There is a potential complication in this argument, arising from the fact that
in English, raising of subjects from complement TPs – as opposed to raising
from complement SCs – is not sensitive to the categorical-thetic distinction
(Basilico 2003):
(29) a. John was thought [TP e to be intelligent]
b. John was thought [TP e to have left]
The question is why the distinction between categorical and thetic predications
should not be similarly neutralized in K/J SOR, as the complement clause in SOR
constructions in the languages is clearly not a Small Clause.
The reason that the distinction is neutralized in raising from nonfinite comple-
ment TPs in English is attributable to the fact that all subjects in English must
occupy SpecTP, perhaps due to a ‘strong’ EPP feature that T has. Assuming that
the event argument cannot satisfy the EPP in SpecTP (Basilico 2003), the lower,
Thetic, Subject must move to it. Once the lower subject is in SpecTP, there is
no obstacle to further movement of the subject to the upstairs clause, because
SpecTP is the highest A-position in the embedded domain. This is why the distinction
is neutralized when a TP is selected by SOR verbs in English.15
In Korean (and Japanese), by contrast, there is ample evidence that subjects do
not undergo obligatory EPP-driven raising out of vP/VP to SpecTP. One particularly
88 JAMES H. YOON
striking confirmation that the surface position of subjects is within vP is the

following paradigm (Yoon 1994; Takano 2004):16
(30) a. [CP[TP[vP [vP John-i apeci-ka chencay-i]-ko [vP Mary-ka emeni-ka mi.in-i]]-ess]-ta]
J- NOM father-NOM genius-COP-CNJ M-NOM mother-NOM beauty-COP-PST-DECL
‘John’s father was a genius and his mother was a beauty.’
b. [CP [TP [TP John-i apeci-ka chencay-i-ess]-ko [TPMary-ka emeni-ka mi.in-i-ess]]-ta]

J-NOM father-NOM genius-COP-PST-CONJ M- NOM mother-NOM beauty-COP-PST-DECL
In verbal coordination, tense and mood need not be expressed on all conjuncts.
In particular, the nonfinal conjunct need not carry tense, in which case the tense
expressed on the final conjunct has distributive scope over the entire coordinate
structure.17 The above researchers interpret this state of affairs to mean that
the structure in (30a) involves coordination of vP, rather than TP. Now, what is
relevant for us is that in this analysis the subject of the initial conjunct is within
the vP. This must be possible since there is no obligatory raising of the subject to
SpecTP. Notice that the subject in (30a) is a Categorical Subject. This means that
Categorical/Major Subjects and Thetic/grammatical subjects are both contained
within the vP.
To summarize, we have argued in this section that the restrictions on
embedded predicates when an embedded subject undergoes SOR imply
that embedded subject raising is also the raising of the Major Subject. We
argued that even in clauses with one subject constituent, there can be two
distinct subject positions – the higher Major Subject position and the lower
grammatical subject position. We interpreted the restriction against raising
embedded Thetic Subjects in SOR to mean that the Categorical Sub-
ject occupies a position higher than a Thetic Subject. To account for why
Thetic Subjects cannot undergo SOR, we adopted the proposal that both
categorical and thetic sentences possess two subject positions. That is, both
types of sentences have Major Subjects. The only difference is in what the
Major Subject is. In categorical sentences, it is the DP on which the Sentential
Predicate is predicating a property. In thetic sentences, it is a spatio-temporal
event argument which cannot undergo SOR but can nevertheless spoil raising
of the lower subject in virtue of its position.
In the next section, we show that the hypothesis that the embedded Major Sub-
ject raises in SOR allows us to account for another category of apparent problems
for the raising analysis of SOR in Korean (and Japanese).
3.3 Locality violations
The lack of locality between the raised nominal and the gap/pronoun in the
embedded clause in SOR (31) can also be accounted for under the hypothesis that
what raises in SOR is the embedded Major Subject.
(31) a. Na-nun Yenghi-luli [[ ei/kunye-ka e ha-nun] il]-i (=5)

I-TOP Y-ACC she-NOM do-ADNOM work-NOM
mopemcek-ila-ko sayngkakhanta
exemplary-COP-COMP think
‘I think of Yenghi that the things she does are exemplary.’
b. Na-nun Cheli-lul hangsang kunyesek-i taytanhan malssengkkwuleki-la-ko (=6)

I-TOP C-ACC always the.guy-NOM extreme troublemaker-COP-COMP
sayngkakhayssessta
had.thought
‘I had always considered Cheli to be a real troublemaker.’
We cannot do justice to the vast amount of literature on the structure and derivation
of MNCs in Korean (and Japanese). Among many debated issues is the question
of whether Major Subjects are base-generated or derived by movement. We believe
that there are good reasons to believe that they are base-generated (Heycock 1993;
Doron and Heycock 1999; Yoon 1987, 2004a). The most straightforward evidence
for base-generation comes from MNCs where the initial NP is co-indexed with
a gap or a resumptive pronoun within an island, or where the Major Subject
is not co-indexed with any constituent within the Sentential Predicate, as we see
below.18
(32) a. Yenghi-ka [[ e/kunye-uy apeci-ka ha-si-nun] sa.ep]-i

Y-NOM she-GEN father do-HON-ADNOM business-NOM
manghay-ss-ta
go.bankrupt-DECL
‘As for/it is Yenghi (that) the business her father was running went bankrupt.’
b. Enehak-i chwuycik-i yocum elyep-ta

linguistics-NOM employment-NOM these.days difficult-DECL
‘As for/it is in linguistics (that) getting a job is difficult these days.’
If we hypothesize that what raises in SOR is the Major Subject, then we have
a natural explanation for why the relation between the raised nominal and the
embedded gap/pronoun can violate Subjacency. This is a property attributable to
the MNC construction that underlies SOR. It is the co-indexation between the
Major Subject and the constituent internal to the Sentential Predicate that is non-
local. Movement of the Major Subject in SOR abides by Subjacency.19
3.4 Interpretive properties of the raised nominal
In section 1 we noted that a raised nominal differs from its unraised counterpart in
terms of a number of interpretive properties. We suggested that this is potentially
a problem for the raising analysis of SOR (see Davies 2005 for a similar conclusion
with respect to Madurese). However, if what we have argued thus far is correct, the
wrong comparison was being made. The correct pairing of raised vs. non-raised
90 JAMES H. YOON
structures should compare the properties o f the raised nominal with those of the
unraised embedded Major (Categorical) Subject. Once we do so, we see that a
number of interpretive properties of the raised nominal that appear to militate
against the raising analysis will be shown to be those that the unraised Major
Subject has independently of raising. This line of explanation extends to most of
the interpretive properties noted earlier. It does not, however, account for all of the
discrepancies in interpretation, because unraised and raised Major (Categorical)
Subjects can differ interpretively. We attribute the remaining differences to the
effect that movement has on interpretive properties, in particular, the ability to
reconstruct (lower) in movement chains.
3.4.1 Lack of idiomatic reading
We noted earlier that raising an idiom chunk (subject) leads to the loss of the
idiomatic reading and pointed out that this is a potential problem for the movement
analysis of raising. Once we reinterpret SOR as raising of the embedded subject,
however, this turns out not to be a problem anymore.
The fact that idiomatic readings disappear can be attributed to the Major Subject
status of the raised nominal. A Major (Categorical) Subject is the subject of a
Sentential Predicate. As such, it does not make sense to say something about
(attribute some property to) a Major Subject that fails to denote or otherwise sets
conditions on reference, such as an idiom chunk. It is not an accidental fact that
most, if not all, idioms that have sentential form (Sentential Idioms) are used to
describe or comment on a situation or an event, rather than predicating a property
of an individual. Another way of putting it is to think of sentential idioms as
thetic sentences predicated of an event argument. Since SOR is restricted to
Categorical Subjects, we can understand why idiom chunks that are part of sentential
idioms will fail to undergo raising.
3.4.2 Other interpretive differences attributable to major subject raising
In addition to differences in idiomatic interpretation, we noted the following

interpretive differences between raised and unraised nominals:
(33) a. Raised bare plural subjects are interpreted generically (23, 25).
b. Raised nominals do not reconstruct into the Sentential Predicate for scope (11).
c. Raised nominals do not reconstruct into the Sentential Predicate for variable
binding (12).
d. Raised indefinites prefer to be interpreted specifically and as presupposed in SOR

contexts (8).
e. Raised nominals are interpreted de re in SOR contexts (9, 10)

The first three properties are directly attributable to the fact that what undergoes
raising in SOR is an embedded Major (Categorical) Subject. We already
discussed the generic reading of raised bare plural subjects. In conjunction
with the evidence coming from the restriction on the embedded predicates, we
took this to be evidence that what undergoes raising is an embedded Major
Subject, which, qua a Categorical Subject, is interpreted generically when it
is a bare plural. Existential readings are possible only for the unraised Thetic
Subject, which cannot undergo SOR in Korean for reasons discussed in the
previous section.
The second and third properties are also attributable to the Major (Categorical)
Subject status of the raised nominal. Major Subjects always take wider scope
than constituents internal to the Sentential Predicate. The reason for this is that a
Major Subject is directly merged into its position rather than derived by movement
from within the Sentential Predicate and because reconstruction is contingent
on Chains. The wide scope of the Major Subject over constituents within the
Sentential Predicate is illustrated below:20
(34) a. Sey-myeng-uy haksayng-i(MS) pwumo-ka(GS) enu kyoswu-eykey-na

three-CL-GEN student-NOM parents-NOM which professor-DAT-ever
sokay-toy-ess-ta
introduce-PASS-PST-DECL
→ three > every, *every > three
‘Three specific students’ parents were introduced to every professor.’
b. Sey-myeng-uy haksayng-uy pwumo-ka(GS) enu kyoswu-eykey-na

three.CL-GEN student-GEN parents-NOM which professor-DAT-ever
sokay-toy-ess-ta
introduce-PASS-PST-DECL
→ three > every, every > three
‘Three specific students’ parents were introduced to every professor.’
‘To each professor, three students’ parents were introduced.’
Since scope tracks surface hierarchy in sentences in active voice, (34) has the
predicate in the passive voice. The Major Subject, even though related to the
grammatical subject, does not scope below it in (34a). By contrast, in (34b), with
only a grammatical subject, sey-myeng-uy haksayng (‘three students’) can scope
out of the grammatical subject, taking wider scope relative to it.
Major Subjects do not reconstruct into Sentential Predicates for variable
binding, again because they are merged directly in their surface position. The
sentence in (35a), though less than perfect, allows a bound reading of caki.
This reading is much more difficult in (35b), where caki is contained within a
Major Subject.
(35) a. ?Caki sensayng-uy chwuchense-ka citohaksayngtul-eykey kakkak

self teacher-GEN letter-NOM advisees-DAT each
kongkay-toy-eyahay-ss-ta
release-PASS-must-PST-DECL
‘Their teachers’ reference letters had to be released to each student.’
92 JAMES H. YOON
b. *Caki sensayng-uy chwunchense-ka(MS) wenpon-i(GS) citohaksayngtul-eykey

self teacher-GEN letter-NOM original-NOM advisees-DAT
kakkak kongkay-toy-eyhay-ss-ta
each release-PASS-must-PST-DECL
‘The originals of their teacher’s reference letters had to be released to each advisee.’
The difficulty of raised nominals to reconstruct for bound variable readings is

directly of a piece with this fact.
The fourth property, the preference for a raised indefinite to be interpreted
as specific, can also blamed on Major Subject raising (cf. J-M Yoon 1989). The
following examples show that an indefinite Major Subject is interpreted strongly
as specific:21
(36) a. Etten haksayng-i(MS) apeci-ka(GS) hakkyo-ey cacwu osin-ta

certain student-NOM father-NOM school-to often comes-DECL
→ specific/*?nonspecific
‘Some (specific) student’s father often comes to school.’
b. Etten haksayng-uy apeci-ka(GS) hakkyo-ey osi-ess-ta

certain student-NOM father-NOM school-LOC come-PST-DECL
→ specific/nonspecific
‘Some student’s father came to school.’
Similarly, in contexts of amount quantification, a Major Subject is preferentially

interpreted in the presuppositional (partitive) reading:
(37) a. Myes-myeng-uy haksayng-i(MS) pwumo-ka(GS) chotaytoyessni?

how.many-CL-GEN student-NOM parents-NOM were.invited
→ presuppositional
‘(Among the students) how many of them have rich parents?’
b. Myes-myeng-uy haksayng-uy pwumo-ka(GS) chotaytoyessni?

how.many-CL-GEN students-NOM parents-NOM were.invited
→ cardinal/presuppositional
‘How many of the students/how many students had their parents invited to the
event?’
We come to the final property – the availability of de re readings. This is not a

property that can be blamed on the Major Subject status of the raised nominal, as
the following sentence shows:
(38) John-un caki anay-ka(MS) moksoli-ka(GS) cohta-ko sayngkakhayssta

J-TOP self wife-NOM voice-NOM good-COMP thought
‘John thought that his wife’s voice was sweet.’
Suppose John’s wife is on the radio being interviewed. John listens to the interview
in his car on his way to work but does not realize it is his wife. However, he does note
the familiar melodious tone and thinks that the voice is sweet. The above sentence
is not a very felicitous description of John’s thought. This means that the Major
(Categorical) Subject caki anay-ka (self’s wife-nom) is not easily interpreted de re

in the above sentence. Hence, the availability of de re interpretation with raised
nominals is not attributable to the Major Subject status of the raised nominal.
We propose that de re readings on raised nominals is a consequence of
raising/movement instead. May (1977) noted that A-movement can facilitate
de re interpretations:
(39) a. It is likely for a unicorn to be discovered
b. A unicorn is likely [t to be discovered]
While (41a) has primarily the de dicto interpretation (where the speaker is not
committed to believing in the existence of unicorns), a de re reading is salient in
(41b) (i.e. there exists a unicorn which is likely to be discovered).22
Modeled on the scope altering property of A-movement, we can account
for the availability of de re readings on raised nominals in Korean SOR. SOR
places the raised embedded Major Subject in a position where it can scope over
the matrix intensional verb (SpecVP or a higher derived object position). This
is why SOR allows de re readings, compared to structures where SOR does not
take place in which the matrix intensional predicate unambiguously scopes over
the embedded Major Subject.
To summarize, we have shown in this section that the interpretive differences
that seem to jeopardize the raising analysis of SOR receive an explanatory account
under the hypothesis that the embedded Major Subject undergoes raising. All but
one of the properties is directly attributable to the embedded Major Subject status
of the raised nominal. And the final interpretive property, while not predicated on
the Major Subject status of the raised nominal, does not endanger it either.
In the next section, we turn to the discussion of an alternative analysis of SOR – the
Proleptic, or Major Object, Analysis where the Korean (Japanese) SOR is not taken
to involve raising but base-generation of the ‘raised’ nominal in the upstairs clause
from which it is co-indexed with a variable in the embedded clause. We will argue
that while the analysis can explain a number of properties of SOR in K/J we have
discussed thus far, it suffers from theoretical and empirical drawbacks.
4. THE PROLEPTIC OBJECT (MAJOR OBJECT) ANALYSIS OF SOR
As noted in section 1, a number of researchers have responded to the differences

between canonical SOR constructions in languages like English and those in Korean
and Japanese by pursuing an alternative analysis where the SOR construction in
the latter languages does not involve raising of an embedded subject but instead a
base-generated nominal in the upstairs clause that is linked to a gap or a pronoun
in the embedded clause. A hallmark of this line of inquiry is the following: (1a)
and (1b) are not derivationally related.
As is easy to see, base-generation/co-indexation analyses can deal with the
following problems facing the raising/movement analysis:
94 JAMES H. YOON
(i) Violations of movement locality between the ‘raised’ nominal and the
embedded clause gap/pronoun
(ii) The possibility of resumptive pronouns in the embedded clause linked to
the raised nominal
(iii) Wide scope of the raised object over embedded clause constituents
(iv) Nonsubject raising
(v) Failure of idiom interpretation
(vi) Additional interpretive differences
(i) is expected since the connection between the raised nominal and the embedded
clause gap/pronoun is not one of movement. (ii) is attributable to this factor as well.
(iii) is a consequence of base-generation. Since the nominal is never in the lower clause
at any point in the derivation, it is expected to scope wider than embedded clause con-
stituents and not reconstruct into the embedded clause. (iv) is expected since there is
no movement link between the embedded constituent and the matrix nominal. (v) is
expected since the idiom chunk will not form a constituent with the embedded clause.
(vi), while not directly due to base-generation, is something that can be expected given
the different base positions of the raised nominal and the embedded subject.
Though it seems at first glance that base-generation analyses are equal in cover-
age to the Major Subject raising account we defended earlier, once we try to flesh
out the details of the analyses, we see that there are nontrivial difficulties. There are
different versions of the base-generation account. The first option is to view Korean
(Japanese) SOR to be equivalent to object control. The second is to take it to be a
type of raising construction that involves base-generated Chains, i.e. copy raising
(Moore 1998; Potsdam and Runner 2001). The third and final alternative is to take
SOR to instantiate what has sometimes been called Prolepsis (Takano 2003; Davies
2005). The line of analysis that Hoji (1991, 2005) calls the Major Object analysis
seems closest to the third. I shall therefore class it as a proleptic object analysis.
The third analysis is the most commonly proposed base-generation alternative to
raising accounts of SOR. Therefore, I shall choose this account and compare it with
the Major Subject raising account defended earlier. Proponents of this type of analy-
sis include Hoji (1991, 2005) and Takano (2003) for Japanese, and K.-S. Hong (1990,
1997), P.-Y. Lee (1992), and J.-G. Song (1994) for Korean. We will argue below that the
class of analyses faces nontrivial difficulties. First of all, theoretical problems crop up
concerning how the Proleptic (Major) Object gets its theta-role. Secondly, empirical
problems arise from what I call the ‘persistence of low properties’. The raised nominal
displays certain properties that could only have been determined in the lower clause.
The Proleptic (Major) Object analysis has no way of coping with such facts.23
4.1 Theta-role of the Proleptic (Major) Object
One salient difficulty with the Proleptic (Major) Object analysis (abbreviated
MOB henceforth, following Hoji 1991, 2005) has to do with the question of how
the base-generated MOB gets its theta-role. As far as I can tell, the following
options suggest themselves, since SOR is clearly distinct from object control where
a matrix object receives a theta-role from the matrix predicate.
Thematic licensing of MOB:

a. MOB obtains its theta-role through Chain Formation, as in copy raising
b. MOB is an adjunct and does not receive a theta-role from the matrix verb
c. CP becomes predicate and assigns a theta-role to MOB (perhaps jointly
with matrix V/V’– see below).
There may be other options, such as the possibility of a constructionally deter-
mined theta-role.24 I will not consider this alternative, as it is essentially nonex-
planatory. Of the options specified above, (a) and (c) take the MOB to be an
argument occupying an A-position, while (b) takes it to be an adjunct.
The first option is unworkable since the connection between the MOB and the
embedded constituent nonlocal, whereas in copy raising it is local (Moore 1998;
Davies 2005).25 In addition, the SOR construction in Korean/Japanese does not
possess other properties of CR constructions noted in Davies 2005.26
The second analysis bypasses the problem of theta-roles altogether, by assuming
that the MOB is an adjunct. Since (at least in Korean) nominal adverbs can be case-
marked (Kim and Maling 1993, and others), there is no prima facie difficulty for this
analysis. However, the analysis is falsified (at least for Korean) by the fact that unlike
adjuncts, the MOB undergoes Passive, a behavior restricted to arguments.
(40) a. Ceketo John-un Mary-lul minye-lako sayngkakhay-ss-ess-ta

at.least J-TOP M-ACC beauty-COP.COMP think-PERF-PST-DECL
‘At least John believed Mary to be a beauty.’
b. Maryi-ka ceketo John-eykey-nun ei minye-lako sayngkak-toy-ess-ess-ta

M-NOM at.least J-by-TOP beauty-COP.COMP think-PASS-PERF-PST-DECL
‘Mary was believed to be a beauty by John at least.’
Adjuncts do not undergo Passive even when they are nominal, as is well known:
(41) a. John lectured all day yesterday
b. *All day yesterday was lectured by John
Neither does the P(roleptic) Object in English. Passive of (44a) is bad, with or
without the preposition accompanying the P-Object.
(42) a. John believed about/regarding Mary that she was a genius
b. *(Regarding/about) Mary was believed (by John) that she was a genius
In this respect, Korean is like Madurese where the P-Object also undergoes a
passive-like alternation (Davies 2005).27
(43) a. Hasan e-kera Siti bari’ [pro melle motor]

H Ov-think S yesterday he Av.bought car
‘Hasan was thought by Siti to have bought a car yesterday.’
96 JAMES H. YOON
b. Siti ngera Hasan bari’ [pro melle motor]

S Av.think H yesterday he Av.bought car
‘Siti thought that Hasan bought a car yesterday.’
The third and final analysis is closest to the suggestion of J.-G. Song (1994) and
O’Grady (1991), though neither develops the analysis in full detail. And, as the
following quote shows, Hoji (2005) seems to be adopting a similar analysis.
‘What thematic role, if any, does the MOB receive, and how?’….We might
assume that due to the formal (though presumably not structural) property
of NP-o in the ‘construction’ in question …, the grammar gives the following
instructions to the language user:
a. NP-o in the ‘construction’ in question, i.e. MOB, denotes some entity about
which one can hold some belief/assumption/judgment, etc. (depending
upon the predicate used).
b. (What appears to be) the CP complement of the verb in the ‘construction’
in question denotes a property that can be attributed to some entity, rea-
sonably and meaningfully.’ (Hoji 2005:18–19)
The key idea seems to be that the CP complement in the MOB construction is prop-
erty-denoting. In other words, it is a Sentential Predicate. Let’s indicate the predicate
status of the complement clause(= ZP) by a null Op(erator) binding a variable (indi-
cated by ei). We assume crucially that the abstraction in question is distinct from con-
trol and can be nonlocal. In addition, it is not accomplished by movement either.
The proposal that the embedded clause in SOR is property-denoting is
something we have argued earlier. Therefore, this version of the MOB analysis is
indistinguishable from the Major Subject raising analysis in this regard (as Hoji
points out, personal communication). The difference between the two analyses
resides in how the Sentential Predicate assigns its theta-role. In the MOB analysis
as we construe it here, the Sentential Predicate assigns a theta-role directly
(perhaps jointly with V) to the MOB base-generated in the matrix clause. In the
Major Subject analysis we defended earlier, the Sentential Predicate assigns its
theta-role to the Major Subject, which then undergoes raising to the matrix VP.
The two analyses are illustrated below, where the solid arrow indicates theta-role
assignment by the Sentential Predicate and the dotted arrow, movement.28
(44) a. MOB Analysis: Major Subject Raising Analysis:
VP VP
?? V’ DP V’
DP
ZP V [e] XP V
Opi Z’ ZP
DP
Opi Z’
ei
ei
The crucial differences between the two analyses boil down to the answers to the
following questions:
a. Is the raised nominal related to a Subject(-like) position in the embedded
clause?
b. Is the raised nominal a constituent of the embedded clause at any stage
of the derivation? That is, is there evidence for a movement connection
between the matrix DP and the embedded subject position?
The arguments in section 3 pointed to the necessity to implicate a Subject(-like) posi-
tion in all felicitous instances of SOR – the embedded Major (Categorical) Subject
position. Since the MOB analysis does not implicate such a position, it is difficult
to see how the range of facts considered in section 3 could be accommodated in this
approach. The only option open under the MOB analysis is to impose the ‘character-
istic property’ condition on the Sentential Predicate that gives the MOB its theta-role,
perhaps as a constructional property (see quote from Hoji above). Though such a
move begs the question of why the ‘characteristic property’ condition holds for the
MOB construction and MNCs but not other constructions involving Sentential Pred-
icates (such as relative clauses), once the condition is adopted, there is a way for the
MOB analysis to incorporate many of the restrictions on SOR we blamed on Major
Subject raising. Thus, the relevance of the ‘characteristic property’ condition does not
by itself necessitate the presence of a Major Subject position.
In the following sections, we provide evidence that shows directly that there
is an embedded Major Subject position which is connected to the matrix DP via
movement. Unlike the interpretive restriction on embedded predicates, this cat-
egory of evidence is difficult, if not altogether impossible, to incorporate into the
MOB analysis, which posits that the MOB is base-generated in the upstairs VP in
construction with a property-denoting embedded CP.
4.2 Proper binding condition
4.2.1 PBC in SOR vs. Object Control
The argument that SOR involves an embedded Major Subject and that the rela-
tion between the Major Subject and the matrix DP is one of movement comes
from facts pertaining to the Proper Binding Condition (PBC). That a trace but
not PRO induces PBC effects is well known, and is supported by the following
contrast:
(45) a. *It is [ti to be intelligent] that Billi seems (to all of us)
b. ?It is [PROi to succeed in business] that Billi (badly) wants
Capitalizing on this difference, Tanaka (2002), following Kuno 1976 (see also
Y.-H. Kim 1985 for Korean SOR), argues that the contrast between preposing the
complement clause in an SOR construction and in an object control construction
is due to the nature of the empty category – it is a trace of the subject in SOR
98 JAMES H. YOON
while in object control, it is a PRO/pro co-indexed with the matrix subject.

A trace is subject to the PBC, and that is why there is a contrast between SOR
and control.
(46) a. ??[t acwu ikicek- ila]-ko na-nun Cheli-lul sayngkakhanta

very selfish-COP-COMP I-TOP C-ACC think
‘I consider Cheli very selfish.’
b. [pro cip-ey kala]-ko na-nun Cheli-eykey/lul seltukhayssta

home-LOC go-COMP I-TOP C-DAT/ACC persuaded
‘I persuaded Cheli to go home.’
This argument is predicated on the existence of a subject position in the lower

clause, a possibility that is denied in the MOB analysis. To the extent that it is
successful, we have evidence for the existence of a low subject position and a
movement link between the subject and the raised nominal.
Nonetheless, Hoji (2005) has recently questioned the validity of the PBC argu-
ment for the raising analysis of SOR. On the basis of acceptability ratings given
by speakers, he claims that the prediction that PBC violations of SOR should be
unacceptable is not supported. The reason is that speakers judge PBC-violating
sentences with SOR as more acceptable than PBC violations with Scrambling but
not SOR. PBC violations in SOR (48a) and in Scrambling (47a) are given below
(Japanese data, from Hoji 2005).
(47) a. *[John-ga ei nigeta-to]j tyuugoku-nii keisatu-wa ej danteisita

J-NOM escaped-COMP China-to police-TOP concluded
‘The police concluded that John had escaped to China.’
cf.
b. Keisatu-wa [John-ga tyuugoku-ni nigeta-to] danteisita

police-TOP J-NOM China-to escaped-COMP concluded
(48) a. ??[ei tyuugoku-ni nigeta-to]j John-oi keisatu-waej danteisita

China-to escaped-COMP J-ACC police-TOP concluded
‘The police concluded that John had escaped to China.’
cf.
b. Keisatu-wa John-o tyuugoku-ni nigeta-to danteisita

police-TOP J-ACC China-to escaped-COMP concluded
Based on an average of raw scores across different subjects (ranging from

+2 to −2), Hoji concludes that the prediction that PBC violations with SOR
should be ungrammatical is falsified (average +0.28 for 48a), while that for PBC
violations with simple Scrambling is supported (average −1.56 for 47a).
Though Hoji is to be commended for seeking experimental confirmation of
intuition-based grammaticality judgments, there are quite a few shortcomings in
his analysis. First, the results of simple magnitude estimation are hardly telling
without further statistical analysis. Second, there is a principled reason why
speakers give a more generous rating to (48a) compared to (47a).
Many of the matrix verbs that govern SOR have a usage where they take the
Acc-marked DP as an argument. The argument can be marked by an adposition, as
in English, but unlike English, Acc-marking is also possible. This is shown below.
(49) a. Ne-nun Cheli-lul/Cheli-eytayhay ettehkey sayngkakha-ni?

you-TOP C-ACC/C-regarding how think-Q
‘What do you think about/of Cheli?’
b. Na-nun Cheli-lul/Cheli-eytayhay kunyesek-i tollassta-ko sayngkakhay

I-TOP C-ACC/C-regarding that.guy-NOM crazy-COMP think
‘I think Cheli is crazy.’
When such structures combine with ko-marked clauses, as in (49b), a surface

string that is identical to that derived by SOR can arise. I submit that this
is a genuine Prolepsis (Major Object) structure, and this is what underlies the
ameliorated judgments that speakers are giving for the PBC cases in (48a).
Does this mean that there is no way to discriminate between SOR and Prolepsis
parses in all cases? Fortunately, the answer is no. There are sentences that could
only have been derived by SOR. When such structures are examined, PBC effects
surface robustly, as expected.
The Prolepsis parse of SOR structures is possible only when the Proleptic
argument is a DP (and animate, preferably). Now, as we shall see in the next
section, embedded non-nominative subjects can undergo SOR, showing up
with Acc-case doubling the dative or locative assigned in the lower clause (50b).
When PBC violations are induced on such structures, the results are uniformly
bad (50a). Speakers I have consulted all agree that (50a) is much worse than (the
Korean equivalent of) (48a). The reason is that being an inanimate and a PP,
yeki-pwuthe-lul in (50a) does not stand a chance of being parsed as a proleptic
object.
(50) a. *[ei nay ttang-ila-ko]j yeki-pwuthe-luli na-nun ej mitnunta

my land-COP-COMP here-from-ACC I-TOP believe
‘I believe my land begins from here.’
b. Na-nun yeki-pwuthe-luli [ei nay ttang-ila-ko] mitnunta

I-TOP land-from-ACC my land-COP-COMP believe
‘I believe my land begins from here.’
The contrast between SOR (48a) and Scrambling (47a) can be explained along
the same lines. Note that the constituent that is responsible for the PBC viola-
tion in (49a), tyuugoku-ni, is an inanimate PP, while that in (48a), John-o, is an
animate DP. Since proleptic objects must be DPs (that are preferably animate),
it is not surprising that speakers perceive a difference between the two. Speakers
are able to parse (48a) as a structure with a proleptic object, while that parse is
unavailable for (47a). It is this difference that is responsible for the discrepancy
in judgments.
In sum, while applauding efforts to ground intuition-based claims of accept-
ability on a firmer footing, I do not believe that the results of Hoji 2005 show that
100 JAMES H. YOON
PBC violations of SOR are grammatical. At most, they indicate that some
SOR sentences may optionally have a parse as Prolepsis constructions. It is
this factor that interferes in the judgments of speakers. When this is controlled
for, PBC violations of SOR are robustly judged as ungrammatical.
4.2.2 PBC with Major Subjects
Tanaka (2002), like others who assume a raising analysis of SOR, posited that
the raised object moves directly from within the embedded clause in SOR.29 If
his analysis is on the right track, we expect PBC effects to disappear if the gap
position in the embedded clause is filled by a pronoun or if the embedded clause
does not contain a gap (when the Major Subject of a non-gap/adjunct-type MNC
is raised, as in 51b). This is because there would be no trace within the embedded
clause that could incur a violation of PBC in such instances. However, this predic-
tion is not borne out. Preposing the complement clause results in ill-formedness
even when the complement clause contains a resumptive pronoun or appears to
lack a gap altogether.30
(51) a. *?[ku-uy apeci-ka pwuca-la-ko] na-nun Cheli-lul sayngkakhanta

he-GEN father-NOM rich-COP-COMP I-TOP C-ACC think
‘I think that Cheli’s father is very rich.’
cf.
Na-nun Cheli-lul ku-uy apeci-ka pwuca-la-ko sayngkakhanta
I-TOP C-ACC he-GEN father-NOM rich-COP-COMP think
b. *[maykcwu-ka choyko-la-ko] na-nun yelum-ul sayngkakhanta

beer-NOM best-COP-COMP I-TOP summer-ACC think
‘I consider summer to be the best time for (a cold) beer.’
cf.
Na-nun yelum-ul maykcwu-ka choyko-la-ko sayngkakhanta
I-TOP summer-ACC beer-NOM best-COP-COMP think
Why should this be so? Our analysis provides a straightforward answer. Since we
assume that SOR is derived by movement of the Major Subject, the fronted clause
will contain the trace of the Major Subject – regardless of whether the gap position
internal to the Sentential Predicate is filled by a pronoun, or whether the Sentential
Predicate lacks a gap altogether. It is the trace of the raised Major Subject within
the preposed constituent that incurs a PBC violation. The analysis of (51a) and
(51b) incorporating the trace of the embedded Major Subject is shown below.31
(52) a. *?[ti [kui-uy apeci-ka pwuca-la-ko]] na-nun Cheli-luli sayngkakhanta

he-GEN father-NOM rich-COP-COMP I-TOP C-ACC think
‘I think that Cheli’s father is very rich.’
b. *[ti [maykcwu-ka choyko-la-ko]] na-nun yelum-uli sayngkakhanta

beer-NOM best-COP-COMP I-TOP summer-ACC think
‘I consider summer to be the best time for (a cold) beer.’
In sum, the robustness of PBC effects in all genuine instances of SOR constitute
strong evidence that (i) the embedded Major Subject position exists in SOR and
(ii) the connection between the embedded Major Subject and the SOR nominal is
one of movement, rather than co-indexation. It is not clear how the Major Object
analysis can handle this, especially when we consider that genuine examples of
Prolepsis are not sensitive to the PBC.
4.3 Persistence of low properties
Recall that the MOB analysis does not assume that the raised nominal in SOR
is a constituent of the embedded clause at any stage of the derivation. How-
ever, the following suggests otherwise. While most cases of SOR involve a case
alternation between nominative and accusative, a raised nominal in SOR can
sometimes retain the case assigned in the embedded clause. This happens when
a non-nominative subject of the embedded clause is raised and accusative case
stacks on top of the inherent case assigned in the embedded clause. In such
cases, we assume that what undergoes raising is the Nom-stacked subject of
the embedded clause. A Nom-stacked subject has been argued by Yoon (2004b)
to be a Major Subject, and hence, available for raising in SOR. The analysis is
shown below schematically:
(53) a. Na-nunCheli-hantheyi-(man)-ul [ ti(MS) [mwuncey-ka issta-ko]] mitnunta

I-TOP C-DAT-(only)-ACC problem-NOM exist-COMP think
‘I think that only Cheli has problems.’
a’. Cheli-hanthey-(man)-i(MS) mwuncey-ka issta

C-DAT-(only)-NOM problem-NOM exist
‘Only Cheli has problems.’
b. Na-nun yeki-pwuthei-lul [ti(MS) [nay ttang-ila-ko]] sayngkakhanta

I-TOP here-from-ACC my land-COP-COMP think
‘I consider from about here to be my property.’
b’. Yeki-pwuthe-ka(MS) nay ttang-ita

here-from-NOM my land-COP
‘From about here is my property.’
Now, since the matrix predicates cannot take Dat-marked complements, the only
source of dative (and locative) on the raised nominal is the embedded clause.
The MOB analysis has no way of dealing with this paradigm. This is because
nonmovement co-indexing does not involve case transmission.32 This is the first
example of the ‘persistence of low properties’ in SOR.
A second example of the persistence of low properties can be found in the de
dicto/de re readings. Recall that while many interpretive properties of the raised
nominal in SOR are attributable to its Major Subject status, the preference for
de re reading was argued not to be reducible to this factor. We argued instead
that de re readings in SOR arise as a consequence of movement.
102 JAMES H. YOON
Now, what is interesting is that while de re readings are salient in SOR contexts,
de dicto readings are by no means ruled out. They are just less salient. By contrast,
when SOR does not take place, de re readings are not possible. The relevant exam-
ples are repeated below:
(54) a. John-un caki anay-lul totwuk-ila-ko sayngkakhayssta → de re > de dicto

J-TOP self wife-ACC thief-COP-COMP thought
‘John thought his wife was a thief.’
a’. John-un caki anay-ka totwuk-ila-ko sayngkakhayssta → de dicto, *de re

J-TOP self wife-NOM thief-COP-COMP thought
b. John-un Cheli-lul Tongswu-la-ko sayngkakhayssta → de re > de dicto

J-TOP C-ACC T-COP-COMP thought
‘John thought Cheli was Tongswu.’
b’. John-un Cheli-ka Tongswu-la-ko sayngkakhayssta → de dicto, *de re

J-TOP C-NOM T-COP-COMP thought
In contexts of control, where the controller nominal is not a constituent of the

embedded clause at any stage of the derivation, de dicto readings are not possible,
whereas in contexts of movement, the reading is possible, though less salient. This
is shown below:33
(55) a. A unicorn is likely [t to be discovered] → de re > de dicto
b. A unicorn is anxious[PRO to be discovered] → de re, *de dicto
Now, if the raised nominal in SOR is base-generated in the matrix VP as a

MOB, we predict that sentences with raising should only have the de re read-
ing (matrix VP reading), since the de dicto reading (embedded clause reading)
arises when the DP scopes below the matrix verb, but there is no stage in the
derivation of these sentences where the MOB DP is in such a position. This
prediction is not supported, as we have seen. We thus take the availability of
de dicto readings in SOR to be due to the low origin of the SOR nominal, as
argued in this paper.
In sum, a key prediction of the MOB analysis – that the raised nominal
should not evince any signs of having been derived from a lower position
– is falsified. The persistence of ‘low’ properties argues strongly that the
raised nominal is a dependent of the embedded clause at some point in the
derivation.
5. CONCLUSIONS
In this paper we have argued that a number of properties that look problematic
for a raising analysis of SOR in Korean and Japanese can be accounted for under
the assumption that what undergoes raising is a high subject of the embedded
clause – the Major (Categorical) Subject. We have compared the Major Subject
raising analysis with a base-generation analysis and demonstrated that while the
two are largely comparable in empirical coverage, the latter faces empirical and
theoretical difficulties.
There are some broader implications of our analysis. One consequence of the
analysis has to do with locality constraints on A-movement. While SOR in K/J
seems to flout known constraints on A-movement, in particular, SSC/Relativ-
ized Minimality, we have shown that a closer investigation of the relevant facts
allows us to maintain SSC as a constraint on A-movement. This is because there
was no genuine raising of embedded constituents over a subject. In the absence
of comparable investigations in other languages, we are not sure whether A-
movement can never escape the domain of a subject in any language. However,
at least one language that was considered to be an apparent counterexample (in
J.-M. Yoon 1991, for example) is no longer a problem.
A second implication of our analysis is that SOR may indeed implicate
a higher subject position. This idea has been pursued in different forms by
a number of researchers who have investigated noncanonical instances of
SOR (Massam 1985; Deprez 1992; Ura 2000; Bruening 2001, and others).
Many such proposals utilize the higher position as an escape hatch for move-
ment (Ura 2000), or posit mandatory raising from the higher subject position
(Bruening 2001), with the result that the structure posited as the underlying
source of SOR with the high subject position never surfaces overtly. We have
rectified this deficiency in this paper. We have shown that in Korean and Japa-
nese, such clauses do occur as complements of SOR verbs overtly, without
SOR, because SOR is optional in Korean and Japanese.34
Interestingly, while our analysis offers concrete proof that a higher subject posi-
tion can be implicated in SOR in some languages, it does not support the view that
a higher subject position functions as an escape hatch. The connection between the
higher subject position and a constituent within the embedded clause that is co-
indexed with it was shown to be not one of movement. That is, there is no Chain
linking the raised subject, the embedded high subject and the constituent internal to
the embedded clause that is related to the high subject position. This is unexpected
on an account like Ura’s (2000) that posits multiple specifiers. It remains to be seen
how general this result is and whether other languages that have been reported to
allow raising of embedded nonsubjects in SOR can be similarly reanalyzed.
* The research reported in this chapter has been presented in various incarnations at the University
of Illinois, Urbana-Champaign (2003), Indiana University (2003), Seoul National University (2004),
the summer international conferences of the Linguistic Association of Korea and the Modern
Linguistic Society of Korea (2004), the Workshop on Japanese and Korean Linguistics at Kyoto Uni-
versity (2005), and at the LSA workshop ‘New Horizons on the Grammar of Raising and Control’
(2005). A preliminary version of the work was also presented at the Workshop on Formal Altaic
Linguistics (2003) and appears in the proceedings. I would like to thank the audiences at these venues
for their critical feedback. Special thanks go to Cedric Boeckx, Youngju Choi, Hajime Hoji, Kisun
Hong, JuHyeon Hwang, Ji-Hye Kim, Soowon Kim, Yoshihisa Kitagawa, Peter Sells, Keun Young
Shin, Yukinori Takubo, and Yunchul Yoo whose input and criticism shaped and guided the paper. Bill
Davies and Stan Dubinsky deserve special thanks for putting together a timely project on a topic of
perennial interest to linguists as well as for their editorial comments and feedback.
104 JAMES H. YOON
1
The assumption that English SOR is canonical is by no means uncontroversial, as it may well
be an accident of the history of inquiry into the construction. However, this position is widely
accepted.
We will be using the term ‘SOR’ to designate the construction in question, without necessarily com-
mitting to a movement/raising analysis. What is at stake in this paper is the choice between SOR/
ECM/Long-distance Agree on the one hand, which all posit that the Acc-marked nominal is the-
matically (though not formally) licensed in the embedded clause, and base-generation analyses on the
other, which posit that the Acc-marked nominal bears no thematic relation to the embedded clause.
2
Ura (2000) attempts to allow nonsubject raising through the proposal of multiple specifiers, where
languages with an extra specifier of TP allow nonsubjects to utilize the position as an escape hatch.
In that it implicates a higher, subject-like position, the analysis is similar to the proposal in this
paper. However, we do not take the higher subject to be an escape hatch for movement, as will
become clear.
3
Bruening (2001) claims that for his Japanese informants, idiomatic readings survive under SOR.
However, the idioms that he employed are not truly opaque in that the subject position of the
sentential idioms contains a non-idiomatic slot. For example, the subject of the idiom X-no kao-
ga hiroi (‘X’s face is wide’) has a non-idiomatic slot. Besides, the judgment on whether idiomatic
readings survive even with these idioms is controversial. A number of my Japanese informants
(T. Nakamura, Y. Horikawa, K. Fujioka) report judgments that contradict Bruening’s.
4
Another standard diagnostic of raising, the possibility of expletives, cannot be tested in Korean and
Japanese as the languages lack expletives.
5
We eventually part with tradition, however. As we shall argue later, we generalize the notion of
Major Subject so that even in sentences with only one overt subject constituent the constituent
may be either a Major Subject or a grammatical subject. We shall also argue that there are two
distinct types of Major Subjects – a nominal (DP) subject and a non-nominal ‘event’ argument.
6
That is, sentences expressing a Categorical Judgment, in the terms we shall use subsequently. Com-
plements of SOR verbs need not always express a Categorical Judgment. Thetic judgment sen-
tences are also possible. However, when the latter type of clause occurs as complement, SOR is
prohibited, as we shall see in detail subsequently.
7
Major Subjects that do not bind a variable within the Sentential Predicate occur in the so-called
Non-gap (or Adjunct-type) MNCs, as we shall see.
8
For example, there is no mention of nonsubject raising in J.-S. Lee 1992, and many objected to J.-M.
Yoon’s (1991) claim that embedded objects can undergo SOR in Korean. However, as K.-S. Hong
(1997) pointed out, these objections did not take into account the intricate factors we discuss here.
For Japanese, neither Tanaka (2002) nor Bruening (2001) report such data, while Takano (2003)
and Hoji (2005) make much of such data as an argument against SOR being raising.
9
The DP that functions as the Major Subject could not have been derived by scrambling in (17), since
the predicates are Acc-assigning predicates. The alternative that takes Nom-case to be the realiza-
tion of focus (Schütze 2001) does not work either, as Yoon (2004a, b) argues in detail. For one, if
Nom-case on the initial DP in (17) is a realization of focus, there is no reason why (18) should be
bad compared to (17a, b), since Nom-case has the option of being interpreted as a focus marker
in all three sentences and focus is not restricted by the ‘characteristic property’ condition which
restricts felicitous Sentential Predicates in MNCs.
10
The constraints on the MNCs in (17) support the view that the initial nominal in these structures is
a subject, rather than a Topic or Focus (as claimed by some). In general, the presence of a salient,
animate subject does not interfere in the topicalization (or focalization) of nonsubject constituents,
because the two belong to different systems (A vs. A’ system). In the case at hand, both Major
and grammatical subjects are subjects (belonging to the A system), and that is why the two are in
potential conflict.
11
Note again that the interpretation of the bare plural embedded subject in the felicitous raising sentence
(26) is generic, not existential, confirming the property-denoting nature of the embedded predicate.
12
Note that in this analysis all Categorical Subjects are treated as Major Subjects in terms of their position
in the clause structure. In this we are following the suggestion of Diesing (1992) who posits that subjects
in construction with individual-level predicates (which we are reinterpreting as subjects of categorical
judgment) occupy a higher subject position, binding a PRO in the lower subject position.
An alternative worth exploring is that in sentences with a single overt Categorical Subject, the
subject occupies a lower subject position (the position of the grammatical subject). Under this
alternative, there could be some Categorical Subjects that are not Major Subjects. As far as we can
tell, the subsequent argument is not affected by the existence of this alternative.
13
The question that remains is whether the contrast in extraction of embedded subjects can be rep-
licated with verbal SCs. In other words, as pointed out by a reviewer, do verbal SCs construed as
expressing a categorical judgment permit the embedded subjects to move?
The answer to this question hangs first of all on whether there are such SCs in the first place. The
sentence in (i) below might be a relevant example:
(i) The city made firemen wear protective gear
(ii) *Firemen were made [t wear protective gear]
If (i) is a verbal SC expressing a categorical judgment, the generalization that embedded categori-
cal subjects can move is falsified. We must assume that in verbal SCs, there are factors over and
above the thetic-categorical distinction that prevent subjects from moving. In adjectival SCs, by
contrast, these factors must not be at play.
14
Given that minimality is relativized to the type of intervener, what he is calling a Topic must be in
an A-position in order to block A-movement. That is why we take his proposal to be equivalent to
positing an extra subject position.
15
What then happens to the event argument of thetic sentences? Following Basilico 2003, we assume
that it is bound off by T. Unlike Basilico 2003, however, we assume that such binding takes place
only as a ‘last resort’, that is, when the event argument cannot occupy the higher subject position.
16
The assumption of lack of raising to SpecTP is not uncontroversial. Miyagawa (2001) assumes that there
is an EPP-driven raising in Japanese (and Korean). However, even in his account, the subject is not the
only constituent that can fill the EPP need of SpecTP. Object scrambling is another way to satisfy that
need. Therefore, the point that there is no obligatory raising of subjects to SpecTP can be maintained.
17
D.-H. Chung (2004) questions whether the untensed conjunct in (32a) is a vP. He argues that it is at
least a TP, with a null tense that is interpreted as anaphoric to that on the final conjunct. W.-S. Lee
(2003) argues that there are non-trivial problems with this proposal.
18
Again, versions with a gap in place of pronouns are degraded.
19
In a similar vein, Han and Kim (2004) argue that Subjacency-violating ‘double relative clauses’
in Korean should be analyzed as the Subjacency-observing relativization of Major Subjects of
MNCs where the Major Subject is coindexed with a constituent within an island.
20
A reviewer questions how the Major Subject sey-myeng-uy haksayng-i in (36a) can be a Categorical
Subject. A number of facts suggest that it is, despite the fact that the lexical predicate of the
embedded clause is not a stage-level predicate. First, the indefinite subject must be interpreted
in the specific sense, whereas in (36b) it need not. Second, the Sentential Predicate is construed
as stating a property. The property in question is that of the students’ parents being introduced to
professors. Though arguably not a characteristic property, it is nonetheless a property that dis-
tinguishes in a given context certain students from others. It is a ‘characterizing’ property, in the
terminology of Yoon 2004b. Similar remarks are applicable to the Sentential Predicates in (38)
and (39).
21
To highlight the interpretive differences between Major (Categorical) Subjects and grammatical
subjects, we have employed MNCs with two overt subjects. However, there still is a consistent
though subtle interpretive difference in clauses with one subject. A Categorical Subject differs from
a Thetic Subject in terms of its preference for specific interpretation.
(i) a. Etten haksayng-i pang-eyse naka-ss-ta (thetic sentence)

certain student-NOM room-from leave-PST-DECL
‘Some student (specific/non-specific) walked out of the room.’
b. Etten haksayng-i totwuk-i-ta (categorical sentence)

certain student-NOM thief-COP-DECL
‘Some (specific/*?non-specific) student is a thief.’
106 JAMES H. YOON
22
The results would follow if scope were read off Chains created by A-movement and not determined
by an independent QR (Hornstein 1995). If QR can assign scope, and if it is not clause-bounded,
there is nothing to rule out the indefinite in (41a) from undergoing QR to the matrix clause,
outscoping the matrix predicate and yielding the de re reading.
23
As we shall see below, Korean does possess genuine Prolepsis structures, and some SOR structures
can be parsed as a Prolepsis structure. However, the claim that all instances of SOR involve Prolepsis
cannot stand.
24
Kotzoglou and Papangeli (2005, and this volume) propose a proleptic object analysis of Greek
SOR/ECM but claim that the proleptic object receives a ‘weak’ theta-role from the matrix predi-
cate. The latter assumption seems motivated by the desire to differentiate object control verbs from
those that take part in SOR/Prolepsis. Option (c) does away with the need to posit such a theta-role
and hence, seems to be a more desirable implementation of the Prolepsis analysis.
25
The nature of locality of Chain Formation in copy raising is not a matter on which there is consen-
sus. We are assuming that the same constraints holding on overt NP/A-movement, that is, the TSC
and SSC/Relativized Minimality, are applicable to copy raising. However, Moore (1998) argues on
the basis of Turkish CR that the TSC is inapplicable to CR. And if Ura (2000) is correct, the exist-
ence of a higher Subject ‘escape hatch’ might allow certain types of CR to circumvent SSC/RelMin
violations, yielding ‘super raising’.
Nonetheless, in no reported cases does CR take place out of islands, whereas we have seen that
the MOB in Korean and Japanese can be related to a constituent within an island in the embed-
ded clause. This, together with properties discussed in the previous section, indicates that the CR
analysis of K/J SOR is not viable.
26
Davies (2005) notes that only subjects can be involved in CR and that idiomatic readings are
retained. As we have seen, Korean (Japanese) SOR is not restricted by these properties.
27
The result of the A-movement/GR alternation test implies that while the P-Object in English may be
an adjunct, that in Madurese is not, and thus, that not all Prolepsis is created equal.
Davies’s (2005, and personal communication) solution to the theta-role of the P-Object in Madurese
is that a theta-role is assigned to the object in the matrix clause in the same way that the proleptic
object is presumably assigned a role in the English construction. The proleptic objects can always
appear as PP, where it presumably gets the necessary theta role. This is supported by the fact that in
Madurese, when the proleptic object appears as subject or object, morphology on the verb required
with prepositional objects is required (except for one verb).
Unlike Madurese, however, we cannot assume that the MOB in Korean receives a theta-role from
the matrix predicate, given that there is clear evidence of a lower clause origin for the nominal.
28
The two options are not unlike two ways of thinking of Small Clauses, as sub-clausal versus clausal,
except that here the clause that is posited not small, but ‘super’-sized.
29
Since he was assuming that only embedded subjects undergo SOR, he proposed that SOR pro-
ceeded through an A-position type ‘edge’ in the embedded CP in Japanese.
30
Again, we need to control for a Prolepsis parse of the relevant sentences. For example, assuming
that kiekhata is an SOR verb, K.-S. Hong (personal communication) pointed out that the following
is not as degraded as (53a) and (53b).
(i) ?[ku-uy apeci-ka pwuca-yessta-ko] na-nun Cheli-lul kiekhanta

he-GEN father-NOM rich-was-COMP I-TOP C-ACC remember
‘I remember Cheli’s father as (being) rich.’
We contend that this is because Cheli-lul in (i) is more easily parsed as the proleptic object of the matrix
verb than in (53). The following contrast suggests that this line of thinking is on the right track:
(ii) a. Na-nun cinan-cwu-pwuthe-lul nalssi-ka coaciessta-ko kiekhanta

I-TOP last-week-from weather-NOM became.better-COMP remember
‘I recall the weather started improving since last week (not this week).’
b. *[nalssi-ka coaciessta-ko] na-nun cinan-cwu-pwuthe-lul kiekhanta
weather-NOM became.better-COMP I-TOP last-week-from-ACC remember
Compared to (i), (ii b) is distinctly worse. Why should this be? The reason is clear: while Cheli can
be construed as the proleptic object of the verb kiekhata (remember), ecey-pwuthe cannot. The
ill-formedness of (ii b) stems from the trace of the Major Subject which violates PBC.
31
An issue that remains is why PBC seems to hold in the fronting of predicative constituents in
Japanese/Korean whereas it does not appear to in remnant topicalization in other languages
(e.g. German), as pointed out by Laurent Dekydspotter. I do not have a satisfactory answer to this
question, but to the extent that PBC holds in K/J, I can use it to diagnose the fine structure of the
preposed complement clause in SOR.
32
Not usually, that is. It has been reported that obligatory control (OC) can involve case sharing in
some languages. In cases of OC, however, the case of the controller (the higher nominal) propagates
to the controllee (the lower nominal). However, in the examples shown here, the case propagates
from the lower position to the higher position, as the matrix verb does not govern Dat/Loc-case.
33
This is true of English SOR/ECM as well. Both wide scope and narrow scope readings are possible
for nobody in (i), though the wide scope reading may be more salient.
(i) Dulles believed nobody to be a spy before Johnson did

= There is nobody that Dulles came to believe to be spy before Johnson did (nobody > believe)
= Before Johnson came to believe that nobody is a spy, Dulles came to that belief (believe > nobody)
34
This is possible because the embedded Major Subject can get Nom-case and since there is no ban on
multiple case assignment in the languages (Yoon 1996).
III
RAISING OR CONTROL IN GREEK

GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
NOT REALLY ECM, NOT EXACTLY CONTROL:

THE ‘QUASI-ECM’ CONSTRUCTION IN GREEK*
1. INTRODUCTION
In this paper, we discuss Greek constructions with an accusative DP, which

appears in a syntactic position where it could, in principle, be assigned a thematic
role from the embedded clause and accusative case from the main verb, as in (1a)
and (1b).
(1a) i epivates perimenan ton kapetanio

the passengers-NOM expected-3PL the captain-ACC
na ferthi me aksioprepia
SUBJ behave-3SG with dignity-ACC
‘The passengers expected the captain to behave with dignity.’
(1b) o petros ithele ti maria na traghudhai

the Peter-NOM wanted-3SG the Maria-ACC SUBJ sing-3SG
oli mera
all day
‘Peter wanted Maria to sing all day long.’
The Greek examples in (1) resemble well-known cases of Raising-to-Object (RtO)/

Exceptional Case Marking (ECM) (Postal 1974; Chomsky 1981, among others), where
the embedded subject is assigned accusative case by the matrix verb, as in English (2).
(2a) The passengers expected the captain/him to behave with dignity.
(2b) Peter wanted Maria/her to sing all day long.
The embedded verbal element attracts special attention in the Greek examples.
Specifically, the embedded clauses are finite in Greek, and, as a consequence, they
may allow for the subject to surface in nominative, as in the parallel examples (3a)
and (3b).
(3a) i epivates perimenan o kapetanios

the passengers-NOM expected-3PL the captain-NOM
111
© 2007 Springer.
112 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
(3b) o petros ithele i maria na traghudhai

the Peter-NOM wanted-3SG the Maria-NOM SUBJ sing-3SG
oli mera
all day
‘Peter wanted Maria to sing all day long.’
The same effect can be observed with a host of matrix verbs in Greek, such as ipologhizo
‘estimate’, pistevo ‘believe’, ksero ‘know’, theoro ‘consider’, among others.
Interestingly, the nominative alternate is not attested in a language like English,
which exhibits a prototypical instance of ECM, as can be easily observed in (4):
(4a) *The passengers expected he to behave with dignity.
(4b) *Peter wanted she to sing all day long.
Of course, English permits nominative case-marked subjects with indicative that-

clauses:
(4c) The passengers expected that he would behave with dignity.
The same applies to Greek indicative clauses, as illustrated in (5).
(5) i epivates perimenan oti o kapetanios

the passengers-NOM expected-3PL that the captain-NOM
tha ferotan me aksioprepia
would behave-3SG with dignity-ACC
‘The passengers expected that the captain would behave with degnity.’
These examples are not identical to the examples in (3). In particular, Greek has a
nominative/accusative alternation in the same (non-indicative) linguistic context,
as illustrated in (1) and (3). This is not attested in a language like English, where
ECM alternates only with an embedded that- clause. This situation raises a number
of questions, concerning the similarities and differences between the Greek and
English data, leading to the ultimate goal of this paper, namely the suggestion
of a possible explanation for the alternating pattern in Greek with reference to
thematic information, case properties, and semantic control. Our discussion seeks
to contribute new insights to the overall issue of control and RtO/ECM at the
interface of syntax and semantics.
First, we present the main issues raised by the Greek ‘quasi-ECM’
construction. To be more accurate, we discuss the syntactic analysis of this con-
struction, and its semantic implications for the thematic information that is real-
ized. Second, we present a number of arguments against an analysis of ‘regular’
ECM. This leads to the conclusion that Greek involves finite embedded clauses
with a phonologically empty pronominal subject (pro). The evidence indicates
that the accusative DP is case-marked by the matrix verb, and it presumably also
receives the internal thematic role of the main verb. Next, we present a possible
explanation for the Greek examples that is based on case properties, along the
NOT REALLY ECM, NOT EXACTLY CONTROL 113
lines of Reinhart and Siloni (2005) and on semantic control, as suggested also
by Philippaki-Warburton and Catsimali (1999) and Spyropoulos (this volume).
Last, we discuss a number of theoretical consequences of our analysis with
respect to the universal nature of ECM and its implications for the semantic
properties of the verb and its arguments.
2. MAIN ISSUES
In this paper we argue that the accusative-marked ‘quasi-ECM’ DP in Greek is

in fact an argument of the main verb throughout the derivation, in constructions
such as (2).1 ‘Quasi-ECM’ in Greek is thus a subcase of object control, as illus-
trated in (6).
(6) . . . [vP perimenan [DP ton kapetanio]i [CP proi na ferthi me aksioprepia ]]
Iatridou (1993), Varlokosta (1994), and Alexiadou and Anagnostopoulou (1997)

provide detailed analyses of clear cases of object control in Greek. Our aim, in this
paper, is to explore how ‘quasi-ECM’ may be captured in a similar way. The spe-
cial characteristics of ‘quasi-ECM’ may then be attributed to semantic factors that
indicate the presence of an extra theta-role2 and a semantic operation of control.
In particular, we suggest an operation of semantic control with the control-
led element being an embedded pro, in the sense of Philippaki-Warburton and
Catsimali 1999 and Spyropoulos (this volume). Moreover, we discuss the presence
of an extra theta-role that is realized by the accusative DP and its implications
for the operation of semantic control. The crucial factor for the appearance of
‘quasi-ECM’ is, thus, attributed to the kind of thematic information that is avail-
able for the verbs of a given language. This is linked to the components of accusa-
tive case that characterize verbal elements, along the lines of Reinhart and Siloni
2005. According to them, accusative case consists of two components; inherent
accusative is universally attested, while structural accusative is parameterized. ‘If
a language has no structural case, it should not have ECM constructions,’ they
argue (Reinhart and Siloni 2005:430). We believe that this prediction is confirmed
by the lack of ‘true ECM’ in Greek, a language that does not have structural case,
as discussed in more detail in section 3.
3. CASE-ASSIGNMENT IN SUBJUNCTIVE CLAUSES
Let us now turn to the contrast between (1) and (3). The alternation between a
nominative subject in (3) and ECM-like structures in (1) raises the question of
how the subject of the embedded subjunctives gets its case in Greek. Indeed, the
case marking of the thematic subject of the secondary clause lies at the heart of
the ECM-account.
The traditional analysis of ECM clauses such as (7) attributes accusative
case marking to the fact that nonfinite clauses cannot assign nominative case to
their subjects. In other words, the absence of finiteness signals the appearance of
RtO/ECM as a last resort, case-marking mechanism:
(7) I expected him to like ice-cream.
(8) [TP h?i [Tdef to] [vP ti like ice-cream]]]
It is generally assumed that the raised pronoun is assigned accusative by the matrix
verb, as illustrated in (9) (h? in (8) is taken to indicate the as-yet caseless pronoun
him).
(9) . . . [vP expected [TP himi [Tdef to ] [vP ti like ice-cream]]]]
For some researchers the ECM-subject DP raises further to the periphery of the
matrix vP (with the main verb raising even higher to a superordinate vP-shell, as
suggested by Koizumi 1993; Bošković 1997; Lasnik 1999):
(10) . . . [vP expectedj [vP himi tj [TP ti [Tdef to] [vP ti like ice-cream]]]]]
A similar analysis has been suggested for the Greek quasi-ECM construction
(Kakouriotis 1980). Such an analysis relies mainly on the claim that subjunctives
in Greek are the counterparts of infinitivals, an idea, which appears also in Iatri-
dou 1993, Terzi 1992, and Varlokosta 1994.3
In recent analyses, nonfiniteness has been attributed to the lack or the defective
nature of agreement, as suggested by George and Kornfilt (1981)and Chomsky
(2000, 2001). Along similar lines, researchers have also explained the lack of
finiteness by reference to tense, as suggested by Iatridou (1993) and Martin (1996,
2001). The use of finite verbs in Greek embedded subjunctives is, therefore, a first
indication that agreement and tense are not defective and, hence, there is no need
to resort to ECM. Indeed, the presence of nominative-marked subjects signals the
potential of subjunctives to assign case to their subjects.
3.1 Evidence for a nominative pro in the embedded clause
This section deals with the empirical evidence against ECM. The evidence sig-
nals the appearance of a phonologically empty pronoun (pro), which excludes
the option of assigning accusative by the matrix verb to a DP that would be
viewed as the subject of the embedded verb (regular ECM). This becomes neces-
sary mainly due to reasons of case agreement and restrictions on object clitic
doubling, as we will see below.
Evidence for the claim that a nominative-marked pro occupies the sub-
ject of embedded ‘quasi-ECM’ clauses comes from the fact that an embedded
emphatic modifier or intensifier obligatorily surfaces in nominative, even when
its overt matrix antecedent is assigned accusative (Philippaki-Warburton and
Catsimali 1999):
(11a) i epitheorites ithelan to jani

the inspectors-NOM wanted-3SG the John-ACC
na lisi monos tu/*mono tu to provlima
SUBJ solve-3SG alone-NOM/-*ACC his the problem-ACC
‘The inspectors wanted John to solve the problem on his own.’
(11b) sta teleftea lepta oli ithelan to

in-the last minutes everybody-NOM wanted-3PL the
rivaldo na ektelesi ekinos/*ekinon to penaldi
Rivaldo-ACC SUBJ shoot-3SG he-NOM/-*ACC the penalty-ACC
‘In the last minutes everybody wanted Rivaldo to shoot the penalty himself.’
In (11a) and (11b) the modifiers monos tu ‘alone’ and ekinos ‘him/himself’ sur-
face in nominative case, although their overt matrix antecedents are case-marked
accusative. Since such modifiers cannot appear in isolation (and with a case of their
own choice), we may suppose that they agree with a covert pro in the embedded
subject position. This pro may be further linked to the matrix-accusative DP via
co-indexation.
A similar pattern is observed with the behavior of the pronoun o idhjos, which
always displays overt agreement. As illustrated below, the element o idhjos may
appear in nominative, but not in accusative, indicating the existence of a nominative
pro, with which it agrees:
(12) i aliki theli to jani na kopsi

the Aliki-NOM want-3SG the John-ACC SUBJ cut-3SG
tin turta o idhjos/*ton idhjo
the cake-ACC the-same-NOM/-*ACC
‘Aliki wants John to cut the cake himself.’
The crucial observation here is not that the pronoun o idhjos may surface in
nominative, but that it cannot surface in accusative. If the case-assignment of
optional modifiers were a matter of free choice, we might have expected the
option of the modifier surfacing in accusative, thus agreeing with the overt
DP. However, it seems that the embedded pro is a closer antecedent to the pro-
noun, imposing its nominative case.
A further argument for the existence of a pro derives from the behavior of
secondary predicates. A secondary predicate in Greek must obligatorily agree in
case with its subject. In (13a), the adjectival element may appear in nominative,
but not in accusative, whereas in (13b), the secondary predicate may appear in
accusative, but not in nominative:
(13a) o janis ine arostos/*arosto

the John-NOM is sick-NOM/-*ACC
‘John is sick.’
(13b) theorusame to jani *arostos/arosto

considered-1PL the John-ACC sick-*NOM/-ACC
‘We considered John sick.’
However, even when the alleged ECM DP surfaces in accusative case, the embedded
secondary predicate bears nominative. This gives rise to the contrast between (13b)
and (14b):
(14a) perimena o janis na ine arostos/*arosto

expected-1SG the John-NOM SUBJ be sick-NOM/-*ACC
‘I expected John to be sick.’
(14b) perimena to jani na ine arostos/*arosto

expected-1SG the John-ACC SUBJ be sick-NOM/-*ACC
‘I expected John to be sick.’
The above set of data shows that the secondary predicate agrees with a nomina-
tive subject in both (14a) and (14b). The nominative subject is realized overtly in
(14a). As for (14b), the nominative subject of the secondary predicate is assumed
to be a pro. If such a pro exists, then the embedded clause cannot be infinitival.
Similarly to (11) and (12), the possibility of the secondary predicate surfacing in
accusative is ruled out, due to locality constraints, with pro counting as the closest
antecedent.
3.2 Agreement and tense
Having argued for the existence of a subject pro in Greek embedded subjunc-
tives, let us now proceed to further evidence for the finiteness of such clauses
(and, therefore, for the lack of RtO/ECM in Greek). As already mentioned, lack
of finiteness has been explained in the literature as stemming from a deficiency
in either agreement or tense. We will show that neither of the two may be held
responsible for the behavior of embedded subjunctives in Greek.
Greek lacks any morphological evidence that would lead to the claim that
subjunctives are nonfinite. As far as agreement is concerned, we can see below that
subjunctives inflect for all persons and numbers:
(15) o petros perimene emena na fa-o to paghoto eat-1SG

esena na fa-s eat-2SG
afton na fa-i eat-3SG
emas na fa-me eat-1PL
esas na fa-te eat-2PL
aftus na fa-ne eat-3PL
the Peter-NOM expected me/you/him/us/you/them SUBJ eat the ice-cream
‘Peter expected me/you/him/us/you/them to eat the ice-cream.’
This is the first indication that any analysis which attributes case-assignment to
agreement would judge Greek subjunctives as finite.
It has also been suggested that tense is a decisive factor for nominative case-
assignment. Iatridou (1993) argues that (some) Greek subjunctives are nonfinite,
due to the lack of morphological past/non-past tense specification that they exhibit.
As shown in (16b), they cannot appear in past tense, while they are legitimate in
present tense (16a).
(16a) evala ton kosta na tighanisi psaria

made-1SG the Kostas-ACC SUBJ fry-3SG fish-ACC
‘I made Kostas fry fish.’
*
(16b) evala ton kosta na tighanize psaria
made-1SG the Kostas-ACC SUBJ fried-3SG fish-ACC
‘I made Kostas fry fish.’
According to Iatridou (1993), the environments in which subjunctives in past tense

are ruled out seem to be relevant for the nonfiniteness of the clauses in question
and for the lack of nominative case on the subject.
However, evidence against this account can be derived both from the fact that
(seemingly tenseless) subjunctives may tolerate nominative subjects4 and from the
fact that (fully tensed) indicative contexts show apparent ‘quasi-ECM’ properties.5
Our conclusion is, therefore, that ‘quasi-ECM’ is independent of case-assignment,
at least in the traditional terms.
4. AGAINST ECM IN GREEK
Looking into our evidence so far, we have shown that embedded subjunctives dis-
play case marking of their subject in Greek. The verb of the embedded clause is
associated with a nominative subject, like regular verb-subject constructions,
presumably through Agreement, Tense or Inflection. This differentiates RtO/
ECM in Greek from other languages, as it illustrates that the subject DP has case
and, therefore, need not resort to ECM by the main verb.
Taking this into consideration, let us argue that the DP in accusative is the
object of the main verb throughout the derivation, from which it also receives its
thematic role. In the next section, we present the theoretical support for this claim,
based on Reinhart and Siloni 2005, who argue for a systematic interdependence
between ECM and the case properties of different languages. In addition, we pro-
vide empirical evidence from Greek, indicating that the accusative DP receives the
internal thematic role of the main verb and, hence, occupies the object position.
4.1 Reinhart and Siloni’s theory on ECM and case
The aim of this section is to discuss whether the lack of ECM in Greek falls into
any cross-linguistic patterns. In particular, we follow Reinhart and Siloni (2005),
who argue that ECM is linked to the case properties of languages. According to
their analysis, ECM constructions are dependent on the presence of a specific case
component in a given language, what they name the structural case component.
In the absence of ECM, it is possible to suggest that the embedded clause has a
nominative pronominal element that is phonologically empty (pro) as its subject.
We first turn to the details of their analysis.
Reinhart and Siloni (2005) argue that case is parameterized across languages
along the lines of the Case Parameter, which recognizes two components of case,
namely a thematic component and a structural component, that require checking
under specific conditions: ‘Accusative Case has two components to be checked:

a thematic and a structural component. The thematic component is the imple-
mentation of the theta-criterion and, therefore, is universal. Structural Case is
parameterized’ (Reinhart and Siloni 2005:428). Case is, thus, viewed as a dual
system. The main idea is that languages display the thematic component of case
universally; a component, which enables the association of case properties with
thematic information of verbs and their arguments. The structural component
that is parameterized across languages is viewed as a syntactic requirement that
is associated with a number of subsequent linguistic facts, such as the appearance
of ECM, the selection of auxiliaries in verbal alternations, and the marking of
definiteness or animacy.
For example, Greek behaves on a par with Spanish and Romanian, in the use
of the auxiliary have across all instances of verbal alternations. In other languages,
such as Italian, French, Dutch, and German, the formation of unaccusatives
enforces the selection of the auxiliary be. Examples of reflexive and unaccusa-
tive verbs are given below. In both instances, Greek resorts to the auxiliary have,
presumably because the language has no structural case, as Reinhart and Siloni
(2005) suggest.
(17a) o janis echi ksiristi (reflexive)

the John-NOM has be-shaved
‘John has shaved himself.’
(17b) o janis echi pesi (unaccusative)

the John-NOM has fallen
‘John has fallen.’
Auxiliary selection is possibly a strong indication that Greek lacks structural

accusative case.
As suggested in Reinhart and Siloni for Hebrew and Spanish, other types
of evidence, such as the association of definiteness, animacy, or specificity with
a preposition, also do not apply in Greek. This observation is possibly linked to
the morphological properties of the language, where morphological case seems to
play an important role on a number of phenomena. For example, clitic doubling
can be argued to depend on the availability of either morphological case on DPs
(Greek) or prepositional heads (Spanish) (Papangeli 2000, 2004). If this is correct,
then the lack of definiteness, animacy, or specificity marking by Greek preposi-
tions can also be attributed to issues of morphology that prevent the association
of this phenomenon with syntactic case (i.e. Reinhart and Siloni’s structural case
parameter). This is a possible option in languages like Spanish, that make use of
prepositions not only for marking definiteness, animacy, or specificity, but also
for clitic doubling.6
As for ECM, Reinhart and Siloni (2005) present the following argumenta-
tion: the thematic accusative is checked with the Inflectional Projection, leaving
the subject of the clausal complement without case in the absence of structural
accusative. This immediately predicts, as Reinhart and Siloni point out, that the
subject of a clausal complement cannot be realized in a construction that lacks

structural accusative, which would otherwise be checked with it. It is only the
dual system of both thematic and structural case that enables languages such as
English to display ECM on infinitival subjects. In other words, case cannot be
checked ‘exceptionally’, that is, the accusative of the main verb cannot be checked
with the subject of the embedded verbal element due to Reinhart and Siloni’s sug-
gestion that inherent case is checked by the IP and, in the absence of structural
case, there is no case residue to be checked by the DP. They provide the well-
known English example ‘I expect [him to win]’, where the IP checks the thematic
accusative component, while the nominal element ‘him’ checks the structural accu-
sative component. An immediate question arises: Is there an alternative analysis
for ‘quasi-ECM’ in Greek?
It is argued that, in the absence of ECM, the accusative DP is a regular object
of the main verb, receiving both its case and its thematic role from it. As for the
embedded clause, an empty phonologically pronominal element (pro) is assumed
to occupy the subject position. This is, in a sense, a more traditional analysis,
resorting to two clauses with independent case properties, rather than assuming a
DP that is theta-marked from the embedded verb, while being case-marked from
the main verb. The following Greek example strengthens this view (from Philippaki-
Warburton 1987):
(18a) ton idhan na tu dhinun ta

him-ACC saw-3PL SUBJ him-GEN give-3PL the
lefta
money-ACC
‘They saw him being given the money.’
In the above example, the object of the main verb, realized by the accusative
clitic ton ‘him-ACC’ is co-referential with the indirect object of the embedded
verb, realized by the genitive clitic tu ‘him-GEN’. This immediately excludes the
formation of ECM and, furthermore, shows that ECM is not forced in relevant
examples. Of course, the group of perception verbs could be excluded from
regular ECM verbs. However, similar examples are generated with any kind of
verb, as illustrated below:
(18b) ton perimenan na tu dhinun

him-ACC expected-3PL SUBJ him-GEN give-3PL
(ta) lefta
(the) money-ACC
‘They expected him being given the money.’
(=‘They expected that he would be given (the) money by them.’)
In the next section, we present a set of data which supports the lack of ECM.
First, we observe the behavior of the accusative DP in the periphery of the clause,
showing that it can appear on the left of an adverbial of the main clause, a possible
option for objects of the main verb, but not for subjects of the embedded clause.
Second, we look into clitic-doubling constructions, showing that the accusative
DP may be doubled by a clitic, a characteristic behavior of objects, but not of

subjects. The behavior of negative polarity items also points in the same direction,
namely that the accusative DP is an object of the main verb. Last, we discuss doubling
of a clause (CP-doubling in Greek), which leads to the conclusion that the construc-
tion is better analyzed as an instance of doubling of the accusative object DP.
4.2 Further empirical evidence
Given that the RtO/ECM accounts have been shown to be inadequate for Greek, the
following evidence of the matrix behavior of the DP leads us to the conclusion that
the matrix position of this DP is a base-generated one. We first look into adverbial
modification.
The nominative DP-subjects of the embedded predicate cannot surface on
the left of matrix adverbial material; the same test as in Postal 1974 and Zidani-
Eroglu 1997. On the contrary, this is possible with accusative-marked DPs in
‘quasi-ECM’. We may thus conclude that the accusative DP is the object of the
main verb instead of the subject of the embedded verbal form.
(19) o petros perimene *i sofia/ti sofia

the Peter-NOM expected-3SG *the Sofia-NOM/the Sofia-ACC
me laxtara na dhechti tin protasi ghamu
with desire SUBJ accept-3SG the proposal-ACC wedding-GEN
‘It is with desire that Peter expected Sofia to accept the wedding proposal.’
(matrix reading of PP)7
Along similar lines, it is observed that post-verbal ECM subjects may be doubled
by a clitic in the main clause. This resembles regular object clitic-doubling con-
structions and presumably indicates that the post-verbal ECM subject is linked to
the position of the object of the main verb:
(20a) *i astinomia ithele na paradhothi ton kakopio

the police-NOM wanted SUBJ surrender-3SG the bandit-ACC
‘The police wanted the bandit to surrender.’
(20b) i astinomia ton ithele na paradhothi ton

the police-NOM him-CLITIC wanted SUBJ surrender-3SG the
kakopio
bandit-ACC
‘The police wanted the bandit to surrender.’
We thus assume for (20b) that the clitic is the argument of the matrix clause
and the accusative DP is dislocated. This explains the contrast between (20b)
and (21), where the DP ti maria cannot occupy a dislocated position, as it precedes
material of the embedded clause.
(21) *o jorghos tin ithele na taisi

the George-NOM her-CLITIC wanted-3SG SUBJ feed-3SG
ti maria to jaidharo
the Maria-ACC the donkey-ACC
‘George wanted Mary to feed the donkey.’
So, (22) with the DP ti maria dislocated (following an intonational break)

is grammatical.
(22) o jorghos tin ithele na taisi

the George-NOM her-CLITIC wanted-3SG SUBJ feed-3SG
to jaidharo ti maria
the donkey-ACC the Maria-ACC
‘George wanted Mary to feed the donkey.’
Next, let us turn to Negative Polarity Items in Greek, which are licensed only
when found in the scope of affective (or downward entailing, according to Gian-
nakidou 1998) operators. The licensing of negative polarity items is dependent
on the syntactic position of the given elements. It is possible to show whether
the DP in ‘quasi-ECM’ occupies the complement position of the main verb,
that is, the canonical object position, by looking into the effects of the scope of
negation:
(23a) o manos *(dhen) perimene KANENA

the manos-NOM NEG expected-3SG nobody-ACC
‘Manos did not expect anyone.’
(23b) *(dhen) tu milai KANIS

NEG he-GEN talks-3SG nobody-NOM
‘Nobody talks to him.’
Note here that NPIs that are licensed in their clause (by the c-commanding Neg
particle dhen) may move freely either to the periphery of the clause or to other
clauses that appear in higher positions:
(24a) KANIS nomize i fani oti dhen

nobody-NOM thought-3SG the Fani-NOM that NEG
aghorase efimeridha
bought-3SG newspaper-ACC
‘Fani thought that nobody bought a newspaper.’
(24b) o janis pisteve oti (KANENA) dhen aghapai

the John-NOM believed-3SG that (nobody-ACC) NEG loves-2SG
i sofia (KANENA)
the Sophia-NOM (nobody-ACC)
‘John believed that Sophia did not love anybody.’
Accusative case-marked subjects in Greek ‘quasi-ECM’ cannot be licensed by

embedded negation, while nominative subjects can be licensed in such configura-
tions, as illustrated by the following minimal pair:
(25) o manos perimene KANIS/*KANENA na min

the Manos-NOM expected-3SG nobody-NOM/*-ACC SUBJ NEG
tu milai
him-GEN speak-3SG
‘Manos expected nobody to speak to him.’
This shows that the accusative-marked DP occupies the object position of the
main clause rather than being an ECM subject. If it were an embedded subject,
then at some point in the derivation it would have been found in the scope of the
negative particle min and it would have been licensed, in much the same way that
the accusative object is licensed in (24b). However, this is not an option here.
Lastly, let us look into CP-doubling by a clitic in Greek. As we observe, CP-
doubling is ruled out in the case of ‘quasi-ECM’, which is possibly linked to the
position of the accusative DP being the object of the main verb rather than the subject
of the embedded clause. The clitic is thus prevented from doubling the embedded CP
and tends to be associated with the accusative argument of the main verb, like other
instances of object clitic doubling. First we look into the examples of CP- doubling:
(26a) toi perimena [o janis oti tha aghapisi

it expected-1SG the John-NOM that will love-3SG
ti maria]i
the Maria-ACC
‘I expected it that John will love Maria.’
(26b) toi perimena [o janis na aghapisi ti

it expected-1SG the John-NOM SUBJ love-3SG the
maria]i
Maria-ACC
‘Quasi-ECM’ contexts exclude this kind of doubling:
(27a) *toi perimena [ton jani oti tha aghapisi

it expected-1SG the John-ACC that will love-3SG
ti maria]i
the Maria-ACC
(27b) *to perimenai [ton jani na aghapisi ti

it expected-1SG the John-ACC SUBJ love-3SG the
maria]i
Maria-ACC
‘I expected John to love Maria.’
If the accusative-marked DP forms a constituent with the embedded clause, then

there is no reason why such doubling would be ungrammatical. The ungrammaticality of
(27) resembles the ungrammaticality that is attested in similar instances of object

control, as illustrated below:
(28) *to episa ton jani na fighi

it persuaded-1SG the John-ACC SUBJ leave-3SG
‘I persuaded John to leave.’
The above constructions, with the verbs perimeno ‘expect’ and pitho ‘persuade’,
are grammatical if the clitic doubles the accusative DP:
(29a) toni episa ton janii na fighi

him persuaded-1SG the John-ACC SUBJ leave-3SG
‘I persuaded John to leave.’
(29b) toni perimenai ton jani na fighi

him persuaded-1SG the John-ACC SUBJ leave-3SG
‘I expected John to leave.’
The accusative DP behaves, once again, like an object of the main verb. In order to
maintain this conclusion, we need to exclude the option of the DP having moved
to the specifier position of the verbal phrase. As shown below, there is some evi-
dence against overt raising of the DP to a specifier position, namely [Spec, vP].
A further argument derives from the position of matrix subjects. If Greek
‘quasi-ECM’ were indeed a case of overt RtO, then we would expect the DP to
surface in the Spec of the case-assigning v. However, in (30), the accusative DP
appears on the right of an in situ subject.8 Therefore, it cannot have raised to an
outer specifier of the matrix vP.
(30) perimene i niki to luka na pi

expected-3SG the Niki-NOM the Lukas-ACC SUBJ tell-3SG
tin alithia
the truth
‘Niki expected Lukas to tell the truth.’
Having established that the accusative DP in ‘quasi-ECM’ constructions occupies

the object position, which excludes the formation of regular ECM, we need to
examine the kind of analysis that may capture these facts. We thus turn to the
theoretical consequences of the Greek case.
4.3 Theoretical consequences
We first look into the syntactic consequences of our evidence. The idea that
the accusative DP occupies the object position does not comply with Philip-
paki-Warburton and Spyropoulos (1997) for Greek as well as Bruening (2001) for
Passamaquoddy, who have argued that it is the peripheral position of the clitic left-
dislocated (CLLDed) subject DPs in these languages that make them ‘vulnerable’ to
the case-marking properties of matrix verbs. Let us note that under such analyses,
the accusative case-marked DP is neither the genuine subject of the embedded

clause nor an argument of the matrix one. It is just the CLLDed subject of
pro-drop languages, a claim put forth by Barbosa (1995):
(31a) [vP . . . perimena [CP [DP o janis]i . . . proina min kani tetio lathos]]
expected-1SG the John-NOM pro SUBJ NEG make-3SG such mistake-ACC
‘I did not expect John to make such a mistake.’
(31b) [vP . . . perimena [CP [DP ton jani]i . . . proi na min kani tetio lathos]]
expected-1SG the John-ACC pro SUBJ NEG make-3SG such mistake-ACC
‘I did not expect John to make such a mistake.’
The idea that the accusative DP is the object of the main verb that is co-referential
with a pro-subject of the embedded clause leads to important predictions regarding
the behavior of idioms in ‘quasi-ECM’ and the semantic reading that is available in
the relevant configurations.
5. PREDICTIONS
First, we turn to idioms that are often used as a test for syntactic structure and
may thus give us a final confirmation of our claims. Next, we examine the inter-
pretation of the accusative DP in Greek ‘quasi-ECM’ constructions.
5.1 Idioms
Idioms can retain their idiomatic reading even after raising of one of their parts to
a superordinate clause. That is, idioms may appear in raising constructions, since
the part which raises is not assigned any extra theta-role. This test has appeared in
the literature usually with the following example:
(32a) The shit has hit the fan. (Simple declarative)
(32b) The shit seems to have hit the fan. (Raising-to-Subject)
(32c) I expected the shit to have hit the fan. (Raising-to-Object/ECM)
(32d) *I persuaded the shit to hit the fan. (*Control)
Our approach predicts that idioms cannot appear in ‘quasi-ECM’ constructions in

Greek, given that these constructions do not involve any movement of DPs through
raising, but rather involve the use of two independent clauses. This is borne out:
(33a) perimena psili/*psilus na tu bun st’ aftia

expected-1SG fleas-NOM/*-ACC SUBJ his get-3PL in-the ears
‘I expected fleas to get into his ears (I expected him to become suspicious).’
(33b) me afta pu eleghe perimena o

with these that said-3SG expected-1SG the
dhjaolos/*ton dhjaolo na ton pari
devil-NOM/*-ACC SUBJ him take-3SG
‘With the things he said I expected the devil to take him (I expected him to be
destroyed).’
We notice, in (33a) and (33b), that subject idioms in Greek retain their idiomatic
reading when their subject surfaces in nominative, but not in ‘quasi-ECM’, that
is, when the semantic subject of the idiom surfaces in accusative. We may thus
conclude that the accusative-marked DP has never been a part of the embedded
clause, neither has it received any thematic role from it.
In the next section, we turn to the interpretation of the accusative DP and the
interesting question whether its syntactic position, which separates it from regular
ECM, also affects its semantic reading. This leads us to the interface of syntax
and semantics, to the extent that this is relevant for ‘quasi-ECM’.
5.2 The interpretation of the accusative DP
The semantic reading of the accusative DP in Greek ‘quasi-ECM’ is approached from

the perspective of its behavior in passive constructions and of the thematic informa-
tion that it involves, an issue that is further explained in sections 5.2.1 and 5.2.2.
5.2.1 Passivization
An immediate prediction is that passivization in the embedded clause of ‘quasi-

ECM’ constructions will yield a semantic effect. This would comply with the well-
known contrast of ECM vs. control, as presented in Postal 1974:
(34a) I persuaded the doctor to examine Mary ≠
(34b) I persuaded Mary to be examined by the doctor.
(34c) I expected the doctor to examine Mary =
(34d) I expected Mary to be examined by the doctor.
Given that ‘quasi-ECM’ in Greek is a subcase of object control, we expect

embedded passivization to yield some semantic contrast. This is indeed confirmed
by the following data, where the emphasis of the expectation of the speaker is
switched in passive constructions, which is indicated in the rough translations of
the examples in (35) and (36).
(35a) perimena to jatro na eksetasi ti maria ≠

expected-1SG the doctor-ACC SUBJ examine-3SG the Mary-ACC
‘roughly: I expected of the doctor to examine Mary.’
(35b) perimena ti maria na eksetasti apo

expected-1SG the Mary-ACC SUBJ be-examined-3SG by
to jatro
the doctor-ACC
‘roughly: I expected of Mary to be examined by the doctor.’
(36a) thelo to jatro na eksetasi ti maria ≠

want-1SG the doctor-ACC SUBJ examine-3SG the Mary-ACC
‘roughly: I want of the doctor to examine Mary.’
(36b) thelo ti maria na eksetasti apo

want-1SG the Mary-ACC SUBJ be-examined-3SG by
to jatro
the doctor-ACC
‘roughly: I want of Mary to be examined by the doctor.’
Further remarks on the interpretation of the accusative DP in Greek ‘quasi-ECM’

constructions involve the discussion of an extra thematic role that is realized by
this DP.
5.2.2 Thematic role
The suggestion that Greek involves two parallel constructions, one in which a DP
in accusative is co-referential with a pro-subject of the embedded clause (‘quasi-
ECM’) and another one, where the same verb takes a clausal complement, predicts
a difference in the reading of the two examples. This is borne out:
(37a=1a) i epivates perimenan ton kapetanio

the passengers-NOM expected-3PL the captain-ACC
‘The passengers expected of the captain to behave with dignity.’
(37b=3a) i epivates perimenan o kapetanios

the passengers-NOM expected-3pl the captain-NOM
subj behave-3SG with dignity-ACC
The interpretation of (37b) is associated with the following reading: ‘the passen-
gers expected the fact that the captain should behave with dignity’, while (37a) has
a more emotive interpretation: ‘the passengers expected that, as for the captain, he
. . .’ or ‘the passengers had the expectation from the captain that he would. . .’, as
observed also by Hadjivassiliou et al. (2000) and Kotzoglou (2002).
Note here that the Greek ‘quasi-ECM’ constructions display semantic simi-
larities to the following example, perhaps marginally accepted in English:
(38) I believe of John that he will win.
Moreover, ‘quasi-ECM’ constructions have a similar reading with constructions

involving a PP argument in Greek:
(39) i epivates perimenan apo ton kapetanio

the passengers-NOM expected-3PL from the captain-ACC
‘The passengers expected on behalf of the captain that he would behave with dignity.’
This leads us to the conclusion that ‘quasi-ECM’ constructions in Greek involve

a weak thematic role meaning something like ‘as for . . . DP’ or ‘on behalf of
. . . DP’. Given that there is no animacy restriction, we take this thematic role
to be encoded as [−m], what Reinhart (2000, 2002) presents as the equivalent
of Pesetsky’s (1995) subject matter role in the case of experiencing derivations.
Note that within Reinhart’s (2000, 2002) Theta System this is an underspecified
cluster. This means that it consists of only one thematic feature [m] specified
with a negative value [−], and permits two realizations: its [−m] (subject matter)
realization as a PP, while also allowing the option of a [−c −m] (theme) realiza-
tion, as an accusative DP, like any other theme argument. Suppose, then, that
the verbs thelo (want), perimeno (expect) have the following thematic structure
in the lexicon:
(40) thelo (want) [+m], [−m], [−c −m]
The verb in (40) selects for three arguments. The first argument involves some
mental state and is syntactically realized in a position external to the verb (the
subject position is usually assumed to be the specifier of some inflectional pro-
jection, IP or TP). There are two other arguments that may be realized inter-
nally (i.e. in the verbal complement position or in a specifier position that is
lower than the subject). The latter encode the subject matter [−m] and the
theme [−c −m]. The following patterns are derived when all arguments are realized
simultaneously:
(41a) egho ithela apo to jatro (PP) na eksetasi ti maria

I wanted (of) the doctor [to examine Maria]
[+m] [−m] [−c −m]
(mental state) (subject matter) (theme)
In (41a) the argument involving mental state appears as the personal pronoun
egho (I), the argument denoting the subject matter is realized by the PP apo to
jatro ((of) the doctor), while the argument that expresses the theme appears as an
embedded clause na eksetasi ti Maria (to examine Maria).
The crucial difference between these verbs, namely verbs of ‘quasi-ECM’ in
Greek (want, expect), and mainstream object control verbs (convince) relies on the
thematic role of DPs such as to jatro (the doctor) in (41), being a subject matter in
the former (want, expect) and a theme in the latter (convince). This difference gives
rise to the possible realization of the PP in (41a), which is ruled out in the case of
regular object control verbs, such as convince, persuade in (41a’).
(41a’) *egho episa apo to jatro (PP) na eksetasi ti maria.

I convinced (of) the doctor [to examine Maria]
‘I convinced the doctor to examine Maria.’
Next, we turn to the realization of ‘quasi-ECM’ and we note how thematic

information is attested in the Greek constructions:
(41b) egho ithela [o jatros (nom) na eksetasi ti maria]

I wanted [(of) the doctor to examine Maria]
[+m] [−c−m]
(mental state) (theme)
(41c) egho ithela to jatro (acc) [na eksetasi ti maria]

I wanted (of) the doctor [to examine Maria]
[+m] [−m] [−c−m]
(mental state) (subject matter) (theme)
In (41b), only two arguments are realized, namely the argument that involves
mental state [+m] and the theme [−c −m]. The option of realizing only parts of
the thematic information that is encoded in the Lexicon for a given verb is gener-
ally attested across languages for different types of verbs, the most prominent
examples being ‘experiencing’ derivations that involve ‘psych’ verbs, such as the
verb worry. As Reinhart (2000, 2002) explains in detail, these verbs involve three
arguments, the experiencer [−c +m], the subject matter [−m], building on Pesetsky
(1995), who first introduces the idea of a subject matter thematic role, and the
cause [+c]. Reinhart attributes the different realizations of these verbs to the option
of expressing only parts of their thematic information, such as the cause and the
experiencer in what is usually called ‘object experiencing’ derivations (‘The doctor
worries Lucie’) or the experiencer and the subject matter in ‘subject experienc-
ing’ derivations (‘Lucie worries about her health’). The analysis of the thematic
information of ‘quasi-ECM’ builds along the main lines of Reinhart’s approach
to experiencing derivations, although the theta-roles that are suggested here for
‘quasi-ECM’ verbs are not identical to the theta-roles of psych verbs. What they
might have in common is the presence of a subject matter argument.
If this line of argumentation is correct, the immediate prediction is that we
should be able to cluster the verbs that display ‘quasi-ECM’ behavior on the basis
of their thematic information. This immediately implies that there exists an entry
of the verb perimeno ‘expect’ in Greek with similar thematic information as the
verb want ([+m], [−m], [−c −m]), crucially involving an argument ([+m]) and a
subject matter ([−m]) that triggers the mental state of the former [+m] argument.
The same thematic structure is assumed for other verbs, such as ipologhizo ‘esti-
mate’, pistevo ‘believe’, ksero ‘know’, theoro ‘consider’, among others. This being
a first attempt to distinguish a group of ‘quasi-ECM’ verbs on the basis of their

thematic information, there is no doubt that further systematic overview of the
thematic clusters is required.
We next turn to the nature of control in Greek ‘quasi-ECM’, which has inter-
esting implications for the grammar of raising and control more in general, espe-
cially when viewed from a cross-linguistic perspective.
6. ‘NOT EXACTLY CONTROL’
Despite the evidence for an extra matrix thematic role in ‘quasi-ECM’ con-
structions in Greek and, therefore, for object control, we observe a number of
differences between ‘quasi-ECM’ and clear-cut object control constructions.
These differences, thoroughly discussed in this section, turn out to motivate an
innovative perspective of control.
First, there is a crucial difference in the thematic information that is realized in
the Greek examples. Object control verbs cannot select a clause as their single argu-
ment, while this was shown to be possible in the ‘quasi-ECM’ examples. Further-
more, object control verbs always realize the subject matter role as a clause. They
thus lack the PP alternate that is attested with verbs of the ‘quasi-ECM’ type.
Second, wh-extraction is banned in Greek ‘quasi-ECM’ domains, while extrac-
tion out of the corresponding control domains is licit:
(42a) ??pjon itheles ton prothipurgho na entiposiasi?

who-ACC wanted-2SG the prime-minister-ACC SUBJ impress-3SG
‘Who did you want the prime minister to impress?’
Corresponding cases of clear-cut object control are attested in Greek, as

illustrated by the following examples:
(42b) pjon epises ton prothipurgho na entiposiasi?

who-ACC persuaded-2SG the prime-minister-ACC SUBJ impress-3SG
‘Who did you persuade the prime minister to impress?’
According to Hadjivassiliou et al. (2000), Greek ‘quasi-ECM’ is an effect of the

reanalysis of a clause beginning with a CLLDed preverbal subject as a DP build-
ing on the idea that preverbal subjects of null subject languages are CLLDed
elements (Philippaki-Warburton 1987; Solà 1992; Barbosa 1995; Alexiadou and
Anagnostopoulou 1998).
(43a) o petros ithele [CP [DP i maria] [CP na kimithi . . .]]

the Peter-NOM wanted-3SG the maria-NOM SUBJ sleep-3SG
‘Peter wanted Maria to sleep.’
(43b) o petros ithele [DP [DP ti maria] [CP na kimithi . . .]]

the Peter-NOM wanted-3SG the maria-ACC SUBJ sleep-3SG
‘Peter wanted Maria to sleep.’
After the reanalysis has taken place, the resulting embedded structure has the
form of a complex-NP and is an island; hence the above contrast. Despite the
observed asymmetry, which remains unexplained under our analysis, we believe
that the benefits of the latter, namely of our analysis of ‘quasi-ECM’ as object
control sentences, have become obvious by the strength of the empirical data that
were presented in the previous sections and that are captured in a systematic way
by our account.
7. CONCLUDING REMARKS
We have argued that, given a dual realization of a number of verbs, allowing either
control or ECM universally, the control option is the only one available in certain
languages, like Greek, due to case reasons. The interaction of case properties and
the availability of ECM was built along the theoretical framework of Reinhart
and Siloni (2005), who suggest that accusative case consists of two components:
the inherent accusative component, which is attested universally, and the struc-
tural accusative component, which is parameterized and determines the appearance
of ECM in a given language. Greek has been argued to lack the structural accusa-
tive component and, hence, the formation of ECM.
The Greek ‘quasi-ECM’ construction displays an operation of semantic con-
trol and requires an extra theta-role that is realized by the accusative-marked DP,
which is a regular object of the main verb and semantically controls a pro subject
of the embedded subjunctive clause. Embedded subjunctives crucially differ from
the infinitival forms that are usually attested across languages and prevent nomi-
native from being assigned to their subject.
Finally, we have argued that the ‘Influence by the matrix verb’-analyses cannot
hold in view of the matrix behavior of the accusative-marked DP as the regular
object of the main verb. The thematic information of the ‘quasi-ECM’ construc-
tion requires further research, the most interesting point being that the accusative
DP is read on a par with PPs in Greek and other languages.
1
Similar suggestions have been made for parallel constructions (proleptic constructions) in Korean
(Byun and Cho 2005), Madurese (Davies 2005, Davies and Dubinsky 2004), and Japanese (Hoji
2005).
2
This theta-role is possibly linked to the theta-role that is assigned to the PPs in examples (i) and (ii).
(i) I want of you to stay at home.
(ii) Perime na apo ton Petro na ferthi me

expected-1SG from the Peter-ACC SUBJ behave-3SG with
aksioprepia
dignity-ACC
‘I expected (of) Peter to behave with dignity.’
3
For the claim that subjunctives are the counterparts of infinitivals in the languges of the Balkans in
general see Joseph 1983 and Terzi 1992.
4
As observed by Philippaki-Warburton and Catsimali (1999), Kotzoglou (2002) and Spyropoulos
(this volume), matrix subjunctives which do not exhibit a past/non-past morphological distinction
may tolerate a nominative:
(i) na klisi tin porta ekinos pu tha mbi telefteos

SUBJ close-3SG the door he-NOM that will enter-3SG last-NOM
‘The one who enters last should close the door.’
5
Philippaki-Warburton (1987) and Theophanopoulou-Kontou et al. (1998) observe that in embedded
declarative environments the alternation of nominative and accusative is permitted irrespective of
the tense specification of the verbal forms.
(i) perimename i eleni/tin eleni oti tha eksorghisti

expected-3PL the Eleni-NOM/Eleni-ACC that will be-furious-3SG
‘We expected that Eleni will be furious.’
The assumption here is that preverbal subjects in Greek are dislocated elements, coindexed with
an argument pro. So, the contrast between the Nom vs. Acc DPs in the above example is attributed
to the antecedent of pro and stems from a corresponding difference in the structures involved (i.e.
the fact that the Acc DPs are arguments of the matrix clause while the Nom DPs are dislocated
elements belonging to the embedded clause).
6
The details of Reinhart and Siloni’s analysis regarding case checking in languages such as Greek are
left aside here, an interesting point being that all instances of case realization are associated with
the thematic component. We concentrate here on empirical facts that seem to relate to this issue,
such as auxiliary selection, ECM, and clitic doubling.
7
Of course both the Nom and Acc counterparts of the DP are licit with an embedded reading of the
PP since in that case the PP occupies a position in the embedded clause and it cannot be used as a
right-boundary element of the matrix one.
8
We follow here most of the literature on Greek in assuming that inverted subjects in Greek remain
vP-internal.
KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
CONTROL IN MODERN GREEK:

IT’S ANOTHER GOOD MOVE*
1. INTRODUCTION: EXPLANATION THROUGH DEDUCTION
The Minimalist Program advances explanatory adequacy to the extent that

stipulative principles and filters of GB are deduced from the smallest number
of simple, ‘natural’ axioms. Fundamental among these are:
(i) Sound and meaning are ineliminable: there are lexical features and properties.
(ii) There is a (recursive) structure building operation: Merge (A & B) produces C.
(iii) The language faculty interacts with external systems: to be usable, the objects of the syn-
tactic component must be legible to the interfaces.
Thus, Minimalist methodology involves deducing as much as possible from

irreducible lexical properties, from the (required) structure building operation,
and from the conceptually necessary interfaces (LF and PF). As Epstein and Seely
(2002:2) in reviewing the Minimalist Program put it: ‘The goal is to minimize
each premise, and the number of them, thereby seeking to maximize explanation
through deduction.’
Our paper is a case study in this reductivist mode of explanation relative to
subjunctive na clauses (henceforth na clauses) in Modern Greek. We are concerned
with properties of the embedded clause in (1)1:
(1) o Yanis theli [na kalesi tus filus tu

the-NOM John-NOM want-3SG/PRES NA invite-3SG/PRES the-ACC friends-ACC his
sta genethlia tu ]
at the-ACC birthday-ACC his
‘John wants to invite his friends to his birthday party’
The key phenomenon involves the fact that in some cases the na clause behaves
like an obligatory control (OC) infinitival, while in other cases it behaves like a
non-obligatory control (NOC) clause, even though the two types of na clauses are
indistinguishable relative to their surface morphology; i.e. they look the same but
they do not behave the same. There is a classic GB analysis that attempts to explain
these behaviors, and one of our goals is to reveal previously unnoted problems with
this analysis (problems which, in fact, carry over to recent Minimalist accounts).
Our central goal, however, is as follows: Within an explanatory and reductivist
Minimalist framework, we attempt to deduce the properties of na clauses appealing
to the following simple, natural lexical property:
133
© 2007 Springer.
134 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
(2a) Certain predicates select a phi-defective Agr (in their complement)
We combine this with the hypothesis in (2b).
(2b) ‘Degrees of Agreement’: Overt agreement morphology and abstract Agr are related via
individual phi-features, not the phi-complex as a whole.
Simply put: If a feature F is present in the surface representation, then F is present

underlyingly. If F is not present on the surface, then F may or may not be present
underlyingly depending on what indirect evidence for F’s existence we find, including,
among other things, the control phenomena that we consider in this paper.
It is an irreducible property of certain predicates that they select defective Agr,
i.e. Agr contains less than the complete set of three abstract phi-features [person],
[gender], [number]. Phi-complete Agr occurs elsewhere. We argue that this simple feat-
ural distinction goes a surprisingly long way in deducing, and hence explaining, in the
sense indicated above, the properties of na clauses; and has consequences beyond.
The paper is organized as follows. In section 2, we present what we consider a clas-
sic GB analysis of na clauses, that of Varlokosta (1993). Section 3 introduces a series
of problems for this GB analysis. Our alternative, reductive Minimalist analysis, is
detailed in section 4. Section 5 presents certain challenges for our approach and
conclusions.
2. A CLASSIC GB ANALYSIS OF na CLAUSES
Varlokosta (1993) ‘Control in Modern Greek’ (CMG) develops a GB analysis of

na clauses.2 Recall that na clauses come in two varieties: those that display OC
properties and those that show NOC3 (cf. Spyropoulos this volume). Relative to the
overt morphology of the verb within the na clause, they are identical. For CMG,
the properties of these two types of na clauses reduce to hypothesized differences
between OC PRO on the one hand vs. pro on the other. CMG first adopts the
contention that PRO is anaphoric (Bouchard 1984) and that it therefore displays
OC properties. Like other pure anaphors then, PRO must have an antecedent; its
antecedent must c-command it; it does not allow split antecedents; and so on.
Alternatively, since pro is pronominal, it displays NOC properties (e.g. it may
refer deictically, it may take split antecedents).These properties are summarized in
Table 1. Given this, the trick is to guarantee that PRO is the subject of OC, while
the subject position of the NOC na clause, if empty, is necessarily pro.
Table 1. OC and NOC Properties of na clauses
The empty subject of the na clause OC na clauses NOC na clauses

Alternates with overt subject No (ia) Yes (ib)
Allows deictic reference No (iia) Yes (iib)
Must have a c-commanding antecedent Yes (iiia) No (iiib)
Allows split antecedents No (iva) Yes (ivb)
Permits both sloppy and strict readings No (va) (only sloppy) Yes (vb)
Allows de se and non-de se readings No (via) (only de se) Yes (vib)
CONTROL IN MODERN GREEK 135
(i) a. *o Yanis kseri na horevi o Vassilis

the-NOM John-NOM know-3SG/PRES NA dance-3SG/PRES the-NOM Bill-NOM
*‘John knows how Bill to dance’
b. o Yanis elpizi na figi (o Vassilis)

the-NOM John-NOM hope-3SG/PRES NA leave-3SG/PRES (the-NOM Bill-NOM)
‘John hopes (Bill) dances’ (good with overt or empty lower subject)
(ii) a. *o Yanis1 kseri [ec2 na horevi]

the-NOM John-NOM know-3SG/PRES NA dance-3SG/PRES
‘John1 knows (how) ec2 to dance’
b. o Yanis1 elpizi [ec1/2 na figi]

the-NOM John-NOM hope-3SG/PRES NA leave-3SG/PRES
‘John hopes to leave’
(iii) a. *o filos tu Yani1 kseri [ec1 na horevi]

the-NOM friend-NOM the-GEN John-GEN know-3SG/PRES NA dance -3SG/PRES
‘John1’s friend knows how ec1 to dance’
b. o filos tu Yani1 elpizi [ec1 na figi]

the-NOM friend-NOM the-GEN John-GEN hope-3SG/PRES NA leave-3SG/PRES
‘John1’s friend hopes ec1 to dance’
(iv) a. *o Yanis1 nomizi oti i Maria2 kseri

the-NOM John-NOM think-3SG/PRES that the-NOM Mary-NOM know-3SG/PRES
[ec1+2 na voithisoun o enas ton allo]
NA help-3PL/PRES each other
‘John1 thinks that Mary2 knows ec1+2 to help each other’
b. o Yanis1 nomizi oti i Maria2 elpizi

the-NOM John-NOM think-3SG/PRES that the-NOM Mary-NOM hope-3SG/PRES
‘John1 thinks that Mary2 hopes ec1+2 to help each other’
(v) a. o Yanis kseri na horevi, to idhio ke o Vassilis

the-NOM John-NOM know-3SG/PRES NA dance-3SG/PRES, the same and the-NOM Bill-NOM
‘John knows how to dance, so does Bill’
(= Bill knows how Bill to dance. Not: Bill knows how John to dance)
b. o Yanis elpizi na horevi, to idhio ke o Vassilis

the-NOM John-NOM hope-3SG/PRES NA dance-3SG/PRES the same and the-NOM Bill-NOM
‘John hopes to dance and so does Bill’
(= Bill hopes that Bill will dance or Bill hopes John to dance)
(vi) a. o Atichis kseri na ektimisi to metalio

the-NOM Unfortunate-NOM know-3SG/PRES NA appreciate-3SG/PRES the-ACC medal-ACC
‘The Unfortunate knows to appreciate the medal’
b. o Atichis elpizi na kerdisi to metalio
the-NOM Unfortunate-NOM hope-3SG/PRES NA win-3SG/PRES the-ACC medal-ACC
‘The Unfortunate hopes to win the medal’
Quite rightly, CMG attempts to derive this distribution from independent

properties of the different na clauses, combined with the hypothesized differences
between PRO and pro. To this end, CMG observes that NOC na-clauses as in (3)
‘form an independent domain aspectually’ denoting distinct, independent events
relative to the event denoted by the main clause.
(3) o Yanis elpizi na kalesi i Maria

the-NOM John-NOM hope-3SG/PRES NA invite-3SG/PRES the-NOM Mary-NOM
ti fili tis sta genethlia tu
the-ACC friend-ACC her at the-ACC birthday-ACC his
‘John hopes that Mary invites her friend to his birthday party’
The idea is that in (3) there is both a ‘hoping’ event and a separate ‘inviting’
event. OC na-clauses, as in (4), on the other hand, are claimed not to denote an
event independent of the main clause. In (4) there is just one event, a ‘know-how-
to-dance’ event.
(4) o Yanisi kseri na eci / *j horevi

the-NOM John-NOM know-3SG/PRES NA dance-3SG/PRES
‘John knows how to dance’
The argument for multiple vs. single events (as the crucial factor distinguishing the
different na-clauses) is supported by the presence of past tense in the embedded INFL
(5) and, in the absence of past tense, by the use of different matrix and embedded
adverbs (6). That is, where the na clause intuitively denotes an event separate from the
main clause, the na clause may have a tense and/or temporal adverbial distinct from
the main clause.
(5) o Yanis elpizi na kalese i Maria

the-NOM John-NOM hope-3SG/PRES NA invite-3SG/PAST the-NOM Mary-NOM
ti fili tis sta genethlia tu
the-ACC friend-ACC her at the-ACC birthday-ACC his
‘John hopes that Mary invited her friend to his birthday party’
(6) tora, o Yanis elpizi na kalesi i Maria

now the-NOM John-NOM hope-3SG/PRES NA invite-3SG/PRES the-NOM Mary-NOM
ti fili tis sta genethlia tu avrio
the-ACC friend-ACC her at the-ACC birthday-ACC his tomorrow
‘Now, John hopes that Mary invites her friend to his birthday party tomorrow’
na-clauses that intuitively do not denote an event separate from the event of the main
clause do not support different tenses (7a), nor different matrix and embedded adver-
bials (7b):
(7a) *o Yanis kseri na kolimbise

the-NOM John-NOM know-3SG/PRES NA swim-3SG/PAST
‘John knows swam’
(7b) *tora, o Yanisi kseri na eci /*j kolimbai avrio

now the-NOM John–NOM know-3SG/PRES NA swim-3SG/PRES tomorrow
‘Now, John knows how to swim tomorrow’
CMG’s next step is to tie this independent property of the different na clauses
to ‘tense sequencing,’ in the sense of Hornstein 1990.4 CMG argues that if the
embedded na-clause denotes an event distinct from the event of the main clause,
Table 2. Varlokosta’s (1993) Analysis
If two independent events If only one event

→ Tense sequencing → No tense sequencing
If tense sequencing If no tense sequencing
→ Verb movement to C → No verb movement to C
If verb movement to C→ If no verb movement to C
Nom case is licensed → Nom case is not licensed
If nom case is licensed → pro If nom case is not licensed → PRO
⇓ ⇓
NOC OC
then tense sequencing is triggered. Tense sequencing, in turn, requires V to C

movement. In this way the embedded V can get close enough to the matrix V to be
governed by, and hence properly dependent on it. Next, CMG hypothesizes that
it is precisely this V to C raising (not tense or agreement) that licenses Nominative
Case. The final step of the argument is to adopt the view that pro must be Case-marked,
but PRO cannot be Case-marked.
To sum up, CMG accounts for OC vs. NOC na clauses by forcing PRO to occur
as the subject of the former, while the subject of the latter, if empty, is pro. This
distribution is tied, ultimately, to event structure. Table 2 reviews the full argument.
3. na CLAUSES AND MINIMALIST METHODOLOGY:

THE CHALLENGE
The GB analysis reviewed above appeals crucially to: (i) the PRO vs. pro distinc-
tion; (ii) the idea that V raising to C ‘licenses’ Nominative Case; and (iii) the claim
that V to C raising is (ultimately) triggered by the presence of multiple events. In
this section we argue that each of these is problematic.
3.1 PRO vs. pro distinction: stipulated distinctions?
CMG adopts the idea that the anaphor PRO5 must be Caseless, while the pronominal
pro must be Case-marked. The problem noted in Landau 2002 is that this stipulates
the distribution of the empty elements. ‘It’s an anaphor and thus cannot be Case-
marked’ and ‘it’s a pronoun and thus must be Case-marked’ do not follow from inde-
pendent properties. The difference cannot be deduced from their phonetic makeup
– they are both null – and it is not deduced in CMG from their formal makeup (i.e.
from the semantic features of PRO, an anaphor, vs. pro, a pronominal.6
Furthermore, as noted in Hornstein 2001, it is a stipulation that anaphoric PRO
has OC properties. It is not clear, for example, why PRO and other, overt, anaphors
must have a c-commanding antecedent, why it only allows the sloppy reading under
ellipsis, and so on. Again, the difference is not deduced, but rather (re)stated.7
The PRO vs. pro distinction is crucial to the CMG analysis, and critically, the
distribution of PRO vs. pro is tied to Case. But the explanatory value is attenuated
given the stipulations required: PRO cannot be Case-marked, pro must be, and only
PRO has OC properties. Of course, it could be that these mechanisms for determining
the distribution of PRO are axiomatic. However, a theory from which these properties
are deduced would be preferred, and such a theory, we argue below, is available.
3.2 V raising to C licenses Nominative Case: An operation is assigning a feature?
CMG states that ‘V raising to C licenses Nominative case marking in na clauses . . .’

(Varlokosta 1993:157). It is not clear, however, that such licensing can be implemented,
in any natural way. Case checking is relational: X (a Case bearer) assigns/checks
the Case of Y (the Case receiver). The Case bearer is a lexical element with formal
features (Case and/or Agreement features). CMG, however, seems to claim that
it is the movement itself that licenses, and therefore, checks Nominative Case.
But an operation is not a licit member of a Case checking relation since an operation
cannot bear features and does not have Case or agreement features to check
against the Case checkee.8 If the operation of V raising to C licenses Nominative
Case and if other Case checking is done through the government (or Spec-head)
relation, then there are two disparate Case checking mechanisms.9 While this
may be the fact of the matter, a single Case checking mechanism is preferred and
we will argue for just such a single, uniform, mechanism below.
3.3 Event/temporal independence does not always correlate with

non-obligatory control
A final challenge for the classic GB analysis of na clauses reviewed above is the
purported correlation between event/temporal independence and NOC. Recall
that according to the CMG analysis a na clause will display NOC if and only if
that na clause denotes an event independent of the main clause event. If the na
clause denotes an independent event, then there is V (to I) to C raising within
that na clause, which licenses Nominative case in the Spec of the na clause (which
in turn allows an overt DP or else pro in the Spec of that lower IP). Since event
structure is linked to temporal structure such that there is an independent event in a
na clause if and only if that na clause constitutes a (matrix) independent temporal
domain, it follows that a na clause will show NOC if and only if it is temporally
independent.10 The CMG analysis is predicated on a direct correlation between
Event/Temporal (in)dependence and control:
(8) a. event/temporal independence iff NOC
b. event/temporal dependence iff OC
However, we argue now that there seems to be evidence that event/temporal

independence does not always correlate with NOC.11
Consider (9):
(9) o Yanis entharini ti Maria na erthi

the-NOM John-NOM encourage-3SG/PRES the-ACC Mary-ACC NA come-3SING/PRES
sta genethlia tu
to the-ACC birthday-ACC his
‘John encourages Mary to come to his birthday party’
Here, we have multiple events. Multiple events are evidenced by the fact that the
na clause may take a temporal adverbial distinct from the matrix:
(10) hthes o Yanis entharine ti Maria na erthi

yesterday the-NOM John-NOM encourage-3SG/PAST the-ACC Mary-ACC NA come-3SG/PRES
avrio sta genethlia tu
tomorrow to the-ACC birthday-ACC his
‘Yesterday, John encouraged Mary to come to his birthday party tomorrow’
Under CMG’s analysis, then, we predict that the na clause of (10) should display
NOC properties. This prediction, however, is false. In fact, we get OC.12 Thus,
the na clause does not allow an overt subject (11a). The empty subject of the na
clause cannot refer deictically (11b) and it must have a c-commanding antecedent
(11c).13
(11) a. *o Yanis entharine ti Maria na erthi

the-NOM John-NOM encourage-3SG/PAST the-ACC Mary-ACC NA come-3SING/PRES
o Vassilis sta genethlia tu
the-NOM Bill-NOM to the-ACC birthday-ACC his
‘John encouraged Mary Bill to come to his birthday party’
b. *o Yanis entharine ti Maria1 na ec2 erthi

the-NOM John-NOM encourage-3SG/PAST the-ACC Mary-ACC NA come-3SING/PRES
sta genethlia tu
*‘John encouraged Mary1 ec2 to come to his birthday party’
c. *o Yanis entharine to filo tis Marias1

the-NOM John-NOM encourage-3SG/PAST the-ACC friend-ACC the-GEN Mary-GEN
na ec1 erthi sta genethlia tu
NA come-3SG/PRES to the-ACC birthday-ACC his
*‘John encouraged Mary1’s friend ec1 to come to his birthday party’
Notice that the argument above is a potential problem for any analysis that ties
NOC to temporal independence. That is, it is perhaps false that if the inflected
[+Agr] embedded clause is temporally independent (= takes own tense and/or
own temporal adverbial [+T]), then that embedded clause will display non-OC
properties. In our case above, the na clause is temporally matrix-independ-
ent, but it is Obligatory Control properties that we find. In short, we have +T
with +Agr associated with necessary OC. But this is precisely what is predicted
NOT to occur in Landau 2004a. For Landau, the feature combination [+T] and
[+Agr] requires [+R], which in turn yields the possibility of NOC. Necessary
OC should arise only when one or the other of [T] and [Agr] features has a
minus specification.
In this section, we have presented three central challenges for what we consider a
classic GB analysis of control in na clauses. Next, we propose an alternative analysis
of na clauses, which, we argue, has greater explanatory depth in the sense specified in
our introduction, has wider empirical coverage and overcomes the above difficulties.
4. EXPLANATION BY DEDUCTION; AGREEMENT BY DEGREES
Recall the Minimalist mode of explanation that we are pursuing. X is explained

to the extent that X is deduced from the fewest simple axioms. Recall also what
the working axioms are:
(12) a. There are ineliminable lexical features/properties.

b. There is a structure building device Merge, which means there is a derivation.
c. There is interaction with external systems. Thus, there are operations (most notably
Agree) that eliminate uninterpretable features. Operations produce objects legible
to the interfaces and do so efficiently.
With this in mind, our research questions are as follows: To what extent can
we deduce the stipulated properties of the GB analysis of na clauses? Can we
overcome the conceptual and empirical problems revealed above? That is, what
is the least we can say beyond (12) and still have the greater empirical coverage
required given the problematic cases that we introduced?
We argue that many of the properties of na clauses can be deduced and that
the problems raised above can be overcome within a highly reductivist Minimalist
framework. As noted earlier, our central proposal is this:
(13) phi-defective Agr is sometimes selected
It is an irreducible lexical property of certain matrix predicates that they select

phi-defective Agr in their subjunctive IP.14 Other predicates select phi-complete
Agr. This lexical property goes a long way in deducing, and hence explaining, the
properties of na clauses.
Our assumption about selection embodied by (13) assumes a particular
reductivist Minimalist framework. The critical independently motivated
postulates are as follows. First, we adopt the probe-goal analysis of Chomsky
(2000, 2001). The basic idea is that uninterpretable features (e.g. phi-features in
I, the EPP feature, and Case features) are valued, and then spelled-out,15 via the
operation Agree whose structural description is as in (14):
(14) X (the attracting ‘probe’) and Y (the ‘goal’) Agree iff:
X and Y Match16 relative to features f

X (derivationally) c-commands Y
Y is the first17 Matching element for X
X and Y are both ‘active’ (i.e. bear an uninterpretable feature)
A further critical assumption of the probe-goal analysis is:
(15) Only a phi-complete Agr may check Case of a DP (modified from Chomsky 2001)18
It follows from (15) that a phi-defective Agr cannot check the Case of a DP. Thus,
in a typical configuration like (16) the structural description for Agree between
the phi-complete ‘probe’ Ic and the ‘goal’ Bill is met.
(16) Ic Bill [VP chased the rabbit]

(Ic = I is complete; i.e. contains all phi-features, person, gender, number)
The structural change is that the phi-features of I are valued by Bill, and the
Nominative Case feature of Bill is valued by Ic (so in the typical case Agree
is symbiotic). In (17), on the other hand, the I of the infinitival is, by hypothesis,
phi defective. Thus, by (15), it cannot value the Case of Bill, resulting in a
non-convergent object.
(17) *I tried Bill Id to chase the rabbit Id = I defective
In this way, Chomsky deduces the GB Case Filter (NPs must have Case by S-structure)
from the legibility requirements of the interfaces (to be usable, the syntax must
produce objects that are legible to the interfaces, hence a Case feature must be
removed in order for the object containing it to be legible to LF).
The second major component of the Minimalist framework that we adopt is
the construal-as-movement analysis of Hornstein (2001). The construal-as-movement
approach is a prime example of reductive minimalism. Hornstein argues that
PRO and control theory (both the classic GB formulation and the null Case
checking analysis of early Minimalism) can and should be eliminated, the
effects of the PRO submodule derived from independently motivated properties of
feature-driven (NP) movement. The lexical element PRO is eliminated, replaced in
effect, by NP-trace and the principles associated with OC (for example, subject vs.
object control) are eliminated and replaced by independently motivated locality
constraints on feature-driven movement.
The attempt to eliminate PRO and control theory, if feasible, is a welcome move.
Hornstein (2001) argues against the control submodule mainly on the grounds that
it is not compatible with fundamental tenets of Minimalism, as formulated in GB
theory. He points out, for example, that control theory appeals to the structural
relation of government.19 But government is not a fundamental X-bar relation
and hence (at least for Chomsky 1993) is not available in minimalism. We take a
somewhat different view on the matter here. Following Epstein 1999, Epstein
et al. 1998, and Epstein and Seely 2006, we argue that appeal to any relation (such
as government) that is defined on trees is less desirable than deriving that rela-
tion from independently motivated, and independently necessary, operations that
build trees (including both Merge and Move). Movement (more specifically in
the present context Attraction) is independently necessary, and to the extent that
constraints on movement can be deduced, and to the extent that movement
can replace control theory, it is desirable to eliminate control as a separate module

of the grammar.
Under the traditional view of control, we have objects like (18) where the
distribution of PRO (it occurs only in Spec of non-tensed domains) is derived
from PRO’s feature makeup (it is a pronominal anaphor – but see footnote 9)
and the Binding Theory (yielding the PRO theorem), and the antecedent of PRO
is determined in the OC case by properties of lexical items. Thus, try is a subject
control verb.
(18) Bill1 tried [PRO1 to leave]
Note the following basic requirements of the traditional control module: formative
PRO, a set of mechanisms for determining the distribution of PRO, a mechanism
for associating PRO and its antecedent (co-indexation), and principles of control.
Under the construal-as-movement analysis, on the other hand, we have (NP)
movement in place of the control submodule. For Hornstein (18) becomes (19)
(irrelevant details supressed):
(19) Bill tried Bill to Bill leave
Bill starts inside the lower VP, where it is associated with the leaver theta-role
and moves to spec of to.20 It then moves up to the Spec of v of tried where it
gets a second theta-role and finally up to spec of the higher T where it checks
Nominative Case. For Hornstein, the properties of OC PRO follow from independently
motivated properties of and constraints on feature-driven movement. Thus,
in relevant respects (19) is an instance of the sort of NP movement that arises
in passive or raising structures.
The final component of the Minimalist framework that we assume is the
derivational approach to syntactic relations initiated by Epstein 1999, and developed in
Epstein et al. 1998 and Epstein and Seely 2002, 2006. Two features of this approach
are important for present concerns: (i) derivations proceed in a strictly bottom-
up fashion and (ii) relations are not defined on already built up trees but rather are
deduced from the fundamental structure building operation (i.e. Merge) itself.
With this much in place, we can now begin to deduce the properties of na
clauses. Consider first the NOC structure in (20a).
(20) a. o Yanis elpizi na figi

‘John hopes to leave’
We hypothesize that the predicate elpizi ‘hope’21 selects a phi-complete Agr in its
complement clause. Combined with the Minimalist framework outline above, this
yields the NOC properties of (20a). Since the lower Agr is phi-complete, it can
check the Nominative Case of a full lexical DP. Thus we get overt subjects as in
(20b) below. Adapting Hornstein 2001, we assume that the subject of this type
of na clause, if empty, is pro.22 Unlike Hornstein, we tentatively assume that this

pro is like any DP in that it must have its Case feature checked, which it can in
this instance since Agr is phi-complete.23 There is little more to say regarding (20a)
other than that pro is licensed. Just like overt pronouns, pro can refer deictically (20c). It
may take a non-c-commanding antecedent (20d) or split antecedents (20e) (overt
and empty pronouns are the same with respect to these properties, as expected
under the null hypothesis). We get both strict and sloppy readings under ellipsis
(20f) where the pro in the subject position of the second conjunct can take either
Bill or I as its antecedent and, finally, as in Hornstein, we get de se and non-de
se readings (20g).24
(20) b. o Yanis elpizi na figi i Maria

the-NOM John-NOM hope-3SG/PRES na leave-3SG/PRES the-NOM Mary-NOM
‘John hopes that Mary will leave’
c. o Yanis1 elpizi na pro2 figi

‘John hopes to leave’ ‘John1 hopes ec2 leaves’
d. i mitera tu Yani1 elpizi na pro1 diavasi

the-NOM mother-NOM the-GEN John-GEN hope-3sG/PRES NA read-3SG/PRES
ena vivlio
a-ACC book-ACC
‘John1’s mother hopes pro1 to read a book’
e. o Yanis1 nomizi oti i Maria2 elpizi

the-NOM John-NOM think-3SG/PRES that the-NOM Mary-NOM hope-3SG/PRES
na pro1+2 diavasun ena vivlio
NA read-3PL/PRES a-ACC book-ACC
‘John thinks that Mary hopes proJohn+Mary to read a book’
f. elpizo na figo, to idhio ke o Yanis

hope-1SG/PRES NA leave-1SG/PRES the same and the-NOM Yanis-NOM
‘I hope pro to leave and John does too’
(= John hopes pro-John leave or John hopes pro-I to leave)
g. o Atichis elpizi na kerdisi to metalio

the-NOM Unfortunate-NOM hope-3SG/PRES NA pro win-3SG/PRES the-ACC medal-ACC
‘The Unfortunate hopes to win the medal’ – de se and non-de se readings
Consider next the OC structure in (21):
(21) o Yanis kseri na kolimbai

the-NOM John-NOM know-3SG/PRES NA swim-3SG/PRES
‘John knows how to swim’
We hypothesize that kseri ‘know’ selects a phi-defective Agr in its complement

clause,25 this Agr bearing only person and number but not gender.26 In the spirit
of Hornstein (though quite different in some details, specified below), we can now
deduce OC properties. Let us first trace the key steps in the convergent derivation
associated with (21), using English glosses, and simplifying, for ease of exposition:
(22) a. John + swim Merge; theta-features checked

b. Id + [John swim] Merge; Id is defective27
c. know + [Id John swim] Merge; selection, i.e. know selects
subjunctive (assume na is in I) and
Id
d. v + [know Id John swim] Merge; selection (v of VP)
e. John + [v know Id John swim] Move; John attracted by v; theta
checking
f. Ic + [John know Id John swim] Merge
g. John + [Ic John know I John swim] Move; John attracted by matrix Ic,
Case checking
One critical step is (22d) to (22e). Since the lower I is, by hypothesis, defective, it cannot
check the Nominative Case of John. Thus, John is still active and hence can be
attracted up to the spec of v in step ‘e’ (for theta-feature checking) and, ultimately,
John moves to spec of the matrix IP to check Nominative Case, yielding (21).28
Recall next the central properties of OC. First, no overt subject is allowed in
the na clause (23).
(23) *o Yanis kseri na kolimbai i Maria

the-NOM John-NOM know-3SG/PRES NA swim-3SG/PRES the-NOM Mary-NOM
‘John knows Mary swim’
This follows since the lower I is defective and cannot check the Case feature of
Mary, thus causing LF crash.
We now ask why must the empty subject of the na clause in (24) have an ante-
cedent. That is, why can’t it refer deictically?
(24) *o Yanis1 kseri na ec2 kolimbai

the-NOM John1-NOM know-3SG/PRES NA ec2 swim-3SG/PRES
‘John1 knows ec2 swim’
The only way to get deictic reference with an empty subject is to generate pro in the
lower subject position.29 However, pro cannot be Case-marked in that position. As we
saw a moment ago, ‘know’ selects a phi-defective I. The issue is not that the empty
element in the subject of the na clause must have an antecedent. Rather, it is that no
DP can occupy the subject position and be convergent. A DP in the lower subject
position must move out if that DP is to be Case-checked. Thus, if John starts in the
lower subject position, as in (22), it moves out for Case reasons, leaving behind a
copy, which by definition is the same thing as John. So there is simply no question of
the copy referring deictically.30 The empty element pro can start in the subject of the
lower clause, but cannot stay there – again, for Case reasons. Pro can move, attracted
through higher positions, to yield (25), and here pro can refer deictically.
(25) pro pro’ kseri [na pro’ kolimbai
But since pro’ is a copy of pro (which means that pro and pro’ are occurrences of
the same thing), there is no question of pro’ referring deictically.
The next question is why must the empty subject of the na clause in (21),
repeated in (26), have a c-commanding antecedent?
(26) o Yanis kseri na kolimbai

the-NOM John-NOM know-3SG/PRES NA swim-3SG/PRES
‘John knows how to swim’
The short story is that the restriction is built into the structural description of the
operation Agree. In a structure like (26), John must have its Case feature checked
as in, for example, the derivation in (22). The Case feature can only be checked
under Agree. Agree requires that a phi-complete element attract John. But X can
probe and attract Y only if X c-commands Y. Thus, (27) is out, since there is no
licit attractor for John.
(27) *o filos tu Yani1 kseri [ec1 na horevi]

the-NOM friend-NOM the-GEN John-GEN knows-3SG/PRES NA dance-3SG/PRES
‘John1’s friend knows how ec1 to dance’
The relevant derivation is given in (28) – recall that derivations are strictly bottom-up.
(28) a. John + dance

b. Id + [John dance]
c. know + [Id John dance]
d. [ gen friend] + [know Id John dance]
Assume that there is some genitive Case element in the DP ‘gen friend’. Since
this element is contained inside the DP, it does not c-command out of that DP. It
hence does not c-command John at point (28d) in the derivation, and thus cannot
participate in the Agree operation relative to John. That is, John at point ‘d’ cannot
be attracted up by the genitive since the genitive does not c-command John.
Note further that the deduction of the c-command requirement is complete
under Epstein’s notion of derivational c-command. As Epstein (2001) notes, the
c-command requirement of the structural description of Agree follows from
the conception of movement as attraction incorporated into the probe-goal anal-
ysis combined with Epstein’s derivational approach to syntactic relation. Epstein
develops the First Law, informally stated as (29):
(29) X may participate in a syntactic relation with Y only if X derivationally c-commands Y.

Thus X can attract Y (where attract is necessary for Agree) iff X (derivationally)
c-commands Y.
As for the other properties of OC, they too emerge. First, why are split antecedents
not allowed with OC? For Hornstein (2001), this follows since we cannot first-Merge
more than one DP into the same position. Consider again the prohibited split
antecedent case, repeated as (30):
(30) *o Yanis nomizi oti i Maria2 kseri

the-NOM John-NOM think-3SG/PRES that the-NOM Mary-NOM know-3SG/PRES
‘John1 thinks that Mary2 knows ec1+2 to help each other’
There is no question of generating PRO in the lower subject position since there
is no PRO. And, as we saw above, pro is also not allowed in the lower subject
position since the lowest I is defective (‘know’ selects Id) and cannot check Case on
pro (which, by the null hypothesis, is like any other DP in needing its Case feature
checked). So, the question is not why can’t the empty element in the lower subject
position of (30) take split antecedents, but rather why can’t we derive something
like (31)?
(31) John thinks Mary knows Mary John to help each other
Hornstein’s answer would be that both Mary and John cannot Merge into the
same lower subject position. We suggest two other reasons. One is that Mary
and John cannot both receive the agent role of help. For the construal-as-move-
ment approach, a DP may get more than one theta-role. However, there is a
vestige of the theta-criterion tacitly assumed; namely, theta-features can only
be assigned once. Thus, merging two DPs into one theta position is prohibited.
Another reason is that (31) would involve an intervention effect. The highest
T could not attract John since Mary intervenes. Once Mary is attracted up to the
spec of the v of thinks, it would be closer to the higher T, thus blocking T from
attracting John.
Note finally that the sloppy only reading of OC (32) can also be derived.
(32) o Yanis kseri na horevi, to idhio ke o Vassilis

the-NOM John-NOM know-3SG/PRES NA dance-3SG/PRES, the same and the-NOM Bill-NOM
‘John knows how to dance, so does Bill’
(=Bill knows how Bill to dance. Not: Bill knows how John to dance)
Our reasoning, which differs from that in Hornstein 2001,31 is as follows. The
sloppy reading of (32) would need a derivation like that in (33):
(33) *John knows John to leave and *Bill does know John to leave
namely, John would start in the second conjunct and be attracted into the first
conjunct. Such attraction is not possible however since no element inside the first conjunct
c-commands John in the second, and the structural description of Agree is not met. The
licit sloppy reading, however, is derived as in (34) where each conjunct has only conjunct
internal attractions that properly meet the structural description of Agree.
(34) John knows John to leave and Bill does know Bill to leave
To summarize so far, we have attempted to deduce the properties of the two

types of na clauses within a highly reductivist Minimalist framework. The central
difference between the na clauses is depicted in (35):
(35) NP verb [ na Idef verb . . .] NP verb [ . . . na Icmpl verb]

OC NOC
Certain predicates select phi-defective I32 and the OC properties then follow. Others
select phi-complete I and NOC properties follow. Beyond this selectional difference,
the properties emerge under a highly reductive, and independently motivated set of
postulates (thus probe-goal, construal-as-movement, and derivational relations are
motivated on completely independent conceptual and empirical grounds).
Notice further that the challenges for the CMG analysis introduced in section 3
dissolve under our approach. First, we do not require the stipulation ‘PRO cannot
be Case checked, but pro must be Case checked.’ In principle, we can treat all NPs
alike. They have an LF uninterpretable Case feature that must be checked. Furthermore,
there is no question of V (to I) to C, i.e. the movement operation, licensing nominative
Case. Rather, Case is uniformly checked under probe-goal.33
Finally, we are not appealing in any direct way to event structure or to notions of
independent temporal domain. For us, (9), repeated here as (36), is not a problem.
(36) o Yanis entharini ti Maria na erthi

the-NOM John-NOM encourage-3SG/PRES the-ACC Mary-ACC NA come-3SING/PRES
sta genethlia tu
‘John encourages Mary to come to his birthday party’
Recall that, according to the standardly used adverb test, the na clause in (36) is tem-
porally independent of the matrix and would seem to encode an independent event.
For Varlokosta, this predicts NOC properties. But (36) displays OC properties. Alterna-
tively, we argue that entharino ‘to encourage’ selects a phi-defective I in its na clause34
and thus cannot check nominative Case in that clause. OC then follows. Assuming a
bottom-up derivation, we first get the na clause in (37) through a series of Merges.
(37) [TP na-Id Mary come to the birthday party]
Since I is defective, the Case of Mary is not checked internal to the na clause. At this
point, we can directly plug into the Hornstein account of object control, whereby
Mary moves to the object position of encourage (ultimately checking a theta-role
and accusative Case), and then the embedded IP is Merged to yield (38) and finally
the matrix subject is Merged (and raises to the matrix IP) to produce (36).
(38) encourage Mary [na Mary come to the party
Alternatively, we could assume a VP-shell structure whereby the na clause in (37)

above is first merged with encourage; then Mary is attracted to the higher object
position, and then verb movement and Merging the higher subject in yields (36).35
(39) Mary encourage [na Mary come to the party]
Consider next (40):
(40) o Yanis parakalese to Vassili na erthi

the-NOM John-NOM plead-3SG/PAST the-ACC Bill-ACC NA come-3SG/PRES
i Maria sta genethlia
the-NOM Mary-NOM to the-ACC birthday-ACC
‘John asked/pleaded Bill Mary come to the birthday party’
In this case, the na clause displays NOC properties. Here, parakalo ‘to ask/plead’
selects phi-complete I. Thus the lower subject position is a Case position, licensing
an overt DP and pro, and from pro we get NOC properties, as detailed above.
When embedded within a highly reductivist framework, we can account for the
properties of na clauses by appealing to an irreducible lexical property (‘defective
I is sometimes selected’). Not only do we account for the NOC vs. OC distinction
in na clauses with no added technicalia, but also have greater empirical coverage
since we can account for the cases troublesome to previous analyses. In short,
some explanatory and descriptive success can be claimed. In the next section we
turn attention to certain implications of and challenges for the proposed analysis.
5. IDEAS AND IMPLICATIONS OF AGREEMENT BY DEGREES
This section briefly considers certain questions for and implications of our analysis,
pointing the way toward areas of potentially fruitful further research.36
5.1 On word order
Varlokosta provides an account of certain word order phenomena within na

clauses. Since with NOC there is overt V to I to C raising, Varlokosta accounts
for the fact that the verb is generally clause initial (specifically that it precedes
the subject) thus:
(41) o Yanis elpizi na horevi i Maria

the-NOM John–NOM hope-3SG/PRES NA dance-3SG/PRES the-NOM Mary-NOM
‘John hopes Mary (will) dance’
For Varlokosta, the na clause in (41) is as in (42) where the verb has moved to I
(picking up the na particle) and then to C (for tense sequencing), yielding the final
word order.37
(42) . . . na horevi John __horevi
Alternatively, we have argued against V to I to C raising, at least for nominative

Case assignment. It is the phi-complete Agr that checks nominative Case and this
does not require V movement. How then do we get the word order facts? There are
a number of options. We could argue that there is V to I to C raising.38 It is just that
this operation is not responsible for nominative Case within the na clause. Rather
Case is still a function of the phi-complete I. The V movement may39 in fact be
triggered by tense sequencing, it just does not have the added by-product consequence
of licensing nominative Case. Alternatively, we could appeal to the basic clause struc-
ture argued for in Philippaki-Warburton (1994) and depicted in (43) where
there is a Mood Phrase above TP, which itself is above vP (AgrP).
(43) MP
na NegP
min TP
Spec T⬘
T AgrP
V Spec Agr⬘
Agr VP
V Spec V⬘
So far we have assumed a non-split T/Infl. Basically, what we do here is develop

our account under a split-Infl combined with the Mood Phrase. If the verb raised
to Agr and then T, but the subject stayed within spec VP (or vP/AgrP), then we
would get the same word order as Varlokosta.40 Notice, however, that preverbal
DP subjects are also possible in subjunctive complements as shown in (44):
(44) o yanis elpizi i maria na erthi sto parti

the-NOM John–NOM hope-3SG/PRES the-NOM Mary-NOM NA come-3SG/PRES to-the party
‘John hopes Mary (will) come to the party’
Although postverbal DP subjects in subjunctive complements constitute the

unmarked option, preverbal DP subjects do not render sentences like (44) ungram-
matical but rather more marked. Assuming that preverbal DP subjects in root clauses
have a Topic interpretation in Greek and thus are not IP-internal elements, occu-
pying Spec,TP but Spec,TopicP in the left periphery, as has been argued in Alexiadou
and Anagnostopoulou 1998 and Spyropoulos and Philippaki-Warburton 2001, then
we see that in embedded subjunctive clauses preverbal subjects are more marked.41 It
is interesting to note that preverbal DP subjects are also allowed in indicative com-
plements as shown in (45) but, according to most speakers, they are less marked
than preverbal subjects in subjunctive.
(45) o yanis kseri oti i maria tha erthi sto parti

the-nom John–NOM know-3SG/PRES that the-NOM Mary-NOM FUT come-3SG/PRES to-the party
‘John knows that Mary will come to the party’
Thus, preverbal DP subjects in subjunctive complements are possible but appear

to be more marked than postverbal subjects and preverbal subjects in indicative
complements.42
5.2 On Nominative Case internal to OC na clauses
As in Varlokosta (1993), though very different in orientation and implementation,

we argue that in an OC na clause as in (46), there is no Nominative Case of a DP
checked internal to that na clause.
(46) o Yanis Kseri na horevi

the-NOM John-NOM know-3SG/PRES NA JOHN dance-3SG/PRES
‘John knows (how) to dance’
However, Philippaki-Warburton and Catsimali (1999) – see also Spyropoulos (this

volume)– argue against a Case theoretic account of OC vs. NOC in na clauses,
proposing that nominative Case can be checked/assigned internal to an OC na
clause. Consider (47), modeled after Philippaki-Warburton and Catsimali 1999.
(47) o Yanis kseri na ine harumenos

the-NOM John-NOM know-3SG/PRES NA be-3SG/PRES happy-NOM/SG/MASC
‘John knows (how) to be happy’
The na clause in (47) displays OC properties. For us, ksero ‘know’ selects a
phi-defective I, which does not check nominative Case. But the problem now,
as Philippaki-Warbuton and Catsimali point out (adopting their argument to fit
our analysis), is that nominative Case is in fact borne by the DP modifier happy
in the example. This is an important observation, and potentially problematic.
Indeed, Philippaki-Warbuton and Catsimali (1999) and Spyropoulos (this volume)
argue that there is a pro in the subject position of the na clause, that this pro subject
does get nominative Case, and that through a form of Case concord matches with
the modifying adjective for this Case. However, there is a potential problem
with such an account. It is not clear how Philippaki-Warbuton and Catsimali
would rule out (48).
(48) *o Yanis kseri na ine harumenos o Vassilis

the-NOM John-NOM know-3SG/PRES NA be-3SG/PRES happy-NOM/SG/MASC the-NOM Bill-NOM
‘John knows (how) Bill to be happy’
If nominative Case is assigned internal to the na clause in (48), then why is

the overt subject Vassilis not licensed? Is there a way to accommodate the
Philippaki-Warbuton and Catsimali phenomena while at the same time maintaining
our basic approach? We believe there is. Consider this proposal:
(49) Phi-defective I cannot check Nominative of a DP, but can check the Nom of an adjective/
nominal modifier.
Why (49) would hold is unclear, but it does yield the desired results. In (48) no
overt subject is allowed and the other OC properties follow as before since I can-
not check the nominative Case of a DP. However this defective I can check the
nominative Case of a DP modifier. In this way, we maintain our earlier results,
while suggesting a path to explore for the new data.
As a brief side note, the current proposal extends to other relevant data. Consider (50):
(50) a. thelo na ine harumenos o Yanis

want-1SG/PRES NA be-3SG/PRES happy-MASC/SG/NOM the-NOM John-NOM
‘I want John to be happy’
b. thelo to Yani na ine harumenos

want-1SG/PRES the-ACC John-ACC NA be-3SG/PRES happy-MASC/SG/NOM
c. *thelo to Yani na ine harumeno

want-1SG/PRES the-ACC John-ACC NA be-3SG/PRES happy-MASC/SG/ACC
d. *thelo na ine o Yanis harumeno

want-1SG/PRES NA be-3SG/PRES the-NOM John-NOM happy-MASC/SG/ACC
This paradigm follows under the assumption that thelo ‘want’ selects phi-complete I
but is also an optional ECM verb (that it can check accusative Case). Then, in ‘a’ the
DP John is checked by the phi-complete I of the na clause. In ‘b’ John is introduced
into the derivation with an Acc Case feature, rather than nominative. Then, John
is active and can value the phi-features of the lower I, but I will not value the Acc
Case of the DP. Since it is still active, John can then be attracted up to the matrix
object position, checking accusative, while the I of the na can check the nominative
of the modifying adjective. On the other hand, ‘c’ and ‘d’ are out since I of the na
clause cannot check the accusative Case of the adjective (rather, it can only check
nominative).
5.3 On temporal / event independence
We have argued against tying nominative Case checking and, more specifically,
NOC to temporal independence, at least for the structures focused on here. As
pointed out, a standard claim for Greek is that if the na clause is temporally inde-
pendent of the matrix, then that na clause displays NOC properties; and that if
the na clause is temporally dependent, then it displays OC properties. Opposing
this, we argued that the na clause in (51) passes the traditionally-used test for
temporal/event independence. That is, the na clause allows a temporal adverbial
(‘tomorrow’) distinct from that of the matrix (‘yesterday’), and yet the na clause
exhibits OBLIGATORY Control, not the predicted NOC, properties.
(51) hthes o Yanis entharine ti Maria na erthi

yesterday the-NOM John-NOM encourage-3SG/PAST the-ACC Mary-ACC NA come-3SG/PRES
avrio sta genethlia tu
tomorrow to the-ACC birthday-ACC his
‘Yesterday, John encouraged Mary to come to his birthday party tomorrow’
Indeed, (51) and (52) seem to be a minimal pair with respect to temporal structure
(the na clause in both takes the same matrix-independent adverbial), and yet (51)
shows OC while (52) shows NOC.
(52) hthes o Yanis parakalese to Vassili na erthi

yesterday the-NOM John-NOM plead-3SG/PAST the-ACC Bill-ACC NA come-3SG/PRES
avrio sta genethlia
tomorrow to the-ACC birthday-ACC
‘Yesterday, John asked/pleaded Bill come to the birthday party tomorrow’
We have adopted the probe-goal analysis of Chomsky (2001), and although

the issue is intricate, an implication of this analysis is that it is agreement
(phi-complete I) that checks/assigns nominative Case, not tense (or event
structure).43
We may want to link phi-(in)complete agreement and nominative Case checking
to tense but it is not clear how to do this. Along with (51) and (52) we find pairs
like (53) and (54), where there is the same temporal relation (namely, temporally
independent) between the embedded and the matrix clause but the control relation
is different: (53) is OC, while (54) is NOC.
(53) o Yanis1 skopevi na ec1/*2 pai stin ellada

the-NOM John-NOM plan-3SG/PRES NA go-3SG/PRES to the-ACC Greece-ACC
‘John is planning to go to Greece’
(54) o Yanis schediazi na pai i Maria

the-NOM John-NOM arrange-3SG/PRES NA go-3SG/PRES the-NOM Mary-NOM
stin ellada
to the-ACC Greece-ACC
‘John arranges that Mary (will) go to Greece’
For us, it is a matter of the selection of phi- complete or defective T, disassociated

from tense/event structure. Thus, cases such as those above are not unexpected.
5.4 On overt inflection, ‘degrees and agreement,’ and learnibility
Pires (2001) pursues the construal-as-movement approach relative to inflected

infinitives in Portuguese, arguing that (i) inflected infinitives = I is phi-complete
= NOC, and (ii) non-inflected infinitives = I is phi-incomplete = OC. So here
we have an overt morphological reflex of syntactic/abstract agreement features.
Informally, the idea is that we can see the phi-completeness of I through the overt
inflection on the predicate:
If there is overt inflection, then I is complete.
If there is no overt inflection, then I is defective.
Greek seems different, however. If we are on the right track, then in Greek, if
we see the overt inflection (in the na clause), then I is complete OR defective.
So, the question is not ‘is there overt inflection or not’ but rather ‘to what
degree is there overt inflection.’ In Greek, there is overt inflection for person and
number, but not gender. Thus, the overt inflection is ‘two out of three phi-features’,
or ‘partial’ overt agreement (it is in this sense that we talk about ‘degrees of
agreement’). Presumably there is a correlation between overt morphology and
abstract agreement features such that if the person feature is overtly present (if it
is pronounced), then it is present abstractly. However, if the feature is not
morphologically present, it may or may not be present in the syntactic representation.
Thus, no overt agreement features are present in the subjunctive in English as in
(55), and yet, the lower I here must be complete to check the nominative Case of he.
(55) the teacher requires he be there at 5:00

(cf. *the teacher thinks he be there at 5:00)
In English, there is no overt inflection on the verb for both OC and NOC
(infinitives). In European Portuguese, only non-inflected infinitives display OC;
inflected infinitives are NOC (but note that inflected infinitives are overtly marked
for two of the three phi-features, person and number). In Greek, the verb is inflected
(person and number) for both OC and NOC. If it is true that overt morphology
reflects abstract agreement (thus if person is pronounced, then the person feature
must be present in the syntactic representation), and if only phi-complete I can

check Case, then we predict that a language that had fully inflected infinitives or
subjunctives would show NOC, not OC. This is a prediction that we have not yet
been able to test, but is a potentially valuable area for further exploration.
Our story also raises an important learnibility issue. How does the child learn
which na clause is OC vs. NOC when the overt morphology is the same, and how
can the child learn this with only positive evidence? We hypothesize that since
the inflection is partial (i.e. person and number, but not gender) the child’s first
assumption is that I is defective. Then she can get positive evidence (she sees the
overt subject in some cases, for example) that I is, for some predicates, in fact
complete, and in this way is able to establish the proper selectional information
for each predicate.
In this paper we have attempted an exercise in explanation by deduction relative
to na clauses in Greek. We have suggested that the standardly noted properties of
OC vs. NOC na clauses, along with a range of (unnoticed) properties problematic
to previous accounts, can be explained within a reductivist Minimalist framework.
We started with the question what is the least we can say? We have answered that
we can go a surprisingly long way with just this: phi-defective Agr is sometimes
selected. We suggested that this gives us explanatory depth, and wider empirical
coverage, and has potentially interesting implications beyond.
* For very helpful, and extensive, comments and discussion from which this paper has benefited
immensely we would like to thank: Gabriela Alboiu, Chris Collins, Samuel Epstein, Gerardo
Fernandez-Salgueiro, George Kotzoglou, David Pesetsky, Irene Philippaki-Warburton, Acrisio Pires,
Dimitris Psychoudakis, Carson Schutze, Vassilis Spyropoulos, Heather Taylor, and anonymous reviewers.
Thanks also to audiences at the North American Syntax Conference, 2003, at the Syntax Support
Group, University of Michigan, 2003, and at New Horizons in the Grammar of Raising and Control
Workshop, the LSA Summer Institute, 2005. Many thanks to Stan Dubinsky and William Davies for
their comments on and interest in this paper. We blame each other for all remaining errors.
1
Of course, ‘Greek na clause’ is informal usage, adopted merely for convenience. There are no specific
constructions; rather, there are lexical features, syntactic operations, and interface conditions that conspire
together to produce syntactic objects. ‘Greek na clause’ has no independent status as such.
2
There are other pre- and early- Minimalist analyses of control in subjunctives in Modern Greek
that, ideally, we would discuss, including Philippaki-Warburton 1987, Terzi 1992, 1997, Iatridou
1988/1993. See Philippaki-Warburton and Katsimali 1999 for excellent commentary on these
works, as well as Philippaki-Warburton and Spyropoulos 2002. We focus on Varlokosta here
mainly given space limitations; we will point out where our discussion of Varlokosta is relevant to
other works as we proceed.
3
See Table 1 for details and examples.
4
The basic idea is simply that the tense properties of the embedded clause are dependent on the matrix.
5
As mentioned earlier, Varlokosta (1993) adopts the idea from Bouchard (1984) that PRO is a pure
anaphor, not the pronominal anaphor of Chomsky (1981). Note that the PRO of classic GB is
arguably LF uninterpretable. If [+anaphoric] entails ‘semantically dependent’ and [+pronominal]
entails ‘able to be semantically independent,’ then an element positively specified for both of these
features (i.e. PRO) is simultaneously dependent and capable of being independent, and this combination,
we contend, is not LF interpretable; likewise, a [+high], [+low] phonetic feature is PF uninterpretable.
See Burzio (1991) on the need to have a semantic definition of anaphors/pronominals.
6
A similar criticism has been leveled against the null case checking analysis of Chomsky and Lasnik
(1993). Hornstein (2001), for instance, asserts that the analysis (namely that PRO, and only PRO,
must check null Case) has a number of problems:
‘The most glaring is that it essentially stipulates the distribution of PRO. First it is designed to
fit only one expression – PRO. Lexical expressions don’t bear null case nor do other phonetically
null expressions such as wh-t or NP-t. Second, only nonfinite Ts can check/assign it. In effect, the
case properties of PRO and nonfinite T are constructed to exactly fit the observed facts. . . .This
comes close to restating the observations; PROs appear in Spec IPs of nonfinite clauses’ (Hornstein
2001:34).
7
Seely (1988) attempts to deduce the OC properties of overt anaphors. The idea is that an anaphor
must check its inherently unspecified phi and referential features with another element (its antecedent);
and the anaphor moves to get close enough to its antecedent for checking. Interestingly, the properties
of OC then follow from the movement itself.
8
The CMG analysis is quite unclear about the implementation of the licensing of Nominative Case
under V to C movement. The problem might be avoided if the set of features that constitute the
complex ‘V inside C’ (i.e. the result, rather than the process, of V to C movement) could somehow
check Nominative Case. Just how this is to be accomplished is a mystery, however, given that V is
standardly associated with Accusative, not Nominative, Case.
9
Another issue that arises is how V-to-C applies in negative subjunctive complements. Negation,
which is assumed to head its own functional projection, NegP, situated below CP or MoodP and
above TP (as shown in (43) below, which we adapt from Philippaki-Warburton 1994), would block
V-movement. The verb would have to move past another head, i.e. Neg, on its way to C and this
would violate the HMC (Head Movement Constraint).
10
In fact, CMG does not explicitly state the precise relation between event structure and temporal
structure. It should be noted that the correlation between temporal independence and NOC is
assumed in much past and current research. See Landau 1999, 2002. It should also be noted that
the phrase ‘temporal independence’ in the above context is somewhat misleading; all subjunctives
are temporally linked to and, in one sense, ‘dependent’ on the matrix (subjunctives after all
representing irrealis tense). It is relative to that association to the matrix that na clauses are claimed
to have the differences reviewed above.
11
Other possible correlations will be considered in section 5.
12
Note first that other verbs that pattern in this way, i.e. that take a thematic object and a subjunctive
clause, and show OC include leo ‘tell’, epitrepo ‘allow’, afino ‘let’, ipochreono ‘oblige’, diatazo
‘order’, vazo ‘put’, simvulevo ‘advise’.
13
As for the sloppy vs. strict reading, note that (i) can only mean: John encourages Mary for Mary to
come, and Bill encourages Mary for Mary to come.
(i) o Yanis entharini ti Maria na erthi sta

the-NOM John-NOM encourage-3SG/PRES the-ACC Mary-ACC NA come-3SG/PRES to the-ACC
genethlia, to idhio ke o Vassilis
birthday-ACC the same and the-NOM Bill-NOM
‘John encourages Mary to come to the birthday party, and Bill does too.’
It is unclear in such cases, where the matrix verb takes a thematic object plus a clausal comple-
ment, just what the strict reading would be. What is clear is that the example above cannot
mean anything like ‘John encourages Mary for Mary to come and Bill encourages Mary for
John to come’.
As for the de se reading, note that this can’t really be tested, at least not with object control cases, since
the thematic object (i.e. the controller of the lower subject) does not have belief states attributed to it.
14
For purposes of exposition, we represent the abstract Agr element as occurring in I(nfl), leaving
open the possibility that I is actually an Agr, a T, or a split head.
15
For detailed discussioin of valuation and spell-out see Epstein and Seely 2002.
16
X, Y Match only if, for every probing feature of X (e.g. person, gender, number), the goal Y bears the
same feature but not necessarily the same value for that feature. Note furthermore that in Chomsky
2001, unlike Chomsky 2000, Case is not borne by the probe and thus there is no Matching with
respect to Structural Case. For instance, phi-complete T values the Nominative Case of a Matching
DP, but T does not itself bear Case. See Epstein and Seely 2006 for further details; see also Pesetsky
and Torrego 2001 for discussion.
17
Y is the first Matching element for X iff X, Y Match, X c-commands Y, and there is no Z, such that
X, Z Match, X c-commands Z, and Z c-commands Y. Basically then locality of movement is built
into the structural description of Agree.
18
We stress that (15) is an independently motivated axiom, see Chomsky 2001, 2002 for extensive
supporting evidence.
19
Government is appealed to in characterizing the Governing Category of an element X; and this
government relation is critical in the ‘PRO theorem;’ Chomsky (1981) attempts to deduce that PRO
can only occur in ungoverned positions.
20
For Hornstein (2001), Bill moves to Spec of to to check the EPP feature of to. But, whether to has an
EPP feature is an open question (see Epstein and Seely 2006 for extensive discussion, and for a develop-
ment of the view that there is no EPP feature or EPP property). For Epstein-Seely there would be no
movement to or through the Spec of to in (19), but the matter is not crucial for present argumentation.
21
Other verbs that select phi complete Agr in their subjunctive na clause include efhome ‘wish’,
protimo ‘prefer’, apofasizo ‘decide’, epimeno ‘insist’.
22
Though the licensing mechanism for nominative Case is completely different, we do adopt the
Varlokosta analysis for NOC na clauses. That is, the subject of such a clause, if empty, is pro, which
must be Case marked.
23
Of course, there must be some pro drop mechanism, which allows pro in MG, but does not allow
pro in English, at least not in tensed clauses. Presumably some Minimalist reworking of the rich
inflection identification of pro can be developed. Thus, Case marking of pro is necessary but not
sufficient. There is also the question of arbitrary PRO, a function, which, in the absence of PRO,
would have to be taken over by pro. See Hornstein 2001 for a last resort mechanism for licensing
pro in the NOC contexts of English. We leave this area for further research.
24
Thus, in (20g) the epithet ‘the Unfortunate’ may refer to the amnesia victim, while pro may refer to
the pre-amnesia victim war hero. See Salmon 1986 and Hornstein 2001.
25
Other verbs that select phi-defective Agr include archizo ‘start’, dokimazo ‘try’, fovame ‘fear’, and
others that will be discussed below.
26
As mentioned above, OC and NOC na clauses are identical with respect to their overt morphology.
Thus the verb of an OC and an NOC na clause shows tense/aspect, and person and number, but not
gender. What we are claiming is that a morphologically incomplete predicate (i.e. one that shows
less that the full set of phi features person, gender, number) can be underlyingly (i.e. abstractly)
complete or incomplete. Note further that there is nothing about the Gender feature specifically
that renders Agr defective. It just happens that in Greek the gender feature is missing. The key
point is that any missing phi feature will render Agr defective. We consider the consequences of
this in greater detail in section 5.
27
Note that if I were complete here, then later in the derivation at derivational point ‘c’ there would be
a selection violation (‘know’ selects defective, not complete I) and the derivation would crash.
28
It is standardly assumed that preverbal DP-subjects in Greek do not occupy the Specifier position
of TP (see Alexiadou and Anagnostopoulou 1998 and Spyropoulos and Philippaki-Warburton
2001 for extensive discussion and distributional facts supporting this view). Furthermore, based
on interpretational, distributional, and binding facts, the authors argue that preverbal subjects in
Greek occupy an A-bar position in the left periphery of the clause, either Spec,TopicP or Spec,
FocusP. Following this proposal, we argue that the preverbal DP-subject John in (21) will check
nominative case in Spec, TP and then move out to a higher peripheral position.
29
Following Hornstein, we assume that there simply is no PRO. We adopt the arguments in Hornstein
2001 for this and have added another argument. Recall footnote 9 which suggested that the clas-
sic pronominal anaphor PRO is LF uninterpretable. The only remaining empty element with the
potential to refer deictically, then, is pro.
30
For extensive discussion of the consequences of copy theory see Epstein and Seely 2006 and Nunes
1999.
31
Hornstein (2001) makes crucial use of sideward movement. We take a strict version of attraction,
requiring c-command, which, combined with Epstein’s First Law, prohibits many instances of
sideward movement. This difference between our approach (without sideward movement) and
Hornstein’s is significant with far-reaching consequences. Space limitations prohibit a detailed
comparison here, but see Seely 2002 for discussion of this important issue.
32
Recall that we assume that abstract Agr features are in I, abstracting away from the categorical
status of I (as T or Agr or split).
33
There is the question of precisely what motivates movement, whether the EPP feature (as in Chomsky 2001)
or Case (as in Epstein and Seely 2006). We have been assuming the latter here, but the matter is open. See
Alexiadou and Anagnostopoulou 1998 for an account that in Greek the EPP is satisfied via V-raising.
34
Other verbs that pattern in this way (i.e. that take a thematic object, a subjunctive IP, and select
phi-defective I in the IP) include afino ‘allow/let’, kano ‘make’, ipochreono ‘oblige’, diatazo ‘order’,
vazo ‘put’, simvulevo ‘advise’, and apotrepo ‘dissuade’.
35
Choosing between these alternative depends on the status of ‘sideward’ (see footnote 28).
36
There are a number of questions that we will not be able to consider here. One, for example, is the
matter of partial control, detailed by Landau. One interesting fact is that the OC control cases like
(36) above (‘John encouraged Mary to come to the party’) allow a type of split control, namely,
[John encouraged Mary for John and Mary to come] where the verb takes plural agreement. But
partial control in Landau’s sense does not emerge here. (i) cannot mean [John encouraged Mary
for Mary and someone else to meet].
(i) o yanis entharine ti maria na sinandithun
the.NOM John.NOM encourage.3SING.PAST the.ACC Mary.ACC SUBJ meet.3PL.PRES
See Hornstein 2003 for a response to Landau’s criticism, based on partial control, of construal as
movement.
37
With OC, there is no V to C movement for Varlokosta, but, in fact, the effects of V to C are not
visible (relative to the subject) since the subject is never overt in OC clauses. Thus, there could in
principle be V movement in OC structures as well.
38
Notice, however, that Philippaki-Warburton (1998) independently argues that V-to-C movement
does not apply in Greek indicative and subjunctive, ‘since such a movement has no morphological-pho-
netic consequences’ (p. 179) in that V-movement is not triggered by affixes that need to be attached
on V. As Philippaki-Warburton points out, the preverbal particles na and tha (future marker) can-
not motivate V-movement because they are not part of the verbal morphology. They are separate
elements which head their own functional projections, MoodP and T(Future)P, and are merged
into the derivation from the lexicon. Thus, there is no independent evidence for V raising to C in
Greek subjunctives.
39
An anonymous reviewer points out that the question may not be why the verb in subjunctive raises
to C but why the subject does not raise higher than the verb in embedded subjunctives. We address
this issue below.
40
Alternatively, it may be that Case checking is done in situ in Greek and movement is functional (i.e.
for Topicalization/Focus). See, for example, Philippaki-Warburton 1999.
41
The issue is whether preverbal DP subjects in subjunctive are interpretated as Topics. That is,
whether they exhibit the interpretational effects that are associated with A-bar subjects, namely the
scopal properties of quantificational and indefinite subjects.
42
Although it is not our goal to account for this difference in this paper, we could assume that what
renders (44) more marked is the movement of the DP subject past the Mood marker na. If the
preverbal subject is a Topic and has to move to Spec, TopicP in the left periphery (see Roussou 2000
for a description of a more refined CP domain including TopicP and FocusP), it has to move past
MP, which hosts the marker na as shown above in (43). This is not the case, however, in indicative
complements; as has been argued in Philippaki-Warburton 1994, indicative has a zero exponent
thus the head of MP is empty. The DP subject on its way to Spec,TopicP does not cross any other
element. As shown in (45) the preverbal subject surfaces after the complementizer. When the preverbal
subject surfaces before the complementizer, it is more marked as in the case of subjunctive.
43
This is an oversimplification. In fact, Chomsky (2001) sketches an approach whereby C is always
phi-complete. If C is selected then the T/I that it selects will also be phi-complete and hence check
Case. T/I can be defective only if C is not selected. Thus, there is a relation between tense (assuming
that C is required for independent tense) and phi-(in)completeness.
VASSILIOS SPYROPOULOS
FINITENESS AND CONTROL IN GREEK∗
1. INTRODUCTION
Finiteness has been considered to be the most prominent syntactic property of

clauses because it determines whether certain syntactic operations can apply across
a clause boundary. More specifically, finiteness has been used as a cover term for
the ability of a clause to check nominative case on its subject. In such a way, finite-
ness regulates the distribution of a nominative subject, either an overt DP-subject
or a null subject pro. In a finite clause, the subject is able to check its nominative
case and thus becomes inaccessible to further computation, whereas in a nonfinite
clause the subject cannot check its nominative case, and can either be a PRO result-
ing in control structures or be targeted by a higher probe in raising constructions.
According to standard approaches to finiteness within the Principles & Parameters
and early Minimalist frameworks (Chomsky 1981, 1993, 1995), nominative case is
checked by a finite T (= INFL)1 functional category; T is able to check (nominative)
case (i.e. it is finite) when it is fully specified for Tense and Agreement. In the most
recent minimalist approaches (Chomsky 2000, 2001a, b), nominative case checking
is considered to be the by-product of subject-agreement valuation in T. Such an
approach implies that finiteness is a property of agreement.
However, the correlation between finiteness and control breaks down, when we
consider languages that exhibit control with finite clauses,2 such as Greek.3 Greek lacks
the verbal category of infinitive,4 so complement clauses involve either an indicative or
a subjunctive verb form. Control constructions in Greek typically involve a subjunctive
complement clause. Crucially, Greek subjunctive employs the same verb form as the
indicative, which fully inflects for tense and subject agreement. Thus, morphologically
speaking, subjunctive is as finite as the indicative. If control is a property of a syntactic
category PRO (Chomsky and Lasnik 1993; Martin 1996, 2001; Landau 2000) or the
by-product of multiple theta-role assignment/checking in a movement/Agree opera-
tion (Hornstein 1999, 2000, 2001, 2003, 2005a, b, forthcoming5; Manzini and Rous-
sou 2000), how is control licensed in finite environments such as Greek subjunctive
complement clauses?
There are three possible approaches to the aforementioned issue: (a) maintain the
strict association between control and nonfiniteness and assume that subjunctives in
control environments are syntactically nonfinite, because they involve a defective
T, and that they, thus, license a PRO subject or a movement/Agree operation (Terzi
1992, 1997; Iatridou 1993; Varlokosta 1993, 1994; Roussou 2001; Kapetangianni and
159
© 2007 Springer.
160 VASSILIOS SPYROPOULOS
Seely 2003, this volume); (b) maintain the finiteness of the subjunctive and assume
that control can be established with other null or overt elements as well, such as pro,
overt pronouns, or even DPs, under special syntactic, semantic, or even pragmatic
requirements (Joseph 1992; Philippaki-Warburton 1987; Philippaki-Warburton and
Catsimali 1999; Spyropoulos and Philippaki-Warburton 2001); or (c) maintain both
the finiteness of the subjunctive and the strict association between control and PRO
and assume that PRO can be case-marked6 and that its distribution with respect to
pro and overt DP-subjects derives from the referential properties of both the PRO
and the subjunctive C and T heads (Landau 2004a, 2006).
The evaluation of these approaches boils down to the following question: Is
control solely a property of specific syntactic devices such as the category PRO or
the multiple theta-role assignment/checking in a movement/Agree operation? That
is, what kind of syntactic elements are allowed to be controlled? In this paper, we
argue that Greek subjunctive clauses are always finite, so they check nominative
case on their subjects even in obligatory control (OC) constructions. We show that
control can be attested in constructions where no PRO (or its movement/Agree
equivalent) can be licensed, such as control over an overt pronoun or a DP-sub-
ject, or even over an object clitic. We also argue that the control pattern is mainly
determined by the licensing of the temporal properties of the subjunctive comple-
ment and also by the semantic requirements of the main predicate. Thus, we put
forward the hypothesis that, in order to account for cases of finite control, the
devices that give rise to control should be enriched with other syntactic dependen-
cies or even semantic and pragmatic requirements, which crucially do not require
the nonfiniteness of the embedded clause (see also Roussou 2005, in preparation).
2. GREEK SUBJUNCTIVES: STRUCTURE, DISTRIBUTION,

AND TEMPORAL PROPERTIES
In Greek, the indicative vs. subjunctive distinction is not marked on the verbal
inflection. Indicative and subjunctive share the same verb forms. This indicates
that both moods involve a functional category T in their clause structure with the
same feature specification to which the verb form moves overtly. Subjunctive is
marked by the subjunctive particle na/as which occupies a M(ood) functional cat-
egory. In addition, indicative and subjunctive select for different negation particles
(ðe for the indicative – mi for the subjunctive) and subjunctive is incompatible
with the future particle qa (1). We adopt the morphosyntactic structure illustrated
in (2):7
(1) a. indicative
(ðe) (θa) erθis
NEG FUT come-2SG
b. subjunctive
na (min) erθis
SUBJ NEG come-2SG
FINITENESS AND CONTROL IN GREEK 161
(2) [CP C [MP M [NEGP NEG [FUTP FUT [TP T [VPV ]]]]]]
a. indicative: (oti) Ø (δen) (θa) γrafi ti

that not will be writing
b. subjunctive: Ø na/as (min) γrafi ti

SUBJ not be writing
The subjunctive in Greek can be found both in main and embedded (complement
and adjunct) clauses:8
(3) a. main clause

na erθi (o janis)
SUBJ come-3SG the John-NOM
‘John/he should come’ or ‘Let John come’
b. complement clause
elpizo na erθi (o janis)
hope-1SG SUBJ come-3SG the John-NOM
‘I hope that John/he will come’
c. adjunct clause
efiγan protu na erθi (o janis)
leave-3PL before SUBJ come-3SG the John-NOM
‘They left before John/he came’
In this paper, we focus on subjunctive complement clauses. There are three types
of subjunctive complement clauses with respect to their temporal properties:9
(a) independent subjunctives (IS), with full temporal properties; (b) dependent
subjunctives (DS), which exhibit a fixed temporal reference; and (c) anaphoric subjunc-
tives (AS), with no temporal properties at all.
The main property of IS is that there is no restriction on the tense morphol-
ogy of the verb form. Thus, independent subjunctive complements can employ
both past and nonpast verb forms. In addition, this tense morphology is fully
referential, so that the past forms denote anterior time reference, whereas the
nonpast verb forms denote either simultaneous or posterior time reference:
(4) a. past morphology → anterior

elpizo na eftase soos
hope-1SG SUBJ come-PAST.3SG safe-NOM
‘I hope that he arrived safe’
b. nonpast morphology → simultaneous/posterior

elpizo na ftasi soos
hope-1SG SUBJ come-NONPAST.3SG safe-NOM
‘I hope that he arrives/will arrive safe’
This kind of subjunctive is selected by predicates such as elpizo ‘hope’, perimeno

‘wait, expect’, pistevo ‘believe’, etc.
(5) a. elpizo na erθi

hope-1SG SUBJ come-3SG
‘I hope that he will come’
b. perimeno na erθi
wait-1SG SUBJ come-3SG
‘I am waiting for him to come’
c. pistevo na erθi
believe-1SG SUBJ come-3SG
‘I believe that he will come’
Not surprisingly, the subjunctive in such complement clauses can be substituted

for by the indicative:
(6) a. elpizo oti θa erθi

hope-1SG that will come-3SG
‘I hope that he will come’
b. perimeno oti θa erθi

wait-1SG that will come-3SG
‘I am waiting for him to come’
c. pistevo oti θa erθi

believe-1SG that will come-3SG
‘I believe that he will come’
DS are selected by predicates such as qelo ‘want’, protimo ‘prefer’, kataferno

‘manage’, kanonizo ‘arrange’, prospaqo ‘try’, parotrino ‘urge’, anagazo ‘force’, piqo
‘persuade’, zitao ‘ask’, apagorevo ‘forbid’, apofasizo ‘decide’, skopevo ‘intend’,
ðiatazo ‘order’, epitrepo ‘allow’, ipoxreono ‘oblige’, leo ‘tell’, ðiatazo ‘order’, sim-
vulevo ‘advise’, enqarino ‘encourage’, voiqao ‘help’, afino ‘let’. Their main char-
acteristic is that they have a fixed temporal reference which is imposed by the
semantics of the matrix predicate and it is mainly future-oriented (irrealis) or,
sometimes, simultaneous with respect to the matrix one. As a reflex, the verb form
cannot inflect for past morphology (7a) and it cannot be modified by a past tense
adverb (7b):10
(7) a. o janis epise ti maria na erθi /*irθe

the John-NOM persuaded-3SG the Maria-ACC SUBJ come-NONPAST.3SG/come-PAST.3SG
‘John persuaded Mary to come’
b. o janis epise ti maria na erθi

the John-NOM persuaded-3SG the Maria-ACC SUBJ come-NONPAST.3SG
tora /avrio /*xtes
now/tomorrow/yesterday
‘John persuaded Mary to come’
AS are selected by predicates such as ksero ‘know how’, tolmo ‘dare’, maqeno ‘learn
how’, ksexnao ‘forget’, qimame ‘remember’, arxizo ‘begin’, stamatao ‘stop’, sinexizo
‘continue’, vlepo ‘see’, akuo ‘hear’, etc. The event time of AS is identical with the
matrix one, which means that the temporal reference of AS is anaphoric to that of
the matrix clause. As a consequence AS cannot be modified by a temporal adverbial
that is future- oriented with respect to the matrix clause temporal reference:
(8) a. i zoii emaθe na kolimbai (*avrio)

the Zoe-NOM learned-3SG SUBJ swim-3SG tomorrow
‘Zoe has learned how to swim (*tomorrow).’
b. o nikos kseri na xorevi tsamiko (*avrio)

the Nikos-NOM know-3SG SUBJ dance-3SG tsamiko tomorrow
‘Nikos knows how to dance tsamiko (Greek folk dance) (*tomorrow).’
c. iða to jani na erxete (*avrio)

saw-1SG the John-ACC SUBJ come-3SG tomorrow
‘I saw John coming (*tomorrow).’
The anaphoric temporal reference is also reflected on the aspectual restrictions

imposed on AS. Thus, the embedded verb form cannot vary for aspect, and it can
be either imperfective or perfective depending on the selectional requirements of
the matrix predicate:
(9) kseri na kolimbai /*kolimbisi

know-3SG SUBJ swim-IMPERF.NONPAST.3SG/swim-PERF.NONPAST.3SG
‘He knows how to swim’
3. CONTROL IN GREEK
Subjunctive complements may be either controlled or noncontrolled. In noncon-

trol situations the null subject of the subjunctive complement is able to establish
independent reference, even if it shares the same features with a potential controller
in the matrix clause. On the other hand, controlled subjunctives may exhibit either
exhaustive or partial control.11 In partial control situations, although the null sub-
ject is interpreted as coreferent with a potential controller in the matrix clause when
they share the same feature specification, control can be suspended by means of an
overt DP-subject or a strong pronominal subject. In addition, given the appropriate
context, a null subject is able to establish independent reference when it has different
feature specification from that of the potential controller.
Thus, subjunctive complements exhibit a tripartite control pattern: no control
– partial control – exhaustive control. Interestingly, this control pattern coincides
to a great extent with the temporal properties of subjunctive complements as
described above.12 Thus, IS exhibit no control, DS partial control, and AS exhaus-
tive control.
3.1 Pattern 1: no control
IS typically exhibit no-control properties. Such clauses involve either an overt

nominative DP-subject (10a) or a null subject13 (10b–d). Null subjects in IS are
not controlled: Although they can be coreferent with a matrix clause argument,
as in example (10b), they can also establish independent reference as in examples
(10c, d). Crucially, in example (10d) the null subject can establish its own refer-
ence, even if it has the same feature specification as a potential controller in the
matrix clause:
(10) a. i maria elpizi na ftasi o janis stin ora tu

the Maria hope-3SG SUBJ arrive-3SG the John-NOM in-the time his
‘Maria hopes that John will arrive in time’
b. ta peðiai elpizun na ftasun [eci] stin ora tus

the children-NOM hope-3PL SUBJ arrive-3PL in-the time their
‘The children hope to arrive in time’
c. i mariai elpizi na ftasume [ecj] stin ora mas

the Maria-NOM hope-3SG SUBJ arrive-1PL in-the time our
‘Maria hopes that we will arrive in time’
d. o janisi elpizi na ftasi [eci/j] stin ora tu

the John-NOM hope-3SG SUBJ arrive-3SG in-the time his
‘John hopes to arrive in time’ or ‘Johni hopes that hej will arrive in time’
3.2 Pattern 2: partial control (PC)
DS typically involve a controlled null subject:
(11) a. o nikosi prospaθise na fiγi [eci/*j]

the Nikos-NOM tried-3SG SUBJ leave-3SG
‘Nikos tried to leave’
b. i maria anagase to nikoi na fiγi [eci/*j]

the Maria-NOM forced-3SG the Nikos-ACC SUBJ leave-3SG
‘Mary forced Nikos to leave’
However, the control pattern attested in these cases does not have the properties
of exhaustive control.14 First, partial control is permitted (12):
(12) episa ti mariai na pane [eci+] ja psonia tin triti

persuade-PAST.1SG the Maria-ACC SUBJ go-3PL for shopping the Tuesday-ACC
‘I persuaded Maria that they should go for shopping on Tuesday’
Second, the controlled null subject can take split antecedents (13):
(13) i mariai epise to janij na pane [eci+j] ja psonia tin triti

the Mary-NOM persuade-PAST.3SG the John-ACC SUBJ go-3PL for shopping the Tuesday-ACC
‘Mary persuaded John that they go shopping on Tuesday’
Moreover, Control Shift is also attested. Example (13) involves a violation of the
Minimal Distance Principle on control (Rosenbaum 1967):
(14) i mariai epise [tus γonis tis]j na pai [eci/*j] sto parti
the Mary-NOM persuade-PAST.3SG the parents-ACC her SUBJ go-3SG to-the party
‘Mary persuaded her parents that she can go to the party’
But, more significantly, an overt DP-subject/pronoun can also be licensed in the

embedded subjunctive.15 Thus, example (15a) involves backward control over an
overt DP-subject; in example (15b) the subjunctive complement has a controlled
overt (strong) subject pronoun:
(15) a. akoma ki an prospaθuse [eci] [na fiγi o janisi apo to xorio…]

even and if try-PAST.3SG SUBJ leave-3SG the John-NOM from the village
‘Even if John has tried to leave the village…’
b. o janisi prospaθuse na fiγi aftosi apo to xorio

the John-NOM try-PAST.3SG SUBJ leave-3SG he-NOM from the village
‘John was trying to leave the village’
Furthermore, the licensing of an overt DP-subject may lead to control suspension:16
(16) a. o kir-jorγos matea prospaθuse na ðioristi

the Mr-George-NOM in vain was trying SUBJ be appointed-3SG
o jos tu stin trapeza
the son-NOM his to.the bank
‘Mr George was trying in vain for his son to be appointed by the bank’
b. o janis epise ti maria na ti sinoðepsi o vasilis

the John-NOM persuaded-3SG the Mary-ACC SUBJ her escort-3SG theVassilis-NOM
sto parti
to.the party
‘John persuaded Mary that Vassilis will escort her to the party’
It should be noticed, that given the appropriate context, control can also be
suspended, even when the subject of the embedded subjunctive is null, with the
condition that there is a possible prominent antecedent in the discourse. The only
formal condition is that the null subject of the subjunctive complement have a
different feature specification from that of its potential controller.
(17) a. o janis prospathise na erθun, ala afti ðen ta kataferan

the John-NOM try-PAST.3SG SUBJ come-3PL but they-NOM NEG them.CL manage-3PL
‘John tried for them to come, but they didn’t make it’
b. i γonis tu jani episan tin epitropi

the parents-NOM the John-GEN persuade-PAST.3SG the committee-ACC
na apalaxti apo ta kaθikonda tu
SUBJ relieve-PASS.NONPAST.3SG from the duties his
‘John’s parents persuaded the committee for him to be relieved of his duties’
In example (17a) the null subject of the subjunctive complement has a 3PL
specification which is different from the 3SG specification of its potential controller
o janis ‘John’ and, thus, it can establish independent reference. In example (17b)
the null subject of the subjunctive complement picks up a non-c-commanding

antecedent, which is included in the potential matrix controller.
There is a variation with respect to the availability of control suspension, which
depends on the semantics of the matrix predicate. Predicates such as qelo ‘want’
permit control suspension more easily (18), whereas with predicates that express
advice/order or intention, such as ðiatazo ‘order’ or skopevo ‘intend’ control sus-
pension seems not to be an option (19).17 However even in the latter case, split-
antecedents and partial control are allowed (20), which is a property of PC:
(18) a. θelo na erθis

want-1SG SUBJ come-2SG
‘I want you to come’
b. θelo na erθi o janis

want-1SG SUBJ come-3SG the john-NOM
‘I want John to come’
(19) a. skopevo na erθo /*erθis /*erθi (*i maria)

intend-1SG SUBJ come-1SG/come-2SG/come-3SG the Maria-NOM
‘I intend to come’
b. i maria ðietakse to niko na tis feri

the maria-NOM order-PAST.3SG the nikos-ACC SUBJ her bring-3SG
[eci/*j]/aftosi/*j/*o janis ena potiri nero
he-NOM/the John-NOM a glass-ACC water
‘Maria ordered Nikos to bring her a glass of water’
(20) a. o janisi mu ipe oti i mariaj skopevi na pane [eci+j]

the John-NOM me-GEN tell-PAST.3SG that the Maria-NOM intend-3SG SUBJ go-3PL
ja psonia tin triti
for shopping the Tuesday-ACC
‘John told me that Maria intends that they go shopping on Tuesday’
b. o loxaγos ðietakse to loxiai na epiteθun [eci+]

the lieutenant-NOM order-PAST.3SG the sergeant-ACC SUBJ attack-3PL
ke na katalavun [eci+] to lofo
and SUBJ occupy-3PL the hill-ACC
‘The lieutenant ordered the sergeant that they attack and occupy the hill’
PC in Greek DS seems to be similar to the no-control pattern of IS. In fact it can be

argued that the difference between the two is due to pragmatic load only, in the sense
that the control pattern of DS looks like a stricter version of the no-control pattern of
IS. However, for the purposes of this paper we will maintain the distinction between
PC and no control on the grounds of the interpretative options of the subjunctive null
subject when it has the same feature specification as a matrix potential controller:
(21) a. o janisi elpizi na ftasi [eci/j] noris

the John-NOM hope-NONPAST.3SG SUBJ arrive-3SG early
‘John hopes that s/he will arrive early’
b. o janisi prospaθise na ftasi [eci/*j] noris

the John-NOM try-PAST.3SG SUBJ arrive-3SG early
‘John tried to arrive early’
Example (21a) involves an IS complement. The null subject of the subjunctive

complement has the same feature specification as the subject of the matrix clause
and it can either corefer with it or establish independent reference. On the con-
trary, the null subject of the DS complement in example (21a) cannot establish
independent reference.
3.3 Pattern 3: exhaustive control (EC)
AS exhibit EC that cannot tolerate obviation, partial control, or split antecedents.
(22) a. i zoii emaθe na kolimbai [eci/*j]

the Zoe-NOM learned-3SG SUBJ swim-3SG
‘Zoe learned PRO/*him to swim.’
b. *i zoii emaθe na kolimbane [eci+]

the Zoe-NOM learned-3SG SUBJ swim-3PL
‘*Zoe learned himself and others to swim.’
c. o janisj ipe oti i zoii emaθe na kolimbane [ec*j+i]

the John-NOM say-PAST.3SG that the Zoe-NOM learned-3SG SUBJ swim-3PL
‘*John said that Zoe has learned them [John and Zoe] to swim.’
Significantly, even in such constructions, an overt DP-subject or a strong subject

pronoun can be licensed in the subjunctive complement, with the condition that
they are controlled.18 Example (23b) involves a controlled strong subject pronoun.
Examples (23c, d) are instances of backward control, with a controlled strong
pronoun (23c) and a controlled overt DP-subject (23d):
(23) a. o janisi kseri na xorevi [eci/*j] kalo tsamiko

the John-NOM know-3SG SUBJ dance-3SG good tsamiko-ACC
‘John knows how to dance tsamiko (folk Greek dance) well’
b. o janisi kseri na xorevi ki aftosi/*j kalo tsamiko

the John-NOM know-3SG SUBJ dance-3SG and he good tsamiko-ACC
‘John knows how to dance tsamiko well too’
c. kseri proi na xorevi aftosi/*j kalo tsamiko?

know-3SG SUBJ dance-3SG he-NOM good tsamiko-ACC
‘Does he know how to dance tsamiko well?’
d. kseri proi/*j na xorevi o janisi kalo tsamiko?

know-3SG SUBJ dance-3SG the John-NOM good tsamiko-ACC
‘Does John know how to dance tsamiko well?’
3.4 The correlation among temporal properties, control, and finiteness
The facts presented above show that Greek exhibits a three-scale control pat-
tern which coincides with the temporal properties of the subjunctive com-
plement. It can be therefore concluded that the temporal properties of the
subjunctive license the control pattern. In this spirit, Iatridou (1993) and
Varlokosta (1994) have proposed that the lack of certain temporal properties
renders the subjunctive complement nonfinite. It has, therefore, been argued
that in control situations the T of the subjunctive clause is defective, so it
cannot check nominative case and as a consequence a PRO is licensed as the
subject of the subjunctive clause. Thus, the correlation between the lack of
temporal properties and control is proposed to be indirect and to be mediated
by the nonfiniteness of the subjunctive clause. These approaches maintain the
strict association between control and nonfiniteness, which is also found in the
defective T analyses proposed by Kapetangianni and Seely (2003, this volume)
and Terzi (1992, 1997). According to such an association, given that control is
a property of a special-category PRO or the by-product of multiple theta-role
checking in a movement/Agree relation, Greek subjunctive clauses in control
situations should be nonfinite and involve a defective T that cannot check
nominative case.
In what follows, we argue for a direct correlation between the licensing of the
temporal properties of the subjunctive clause and the control pattern and for the
dissociation of control from nonfiniteness. In other words, we claim that non-
finiteness is not a requirement for control and we argue that Greek subjunctive
clauses in control situations are finite in the sense that they involve a T which is
able to check nominative case on its subject.
4. SUBJUNCTIVE AND FINITENESS
In order to determine the finiteness of subjunctive complements in control

situations, we need independent evidence that shows that the subject of the
subjunctive complement clause has a nominative case feature checked. Such
evidence derives from two tests (Philippaki-Warburton and Catsimali 1999;
Spyropoulos and Philippaki-Warburton 2001); (a) the distribution of overt
nominative subjects and (b) the case agreement between the subject and its
nominal predicate modifiers.
4.1 Distribution of overt subjects
The first piece of evidence comes from the distribution of overt DP-subjects and
strong pronominal subjects in control subjunctive complements. We have already
mentioned in sections 3.2 and 3.3 that an overt nominative subject can appear in
both DS and AS, which exhibit PC and EC respectively:
(24) DS = PC
a. akoma ki an prospaθuse [eci] [na fiγi o janisi apo to xorio . . .]

even and if try-PAST.3SG SUBJ leave-3SG the John-NOM from the village
‘Even if John has tried to leave the village . . .’
b. o janisi prospaθuse na fiγi aftosi apo to xorio

the John-NOM try-PAST.3SG SUBJ leave-3SG he-NOM from the village
‘John was trying to leave the village’
c. o janisi epise ti maria na ti sinoðepsi o Vasilis

the John-NOM persuaded-3SG the Mary-ACC SUBJ her escort-3SG the Vassilis-NOM
sto parti
to.the party
‘John persuaded Mary that Vassilis will escort her to the party’
(25) AS = EC
a. o janisi kseri na xorevi ki aftosi/*j kalo tsamiko

the John-NOM know-3SG SUBJ dance-3SG and he good tsamiko-ACC
‘John knows how to dance tsamiko well too’
b. kseri proi na xorevi aftosi/*j kalo tsamiko?

know-3SG SUBJ dance-3SG he-NOM good tsamiko-ACC
‘Does he know how to dance tsamiko well?’
c. kseri proi/*j na xorevi o janisi kalo tsamiko?

know-3SG SUBJ dance-3SG the John-NOM good tsamiko-ACC
‘Does John know how to dance tsamiko well?’
The examples in (24) illustrate the licensing of an overt nominative subject in

DS which exhibit PC. Example (24a) involves an overt nominative subject in the
subjunctive complement, which results in backward control. In example (24b) a
strong nominative pronoun, which is coreferent with the matrix subject o janis
‘John’, is licensed as the subject of the subjunctive complement. In example (24c)
control is suspended because the nominative DP o vasilis ‘Vassilis’ is licensed as
the subject of the subjunctive complement.
The examples in (25) illustrate the licensing of an overt nominative subject
in AS, which exhibit EC. In example (25a) a strong nominative pronoun, which
is controlled by the matrix subject o janis ‘John’, is licensed as the subject of the
subjunctive complement. In examples (25b, c) the licensing of a strong nominative
subject pronoun and of an overt nominative DP-subject in the subjunctive complement
results in backward-control constructions.
The licensing of overt nominative subjects in both DS and AS shows that the
T functional head in these clauses is finite and able to check nominative case, even
in EC situations.19 Significantly, given that AS constitute the most defective type of
subjunctive, since they can vary neither for tense nor for aspect and they lack tempo-
ral reference, the licensing of an overt nominative subject in such clauses constitutes
a strong piece of evidence against the association of the lack of temporal properties
with nonfiniteness as proposed by Iatridou (1993) and Varlokosta (1994).
Our conclusion is reinforced by the distribution of overt nominative subjects in the

subjunctive complement of perception verbs such as akuo ‘hear’, vlepo ‘see’. Percep-
tion verbs in Greek take AS complements in a construction which looks like ECM:20
(26) akusa kapion /afton /to jani na aneveni tis skales

hear-PAST.1SG someone-ACC/he-ACC/the John-ACC SUBJ come.up-3SG the stairs-ACC
‘I heard someone/him/John coming up the stairs’
There are good reasons to believe that (26) is not a pure ECM construction and
that the subjunctive complement involves a null subject which is coreferent with
the accusative DP. Leaving aside the issue of ECM-like constructions in Greek,21
what is crucial is that (26) may be alternatively stated as (27) with an overt nomi-
native DP/pronoun licensed as the subject of the AS complement:22
(27) akusa na aneveni kapios /aftos /o janis tis skales

hear-PAST.1SG SUBJ come.up-3SG someone-NOM/he-NOM/the John-NOM the stairs-ACC
‘I heard someone/him/John coming up the stairs’
Example (27) cannot be accounted for by any defective T approach, especially

those that relate finiteness to temporal properties. It cannot be the case that akuo
selects for a subjunctive with defective T in (26) and for subjunctive with a com-
plete T in (27). In both examples the AS complement has no temporal properties,
yet a nominative subject is able to be licensed in example (27).
4.2 Case agreement
In Greek, nominal predicates and predicative nominal modifiers obligatorily agree in

case with the element they are predicated of/modify (Spyropoulos 1998, 1999, 2005a):
(28) a. o janis ine eksipnos /*eksipno

the John-NOM is clever-NOM/clever-ACC
‘John is clever’
b. θeoro to jani eksipno /*eksipnos

consider-1SG the John-ACC clever-ACC/clever-NOM
‘I consider John clever’
c. o janis efiγe telefteos/*telefteo

the John leave-PAST.3SG last-NOM/last-ACC
‘John was the last one who left’
d. i maria xeretise to jani telefteo /*telefteos

the Maria-NOM greet-PAST.3SG the John-ACC last-ACC/last-NOM
‘John was the last one who Maria greeted’
In control subjunctives nominal predicate modifiers that modify the null subject
always appear in nominative case.
(29) a. o janisi prospaθise na fiγi [eci] telefteosi/*telefteoi (DS)

the John-NOM try-PAST.3SG SUBJ leave-3SG last-NOM/last-ACC
‘John tried to be the last one to leave’
b. o janisi stamatise na piγeni [eci] monosi /*monoi sinema (AS)

the John-NOM stop-PAST.3SG SUBJ go-3SG alone-NOM /alone-ACC cinema
‘John has stopped going to the movies alone’
The following set of data show that such modifiers appear in nominative, even if
the potential controller of the subjunctive null subject appears in accusative. We
claim that this is a strong indication that the controlled null subject in the subjunc-
tive clause has a nominative case feature checked.
(30) a. i mariai epise to janij na fiγi [ecj] telefteosj/*telefteoj

the Maria-NOM persuade-PAST.3SG the John-ACC SUBJ leave-3SG last-NOM/last-ACC
‘Maria persuaded John to leave last’
b. i mariai voiθise to janij na fiγi [ecj] telefteosj/*telefteoj

the Maria-NOM help-PAST.3SG the John-ACC SUBJ leave-3SG last-NOM/last-ACC
‘Maria made John leave last’
The nominal predicate modifier telefteos ‘last’, which modifies the controlled null
subject of the subjunctive clause, appears in nominative case, although the control-
ler in the higher clause is in the accusative case. Such a fact indicates the existence
of a null subject in nominative case inside the subjunctive clause; otherwise, the
source of the nominative case of the predicate modifier receives no explanation.23
Additional evidence comes from the examples in (31):
(31) a. iða to janii na aneveni tis skales monosi /*monoi

see-PAST.3SG the John-ACC SUBJ come.up-3SG the stairs-ACC alone-NOM/alone-ACC
‘I saw John coming up the stairs alone’
b. iða to jani na ine stenoxorimenos/*stenoxorimeno

see-PAST.3SG the John-ACC SUBJ be-3SG sad-NOM /sad-ACC
‘I saw John being sad’
As mentioned above perception verbs take AS complements. In the examples in

(31) the nominal predicate modifiers appear in nominative and not in accusative,
indicating the existence of a nominative null subject in the subjunctive clause.
The significance of the test and its results is strengthened by the behavior of such
elements in an earlier stage of Greek. Classical Greek exhibits control structures with
infinitival clauses. Nominal predicate modifiers that modify the controlled null subject
of the infinitival clause appear in the case of the controller (Andrews 1971; Morrel
1989; see Spyropoulos 2005b, for an analysis of Classical Greek infinitival syntax).
(32) a. kyrou edeonto o:s prothimotatou pros ton polemon genesthai

Cyrus-GEN begged-3SG so most.willing-GEN to the war-ACC become-INF.AOR
‘They were begging Cyrus to become most willing for war’ (X. HG. I.5, 2)
a′. proj edeonto [kyrou-GEN]i [PROi genesthai o:s prothimotatou-GEN]
b. o de:mos synekho:re:sen auto:i pro:to:i graphe:nai

the public-NOM permitted-3SG he-DAT first-DAT write-INF.PASS.AOR
‘The public permitted him to register first’ (Aeschin. III.186)
b′. o de:mos synekho:re:sen [auto:i-DAT]i [PROi graphe:nai proto:i-DAT]
In Classical Greek the infinitive is a nonfinite form; infinitival clauses involve a

defective T with no agreement features which is not able to check nominative case
on the subject. Thus, the nominal predicate modifier appears in the case of the
controlling DP in the matrix clause. In Modern Greek, the nominal predicate modi-
fier always appears in nominative case, which suggests that the controlled null sub-
ject in the subjunctive clause is always able to have its nominative case checked.
4.3 Implications for control analyses
The evidence presented in the previous sections show that Greek subjunctive clauses
are finite and have a nominative case-marked subject even in EC constructions with
AS complements. This is a welcome conclusion given the recent assumptions about
case checking in the Minimalist Program. Chomsky (2000, 2001a, b, 2004, 2005,
2006) has suggested that case checking is a side effect of agreement valuation. In
other words, when a complete [Agr] probe values its features by targeting a nomi-
nal element in its probing domain, this nominal element has the appropriate case
checked as a side effect. Given that Greek subjunctives involve a complete [Agr]
probe on T, we conclude that the valuation of the features of this [Agr] results in the
nominative case checking on its subject. Thus, the licensing of the inflectional prop-
erties of the verb form in subjunctive clauses imposes the nominative case feature
on their subject. Greek subjunctive clauses are finite simply because their verb forms
carry full agreement inflection.24
The finiteness of the subjunctive has certain implications for the analysis of
control in Greek. First, the controlled null subject in PC and EC constructions
cannot be a caseless or a null case-marked PRO. Second, control in Greek cannot
be the result of a movement/Agree operation as Hornstein (1999, 2000, 2001,
2003, 2005a, b, forthcoming) has suggested. According to such an approach, EC
is the by-product of multiple theta-role checking in a movement/Agree operation
in which the shared DP originates as the subject of the embedded clause and it
is then targeted by the relevant probe in the higher clause. In order for such an
operation to be established, the DP should be active, i.e. it should not have had
its case checked. However, the evidence presented above show that in subjective
complements the finite T always checks the nominative case of the subject. As a
consequence, the subject of the subjunctive complement cannot be targeted by a
higher probe, because it is inaccessible to further computation (Chomsky 2000,
2001a). We should also note that most instances of control in Greek involve PC
or NOC. According to Hornstein (1999) the null subject in such constructions is a
pro. Given that pro in Greek has a nominative case feature checked, the noncom-
plementary distribution between the controlled null subject (i.e. pro) and overt
DP-subjects/subject pronouns in these constructions is naturally derived. In the
next section we discuss the nature of the controlled null subject in Greek subjunc-
tives and we put forward the hypothesis that it is a controlled pro.
5. FINITE CONTROL IN GREEK
5.1 The nature of the controlled null subject
The evidence presented above show that the controlled null subject in subjunctive
complements has a nominative case feature. Landau (2004a) claims that such a
controlled null subject in finite environments is a case-marked PRO. He maintains
that control is a property of a special category PRO and he suggests that PRO can be
case-marked and that its distribution is not derived by its case properties, but by the
referential properties of the C and T heads of the clause, which depend on the temporal
properties of the clause, as these are represented on both heads (see also Landau
2006). Leaving aside the technical details of his analysis, let us take a closer look at the
properties of the controlled null subject in both PC and EC environments.
In PC situations the controlled null subject does not exhibit fundamental
properties of PRO. Thus, its controller need not be local:
(33) i mariai epise [tus γonis tis]j na pai [eci/*j] sto parti
the Mary-NOM persuade-PAST.3SG the parents-ACC her SUBJ go-3SG to-the party
‘Mary persuaded her parents that she can go to the party’
In addition, it can take split antecedents. The crucial fact is that the null subject is
able to take split antecedents even in the absence of a collective predicate:
(34) a. i mariai epise to janij na pane [eci+j]

the Mary-NOM persuade-PAST.3SG the John-ACC SUBJ go-3PL
ja psonia tin triti
‘Mary persuaded John that they go shopping on Tuesday’
b. o janisi mu ipe oti i mariaj skopevi na pane [eci+j]

the John-NOM me-GEN tell-PAST.3SG that the Maria-NOM intend-3SG SUBJ go-3PL
ja psonia tin triti
‘John told me that Maria intends that they go shopping on Tuesday’
It also allows for partial control. Crucially, the null subject is both syntactically
and semantically plural, as evident from the plural subject-agreement of the verb
form. A PRO in a partial-control structure can only be semantically plural (see the
discussion in Landau 2000, 2004a: 833–35):
(35) a. episa ti mariai na pane [eci+]ja psonia tin triti

persuade-PAST.1SG the Maria-ACC SUBJ go-3PL for shopping the Tuesday-ACC
‘I persuaded Maria that they should go for shopping on Tuesday’
b. o loxaγos ðietakse to loxiai na epiteθun [eci+]

the lieutenant-NOM order-PAST.3SG the sergeant-ACC SUBJ attack-3PL
ke na katalavun [eci+] to lofo
and SUBJ occupy-3PL the hill-ACC
‘The lieutenant ordered the sergeant that they attack and occupy the hill’
Finally, Varlokosta (1994) has noticed that the null subject in PC constructions
allows for a strict reading under VP-ellipsis and de re interpretation. Putting all these
facts together, we claim that the null subject in PC situations is not a PRO but a
controlled pro. Landau does not deny the existence of a pro in PC situations, and he
claims that ‘F-subjunctives25 whose null subject is co-indexed with a matrix argument
are systematically ambiguous between a pro-structure with accidental coreference
and a PRO-structure with OC’ (2004: 845). Although this is a logical possibility, we
believe that, since coreference is a possible pro interpretation and the fundamental
properties of PRO are systematically absent, there is no empirical reason to maintain
the existence of an EC structure with a case-marked PRO in such constructions,
especially when there are no visible effects of it.
However, the strong argument against the existence of case-marked PRO in
Greek control subjunctives derives from the distribution of overt DP-subjects and
strong subject pronouns in AS. AS have no temporal properties and exhibit EC.
Crucially, as we have already shown, AS involve a finite T which checks nomina-
tive case on its subject. The controlled null subject in AS cannot be a nominative
case-marked PRO, because, according to Landau’s analysis, the licensing of a case-
marked PRO excludes the licensing of overt subject elements in AS complements
(his own C-subjunctives). But, we have already shown that overt DP-subjects/pro-
nouns are also licensed in AS with EC (see the examples in 25 and 27). Since overt
DP-subjects/pronouns have the same distribution as pro, we conclude that the sub-
ject of AS with EC is a controlled pro.
The hypothesis that the controlled null subject of Greek PC and EC subjunc-
tives is a controlled pro breaks the strict association between control and PRO. It
implies that control is not a property of PRO only and that other categories can
also be controlled. This prediction seems to be justified in Greek, since object
clitic pronouns can also be controlled:
(36) anagasa to janii na toni/*j eksetasi o kenurjos mas jatros

forced-1SG the John-ACC SUBJ him examine-3SG the new our doctor-NOM
‘I forced John that our new doctor should examine him’
The verb anagazo ‘force’ takes a DS complement which exhibits PC. In example
(36) the subjunctive complement has an object clitic pronoun which is controlled
by the object of the matrix clause. Since object clitics in Greek are weak pronomi-
nal forms, example (36) illustrates an instance of control over an overt pronomi-
nal in object position. Notice that example (36) may constitute an indirect piece
of evidence for the availability of control over a pro. Object clitics are weak pro-
nominal elements and their declension paradigm lacks nominative forms. It has
been proposed that the nominative form of object clitics is the subject pro, which
is the corresponding weak subject pronoun (e.g. Philippaki-Warburton 1987). If
controlled object clitics are attested as example (36) illustrates, then in certain
environments a controlled pro can be argued to exist.
We therefore propose that Greek subjunctive complements license a pro subject
which can be controlled. The justification of such a hypothesis relies on addressing
the following two issues: (a) Is there independent evidence for the existence of
controlled pro in Greek? (b) How is control derived, since pro by itself is not refer-
entially restricted?
5.2 Evidence for controlled pro in Greek
There are at least two environments where a controlled pro can be argued to exist
in Greek. Both involve an indicative complement clause. If we assume that the
subject of indicative complements is a pro,26 then these constructions constitute
indications for the existence of a controlled pro.
Perception (akuo ‘hear’, vlepo ‘see’) and knowledge (ksero ‘know how’, maqeno
‘learn’) verbs as well as verbs of beginning and continuing (arxizo ‘start’, sinexizo
‘continue’) may take an indicative complement which is introduced by the comple-
mentizer ke and have an obligatorily controlled null subject:
(37) a. akusa to janii [CL ke anevene [eci/*j] tis skales]

hear-PAST.3SG the John-ACC COMP climb-IMPERF.PAST.3SG the stairs
‘I heard John climbing the stairs’
b. o janisi kseri [CL ke xorevi [eci/*j] kala to tsamiko]

the John-NOM know-3SG COMP dance-3SG well the tsamiko-ACC
‘John knows how to dance tsamiko (Greek folk dance) well’
c. o nikosi arxise
the Nikos-NOM start-PAST.3SG
[CL ke estelne [eci/*j] luluðia sti maria]
COMP send-IMPERF.PAST.3SG flowers-ACC to-the Maria
‘John has started sending flowers to Maria’
The complementizer ke is homophonous with the coordinating conjunction. How-

ever, the structures above do not involve coordination since extraction out of the
element introduced by ke is allowed (if coordination were involved, the examples
in (38) would be ruled out):
(38) a. tii kseri o janis [CL ke xorevi kala ti]?

what-ACC know-3SG the John-NOM COMP dance-3SG well
‘What does John know how to dance well’
b. tii arxise o nikos [CL ke estelne ti sti maria]?

what-ACC start-PAST.3SG the Nikos-NOM COMP send-IMPERF.PAST.3SG to-the Maria
‘What has John started sending to Maria’
Since the null subject of an indicative clause is generally assumed to be a pro, we

conclude that the controlled null subject of the embedded clauses in the examples
above is a pro.
Factive complements in Greek are introduced by the complementizer pu and
involve an indicative verb form (Roussou 1994; Varlokosta 1994; Holton et al.
1997). Perception verbs can also take a factive complement clause with a control-
led null subject.
(39) akusa to janii [CL pu anevene [proi/*j] tis skales]

hear-PAST.3SG the John-ACC COMP climb-IMPERF.PAST.3SG the stairs
‘I heard John climbing the stairs’
Given that factive clauses involve an indicative verb form and indicative clauses
license a pro subject, it can be concluded that the null subject of the complement
clause in (39) is a controlled pro.
5.3 Towards an analysis of control in Greek
In this section we argue that finite control derives from the licensing of the temporal
properties of the subjunctive complement. We thus maintain Varlokosta’s and Landau’s
insights that the licensing of the temporal properties of the clause determines the con-
trol pattern. However, we propose that this licensing regulates the referential possibilities
of the element that realizes the ‘subject’ and not the distribution of its type (i.e. PRO,
pro, pronoun, DP). This distribution is regulated by the finiteness of the clause, i.e. the
ability of its T to check nominative case on its subject. Thus, Greek subjunctives are able
to license a pro or an overt DP/pronominal subject since they are finite. It is the licensing
of the temporal properties of the embedded subjunctive which results in the attested
control pattern, i.e. PC or EC.
We assume that the temporal properties of the clause are represented on the T and
C heads of the structure (Stowell 1982; Boškovic´ 1997; Martin 2001; Chomsky 2001;
Pesetsky and Torrego 2001). We further propose that the T head is morphologically
specified, whereas C carries a [Tense] feature which represents the ‘semantic tense’,
i.e. the temporal properties of the clause. The T head and the [Tense] feature of C are
syntactically related by means of a syntactic operation Agree (see also Varlokosta
1994; Pesetsky and Torrego 2001). By means of this operation, the [Tense] feature of
C acquires morphosyntactic manifestation and the T head its interpretation. Agree-
ment features are assumed to be parasitic to the T head, i.e. they are adjoined to T0.
The feature constitution of the three subjunctive types in Greek is presented in (40):
(40) a. Independent subjunctive

[CP C0[uTense] [MP M0 [TP T0max[T0, Agr] … ]]]
b. Dependent subjunctive
[CP C0[iTense] [MP M0 [TP T0max[T0, Agr] … ]]]
c. Anaphoric subjunctive
[CP C0 [MP M0 [TP T0max[T0, Agr] … ]]]
IS carry a [Tense] feature in C which is unvalued, i.e. it is [uTense]. This unvalued

feature formalizes the transparent temporal properties of IS. The [uTense] feature
of C acquires its value by the morphological specification of T. T carries the feature
specification of the corresponding verbal morphology. If, for instance, the verb form
carries past morphology, T is specified for past; the [uTense] feature of C targets the
T, it acquires this [past] specification and the clause is interpreted as having anterior
time reference.
DS carry a [Tense] feature in C which is valued, i.e. [iTense]. Recall that this type
of subjunctive has a fixed temporal reference which is selected by the matrix predi-
cate as either simultaneous or future-oriented. Matching with this T is reflected by
the morphological restrictions imposed on the tense marking of the verb form; thus,
the verb form cannot be marked for past, but only for nonpast morphology.27
AS carry no [Tense] feature in C since they lack temporal properties. The lack of
[Tense] in embedded C forces the T head to be related with the matrix C. Let us assume
that the [Tense] feature of the matrix C targets both the matrix T and the embedded T.
By transitivity, the two T heads match and acquire the same event time.
Let us now consider how the licensing of the temporal properties of the subjunc-
tive complement affects the referential possibilities of the subject in the embedded
subjunctive. In AS the matrix C targets both the matrix and the embedded T, so
that the two T heads match. We propose that the [Agr] features on the embedded T
match the closest [Agr] probe in the matrix clause as a side effect of the matching
between the two T heads.28 As a result, the subject of an AS cannot vary in its fea-
ture specification from its potential controller in the matrix clause since they share
the same feature specification:
(41) [CP C 0 [Tense] [ … F[Agr] … [CP C 0 [MP M0 [TP T 0max[T 0, Agr] . . . ]]]]]
We also assume that an [Agr] bundle of features acquires the reference of the nominal
element with which it enters in an Agree operation in order to value its features.29 An
immediate consequence would be that the [Agr] features of the embedded T should
match the reference of the closest [Agr] in the matrix clause since they also enter
into an Agree relation. The [Agr] probe in the matrix clause has acquired its refer-
ence by targeting the relevant DP, i.e. the potential controller. In the subjunctive
complement the [Agr] probe of T targets the subject, checks its nominative case,
and acquires its reference. Since the two [Agr] probes match, the result is obliga-
tory coreference, i.e. EC.
(42) … DPi F[Agri ] … [CP C 0 [MPM 0 [TP T 0max[T 0, Agri ] subjecti . . . ]]]
In IS and DS, the complement T is licensed within its clause, since it establishes
a relation with the [Tense] feature of its C head and acquires its interpretation.
Thus, no Agree operation is established with the matrix clause C or T heads
and, consequently, the [Agr] probe of the subjunctive complement T does not
necessarily match a matrix [Agr] probe. As a result, no referential restrictions are
imposed on the [Agr] probe of T in IS and DS complements. It can, therefore,
be argued that no syntactic control is attested in these constructions. The null
subject in IS behaves exactly as the null subject of an indicative clause (see also
the discussion in section 3.1) and it exhibits the referential properties of a pro-
nominal element. In DS the null subject also behaves like a pronominal, but its
reference is more restricted than in IS (see the discussion in section 3.2). Recall
that, in DS, the embedded T is licensed within its clause, but the [Tense] feature
of the C is selected by the matrix predicate. We may suggest that this kind of
selection restricts the reference of the subject of the subjunctive and expresses
syntactically the control property of a predicate. That is we claim that in DS
the PC pattern is apparent and it is the result of certain semantic postulates or
pragmatic requirements imposed on the embedded subject by the matrix predi-
cate when this is a control predicate. Evidence for this hypothesis comes from the
following facts, which were presented and described throughout the paper: (i) the
control pattern exhibited in DS does not have the properties of syntactic control
and resembles pronominal reference; (ii) control suspension is always an available
option, even with null subjects, when certain pragmatic and discourse require-
ments are met; and (iii) the availability of control suspension is determined by
the semantic properties of the matrix predicate as a control predicate (see the
discussion in section 3.2). The exact formulation of such semantic postulates
and pragmatic requirements is a very interesting and complicated issue, which
we leave open for future research (see Huang 2000). What is crucial is that syntax
permits such a behavior by allowing the licensing of a null subject pro or an overt
DP-subject in Greek subjunctive clauses.
6. CONCLUSIONS
The detailed examination of the properties of subjunctive clauses in Greek has

shown that they fall into three types with respect to their temporal properties
and each of these types is associated with a certain control pattern. The main
control pattern is that of PC, and EC is very limited and occurs only with AS.
We maintain the basic insight of Varlokosta’s (1994) analysis for Greek, namely
that the control pattern is related to the temporal properties of the subjunc-
tive clause. Nevertheless, we go a step further to suggest that it is the licensing
of the temporal properties that regulates the control pattern. Such a proposal
has the advantage that it derives the attested pattern without having to assume
the nonfiniteness of subjunctive T in Greek, which is an undesirable specula-
tion on both empirical and theoretical grounds. As a consequence, we maintain
and strengthen the assumption that case is the by-product of agreement valua-
tion. Greek subjunctive shows full subject-agreement and the valuation of these
agreement features by the subject has the side effect of nominative case checking.
Because of this, control in Greek subjunctives cannot be accounted for by a PRO
or a movement/Agree approach. Crucially, the attested control pattern does not
exhibit the properties of control as predicted by these approaches. Thus, Greek
is a language where other syntactic mechanisms30 (licensing of temporal proper-
ties) and semantics/pragmatics reproduce control effects. We suggest that this
constitutes direct evidence for a generalized reductionist approach to control,
according to which the control module should be decomposed and its effects
be accounted for by other modules of the computational system and the gram-
mar in general. By this, we claim that control is not a property only of a special
category PRO or of a specific movement/Agree operation. Such mechanisms
of control surely exist and crucially obey certain conditions and principles. It is
when these conditions are not met, i.e. when the embedded clause is finite, that
the pattern of control obviates the expected properties and other mechanisms
come into play so as to derive the control effects.
*
This paper is an extension of joint work with Irene Philippaki-Warburton, different versions of
which have been presented at the 4th and 5th International Conferences on Greek Linguistics
and at the Workshop on Greek Syntax: The Minimalist Seduction. Many of the ideas developed
here build on the discussion and the issues raised there. This paper has greatly benefited from the
comments of the two editors and one anonymous reviewer. Idan Landau has commented on an
earlier draft of the paper, and his detailed and constructive criticism is kindly acknowledged. We
also thank George Kotzoglou, Anthi Revithiadou, Anna Roussou and Spyridoula Varlokosta for
comments and discussion of the data and the issues raised in the paper. Any errors are our own
responsibility.
Contact information: Department of Mediterranean Studies, 1 Demokratias Av., Rhodes 85100,
Greece. Tel.: + 302241099343, Fax: + 302241099327, E-mail: spiropoulos@rhodes.aegean.gr
1
For the sake of consistency, we will be using the T notation throughout the paper as an equivalent
to the INFL notation.
2
This correlation breaks down for subject-to-subject raising and ECM constructions too. In this
paper, we focus on control and we simply note that Greek lacks subject-to-subject raising as well as
pure ECM constructions (see Hadzivassiliou et al. 2000; Kotzoglou 2002; Philippaki-Warburton
and Spyropoulos 2002; Kotzoglou and Papangeli, this volume).
3
See Philippaki-Warburton 1987, Terzi 1992, 1997, Iatridou 1993, Varlokosta 1993, 1994,
Philippaki-Warburton and Catsimali 1999, Roussou 2001, 2005, in preparation, Spyropoulos and
Philippaki-Warburton 2001. Additionally, finite control is also widely attested in Balkan languages
(see Terzi 1992; Dobrovie-Sorin 1994, 2001; Krapova and Petrov 1999; Krapova 2001; Alboiu
2004a, this volume ).
4
See Joseph 1983, Joseph and Philippaki-Warburton 1987, Holton et al. 1997.
5
Hornstein 2001, 2003 and Boeckx and Hornstein 2003, 2004. For a criticism see Culicover and
Jackendoff 2001 and Landau 2003, 2006, this volume.
6
See also Sigurðsson (1991).
7
See the works by Veloudis & Philippakai-Warburton (1983), Philippakai-Warburton(1994,
1998),Philippakai-Warburton & Spyropoulos (1999, 2004); cf. Roussou 2000 who suggests a Split-
C-Hypothesis with the na particle moving to the lower C head.
8
See Holton et al. 1997 for a complete list of the functions of subjunctive in Greek.
9
Such a tripartite distinction of subjunctive complement clauses with respect to their temporal prop-
erties has also been proposed by Householder et al. (1967), Ingria (1981), Varlokosta (1994) and
Roussou (2004).
10
The verb form in DS may vary with respect to aspect (perfective – imperfective – perfect).
11
There is an issue as to whether Greek controlled subjunctives exhibit the distinction between Exhaus-
tive vs. Partial Control or the more general NOC vs. OC. In sections 3.2 and 3.3, four main diagnos-
tics are used to distinguish between the two attested patterns in Greek controlled subjunctives, namely
(a) partial control, (b) split antecedents, (c) control shift, and (d) control suspension. Given that these
diagnostics have been argued to actually distinguish between Partial vs. Exhaustive Control (Landau
2000, 2001, 2002, 2003, 2004a, b, 2006), we will follow this terminology for the purposes of the paper.
However, we would like to note that there is some uncertainty about where exactly Partial Control leaves
off and NOC starts, in the sense that in many cases they seem to fall together. In addition, it seems quite
hard to develop diagnostics for NOC. Nevertheless, there seem to be some indications that what we refer
to as Partial Control in Greek in this paper is closer to NOC. These are the following: (a) the availability
of a noncontrolled null subject in the subjunctive, although the null subject should have distinct feature
specification (which is overtly manifested on the subjunctive verb form ending) from a potential con-
troller (17a); (b) the availability of control by a non-c-commanding antecedent (17b); and (c) that the
controlled null subject in such constructions can take split antecedents and tolerate partial control with
all predicates, i.e. it need not be the subject of a collective predicate or of a predicate that semantically
forces split antecedents and partial control readings (see the discussion about examples (34) and (35) in
section 5.1). Thanks to the two editors for bringing this issue to our attention.
12
There are a handful of predicates which seem to fall in between these subjunctive and control types. The
most representative example is qelo ‘want’. The subjunctive complement of this verb is semantically a
DS, since it has a fixed temporal reference. Nevertheless, it is possible to inflect the verb form of the
subjunctive complement for imperfective past morphology, in counterfactual statements in which qelo
appears in the conditional tense. In addition, suspension of control is much easier with this predicate
than with all the other predicates that take a DS complement. However, we do not classify it under the
no-control pattern because the null subject of its subjunctive complement can hardly establish inde-
pendent reference when it shares the same feature specification with a potential matrix controller.
13
We use the notation [ec] for the null subject in embedded subjunctives, whenever we make pretheo-
retical observations about the relevant examples and the status of this null subject. The position of
null subjects in Greek is after the verb (Philippaki-Warburton 1987, Alexiadou and Anagnostopoulou
1998, Spyropoulos and Philippaki-Warburton 2002).
14
Kapetangianni and Seely (2003, this volume) question the correlation between DS and PC (=NOC),
based on constructions where DS cannot tolerate control suspension. They claim that such
constructions involve matrix predicates such as leo ‘tell’, epitrepo ‘allow for’, ipoxreono ‘oblige’,
ðiatazo ‘order’, simvulevo ‘advise’, enqarino ‘encourage’. First, we disagree with this observation,
since most of these predicates can participate in constructions that allow for control suspension
(especially under the appropriate pragmatic and discourse situation):
(i) a. ipa sto jani na erθi i maria

tell-PAST.1SG to-the John-ACC SUBJ come-3SG the Maria-NOM
‘I told John that Maria should come’
b. i niki epetrepse sto jani na erθi ke i maria sto parti

the Niki-NOM allow-PAST.3SG to-the John-ACC SUBJ come-3SG and the Maria-NOM to-the party-ACC
‘Niki allowed to John that Maria can come to the party’
c. o proeðros θa ipoxreosi tin kivernisi na paretiθun tris ipurγi

the president-NOM FUT oblige-3SG the government SUBJ resign-3PL three ministers-NOM
‘The president will force the government so that three ministers resign’
Second, as we will show below, even when control is not suspended, the controlled null subject fails
to show exhaustive control (see Landau 2004a). Notice also that control constructions with leo and
epitrepo are problematic for the movement/Agree approach to control because the controller is assigned
inherent dative case which is overtly manifested as a prepositional phrase introduced by se ‘in/to’. To the
best of our knowledge, arguments that are inherently case-marked are not visible for syntactic computa-
tion and inherent case-marking does not trigger a movement or an Agree operation; thus there is no
way to derive sto jani ‘to the John’ by moving it from the embedded subjunctive to the matrix clause.
15
An anonymous reviewer questions the grammaticality judgments of the data presented here on the
grounds that s/he can only interpret verbs like prospaqo ‘try’ as OC (=EC) verbs and s/he cannot
accept sentences with control suspension and overt DPs in the subjunctive clauses such as those
in (15), (16), etc. We would like to note the following: The data reported in the paper have been
checked against a good number of Greek native speakers who live in Greece and speak the stand-
ard variety. Special care was taken with checking the relevant examples with a number of different
control predicates and lexical items. Furthermore, similar data have also been reported by Varloko-
sta (1993, 1994), Philippaki-Warburton and Catsimali (1999), and Roussou (2004, 2005).
16
An anonymous reviewer claims that s/he can only accept control suspension with a causative
reading .S/he also brings to our attention Terzi’s analysis (1997) of control suspension as hidden
causative constructions (see also Terzi 1992) We note the following: First, a possible causative
reading does not necessarily translate into a hidden causative. Second, according to our intutions
and those of our informants, not all control suspension cases have a causative reading E.g. example
(16b) cannot be interpreted as a hidden causative The reader can check this with other examples of
control suspension reported in this paper. Second, we (and our informants) agree with Varlokosta
(1994) that the extraction data that Terzi (1992, 1997) cites as evidence for her own analysis, behave
exactly the opposite, showing that the subjunctive clause is not embedded in a hidden causative
17
However, even in these cases scenarios can be constructed so that control suspension is allowed for.
For instance, the following sentence exhibits control suspension with the predicate ðiatazo ‘order’:
(i) o loxaγos ðietakse to skopo na min

the lieutenant-NOM order-PAST.3SG the guard-ACC SUBJ NEG
perasi kanenas tin pili meta ti mia
pass-3SG no one-NOM the gate-ACC after the one
‘The lieutenant gave the order to the guard that no one should pass the gate after one o’clock’
Thanks to George Kotzoglou for bringing this example to our attention.

18
We note that control can be suspended in coordination
(i) i ana kseri na kaθete ke na ti frodizi o nikos

the Anna-NOM know-3SG SUBJ sit-3SG and SUBJ her take care-3SG the Nick-NOM
‘Anna knows how to do nothing while Nick takes care of her’
(ii) i maria ixe maθi na maγirevi (afti) ke na pleni
the Maria-NOM have-PAST.3SG learned SUBJ cook-3SG she-NOM and SUBJ wash-3SG
ta piata o nikos
the dishes-ACC the Nikos-NOM
‘Mary was accustomed to do the cooking and Nikos the washing up’
19
Examples of backward control such as the ones in (25b, c) could be possibly explained by a move-
ment/Agree approach to control as the result of alternative linearization of the chain created by the
DP o janis. In fact, Alboiu (2004a, this volume) argues for such an account for a similar set of data
in Romanian; see also the analysis proposed by Polinsky and Potsdam (2002) and the criticism by
Cormack and Smith (2002, 2003). Since Greek does not satisfy the EPP by means of overt movement
(Philippaki-Warburton 1987, Alexiadou and Anagnostopoulou 1998, Spyropoulos and Philippaki-
Warburton 2002) it can be suggested that in such constructions the DP-subject has remained in situ
and receives its nominative case by being targeted by the main clause T. Leaving aside the issue of the
phase status of subjunctive clause in Greek, let us take a closer look at example (25a). In this exam-
ple, the DP o janis is an argument of the main clause, and the embedded subjunctive has a distinct
nominative subject, namely a strong pronominal. By assuming that copies cannot surface as strong
pronouns in Greek, we take examples such as (25a) to indicate that the embedded subjunctive has the
ability to assign nominative case even in EC constructions with ASs.
20
An anonymous reviewer notices that perpection verbs such as akuo ‘hear’ and vlepo ‘see’ can
take an indicative complement introduced by oti ‘that’ and she points out this may weaken our
argument for the distinction between IS and DS. It is true that perception verbs can take such
indicative clauses, but when they do so, they no longer are perception verbs. Thus, the verb akuo
with an oti-complement means ‘I am informed that....’, whereas vlepo + oti-complement means
‘I realize that’. This shift in meaning indicates that there are two different lexical enteris of akuo
and velpo with distinct semantics and syntax.
21
See Hadzivassiliou et al. 2000, Kotzoglou 2002, Kotzoglou and Papangeli this volume.
22
An anonymous reviewer points out that s/he does not accept example (27). We know that according to our
intuitions and the intuitions of our informants (see also note 16) example (27) is perfectly grammatical and
that perception verbs can take a subjunctive complement with a nominative subject.
23
Kapetangianni and Seely (2003, this volume) suggest that the predicate modifier has its nominative
case checked by being targeted by the embedded T. However, such a proposal raises the following
issue: If T has the ability to check the nominative case of the predicate modifier, what prohibits it
from checking nominative case on the subject? Kapetangianni and Seely suggest that this is exactly
a property of defective T, i.e. a defective T is able to check the nominative case of only nominal
predicates and not DPs. We believe that such a hypothesis faces the following problems. First, if T
has the ability to check a nominative case feature and probing is a matter of structural relations,
then T should equally check nominative case on all the nominal elements which are inside its prob-
ing domain and enter into an Agree operation with them. In addition, it is evident on the grounds
of agreement mismatches that in Greek nominal predicates/modifiers in general do not enter in
an Agree operation with the [Agr] probe of T (Spyropoulos 2005a). The discussion of this issue
goes beyond the scope of this paper. We briefly mention that in examples such as the following
the nominal predicate has different feature specification for number/gender from the element it is
predicated of:
(i) i maria ine o agelos tu

the Mary-FEM.NOM is the angel-MASC.FEM his
‘Mary is his angel’
(ii) i ekloγes ine i moni lisi

the election-PL.NOM are the only solution-SG.NOM
‘Having elections is the only solution’
If the [Agr] probe of T was able to target both the DP-subject and the nominal predicate and check
their nominative case, the result would be a feature mismatch, since the [Agr] probe would be valued
with conflicting values for gender and/or number. Thus, examples (i–ii) show that nominal predicates
do not enter into an Agree operation with the [Agr] probe of T, so that they can only acquire their
case by means of an Agree operation established with the element they are predicated of.
24
Kapetangianni and Seely (2003, this volume) propose an analysis of control in Greek based on the
movement approach to control (Hornstein 1999, 2000, 2001, 2003, 2005a, b, forthcoming ). They
maintain that nominative case checking is associated with agreement checking and they propose that
T in control subjunctives is defective because the verbal morphology in Greek does not show gender
agreement. According to their hypothesis, verbal morphology in Greek is incomplete, because it
does not manifest the full set of phi-features (i.e. person, number, gender) and incomplete verbal
morphology indicates that the T may be underlyingly complete or defective. Thus, EC is established
when a matrix verb selects for a defective T, which is morphologically manifested in the same way as
a complete T. We believe that it is hard to find a well-motivated empirical reason why lack of gender
agreement marking in Greek results in defective T, since, typologically speaking, Greek does not
show gender agreement in verb morphology in general. In addition, there is crosslinguistic evidence
showing that EC can be established with embedded verb forms that morphologically manifest the
full set of phi-features, including gender. In Standard Arabic, verb forms inflect for the full set of
phi-features, i.e. person, number, and gender. Imperfective verb forms, which overtly inflect for per-
son, number, and gender, are used as complements in EC/raising constructions introduced by a verb
of beginning (examples from Badawi et al. 2004: 427–429).
(i) a. ʔahadatri l-surʕatu ta-zdaadu

started the-speed 3SG.FEM-increase.IMPERF
‘the speed started to increase’
b. al-hadiitu badaʔa yakturu

the-talk began 3SG.MASC-multiply.IMPERF
‘talk has started to multiply’
Kapetangianni and Seely’s approach requires that the embedded T be defective in order for EC/
raising to be established in such examples. The embedded T in the examples in (i) cannot be consid-
ered defective on the grounds of incomplete morphological manifestation on the verb form. Thus,
the correlation between EC and defective T, even in finite environments, seems to break down.
25
Landau’s F-subjunctives coincide with our DSs.
26
Landau (personal communication) suggests that according to his own analysis the controlled subject of
such indicative clauses is a case-marked PRO, because such clauses are tenseless. Thus, these examples
may not constitute strong evidence for the existence of a controlled pro. Testing the exact nature of the
controlled null subject of such clauses is a subtle issue. Evidence against the PRO analysis comes from
the fact that an overt DP-subject can be licensed in such clauses in the same way as in AS:
(i) a. afu kseri [proi/*j] [CL ke xorevi kala o janisi to tsamiko],…

since know-3SG COMP dance-3SG well the John-NOM the tsamiko-ACC
‘since John knows how to dance tsamiko (Greek folk dance) well,…’
b. afu arxise [proi/*j] [CL ke estelne o nikosi luluðia sti

since start-PAST.3SG COMP send-IMPERF.PAST.3SG the Nikos-NOM flowers-ACC to-the
maria],...
Maria
‘since Nikos has started sending flowers to Maria,…’
Given that Landau’s analysis predicts the complementary distribution between PRO and overt DPs
in tenseless complements, the licensing of overt DP-subjects indicates that the null subject in such
indicative clauses is not a PRO.
27
In Greek nonpast verb forms are used for both present and future tenses.
28
Such a proposal implies that AS constitute weak phases, so that they are transparent to operations
established with a matrix probe.
29
For a similar proposal see Landau 2004a.
30
Roussou (2004, 2005, in preparation) claims that control is the by-product of the licensing of a
D feature on the particle na, which lexicalizes the respective feature of T. She proposes that the
interpretation of this D feature is determined by being related to the matrix predicate, which may
impose event unification of feature matching, resulting in control. Disjoint reference is always
available, since full subject-agreement is always present on the verb form.
IV
CONTROL IN ROMANCE
GABRIELA ALBOIU
MOVING FORWARD WITH ROMANIAN BACKWARD

CONTROL AND RAISING*
1. INTRODUCTION
This chapter investigates various licensing constraints imposed on shared

subject arguments in Romanian obligatory control constructions and argues
for an analysis of obligatory control (OC) in this language analogous to that
of raising predicates. On the one hand, the discussion contributes to the cur-
rent debate with respect to whether OC can and should be construed as rais-
ing or not (i.e. the Hornstein–Landau debate1), and on the other hand, the
analysis provides an account of seemingly optional subject dislocation that is
intimately tied to the Theme–Rheme sentence partitioning in Romanian and,
consequently, independent of the control phenomenon per se. With respect
to the first point, I propose that movement out of controlled clauses is a par-
ametrized option made available by UG and governed by well-defined condi-
tions. Specifically, it is available in languages where complements to control
verbs lack phasal status, or can void phasehood, a proviso that guarantees
an active subject goal available to both thematic and non-thematic checking
operations with matrix probes. Regarding the second point, I show that dis-
location of the subject DP, which may but need not occur, is not incumbent
on morpho-syntactic featural requirements related to OC (such as Case or
theta-role valuation), but determined by well-defined semantico-pragmatic
constraints, such as topic and focus movement, construable as OCC features
(Chomsky 2001b) on the various probing heads. This analysis has the merit
of limiting the amount of movement required by reductionist approaches to
OC, accounting for optionality in a systematic manner, and providing ade-
quate empirical coverage of the phenomena under discussion.
The chapter is organized as follows. Section 2 familiarizes the reader
with recent minimalist formalizations of reductionist and non-reductionist
approaches to OC. Section 3 focuses on Romanian subject control construc-
tions and argues for a reductionist solution as the only formalization capable
of doing justice to the data. Section 4 investigates the relationship between
phases, movement, and Case and provides evidence for the non-phasal status
of OC complements and their inability to value Case. Section 5 discusses
the infomation packaging properties available to Romanian and analyzes the
various triggers for PF copy preference. Section 6 is a conclusion.
187
© 2007 Springer.
188 GABRIELA ALBOIU
2. PRO, OBLIGATORY CONTROL, AND MINIMALISM
It is well known that crosslinguistically aspectual, implicative, and modal matrix

verbs select a sentential complement whose external argument has to be corefer-
ential to the matrix clause subject DP, as exemplified in (1).
(1) a. Ericai just started [PROi to take syntax]
b. Philippai tried [PROi to read the new Chomsky]
Given the one-to-one mapping between theta-roles and arguments assumed in

the generative grammar of the 1980s (see theta-criterion of Chomsky 1981),
PRO in (1) is construed as both a semantic and a syntactic subject. Specifically,
the subject theta-position is filled structurally with some brand of null nominal
element ‘PRO’ distinct from any matrix clause DP or trace thereof. In (1), PRO
is constrained to an exhaustive identity with a matrix clause controller, where
coindexation is subject to some c-commanding version of the Minimal Distance
Principle of Rosenbaum (1967). The inherent non-overt nature of PRO is linked to
its compulsory association with Caseless T domains (i.e. infinitival T) – and, later,
null Case (as in Chomsky and Lasnik 1993). The properties of PRO, however, are
known to surpass those seen in (1). For example, PRO also surfaces with a Partial
Control interpretation, as in (2a), or an arbitrary interpretation, as in (2b).
(2) a. Mrs. Dallowayi wanted [PROi+ to meet at 9] (but it was too early)
b. It can be very rewarding [PROarb to do syntax]
PRO then has two arguments working against it: on the one hand, an undesirable
theory-internal flavor, and on the other hand, an unwelcome oscillatory nature.
No wonder it keeps making linguists uncomfortable.
With the advent of minimalism in the 1990s, which sees the collapse of govern-
ment and the elimination of D-structure and S-structure as separate levels of repre-
sentation (Chomsky 1995, 2000, 2001a, b, 2005), the availability of PRO in control
has been questioned by a number of authors, most notably, Boeckx and Hornstein
(2003, 2004), Hornstein (1999, 2000, 2001, 2003), Manzini and Roussou (1999), and
O’Neil (1997). Crucially, these authors also assume that movement out of control-
led clauses is legitimate, which in effect, points toward a synonymous construal of
raising and control, theta-roles notwithstanding. The reductionist view of control is
not only legitimate theoretically but has the added bonus of providing better empiri-
cal coverage than the standard view, given that it can also handle backward control
languages (see discussion in Polinsky and Potsdam 2002). While not everybody agrees
on eliminating PRO (see, especially Landau 1999, 2003, 2004a) and while there may
be conceptual and empirical reasons to maintain PRO in certain cases and for certain
languages, a movement analysis of OC cannot be universally dismissed prior to a
careful crosslinguistic investigation. However, before pursuing such an investigation,
let us first discuss relevant current formalizations of OC.
Recent, minimalist, formalizations of OC pursue either a reductionist (i.e. with-
out PRO) or a non-reductionist (i.e. with PRO) approach. Approaches eliminating
MOVING FORWARD WITH ROMANIAN BACKWARD CONTROL 189
PRO differ primarily as to whether they assume or do not assume movement,

which is in turn linked to whether theta-roles are construed as features or not. Per-
haps the least controversial reductionist minimalist approach is that put forth by
Wurmbrand (1998, 2004), who essentially argues that OC presupposes a mono-
clausal construction, with no PRO. Her analysis is reminiscent of various restruc-
turing analyses that stretch back to Rizzi 1982 and Haegeman and van Riemsdijk
1986. In a nutshell, for sentences like (1), which according to Wurmbrand (ibid.)
are an instance of lexical restructuring labeled ‘semantic control’, the matrix verb
selects a VP complement, as in (3).
(3) John tried [VP to read the new Chomsky]
The most influential syntactic movement analyses of OC belong to Manzini and

Roussou 1999, 2000 and Hornstein 1999, 2000, 2001, 2003, so I will briefly focus on
those. Both of these analyses view theta-roles as features which have to be checked
(i.e. valued) and crucially, both enable a DP to be associated with more than one
theta-role.
Manzini and Roussou propose that theta-roles are aspectual features which
merge in the verbal domain and which associate with a DP. In their system, DPs can
only merge in their Case position and, consequently, can only occur in the inflec-
tional domain. From there a DP will attract as many theta-roles as are in its domain,
essentially, all of the theta-roles up to the next DP. OC is devoid of any PRO, being
simply viewed as attraction of two theta-roles instead of one. This is schematically
illustrated in (4).
(4) a. [TP John T [vP tried [TP to [vP read]]]]
b. [ John [θ1 tried [θ2 read ]]]

D
Hornstein’s approach is more conservative in that it assumes DPs to be merged

in theta-domains rather than in the inflectional domain. However, it is novel in
that is allows for DP-movement into theta-positions on a par with movement into
Case positions. In his system, theta-roles are features that check either by initial
or by second Merge (i.e. via DP-insertion from the Numeration or via DP-move-
ment from within the syntactic tree, respectively). In this approach PRO is simply
a lower unpronounced copy of a moved DP, as in (5), where the pronounced copy
is in bold.2
(5) [TP John T [vP John tried [TP to [vP John v [VP read the new Chomsky]]]]]
The major proponent of the non-reductionist approach to control in

minimalism is Landau. Landau (1999, 2003, 2004a) argues that PRO is crosslin-
guistically present in OC structures and that equating control with raising is a
major mistake. His insights rely heavily on an earlier analysis proposed by Borer
(1989) that he adapts to minimalism. Essentially, OC is seen as an instantiation of
the operation Agree (see Chomsky 2000 et seq.) holding between a matrix probe
190 GABRIELA ALBOIU
and an embedded anaphoric element. This anaphoric element is sensitive to the

specific type of control at stake in the derivation. If involved in ‘Exhaustive Con-
trol’, the anaphoric element is PRO; if involved in ‘Partial Control’, the anaphoric
element is Agr of the embedded clause. The split between Exhaustive vs. Partial
Control depends on whether the selecting matrix predicate obligatorily requires
an identical embedded argument, as in (6), or does so optionally, as in (7b), or
even partially, as in (7c).
(6) Exhaustive Control (EC)
a. Tomi tried [PROi to understand calculus]
b. *Tomi tried [for Mary to understand calculus]
c. *Tomi tried [PROi+ to meet at 9]
(7) Partial Control (PC)
a. Gandalfi wanted [PROi to succeed]
b. Gandalfi wanted [for Frodo to succeed]
b. Gandalfi wanted [PROi+ to meet late at night]
Crucially, for Landau, PRO is present throughout, being ‘active’ for Agree due to its
anaphoric nature and ‘inactive’ for movement given that it is Case-marked with null Case.
I next turn my attention to subject OC constructions in Romanian.
3. TO RAISE OR NOT TO RAISE: EVIDENCE FROM ROMANIAN

OBLIGATORY CONTROL
While in languages like English, the shared argument is constrained to a matrix

clause position, in Romanian it may surface in a number of legitimate slots,
including in the embedded clause, as shown in (8). However, only one PF copy
instantiation of the DP subject is permitted.
(8) (Victor) încearcă (Victor) [să cînte (Victor)

(Victor.NOM) try.PRES.3SG (Victor.NOM) [SBJ SING.3SG (Victor.NOM)
la trombon (Victor)]3
at trombone (Victor.NOM)]
‘Victor is trying to play the trombone.’
Note also, that even if the complement clause in Romanian is a subjunctive (with
agreement morphology) rather than an infinitive, OC still holds, as shown in (9).4
(9) a. Victor încearcă [(* Mihai) să cînte]

Victor try.PRES.3SG [(* Mihai) SBJ sing.3SG ]
‘Victor is trying (*Mihai) to sing.’
b. Victori încearcă [xi să cînte]

Victor try.PRES.3SG [xi SBJ sing.3SG ]
‘Victor is trying to sing.’
Given the availability of agreement morphology present on the subjunctive

and the pro-drop status of the language, the null subject in Romanian OC
constructions (i.e. x in (9b)), has often been claimed to be pro (e.g. Dobrovie-
Sorin 1994; Farkas 1988; Motapanyane 1995), but some authors have argued for
PRO (e.g. Kempchinsky 1986; Terzi 1992). We will see that neither approach can
be maintained.
Clearly, the data in (8) make it difficult to maintain a standard analysis of con-
trol for Romanian and at least the following questions need to be addressed in con-
nection with these data: (i) where does the DP subject originate; (ii) is movement
involved; and if so, (iii) what factors determine the pronunciation site; and (iv) is
there any genuine evidence for PRO or pro?
In the following sub-sections, I investigate the relevance of the various
minimalist proposals presented in section 2 and conclude that none of them can
do justice to the data. I show that the best solution is reductionist in spirit along
the lines of Hornstein, but modified to allow for optional subject displacement
and the relevant semantico-pragmatic effects to be discussed in section 5.
3.1 Wurmbrand’s proposal and Romanian subject control
Wurmbrand’s lexical restructuring analysis cannot be maintained, as (8) shows

evidence for structure beyond VP (i.e. phi-complete T, embedded subject DP,
etc.). Furthermore, clitic climbing – a crucial argument for restructuring – while
present in certain contexts in Romanian, see (10a), is ruled out in OC contexts,
see (10b).5
(10) a. Li-a putut [VP (* îli) vedea]?

CL.3SG.M.ACC-AUX.3SG could.PART [VP (*CL.3SG.M.ACC) see.3SG]
‘Could s/he see him?’
b. Nu (*li)-a încercat [să-li vadă]

not (*CL.3SG.M.ACC)-AUX.3SG try.PART [SBJ-CL.3SG.M.ACC see.3SG]
‘S/he didn’t try to see him.’
Note that functional restructuring along the lines of Cinque 2004 and
Wurmbrand 2004 also needs to be ruled out, as the matrix clause predicate
assigns a subject theta-role, contrary to what functors typically do. Interestingly,
however, restructuring analyses are tempting because they are relevant for
environments that are to be treated as mono-clausal. Once it becomes clear
that OC subjunctives are non-phasal, a clause union of sorts becomes vital.
However, what I hope to have convinced the reader of here is that restructuring
cannot be assumed to apply any lower than the T domain. Consequently, we
need to investigate beyond Wurmbrand’s approach.
192 GABRIELA ALBOIU
3.2 Manzini and Roussou’s proposal and Romanian subject control
A Manzini and Roussou (henceforth M&R) approach has actually been proposed
for Romanian by Dobrovie-Sorin (2001). Essentially, it would work as in (11):
(11) a. Victor încearcă [să cînte la trombon]

Victor try.PRES.3SG [SBJ sing.3SG at trombone]
b.
[ VictorD încearca [θ1 <încearca> [sa cînte θ2 la trombonD <cînte> θ3 ]]]
The shared argument Victor is merged directly in Spec,TP from where it

attracts all the theta-roles up to the next DP.6 Given that it attracts two subject
theta-roles, OC ensues. Leaving aside theory-internal problems with the M&R
approach, such as for example, the fact that it is stipulative to assume the DP
would be interested in attracting more than one theta-feature to begin with, there
are also empirical problems which are more difficult to ignore. Basically, this
analysis only works if the unique DP argument is in the main clause preverbal
position but becomes problematic once we consider data of the type in (12)
where the shared argument is not pronounced in the matrix clause domain but
lower.
(12) a. proi încearcă [să cînte Victori la trombon]

proi try.PRES.3SG [SBJ sing.3SG Victori at trombone]
b. [proD [încearca θ1<încearca> [ sa cînte VictorD [θ2 la trombon <cînte> θ3 ]]]
The reasons are self-evident. First, the obligatory OC interpretation cannot be

accounted for as each DP attracts a single theta-role. Specifically, θ1 checks
against the null pronominal and θ2 checks against Victor and there is no further
permissible coindexation mechanism. Second, (12) represents a Condition C
violation. If anything, assuming the null pronoun is indeed available in the
Numeration, we expect it to be incapable of bearing the same index as the
subjunctive DP subject, contrary to fact. One possibility is to hypothesize that
Romanian is insensitive to Condition C effects. However, example (13) shows
that Condition C effects are, nonetheless, operative in this language, which
amounts to disqualifying the Manzini and Roussou approach as a correct
analysis for Romanian.
(13) pro k / *j ştie [că pleacă Mihaij mîine]

pro k / *j know.PRES.3SG [that leave.3SG Mihaij tomorrow]
‘He k / *j knows that Mihaij will be leaving tomorrow.’
3.3 Landau’s proposal and Romanian subject control
I first discuss Landau and leave Hornstein for last, given that I ultimately adopt a
revised version of Hornstein as the best solution for Romanian.
As mentioned in section 2, Landau’s approach assumes PRO across the board
in OC contexts. However, by definition, standard theories of control have assumed
PRO to be in complementary distribution with overt DPs, which, as shown in (8) and
(12) is not the case for Romanian. Clearly, a PRO analyis cannot do justice to the
data, so it seems stipulative to adopt it. I would, however, like to point out that one of
Landau’s main arguments against a Hornstein-type raising analysis for OC in Eng-
lish comes from the availability of Partial Control with certain matrix verbs in this
language, as seen in (7). This is a viable argument that cannot be ignored. However,
it is an argument that does not apply to Romanian, which lacks the Partial Control
effects seen with English desideratives. Look at (14), with phi-values in bold.
(14) a. * Eu vreau [să plec împreună]

I want.PRES.1SG [SBJ leave.1SG together]
b. * Vreau [să plecăm eu împreună]

want.PRES.1SG [SBJ leave.1PL I together]
c. Eu vreau [să plecăm x împreună]

I want.PRES.1SG [SBJ leave.1PL x together]
‘I want to leave together.’
What (14) shows us is that Romanian requires syntactic plurality of any

semantically plural predicate. Given that syntactic plurality can only be guaranteed
by formal feature checking against a plural value, the x subject in (14c) has to
bear a syntactically plural feature, so is not controlled by the matrix singular
subject eu ‘I’. Rather, it is a referential pro specified as 1.PL. This then explains the
ungrammaticality of (14a, b) but rules out Partial Control which is defined by a
syntactically singular controller of semantically plural predicates and identity in
syntactic agreement between the matrix and the embedded clause predicates.
The data in (14) also show that Romanian OC is always Exhaustive Control
(EC) in the sense of Landau. So, the relevant split for Romanian is between OC
and non-OC (NOC), as in standard accounts, without further refinements for OC.
Specifically, with predicates such as aspectuals and implicatives, OC/EC obtains,
while desideratives instantiate NOC. This dichotomy is illustrated in (15).
(15) a. OC/EC (e.g. aspectuals and implicatives):

proi încearcă [să plece pro i/* j]
proi try.PRES.3 [SBJ leave.3 pro i/* j]
(i) ‘S/he wants (*for them) to leave.’ OR
(ii) ‘They want (*for her/him) to leave.’
b. NOC (e.g. desideratives):

proi vrea [să plece pro i/j ]
proi want.PRES.3SG [SBJ leave.3 pro i/j ]
‘S/he wants (for her/him/them) to leave.’
194 GABRIELA ALBOIU
Note that even if (15a) shows that in the absence of an overt subject, the 3rd
person subject referent can be interpreted as either singular or plural, OC holds.
On the other hand, in (15b), a main clause singular subject can license either an
embedded singular or an embedded plural subject, hence NOC.
The above split yields the expected readings under ellipsis (see Williams 1980).
In particular, OC only allows for a sloppy reading under identity, while NOC
allows for both sloppy and strict readings, as shown in (16). I will argue in section
4 that OC structures are non-phasal while NOC structures are phasal, which
explains their distinct semantics and subject availabilities.
(16) a. sloppy reading only for OC:

Mihai încearcă să-i ajute şi la fel (încearcă ) şi
Mihai try.PRES.3SG SBJ-CL.DAT.3PL help.3SG and at same (try.PRES.3SG) and
Victor.
Victor
‘Mihai is trying to help them and so is Victor (= Victor to help).’
b. sloppy and strict readings for NOC:

Mihai vrea să-i ajute şi la fel
Mihai want.PRES.3SG SBJ-CL.DAT.3PL help.3SG and at same
(vrea) şi Victor.
(want.PRES. 3SG) and Victor
‘Mihai wants to help them and so does Victor (= Victor to help OR Mihai to help).’
To conclude this subsection, I have shown here that OC in Romanian is exclu-

sively EC and that backward control rules out an account based on PRO.
These two facts can only be felicitously captured under some version of a
theta-chain analysis of control.
3.4 Hornstein’s proposal and Romanian subject control
A Hornstein-type analysis would work as in (17), where the subject DP first

merges in the Spec,vP of the embedded clause and subsequently moves to its sec-
ond Merge position in Spec,vP of matrix clause, thus satisfying the external the-
matic roles of both predicates (i.e. θve and θvm, respectively).
(17)
[TP încearca [vP Victor θvm <încearca > [sa cînte [vP <Victor> θve [vP <cînte> la trombon]]]]]
Note, however, that (17) only partially accounts for (8), repeated as (18) with
the copies relevant for the theta-chain in boxes and potential occurrences of the
shared argument in bold:7
(18) (Victor) încearcă ( Victor ) [să cînte ( Victor )
(Victor) try.PRES.3SG (Victor) [SBJ sing. 3SG (Victor)
la trombon ( Victor )]
at trombone (Victor)]
It is important to mention here that all current studies on Romanian (see Alboiu
2002; Cornilescu 2000a; Dobrovie-Sorin 1994; Hill 2002) have argued that the
language is VSO in that the lexical verb undergoes obligatory displacement into
the T domain, while Case is valued via Agree without dislocation to Spec,TP for
classical EPP purposes. Preverbal subjects in this language undergo A-bar rather
than A-movement and are interpreted as topics or – with the relevant prosody – as
contrastively focused constituents. Crucially then, in (18), Victor is not involved in
movement of the English EPP-type when matrix initial. If anything, the Hornstein
account predicts pronunciation of the postverbal copy for VSO language like
Romanian, which we see not to be supported by the empirical facts.
Consequently, while I take an approach which views theta-roles as features in
need of valuation to be not only the correct solution for Romanian, but the opti-
mal one, dislocation not being a prerequisite for either Case, EPP, or theta-valu-
ation.8 Even if A-related properties are satisfied solely via Agree, any analysis of
Romanian OC needs to be capable of providing a coherent account of all avail-
able displacements. I propose that the Romanian data can be captured via a move-
ment analysis of control reminiscent of raising. Specifically, I adopt an approach
which views theta-roles as features that can be valued via Agree on a par with any
other unvalued feature. I argue that, despite agreement with the embedded T, the
embedded DP subject does not value its Case feature in the subjunctive clause and,
consequently, is accessible to matrix clause Agree operations as long as it remains
active. I further argue that whether theta-feature and/or Case feature valuation is
accompanied by movement depends on the presence or absence of relevant seman-
tico-pragmatic triggers for displacement (e.g. focus, de-rhematization, etc.).9
Before developing further a fine-grained analysis to accommodate all of the copy
availabilities seen in Romanian, let us proceed with our discussion of relevant proper-
ties of OC subjunctives that not only enable but crucially require a shared argument.
4. OBLIGATORY CONTROL AND PHASES
Here I discuss morpho-syntactic properties of subjunctive clauses in Roma-

nian OC constructions and conclude that they have properties typical of non-
phasal rather than phasal domains. Evidence for this is based on the absence
of a lexical complementizer, a temporally unsaturated, even if phi-complete, T,
and the incapacity of the embedded domain to value Case on the DP subject.
Given that all of these are properties of C – the phase head – rather than of T
(see Chomsky 2005, 2006), their absence indicates absence of phasehood. The
lack of PIC (Phase Impenetrability Condition) effects only strengthens this
claim, as does the fact that the embedded subject DP remains active for Match
and Agree operations with relevant heads in the matrix clause.
4.1 OC subjunctives lack phasal CP status
In addition to the fact that the subjunctive verb in Romanian shows person and
number agreement with the subject (i.e. synthetic marking on the verb stem as
196 GABRIELA ALBOIU
in other Romance languages), subjunctives also require the obligatory presence

of a subjunctive particle să preceding the verb (i.e. the subjunctive is analytically
marked as in other languages of the Balkans). Most analyses take this particle
to be the highest head of the verbal functional domain and refer to it as Mood,
I, or just T (e.g. Alboiu 2002; Cornilescu 2000a; Isac 2002; Motapanyane 1995;
Pîrvulescu 2001; Rivero 1994; Terzi 1992), but Dobrovie-Sorin (1994) suggests
să is ambiguous between a C and a T element. In fact, under a cartographic
approach to the left-periphery (Rizzi 1997, 2004), this ambiguity could perhaps
translate as T to Fin (i.e. low C) movement. Farkas (1985) actually claims this
movement to be obligatory in the absence of ca and Hill (2003), not only provides
crucial arguments for this movement, but also shows that in the absence of ca,
subjunctives are non-phasal, while in the presence of ca, these domains show
PIC properties and have phasal status. For example, Hill (2003) shows that while
DP movement is licit across să-subjunctives, it is ruled out in ca-subjunctives,
thereby concluding that the latter, but not the former, constitute phasal domains.
In Alboiu 2006, I reach a similar conclusion by looking at movement constraints on
focused negative polarity items across the two types of subjunctives. Given the
availability of a distinct subjunctive complementizer in Romanian, specifically
ca, an unambiguous C element which surfaces to the left of the subjunctive parti-
cle and yields phasal effects, I take să to be at least the highest verbal functor and
at most a low C head and treat it as a genuine non-phasal element. Henceforth,
I use CHIGH notation to represent a phasal domain and CLOW notation to represent
a non-phasal domain. The reader might wonder why I do not simply use the
better established Force vs. Fin(ite) notation of Rizzi 1997, 2004. As mentioned,
Hill (2003) argues that in ca-less subjunctives, să and its verbal host must move
up to the Fin head. The same is argued for infinitives lacking a complementizer:
movement of the infinitive mood particle a and its verbal host proceeds to Fin.
I take this to be essentially correct given that both subjunctives and infinitives
must minimally contain a FinP, the domain responsible for finiteness or lack
thereof (see also Barrie, this volume). Crucially, these are non-phasal domains in
Romanian. However, Hill (2003) also argues, convincingly to my mind, that the
subjunctive complementizer ca, as well as the infinitive complementizer de, are
merged in Fin, and not in Force in Romanian. Nonetheless, the author clearly
shows that these domains are phasal. So it looks like FinP can be both phasal
and non-phasal in Romanian. To put this another way, the phasal status of FinP
is obviated in OC contexts. Possibly when FinP is phasal, the Force domain is
also projected but left empty or is simply merged with ca-Fin and forms a syncre-
tic category with it. Essentially, OC only holds in the absence of phasehood, so
with a non-phasal FinP. To capture this critical distinction, I use CHIGH for phasal
subjunctives (i.e. domains which allow for the lexical complementizer ca) and
CLOW for non-phasal subjunctives (i.e. domains which rule out the complemen-
tizer), even though they both instantiate a FinP.10
From an empirical point of view, as illustrated in Alboiu and Motapanyane
2000, the subjunctive complementizer is obligatory with topicalized material,
(19a), optional with fronted focus, (19b), and subject to idiolectal variation
when nothing precedes să, (19c).
(19) a. Trebuie [*(ca) Mioarai să ajungă ti repede]

must.PRES that.SBJ Mioara SBJ arrive.3SG ti soon.’
‘Mioara must arrive soon.’
b. Vreau [(ca) AZI să pleci (nu mîine)]

want.PRES.1SG that.SBJ today SBJ leave.2SG (not tomorrow)
‘It is today that I want you to leave ( not tomorrow).’
c. Vreau [(? ca) să ningă]

want.PRES.1SG that.SBJ SBJ snow.3SG
‘I want it to snow.’
Given that CHIGH is not always lexicalized as ca, the absence of ca cannot be taken
as synonymous to the absence of CHIGH. However, the illegitimacy of ca is synony-
mous to the absence of a phasal C and denotes a CLOW.
A number of authors have noticed that ca is obligatorily absent in both raising
and OC subjunctive complements (e.g. Alboiu; 2006; Grosu and Horvath 1987;
Dobrovie-Sorin 2001; Rivero and Geber 2004; among others). This is illustrated
in (20).
(20) a. Victor încearcă / pare [(*ca pe Mihai) să-l ajute]

Victor try.PRES.3SG / seem.PRES.3SG [(that PE Mihai) SBJ-3SG.M.ACC help]
b. Victor încearcă / pare [să-l ajute (pe Mihai)]

Victor try.PRES.3SG/ seem.PRES.3SG [SBJ-3SG.M.ACC help (PE Mihai)]
‘Victor is trying to help Mihai/seems to be helping Mihai.’
Consequently, these subjunctives are never CHIGHP but reduced, non-phasal, CLOWP
domains. As shown below, this contrasts with NOC subjunctives, such as desid-
eratives, where the lexical complementizer is optional.
In (21) the desiderative selects a ca-less subjunctive. Given the empirical facts
in (20), (21) is structurally ambiguous between a CHIGHP and a CLOWP domain.
Semantic ambiguity (see the two readings) provides empirical support for this
claim. What is noteworthy, however, is that the OC reading must of necessity
involve a theta-chain (i.e. A-chain). This follows from the fact that Condition C
of Binding Theory would rule out a coindexed referential pro.
(21) prok / *proj vrea [să cînte Mihaij la violoncel]

prok / *proj want.PRES.3SG [SBJ sing.3SG Mihaij at cello]
(i) ‘S/he wants Mihai to play the cello.’ OR

(ii) ‘Mihai wants to play the cello.’
Let us compare (21) with (22), where C is lexicalized as ca, thus ensuring
an unambiguous CHIGHP status to the subjunctive complement. We notice an
asymmetry in readings between (21) and (22). In (22), the OC reading is ruled
out.
198 GABRIELA ALBOIU
(22) pro k / *j vrea [CHIGHP ca mîine să cînte Mihai j la violoncel]

pro k / *j want.PRES.3SG [CHIGHP that.SBJ tomorrow SBJ sing.3SG Mihai at cello]
(i) ‘S/he wants Mihai to play the cello tomorrow.’

(ii) but not: ‘Mihai wants to play the cello tomorrow.’
These data confirm two things: (i) that OC subjunctives are CLOWP domains and
(ii) that A-chains are not permitted across CHIGHP (phasal) boundaries. In the next
two sections, I show that non-phasal domains are neither temporally saturated
nor capable of valuing Case. As such, a clause union analysis of sorts becomes
compulsory.11
4.2 No saturated T in the absence of phasal C
While subjunctive complements do not manifest independent tense on a par

with indicative clauses, their tense properties may or may not be anaphoric
depending on the matrix verb selecting them (see Dobrovie-Sorin 1994; Farkas
1992, for Romanian; Krapova 2001, for Bulgarian; Landau 2004a, for Balkan
languages more generally). Specifically, complements to obligatory OC predicates
are untensed (i.e. bear anaphoric tense). For Romanian, this is illustrated in (23),
where (23a) involves an OC implicative and (23b) involves an OC aspectual predi-
cate. In both instances, subjunctive T is anaphorically related to matrix T as indi-
cated by the absence of distinct temporal deixis.
(23) a. Am reuşit [CLOWP să plec (*mîine)]

AUX.1SG managed [CLOWP SBJ leave.1SG tomorrow]
‘I managed to leave (*tomorrow).’
b. Încep [CLOWP să citesc / *fi citit]

begin.pr es.1sg [CLOWP sbj read.1sg / past read]
‘I’m beginning to (*have) read.’
On the other hand, NOC predicates allow for a distinct tense from that of the
matrix clause, even if dependent on matrix clause T given the irrealis status of
subjunctives in general (see Landau 2004a). This is shown in (24).
(24) Am vrut [CHIGHP să plece Mihai mîine]

AUX.1SG wanted [CHIGHP SBJ leave.3SG Mihai tomorrow]
‘I wanted for Mihai to leave tomorrow.’
In Alboiu 2006, I suggested that temporal deixis is strictly dependent on the

presence vs. absence of the CP domain. This insight goes back to Stowell 1982,
who viewed tense domains as C properties and has received support recently
from novel proposals in Chomsky 2005, 2006. To sum up, OC subjunctives are
non-phasal and, consequently, temporally unsaturated domains.
4.3 A-chains and phases
The non-phasal status of OC and raising subjunctives guarantees that matrix

probes have access to embedded material in these Romanian constructions. In
Minimalism, a DP is active until Case-marked. I take this to be true, at least for
structural Case, though there are other views currently available (e.g. Bejar and
Massam 1999; Ura 2000). Contra earlier generative assumptions (Chomsky 1981;
George and Kornfilt 1981) but in accord with much recent work (Alboiu 2006;
Chomsky 2005, 2006; Pesetsky and Torrego 2001, 2004b; Sitaridou 2002) I take
Case valuation to be a property of phasal domains rather than of agreement. So,
even if subjunctive T is phi-complete, it will only check and value Case once T is
saturated by a phasal C. This entails that embedded subjects in OC constructions
remain active beyond the subjunctive clause and can act as goals for various
A-type Agree operations triggered by matrix probes. The following subsection
provides empirical support for this theoretical claim.12
4.3.1 Nominative as a phasal property: evidence from Romanian emphatics
In Romanian, emphatics cannot be stranded without a pronominal copy. Consider

(25).
(25) a. [Mihai însuşi ] a făcut [vP tsu acest desen]

[Mihai.NOM himself] AUX.3SG done [vP tsu this drawing]
‘Mihai himself made this drawing.’
b. Mihai a făcut [vP [*(el) însuşi ] acest desen]

Mihai.NOM AUX.3SG done [vP [he.NOM himself] this drawing]
‘Mihai made this drawing himself.’
Furthermore, the pronominal copy licensing the emphatic is seen to bear a nomi-
native value. Nonetheless, it is clear that Mihai.NOM and he.NOM form a Case-
chain, given that both are assigned Case via matrix T.
In the presence of two CP phasal domains, where nominative Case is assigned
independently twice (i.e. both in the matrix and in the embedded clause), as
expected, two distinct subject occurrences are permitted in addition to the
emphatic. This is illustrated below.
(26) a. Mihaii regretă [CHIGHP că Victork nu poate]

Mihai.NOM regret.PRES.3SG [CHIGHP that.IND Victor.NOM NEG can]
veni [e1k însuşik]]
come.3SG [he.NOM himself]]
‘Mihai regrets that Victor can’t himself come.’
b. ‘Mihai regretă [CHIGHP că ălak/eli nu

Mihai.NOM regret.PRES. 3SG [CHIGHP that.IND that.one.NOM/he NOM NEG
poate veni [e1k /eli însuşik]]
can come. 3SG [he NOM himself]]
‘Mihai regrets that that guy/he can’t himself come.’
200 GABRIELA ALBOIU
While in (26), there is no Case-chain between main clause and embedded clause
subjects, the grammaticality judgments in (27) force us to conclude differently for
OC constructions.
(27) a. [Victor însuşi] încearcă [CLOWP să facă tsu pizza]

[Victor.NOM himself] try.PRES.3SG [CLOWP SBJ make.3SG tsu pizza]
‘Victor himself is trying to make pizza.’
b. Victori încearcă [CLOWP să facă [eli însuşii] pizza]

Victor.NOM try.PRES.3SG [CLOWP SBJ make.3SG [he.NOM himself] pizza]
‘Victor is trying to himself make pizza.’
c. Victori încearcă [CLOWP să facă pizza [eli însuşii ] ]

Victor.NOM try.PRES.3SG [CLOWP SBJ make.3SG pizza [he.NOM himself] ]
d. * Victori încearcă [CLOWP să facă ăla k / eli pizza [elk / i însuşii]]

Victor.NOM try.PRES.3SG [CLOWP SBJ make.3SG that.one/he.NOM pizza [he.NOM himself]]
What (27b, c) show us is that a stranded emphatic copy may surface in the
embedded clause in either of the two slots available to subjects in OC subjunc-
tives: to the left and to the right of the DP object.13, 14 Crucially, however, an
independent Nominative pronoun is ruled out, as shown in (27d). This proves
that there is a Case-chain established between matrix and embedded subjects in
OC subjunctives and confirms the theoretical predication that Nominative is
not independently available in these non-phasal subjunctive domains, regardless
of phi-specifications.15
In the next subsection, I show that theta-chains also hold across CLOWP domains
but not across CHIGHP, phasal domains.
4.3.2 Dative subjects in (N)OC and raising
Among other things, Landau (2003) takes issue with control-as-raising on the
basis of the behavior of Dative subjects in Icelandic. As a DP-trace, the controlled
position should be Caseless. However, in Icelandic there is a Case mismatch
between a floating quantifier and the matrix controller, a mismatch which is not
observed in raising constructions. This is shown in (28) adapted from Landau
(2003:492).
(28) a. Strákarniri vonast til [að PROi lei ðast ekki öllumi í skóla]
boys-the.NOM hope for [to PRO.DAT to-be bored not all.DAT in school]
‘The boys hope not to be all bored in school.’
b. Strákunumi virðast [ti lei ðast ekki öllumi í skóla]

boys-the.DAT seem [ to-be bored not all.DAT in school]
‘The boys seem not to be all bored in school.’
Consequently, only in (28b) can the embedded subject be a DP-trace and in (28a)
it must be PRO.
Consider (29), on the other hand, which illustrates that in Romanian OC
constructions the moved DP subject retains the Dative Case required by the
embedded predicate both in the presence of the quantifier and when the
quantifier is left floating (to the extent that this yields a felicitous sentence).
This is similar to what is observed in Icelandic raising and not Icelandic OC
constructions.16
(29) a. Tuturor copiilor a reuşit [să le placă

all.DAT.PL kids-the.DAT AUX.3SG managed [SBJ CL.3PL.DAT like
matematica]
mathematics]
‘All the kids managed to like math.’
b. Copiilor a reuşit [să le placă (?tuturor)

kids-the.DAT AUX.3SG managed [SBJ CL.3PL.DAT like all.DAT.PL
matematica]
mathematics]
‘The kids managed to all like math.’
In conclusion, as evidenced by its Case properties, the shared argument could not
have been merged in the main clause domain, so a theta-chain between the matrix
and the embedded subject positions seems appropriate.
Moreover, further investigation into the behavior of Dative subjects reveals
an asymmetry not between raising and OC constructions but between these
and NOC environments. This behavior only strengthens the claim that the
controlled position is part of an A-chain, in this case a theta-chain, as the
readings will show.
Let us consider first the raising and OC data in (30) and (31), respectively.
(30) a. Lui Mihai / *Mihai pare [CLOWP să-i placă şcoala]

Mihai.DAT / *NOM seem.PRES.SG [CLOWP SBJ-CL.3SG.DAT like school.NOM]
‘Mihai seems to like school.’
b. Lui Mihai / *Mihai par [CLOWP să-i placă fetele]

Mihai.DAT / *NOM seem.PRES.PL [CLOWP SBJ-CL.3SG.DAT like girl.PL-the NOM]
‘Mihai seems to like girls.’
c. Copiilor a reuşit [să le placă (?tuturor)

kids-the.DAT AUX.3SG managed [SBJ CL.3PL.DAT like all.DAT.PL
matematica]
mathematics]
‘The kids managed to all like math.’
(31) a. Lui (Mihai) au reuşit [CLOWP să-i placă

3SG.M (Mihai).DAT AUX.3PL managed [CLOWP SBJ-CL.3SG.DAT like
toţi copiii din clasǎ]
all kids.NOM in class]
202 GABRIELA ALBOIU
b. *El au / a reuşit [CLOWP să-i placă

3SG.M.NOM AUX.3PL / AUX.3SG managed [CLOWP SBJ-CL.3SG.DAT like
toţi copiii din clasǎ]
all kids.NOM in class]
(i) ‘He/Mihai managed to like all the kids in his class.’

(ii) and not ‘All the kids in the class managed for Mihai to like them.’
On a par with the raising predicate in (30), the implicative in (31) rules out Nomi-
native on its DP argument, requiring instead that the Dative Case be retained. In
both cases, matrix clause T agrees with the phi-features of the embedded Nomina-
tive DP. Nonetheless, in (31), it is the quirky argument (i.e. the logical subject of
liking17) and not the Nominative that controls. This means that the quirky argu-
ment enters an A-chain with the matrix clause subject theta-domain, even if not
with the matrix clause T domain.
Now look at (32), with a NOC predicate:
(32) a. Eu sper [CHIGHP să-mi priască excursia]

1SG.NOM hope.PRES.1SG [CHIGHP SBJ-CL.1SG.DAT enjoy trip-the.NOM]
‘I hope to enjoy the trip.’
b. Mie sper [CHIGHP să-mi priască excursia]

1SG.DAT hope.PRES.1SG [CHIGHP SBJ-CL.1SG.DAT enjoy trip-the.NOM]
‘Insofar as I am concerned, I hope to enjoy the trip.’
c. Mie sperǎ Ion [CHIGHP să-mi priască

1SG.DAT hope.PRES.3SG Ion [CHIGHP SBJ-CL.1SG.DAT enjoy
lecturile astea]
readings-the.NOM these]
‘Insofar as I am concerned, Ion hopes that I’ll be enjoying these readings.’
In (32b, c) but not (32a), the DP argument with idiosyncratic Dative Case dislo-
cates to the main clause left-peripheral domain. That this is dislocation to a Topic
position is evidenced both by the semantics, as well as by the phi-values on the
main clause desiderative: T agrees with a matrix clause DP and never with the
Nominative DP embedded in the subjunctive. In these cases, the Dative DP does
not control. Rather, there is an independent external argument within the main
clause: eu ‘I’ (in 32a), 1SG pro (in 32b – as evidenced by agreement on the desid-
erative), and Ion (in 32c). This is the outcome of there being two independent CP
(phasal) domains: the main clause CP and the embedded clause CP. These facts
suggest the absence of thematic A-chains across CHIGHP domains.
To sum up, in these two subsections, I have shown evidence for: (i) Case as a
phasal property and (ii) theta-chains across non-phasal (CLOWP) but not phasal
(CHIGHP) boundaries. The first finding argues against Nominative valuation
by subjunctive T in OC constructions, regardless of agreement inflection.
Consequently, on a par with raising constructions, the DP embedded subject
is not only available but necessary to subsequent matrix A-relationships, in the
absence of which the derivation would crash as this D would not get a Case value.
Given this availability, it should not surprise us that thematic A-chains can also be
established. That this is indeed the case is supported by the second finding which,
in effect validates the claim that OC in Romanian is raising. It now remains to
illustrate how the various features are checked in OC, how the derivations converge
without displacements and how we can account for PF copy preference, which is
what I embark on in the next section of this chapter.
5. FEATURE CHECKING AND INFORMATION PACKAGING

IN ROMANIAN OC CONSTRUCTIONS
In this section, I will argue for DP displacement as a semantico-pragmatic conse-

quence, independent of morpho-syntactic requirements for satisfying OC. In the
first subsection, I provide an account of how the various uninterpretable features
are accommodated as far as OC is concerned and in the last subsection, I tease
apart the various PF instantiations of the shared subject.
5.1 Accommodating OC features
Having shown that OC subjunctives cannot satisfy the Case requirements of the
embedded DP subject and that this subject DP cannot be perceived as either pro
or PRO, let us see how this subject is licensed. As previously mentioned, nothing
should prevent an active DP from entering as many Agree operations as there
are probes probing. In principle, this can go on until the active DP becomes inac-
tive, specifically, until its uninterpretable Case feature is valued by an interpretable
counterpart. Note that I am not claiming that the creation of A-chains is synony-
mous to movement. In fact, the Romanian data indicate this not to be the case.
I take the creation of chains to be synonymous to the instantiation of an Agree
operation, with dislocation only triggered by certain special circumstances to be
discussed in the next subsection. So, let us see how these A-chains are formed and
how the various uninterpretable features are accommodated in OC subjunctives.
The sample derivation to be discussed is (33) which shows the shared DP sub-
ject Victor in situ.
(33) [Încearcă [CLOWP să cînte [vP Victor < cînte > la trombon]]]
[try.PRES.3SG [CLOWP SBJ sing.3SG [vP Victor at trombone]]]
Before we proceed a clarification is in order. I formalize anaphoric tense as a uT

feature on T; independent tense, on the other hand, is formalized as i T on T and
is only available in the presence of phasal C heads (i.e. when T is selected by CHIGH).
Let us suppose, following Pesetsky and Torrego’s view (2001, 2004b), that Case is
construed as uT on D arguments. This seems to be supported by the fact that non-
phasal domains lack both Case and deictic tense properties, as argued in section 4.
At stage α in the derivation, extract DP Victor from the Numeration and merge
with v cînte la trombon ‘play the trombone’ to satisfy the external theta-role of
204 GABRIELA ALBOIU
the embedded predicate (i.e. u[θve]18). Next insert T which is phi-complete, having
uninterpretable person and number. See (34).
(34)
T vP
uφ
(uT) Victor v’
iφ, D
uT v VP
u[θve]
play the trombone
Focusing on A-features, in (34), the uninterpretable phi-features in T will probe

for a matching goal and find Victor, thus establishing Agree, chain-formation
and valuation of the phi-set in T. However, given that T is defective (i.e. it is not
selected by CHIGH, so is not temporally deictic and lacks an iT feature), valuation of
Case on Victor is not established and the DP remains active and open to further
Agree operations.19
The derivation proceeds to the next step which is insertion of the inflectional
subjunctive clitic să into the inflectional domain, followed by merge of Fin (CLOW)
and subsequent movement of să-T into CLOW. CLOWP is projected and selected by
the matrix clause predicate încearcă ‘try’. This new syntactic object (VP) is in turn
merged with matrix v which furthermore has an unvalued theta-role to satisfy.
Note that this theta-role can only be satisfied structure internally. Merge of a
distinct DP from the Numeration would prevent the embedded DP subject from
ever checking its Case feature and would thus cause a crash at the interface levels.
The main clause v probe finds Victor, an active goal in the relevant search space,
which, as an inherent D, is capable of satisfying the probe’s thematic require-
ments. See (35).
(35)
v VP
u[θvm]
V CLOWP
try SBJ Victor play the trombone

iφ, D
uT
Crucially, valuation of u[θvm] via the Agree operation in (35) is the necessary and
sufficient condition for the OC dependency. In principle, no dislocation should
be required to satisfy the matrix predicates external thematic role and, in fact, no
dislocation ensues for this purpose.
Next, matrix T merges with matrix vP, and subsequently phasal C (CHIGH)
merges with matrix T (shown here as a syncretic category for ease of exposition),
as in (36).
(36) CHIGH / TP
vP
CHIGH / T
uφ
v VP
iT
u[θvm]
V CLOWP
try SBJ Victor play the trombone

iφ, D
uT
The phasal head guarantees i T both on the main clause and on the embedded
T, as well as Case-licensing of the embedded/matrix DP subject. The uninter-
pretable phi-features in T will probe for a matching goal. Given that Match and
Agree is established with the closest active DP in the c-command domain of
the probe, the goal that meets the required locality conditions turns out to be
Victor. This DP values the matrix unvalued phi-set while simultaneously valuing
it own uT, thus becoming inactive. Feature valuation is now complete and the
derivation converges as desired without any DP dislocation.20 In the process, the
shared DP subject has entered (at least) two A-chains: a thematic chain and a
Case chain.21
5.2 Accounting for PF copies
Having briefly discussed how the morpho-syntactic uninterpretable features

are accommodated in Romanian OC, I now address the issue of copy prefer-
ence. This subsection discusses the various PF instantiations of the shared
argument and shows that pronounced copies are dependent on the presence
of relevant semantico-pragmatic triggers for displacement (e.g focus, de-rhe-
matization, etc.).
I have argued in previous work (Alboiu 1999, 2002, 2004a) that Romanian
exploits syntactic structure to encode sentence pragmatics. Specifically, independ-
ent of formal feature checking, phrases may dislocate for novel interpretive effects
related to the encoding of the Theme–Rheme partitioning within the sentence. In
current theoretical terms (Chomsky 2001b), this displacement is formalizable as
an OCC (occurrence) feature optionally present in the derivation.22 For Roma-
nian, see the schema in (37).
(37) (Topic XP*) – (Kontrast XP) – C/T(OCC) – [vP (OShift) – [vP … v(OCC) …]]
Theme Rheme
206 GABRIELA ALBOIU
While the discussion in this section is by no means exhaustive, a closer look at the
various instantiation of DP copies in OC contexts does show that the pronuncia-
tion site is intrinsically linked to the encoding of information structure, which in
turn explains the apparent ubiquitous behavior of the shared argument.
In the absence of any OCC feature in the derivation, the shared argument
fails to undergo dislocation and is pronounced in situ, in the Spec,vP of the
subjunctive predicate, as illustrated in (33) in the previous subsection. While
from a syntactic viewpoint, the effect is that of backward control, pragmatically
speaking, the DP is part of the presentational, rhematic focus of the embedded
sentence, as illustrated by the dialogue in (38).
(38) Embedded clause-Rheme:

a. Ce e gălăgia asta?
what is noise-the this
‘What’s all this noise?’
b. Încearcă [să cînte [vP Victor / cineva tv+V la trombon]]

try.PRES.3SG [SBJ sing.3SG [vP Victor / someone tv+V at trombone]]
‘Victor / Someone is trying to play the trombone.’
In (38b), the embedded vP is interpreted as new information, there is no DP dis-

placement and, consequently, no OCC feature present in the derivation. That the
DP subject occupies an A-position is indicated by the fact that bare quantifiers
like cineva ‘someone’ is also licensed in this slot.23, 24
The shared argument can also be instantiated as part of the rhematic domain
of the matrix clause. In this case, displacement occurs to the matrix Spec,vP, which
– under the current approach – would be due to the presence of an OCC feature
on the higher v predicate. Such an OCC feature simply spells out the requirement
that the shared DP surfaces in an intimate relationship with the matrix v rather
than the embedded v. Linearization in (39b) is appropriate to (39a).
(39) Matrix clause-Rheme:

a. Ce se întîmplă?
what SE happens
‘What’s going on?’
b. Încearcă [vP Victor/ cineva tv+V / OCC [să cînte [vP tDP tv+V
try.PRES.3SG [vP Victor/ someone tv+V / OCC [SBJ sing.3SG [vP tDP tv+V
la trombon]]]
at trombone]]]
When, the shared argument is the exclusive new information, rhematic focus
in the sentence, it will appear maximally embedded in the subjunctive predi-
cate. In Alboiu 1999, 2002, I argued that maximal embedding of the subject
DP is achieved in situ, by dislocating (i.e. ‘evacuating for focus’) any additional
vP-internal material. Note that this claim is supported by the availability of a
bare quantifier subject. Consequently, in (40b), which is the adequate answer to
(40a), the object DP undergoes ‘object shift’ to the left-edge of vP for pragmatic
purposes (i.e. de-rhematization), formalized as an optional OCC feature on sub-

junctive v (see also discusssion in Alboiu 2004a).25
(40) Presentational Focus argument (DP subject Rheme):

a. Cine încearcă [să cînte la trombon]?
who try.PRES.3SG [SBJ sing.3SG at trombone]
‘Who is trying to play the trombone?’
b. (Încearcă [să cînte [vP la tromboni) [vP Victor / cineva

try.PRES.3SG [SBJ sing.3SG [vP at trombonei [vP Victor / cineva
tv+V / OCC ti]]]
tv+V / OCC ti]]]
In (41), on the other hand, the shared subject argument is known to both speaker
and hearer – as indicated by (41a) – and is consequently interpreted as a Topic. If
visible, in (41b), it surfaces in the matrix sentence preverbal domain, at the left-
edge of the Theme, outside of the main clause predicate Rheme. Given that it is
not initially merged in the Topic domain, it is reasonable to assume that disloca-
tion occurs due to an OCC requirement on matrix C/T domain.26 Given that this
is an A-bar Topic position, the bare quantifier is ruled out.
(41) Matrix predicate-Rheme (DP subject Topic):

a. Mihai, ce face Victor?
Mihai what does.3SG Victor
‘Mihai, what’s Victor doing?’
b. (Victor/*Cineva) încearcă-TOCC [vP tDP tv+V [să cînte

Victor/ * someone) try.PRES.3SG-Tocc [vP tDP tv+V [SBJ sing.3SG
[vP tDP tv+V la trombon]]]
[vP tDP tv+V at trombone]]]
‘Victor/* Someone is trying to play the trombone.’
Last, but not least, the shared argument can be interpreted as contrastively focused
(i.e. Kontrast).27 While there seems to be evidence that contrastively focused con-
stituents trigger operator-variable chains in Romanian (see Alboiu 2003, 2004a),
dislocation is not crucial, the only requirement being heavy prosodic stress. (42)
exemplifies how heavy stress – represented by upper case letters – yields contras-
tively focused readings in all of the previously mentioned slots.
(42) Contrastive Focus argument (DP subject Kontrast):

a. Mihai încearcă [să cînte la trombon]?
Mihai try.PRES.3SG [SBJ sing.3SG at trombone].
‘Is Mihai trying to play the trombone?’
b. (VICTOR) încearcă [vP (VICTOR) [să cînte

(VICTOR) try.PRES.3SG [vP (VICTOR) [SBJ sing.3SG
[vP (VICTOR) la trombon (VICTOR) ]]].
[vP (VICTOR) at trombone (VICTOR) ]]].
‘It’s Victor that’s trying to play the trombone (not Mihai).’
208 GABRIELA ALBOIU
To sum up then, the shared argument of Romanian OC constructions only ‘moves

forward’ to ensure novel semantico-pragmatic effects. The PF copy instantiation
of the subject DP is not incumbent on morpho-syntactic featural requirements
related to OC per se but, rather, is dependent on the interpretation requirements
of the shared argument itself in conjunction with the information packaging strat-
egies afforded by the matrix and the embedded clause domains.
6. CONCLUSIONS
In this chapter, I have discussed properties of subject OC in Romanian and have

argued for a reductionist view of control where the crucial ingredient is a non-
phasal clausal complement. This proviso guarantees an active subject Goal which
may enter into both thematic and non-thematic checking operations with matrix
probes. However, whether dislocation ensues or not seems to be an independent
language specific property. For a language like Romanian, where DP linearization
strategies are intimately tied in with semantico-pragmatic factors (rather than,
for example, the EPP), the shared argument was seen to surface in a number of
distinct positions in accord with relevant interpretations.
I proposed that the optimal solution for Romanian OC relies on the construal
of theta-roles as features which need not check in a sisterhood relationship but can
be valued solely by the operation Agree. Empirical evidence for this comes from
the various copy availabilities, which rules out a PRO analysis, and the interpre-
tive requirements of OC structures with Dative controllers. The analysis adopts
insights from Hornstein (1999 et seq.) but differs from that approach in at least
two ways: first, I claim that theta-roles can be satisfied simply by chain formation
without any dislocation, and second, I show that A-chains cannot cross phasal
CP boundaries. Arguably, control can only be construed as raising (in the sense
of A-chain formation) for mono-clausal domains only (i.e. domains which either
lack phasal status or which can obviate their phasal status, as discussed). I show
that for Romanian, the mono-clausal domain is not as reduced as in proposals by
Wurmbrand (1998, 2004) or Cinque (2004), but can expand all the way up to a
low, non-phasal C domain (i.e. the Fin of Rizzi, as proposed by Hill 2003).
I also showed that any extant asymmetries group together raising and OC
against NOC constructions in Romanian. As would be predicted by an A-chain
analysis, this was seen to hold for all relevant properties. Specifically, phasal
(CHIGHP) vs. non-phasal (CLOWP) status, temporal deixis, Case valuation properties,
A-chain formation, locus of insertion of shared argument, PF occurrence, and
subjecthood properties of quirky Datives.
Equally important is the fact that Case valuation is not synonymous to
inflected T domains but to phasal domains. Empirical support for this proposal
comes from the behavior of emphatic chains, among other things. The absence
of an independent Nominative domain in the subjunctive OC complement not
only rules out a referential pro subject but explains why these predicates cannot
relinquish their OC readings. The Caseless embedded subject will have to associ-
ate with matrix v and subsequently, matrix T, or its Case deficiency will never be
satisfied, a disastrous outcome. The OC reading is forced by the unfinished busi-

ness of embedded arguments.
Consequently, there is no escape from control for predicates selecting reduced,
non-phasal clausal domains, and cheeky enough to flaunt their own theta-role.
*
For various discuss ion and/or comments, I would like to thank Ion Alboiu, Larisa Avram,
Alexandra Cornilescu, William Davies, Stanley Dubinsky, Sam Epstein, Norbert Hornstein, Konstan-
tia Kapetangianni, Idan Landau, Virginia Motapanyane-Hill, Johan Rooryck, David Pesetsky, Daniel
Seely, two anonymous reviewers, as well as the audience of the 2005 LSA Workshop on New Horizons
in the Grammar of Raising and Control. All errors are mine.
1
See discussion in Hornstein 1999, 2000, 2001, 2003, Boeckx and Hornstein 2003, 2004, this volume,
and Landau 1999, 2003, 2004a, this volume.
2
To simplify, in (5), I disregard properties of the embedded T. I also use ‘T’ notation throughout the chap-
ter, despite the fact that there is variation in some of the cited work with respect to ‘T’ vs. ‘I’ notation.
3
The abbreviations used in the Romanian example sentences are: SE: impersonal clitic, AUX: auxiliary,
SBJ: subjunctive, INF: infinitive, IND: indicative, PRES: present tense, PART: participle, NEG: negative,
CL: pronominal clitic, SG: singular, PL: plural, NOM: Nominative case, ACC: Accusative case, DAT:
Dative case, M: masculine, F: feminine. ‘PE’ is a preposition associated with Romanian direct objects
that have an <e> type denotation (see Cornilescu 2000b).
4
These properties are shared with modern Greek (see Alexiadou and Anagnostopoulou 2002; Spyro-
poulos, this volume; Kapetangianni and Seely, this volume).
5
Motapanyane-Hill (personal communication) points out that clitic climbing is independently ruled out
in (10b), given the clitic status of the subjunctive particle and the ban against excorporating from within
clitic domains. Either way, this only strengthens the argument against restructuring as it highlights the
fact that clitics are licensed in the embedded clause. This, in effect, points at least toward a T status of
the subjunctive domain, following assumptions in Kayne 1991 for Romance clitics.
6
Note that (11b) illustrates lexical verb raising to T, a dislocation which is obligatory for Romanian
(see Dobrovie-Sorin 1994).
7
In the embedded subjunctive, both instances of the DP subject Victor are boxed given that they both
occupy Spec,vP, the initial Merge position of the shared argument. As the discussion in Section 5
will clarify, under specific pragmatic constraints, the direct object raises above (and to the left of)
the subject in Spec,vP, seemingly yielding two subject copies within the embedded clause.
8
Unsurprisingly, neither is dislocation a prerequisite in standard raising constructions in this language:
(i) (Mihai) pare [să fie (Mihai) băiat deştept (Mihai)]
(Mihai) seem.3SG [SBJ be.3SG (Mihai) boy smart (Mihai)]
‘Mihai seems to be a smart guy.’
9
An anonymous reviewer expressed concern that the proposed analysis is merely ‘an exercise’ (see
also Kapetangianni and Seely, this volume) and worried about ‘independent evidence’ for it. I hope to
have shown that none of the previous analyses work, and that the proposal here is not only theoreti-
cally coherent but also empirically comprehensive. It seems to me that if we indeed aim at doing justice
to the multifaceted aspects of Romanian subject control, there simply is no other road to take.
10
Note that Barrie (this volume) argues that FinP is a phase in English based on the behavior of
wh-infinitivals. Crucially, extraction of the subject DP is argued to block extraction of a wh-phrase
as the escape hatch for movement [Spec,FinP] has already been used in the A-movement operation.
However, phases are known to block A-movement (Chomsky 2000 et seq.), a constraint I take to
be correct, as well as to hold crosslinguistically. Hornstein (2000:137), being equally aware of this
problem, especially given that in English many of the OC contexts are structurally more complex
than canonical raising constructions (recall the classical CP vs. IP split), argues as follows, ‘assume
that some mechanism, say incorporation, can void the CP phase derivationally …’. Crucially, for
A-movement to occur, the phase status has to be obviated. One could speculate some mechanism
whereby movement of a DP with unvalued features (say, Case) to the left edge of the phase would not
only block A-bar movement (as discussed by Barrie) but would also guarantee obviation of the phase.
210 GABRIELA ALBOIU
11
See also Hill 2003, following Roberts 1997.
12
At this point I can clarify why this chapter refers exclusively to subject, as opposed to object, con-
trol. In contrast with subject control, object control is irrelevant to the present discussion as the
matrix verb selects a non-anaphoric (CHIGHP) subjunctive, rather than an anaphoric CLOWP. This is
shown in (ia) where, furthermore, we can also notice the availability of Nominative Case (bolded
pronoun), as expected in view of the phasal status of these subjunctives. Given the pro-drop nature
of Romanian, (ib) is equally unsurprising.
(i) a. Li-am rugat pe Ioni [CHIGHP ca mîine
CL.3SG.M.ACCI-AUX.1SG asked PE Ioni [CHIGHP that.SBJ tomorrow
Să plimbe eli cîinele]
SBJ walk.3SG 3SG.M.NOMi dog-the]
‘I asked John to walk the dog tomorrow.’
b. Li-am rugat pe Ioni [CHIGHP să plimbe
CL.3SG.M.ACCi-AUX.1SG asked PE Ioni [CHIGHP SBJ walk.3SG
proi cîinele]
proi dog-the]
‘I asked John to walk the dog.’
13
See section 5 for a discussion of these subject positions.
14
The same grammaticality judgments obtain with other OC predicates such as reuşeşte ‘manages’ or
with a raising predicate like pare ‘seems’.
15
Note that this conclusion forces us to renounce pro which presupposes independent Nominative
valuation (see issue raised in section 3).
16
That ‘manage’ is a control predicate rather than a raising predicate is evidenced by its properties in
(i) and (ii) below. Both examples illustrate thematic restrictions: (i) shows sensitivity to the seman-
tics of the DP argument, while (ii) shows that the passivized complement of the implicative is not
truth-conditionally synonymous with its active counterpart. These classical tests make it clear that
‘manage’ assigns an external theta-role in Romanian.
(i) a. Victor a reuşit să plece.
Victor AUX.3SG managed SBJ leave.3
‘Victor managed to leave.’
b. *Apa a reuşit să fiarbǎ.
water AUX.3SG managed SBJ boil.3
*‘The water managed to boil.’
(ii) a. Victor a reuşit să-l ajute pe Mihai.
Victor AUX.3SG managed SBJ-CL.SG.M.ACC help.3 PE Mihai
‘Victor managed to help Mihai.’
b. Mihai a reuşit s fie ajutat de Victor.
Mihai AUX.3SG managed SBJ be.SBJ.3 helped by Victor
‘Mihai managed to be helped by Victor.’
Where, (ii a) is ≠ from (ii b).
17
Note that this particular subjunctive predicate does not itself inflect for agreement.
18
This is, of course, a D feature.
19
Recall that the lexical verb raises to the T domain but I do not show this here as it is irrelevant to
our discussion.
20
Recall that Romanian lacks movement to Spec,TP for EPP-type purposes (see section 3.4).
21
The ‘at least’ specifications stems from the fact that, arguably, the embedded phi-feature chain is
also an A-chain. However, given that this chain is irrelevant to the DP, I do not focus on it.
22
Where, following Chomsky 2001b, OCC is a requirement that a phrase must be an occurrence (i.e.
sister) of some probe and that this sisterhood relationship must license novel interpretations.
23
Cinque (1991) argues that these quantifiers cannot occupy A-bar positions. Consequently, I take
their occurrence throughout to indicate A-domains.
24
Note that I only show movement of the lexical verb to T (via traces) where this movement is relevant
for the interpretation of the shared DP argument. Specifically, in (38b) and (40b), this is indicated for
the embedded lexical verb but not for the main clause verb, while in (39b) and (41b), it is indicated in
both cases and in (42b), it is not indicated at all.
25
Note that other Romance languages also seem to allow for vP-adjoined object raising with specific
semantico-pragmatic and syntactic properties: for Catalan, see discussion in Vallduví 1995, for
Portuguese, see Costa 1999, and for Spanish, see discussion in Ordóñez 1998. It seems then that the
v-related OCC feature is available more consistently within Romance. Alternatively, Belletti (2004,
2005) proposes dislocation to an IP-internal focus position.
26
See discussion in Alboiu 2002 against independent Topic and Focus projections in Romanian.
Under the cartographic approach, dislocation would proceed to Spec,TopP, with the OCC require-
ment as a property of the Topic head. Nothing crucial hinges on this distinction.
27
Clarification of concepts is required at this point. New information/presentational/rhematic focus is
to be kept distinct from contrastive focus discussed so far. The former category of focus covers material
that represents information newly introduced in the discourse and is the opposite of given/old informa-
tion, realized by the theme. Contrastive focus, on the other hand, is presupposed information, part
of what is given and consequently, part of the thematic domain. The distinct semantico-pragmatic
properties are paralleled by distinct syntactic properties, as shown in Table (i).
Table i. Contrastive Focus vs. Rhematic Focus
Affects truth-
A-bar chain [Foc] formal Prosodic functional
effects feature marking values of S
Contrastive focus + + + +
Rhematic focus − − − −
CILENE RODRIGUES
AGREEMENT AND FLOTATION IN PARTIAL AND

INVERSE PARTIAL CONTROL CONFIGURATIONS
1. INTRODUCTION
Traditional classifications of control distinguish two types: obligatory control

(OC) and non-obligatory control (NOC). Recently, it has been demonstrated
that OC should also be divided into two types: exhaustive and partial control (cf.
Landau 2000). This more refined view of OC is taken to be evidence against the
movement theory of control (MTC – cf. Hornstein 2001), the argument being that
partially controlled PRO does not behave like a trace. In this paper, I contribute to
advancing our understanding of control by looking at agreement and flotation in
control configurations and developing an analysis for partial control that results
from movement plus stranding.
The paper is organized as follows: section 2 discusses cases of past participle agree-
ment in control configurations, showing that exhaustively and partially controlled
PRO does not trigger φ-feature agreement independently of its controller. In view
of this, section 3 evaluates the Movement and the Agree-based theories of control,
concluding that partial control cannot be the result of Agree. Section 4 suggests that
partial control involves movement plus stranding of an associative null pronoun. In
section 5, inverse partial control in Spanish is introduced, and I show that it also fits a
movement and stranding analysis. Section 6 is dedicated to the conclusion.
2. CONTROL AND AGREEMENT
In Romance, ‘epicene’ DPs like victim are invariably feminine ([+Fem], henceforth),
though semantically they can refer to either male or female entities. Thus, when an
epicene noun is combined with a participial form, the latter triggers feminine gender
agreement. This is observed in Italian, Portuguese, and Spanish.
(1) a. La vittima fu aggredita/*aggredito dai fascisti (It.)

the victims-FEM was.3SG attacked-FEM/*MASC by fascists
b. La víctima1 fue atacada1/??atacado1 en la calle (Sp.)

the victim-FEM was-3SG attacked-FEM/-MASC in the street
c. A vítima1 foi atacada1/??atacado1 na rua (Port.)1

the victim-FEM was-3SG attacked-FEM/-MASC in.the street
‘The victim was attacked by (the) fascists/on the street.’
213
© 2007 Springer.
214 CILENE RODRIGUES
It is well known that past participle agreement is subject to locality. A DP α agrees

with a past participial form β only if α is the closest DP that c-commands β. The
French data in (2) and (3), for instance, show that a past participle agrees with a
direct object only when the later moves to the left of the former. Researchers have
taken this contrast to reveal that a past participle agreement is a spec–head rela-
tion. That is, in order to trigger φ-feature agreement on a participial form, a DP
has to move to/through the specifier of the maximal projection that hosts the past
participle (cf. Kayne 2000; Sportiche 1996, 1998)
(2) a. Jean a peint(*e) la porte (French)

Jean has-3SG painted-(*FEM) the door
‘Jean painted the door.’
b. La porte1 que Jean a peint(e) t1

the door that Jean has-3SG painted-(FEM)
‘The door that Jean painted’
(3) a. Jean l’a peint(e)

Jean it has-3SG painted(FEM)
‘Jean has painted it.’
b. Jean [le1 a [t1 [peinte … t1]]]
In Chomsky 2001a, although agreement is not a spec-head relation, past participle

agreement is taken to be the result of a local Agree relation as shown in (4):
(4) a. There were believed to have been caught several fish
b. [α caught φF [DO several fish ]φF]

Agree
This locality constraint is, at first sight, violated in sentences like (5), in which the
epicene noun is within a different clause than the past participle, yet gender agreement
is obligatory. Crucially these sentences were judged in a context in which the DP
the victim was taken to refer to a male entity.2
(5) a. La vittima1 ha cercato di essere trasferita1/??trasferido1 (It.)

the victim-FEM had-3SG tried of be-INF transferred-FEM/-MASC
alla stazione di polizia de College Park
to.the station of police of College Park
b. La víctima intentó ser transferida /??transferido (Sp.)

the victim-FEM tried-3SG be-INF transferred-FEM/-MASC
c. A vítima1 tentou ser transferida1/??transferido1 (Port.)

the victim-FEM tried be-INF transferred-FEM/-MASC
para a delegacia de polícia de College Park
to the station of police of College Park
‘The victim (semantically masc) tried to be transferred (to the police station at College Park)’
AGREEMENT AND FLOTATION IN PARTIAL AND INVERSE 215
These are all instances of exhaustive control in the sense of Landau 1999, 2001,
where the controllee (PRO) and the controller (i.e. a [+Fem] DP]) are identical
in reference. Therefore, if past participle agreement is local, we may conclude
that exhaustively controlled PRO is not in itself contrastive for gender agreement.
That is why agreement with the controller can occur under PRO. Hence, exhaus-
tively controlled PRO is similar to DP-traces (cf. (6) ), which also lack independent
contrastiveness for gender agreement.
(6) a. La vittima sembra essere ferita/*ferito (It.)

the victim-FEM seems be-INF injured-FEM/MASC
b. La víctima parecía estar herida/*?herido (Sp.)

the victim(semantically male) seemed-3SG be-INF injured-FEM/-MASC
c. A vítima pareceu estar ferida/*?ferido (Port.)

the victim seemed-3SG be-INF injured-FEM/-MASC
‘The victim (semantically male) seemed to be injured.’
Interestingly, agreement is also obligatory in partial control (cf. Landau 1999, 2000;
Wurmbrand 2001), where the controller is not identical to PRO in reference, being
interpreted as a strict subset of the set denoted by PRO (viz. the acceptance of plural
predicates within the controlled clauses). In (7), for instance, the adjectival secondary
predicate agrees in gender and number with the [+Fem] DP. Let me emphasize that
number agreement is also observed in these constructions. The secondary predicate
records plural agreement only if the controller is syntactically plural (cf. (7b) ).
(7) a. A vítima quer se encontrar bêbada/ *bêbadas/ (Port./Sp.)3

the victim-FEM.SG wants-3SG SE meet-INF drunk-FEM.SG/drunk-FEM.PL
*bêbado/ *bêbados
drunk-MASC.SG/drunk-MASC.PL
‘The victim wants to meet (with somebody else) drunk’
b. As vítimas querem (se) encontrar *bêbada/ bêbadas/

the victims-FEM want .3R PL SE meet-INF drunk-FEM.SG/drunk-FEM.PL
*bêbado/ *bêbados
‘ The victims(semantically males) want to meet drunk’
Thus, we may generalize and say that (exhaustively or partially) obligatory-control-

led PRO does not trigger φ-feature agreement independently of its antecedent.
By contrast, non-obligatory-controlled PRO might trigger φ-feature agreement
on its own. In (8), agreement with the epicene noun is disallowed by some Italian
speakers, whereas in Portuguese, it is quite degraded. In Spanish, it seems to be gen-
erally optional. Notice that, as in the above data, these sentences were all judged in
a context in which the victim was known to be a male.
(8) a. La vittima ha detto che essere *portata/portato (It.)

the victim-FEM has-3SG said that be-INF brought-FEM/-MASC
alla stazione di polizia non era una bona idea
to.the station of police not was-3SG a good idea
b. La víctima dijo que ser ?tranferida/transferido a (Sp.)

the victim-FEM said-3SG that be-INF transferred-FEM/-MASC to
otra ciudad no es buena idea
another city not is-3SG good idea
c. A vítima disse que ser ??tranferida/trasferido (Port.)

the victim-FEM said-3SG that be-INF transferred-FEM/-MASC
para outra cidade não é uma boa idéia
to another city not is-3SG a good idea
‘The victim(semantically male) said that being brought to the police
station/being transferred to another city is not good idea’
It is possible to interpret the epicene DP as the subject of the infinitival clause,

yet syntactic agreement is at best optional. Therefore, it is arguably the case that
in non-OC configurations, an agreeing element within the c-command domain of
PRO, agrees with PRO.
By way of comparison, we conclude that non-obligatory-controlled PRO
patterns with the Romance finite null subject pro, which is also able to establish
agreement independently of its antecedent. Similar to agreement in an NOC
configuration, we may observe that in (9), [+Fem] agreement is unacceptable
under an embedded finite null subject (ie. pro), on the reading in which the victims
are males. Again in European Portuguese and Spanish, the agreement seems to be
optional, but some speakers of Italian disallow it.
(9) a. La vittima1 ha detto che pro1 era *stata aggredita/ (It.)

the victim-FEM has-3SG said-3SG that was-3SG been-attacked-FEM
stato aggredito in strata
been attacked-MASC in street
b. La víctima1 dice que pro1 fue atacada/atacado en la calle (Sp.)

the victim said-3SG that was-3SG attacked-FEM/MASC in the street
c. A vítima1 disse que pro1 foi atacada/atacado na rua (Port.)4

the victim said-3SG that was-3SG attacked-FEM/MASC in. the street
‘The victim(semantically male) said that he was attacked on the street.’
Therefore, the contrast between obligatory-controlled and non-obligatory-controlled

PRO leads us to conclude that only non-obligatory-controlled PRO is contrastive
for agreement features. By contrastive, I mean that non-obligatory-controlled PRO
is able to trigger agreement on a past participle form, independently of the
syntactic agreement features of the matrix subject. This stands in stark contrast
to obligatory-controlled PRO, which is syntactically dependent on the matrix
subject in this regard.
Notice that the cases of OC we have examined thus far are all cases in which
there could have been a syntactic mismatch between the gender feature of the
controller and the gender feature of PRO. Since the controller is semantically
masculine but syntactically feminine, PRO could, in principle, be semantically
and syntactically masculine. This mismatch, however, does not happen. It does
not happen even in cases of partial control, where PRO is apparently referen-
tially independent of its controller. (PRO refers to a semantically plural entity,
whereas the controller refers to a singular entity.) The partial control in (7) also
shows us that there is no syntactic mismatch between the number feature of the
controller and the number feature of PRO, even though this mismatch exists
semantically. Therefore, obligatory-controlled PRO behaves as though syntacti-
cally inert for feature agreement even when it has its best chance of demonstrat-
ing its independence. In what follows, I discuss the theoretical implications of
this finding.
3. THEORETICAL IMPLICATIONS OF AGREEMENT UNDER CONTROL
Within minimalism, OC has been analyzed as an instance of NP-movement in

Hornstein (1999, 2001) and as the result of Agree in Landau (1999, 2000). The
movement analysis argues that obligatory-controlled PRO is a trace of the
controller (cf. (10a), whereas noncontrolled PRO is pro, as represented in (10b)).
(10) a. [John1 tried [t1 to win]]
b. [John thinks [ that [pro winning] is crucial]]
The Agree analysis, on the other hand, claims that control results from structures
like (11a), in which PRO agrees with the matrix T, thus inheriting the φ-features of
the controller.5 NOC (or, more simply, noncontrol) structures are like (11b), where
PRO occurs inside an island, being thus unable to agree with an external head.
(11) a. [TP T [VP John tried [PRO to win]]]

Agree I
Agree II
b. [TP T [VP John thinks [that [PRO winning] is crucial]]]

Agree
Agree
*
In principle, either of these two analyses can account for the contrast in agree-
ment between non-obligatory-controlled and exhaustive-controlled PRO. Under
the movement analysis, in exhaustive control configurations, the controller
controls agreement because it starts the derivation as the subject of the embed-
ded clause, agreeing, thus, with the embedded past participle/quantifier. In non-
control configurations, agreement with the matrix subject is voided because the
null pronominal subject of the infinitival clause (i.e. pro) agrees with the past
participle.
Under the Agree analysis, in exhaustive–obligatory control configurations,
the controller defines the agreement because the past participle/quantifier
φ-agrees with PRO, which in turn φ-agrees with the controller. In NOC, agreement
with the matrix subject is voided because the agreeing form (past participle) is
within an island (a clausal subject), which, according to Landau, blocks agreement

with external probes.
Hornstein’s and Landau’s proposals for NOC are roughly variants of one
another: subject islands block move/agree. However, the data presented here shows
that their analyses for OC are theoretically distinguishable. To account for the agreement
pattern observed in OC, the Agree analysis must include a feature-sharing principle
(cf. Frampton et al. 2000; Pesetsky and Torrego 2004a), stating that if a constituent
α φ-agrees with PRO and PRO φ-agrees with the controller (even if they differ in
semantic features), then α also φ-agrees with the controller. As shown in (12) without
a feature-sharing principle, Landau’s Agree theory needs cumbersome derivations
for OC configurations displaying past participle agreement. Before moving to the
spec of the embedded TP, PRO φ-agrees with the past participle (Agree I ), and,
since PRO’s gender feature is either [+Fem] or [+Masc], the past participle is
valued either as [+Fem] or [+Masc]. However, when Agree II andI III happen, PRO
Agrees with T, inheriting all the φ-features of the controller (la vittima), including
[+Fem] gender. Now, in order to guarantee that the past participle will also come
out as [+Fem], (12) must apply a countercyclic Agree operation (Agree IV ), which
rematches the φ-features of PRO with the φ-feature of the past participle.
(12) [TP T [VP La vittima1 ha cercato [PP di [TP PRO1 [T’ [essere, T] [ t1 trasferita]
Agree II Agree I
Agree III Agree IV
Thus, as stated above, in order to avoid countercyclicity, an Agree-based theory of

control would need to adopt a version of transitivity in feature sharing. Under a
movement theory, such a feature-sharing principle is superfluous.
Let us now consider agreement in partial control. Landau argues that OC
cannot be subsumed under movement given the existence of partial control,
where the controller is understood as a strict subset of the set of entities denoted
by PRO. Landau suggests that partial control is a special case of Agree in which
PRO is lexically specified with a plural semantic number feature, and even
though PRO and the controller share the same set of valued syntactic φ-fea-
tures, the semantic number feature of PRO does not enter into an Agree relation
with features of the controller. This analysis seems to be inconsistent with the
agreement pattern observed above. Let us analyze (13), taking Landau’s theory
into consideration.
(13) a. A vítima quer se encontar bêbada/ *bêbadas/ (Port.)

the victim-FEM.SG wants-3SG SE meet-INF drunk-FEM.SG/drunk-FEM.PL
*bêbado/ *bêbados
‘The victim wants to meet (with somebody else) drunk’
b. As vítimas querem (se) encontrar *bêbada /bêbadas/

the victims-FEM want .3R PL SE meet-INF drunk-FEM.SG/drunk-FEM.PL
*bêbado/ *bêbados
‘The victims(semantically males) want to meet drunk’
For Landau, PRO in (13) has a semantically plural number feature. However, even
though PRO is semantically plural and masculine in (13), it cannot license syn-
tactic gender and number agreement on an adjective. This is surprising, since, in
Romance a semantic number or gender feature on a pronominal element (which
Landau’s PRO is supposed to be) does trigger syntactic agreement on adjectives,
as discussed by Costa and Pereira (2005) and D’Alessandro (2004). In (14), for
instance, the plural agreement features of the adjectives are trigged by the pro-
nouns a gente (European Portuguese) and si (Italian), which are syntactically sin-
gular but semantically plural.6 The same is true of gender agreement: the adjectives
are either masculine or feminine, depending on the referents of the pronouns.7,8
(14) a. A gente está cansados/cansadas (Eur. Port.)

we-SG is-3SG tired-MASC.PL/tired-FEM.PL
‘we, the girls, are tired.’
b. Se si è belli/belle, si è di solito (It.)

if Si is-3SG beautiful-MASC.PL /beautiful-FEM.PL si is-3SG often
anche ricchi/ricche
also rich-MASC.PL/rich.FEM.PL
‘If one is beautiful, one is usually also rich.’
Thus, given that semantic number and gender on pronouns ordinarily triggers
agreement on adjectives, the fact that in (13) the adjective agrees in gender and
number with controller and not with PRO suggests that partially controlled PRO does
not differ from its controller in semantic number and gender features. If PRO in
(13a) had a different set of semantic number and gender features, we would expect
it to be able to license syntactic plural and masculine gender agreement on the
secondary predicate.
In sum, neither the movement theory nor the Agree analysis of control
provides a full explanation for why partially controlled PRO behaves as it does.
The Agree analysis fails to explain why this empty category is not contrastive for
gender and number agreement, whereas the movement analysis as proposed by
Hornstein does not explain the semantics of this category. In what follows,
I suggest partial control should be derived from movement plus stranding of a
null associative pronoun.
4. PARTIAL CONTROL AS MOVEMENT PLUS STRANDING
Partial control is clearly puzzling. Its syntax (e.g. the agreement facts discussed
earlier) suggests that it should be treated on par with exhaustive-controlled PRO.
On the other hand, its semantics (the fact that it is semantically plural even when
the controller is semantically singular) points in the opposite direction; that it
should be analyzed on a par with non-obligatory-controlled PRO. Nevertheless,
this dual behavior of partially controlled PRO may result from its complex syntax.
My suggestion is that partial control is better characterized as involving a complex
DP, with the adjunction of a null pronoun (pro) to a DP that is understood as
the controller. The proposal is that partially controlled PRO has the underlying
structure in (15), in which a null pronoun adjoins to a DP that moves leftwards,
stranding the adjoined pronoun.
•••
(15) DP
pro DP
Sentences like (16) have the structure sketched in (17). A null pronoun (pro)
adjoins to the DP the victim, forming a complex DP, which is then merged in the
embedded [spec VP].9 After that, the DP the victim moves to the embedded [spec
TP] to satisfy the EPP, stranding pro in [spec VP]. At the matrix level, the DP the
victim moves to the matrix [spec VP] to check the external θ-role of want and then
to [spec TP] where it is spelled-out.
(16) A vítima quer encontrar bêbada (Port.)

the victim-FEM.SG want-3SG meet-INF drunk-FEM.SG
‘The victim wants to meet drunk’
(17) CP
C TP
[DP a ví tima] T’
TP VP
t V’
quer TP
t T’
T VP
VP bêbada
DP encontrar
pro t
According to this analysis, the controller is interpreted as a subset of the set of

referents denoted by the embedded null subject because it shares the embedded θ-role
with an empty pronominal category (pro). However, only the controller triggers syn-
tactic agreement on the embedded secondary predicate. Since secondary predicates
are predicated of the DP in the closest spec of TP (cf. Koizumi 1993), in (17) the
secondary predicate agrees with a copy of the controller in the embedded [spec TP].10
Thus, the fact that pro remains stranded in [spec VP] has the following semantic con-
sequences: the secondary predicate must hold only for the controller. Example (16)
means that the victim wishes that she herself will be drunk during the meeting, inde-
pendently of whether the other participants of the meeting will be drunk or not.
The Movement plus stranding analysis proposed here raises at least three impor-
tant questions. First, how does the adjunction process in (15) work such that in (17)
the agent of the embedded event is semantically understood as plurality of entities
that includes the victim? Second, why is pro stranded in the embedded [specVP]? Third,
what happens in British English where partial control allows syntactic mismatches as
in (18a), which are not allowed in other syntactic environments (e.g. (18b) ).
(18) a. The chair preferred to consult each other before the vote.
b. *The chair invited each other to vote. (cf. Landau 2000).
I suggest that pro in (15) is an associative plural pronoun akin to the associative
morphemes -tati in Japanese and -men in Mandarin Chinese (See Nakanishi and
Tomioka 2002 on tati and Li 1999 on men).11
(19) a. Yamada-sensei-wa Taroo-tati-o syokuzi-ni syootai-sita (Japanese)

Yamada-teacher-TOP Taro-PL-ACC meal-to invited
‘Professor Yamada invited Taro and those in his group for dinner.’
b. Xiao Qiang-men shenme shihou lai? (Chinese)

Xiao Qiang-PL what time come
‘When are XiaoQiang and the others coming?’
As the English translation of (19a and b) indicates, attachment of an associate

plural to DP yields a plural reading of that DP. In (19a), for instance, Taroo-tati
means Taro and others, a group represented by Taro. Thus, I suggest we understand
pro in (15) as null associative morpheme that attaches to a DP yielding a plural
reading of that DP.
Interestingly, however, the ungrammaticality of (20) suggests that the associa-
tive pro cannot occur in a matrix clause, even with predicates compatible with a
collective interpretation.
(20) *The victim meets drunk.
Researchers on partial control have argued that partial control is restricted to

infinitival clausal domains embedded under certain verbs. Landau (2000), for
instance, presented a detailed account of the syntactic distribution of partial and
exhaustive control, concluding that exhaustively controlled clauses are embed-
ded under implicative, aspectual, and modal predicates, whereas partial control
occurs with factive, propositional, desiderative, or interrogative verbs. He further
suggests that exhaustive and partial control infinitival clauses differ in that the
former are untensed, whereas the later are tensed. Hence, only partially controlled
complements may express an event that does not coincide in time with the event
described by the matrix clause:
(21) a. *Last week John tried to leave yesterday/tomorrow.
b. Last week John wanted to meet yesterday/tomorrow.

Landau takes this tense distinction to be responsible for the difference in reference
between partial and exhaustive control, as the scheme in (22) shows. In exhaustive
subject control (22a), PRO first agrees with the embedded T-Agr, which is
φ-deficient. The matrix T φ-agrees with the matrix subject DP and with PRO;
consequently, PRO inherits all the φ-features of the matrix subject, including
semantic number features. Conversely, in partial subject control, in virtue of
its [Tense] feature, the embedded T-Agr moves to C, after having agreed with PRO.
This T-to-C movement blocks an Agree relation between the matrix T and PRO. It is
further assumed that when functional categories are probes they need not agree
with their goals in semantic features. Hence in (22b), T-Agr might not inherit
the semantic number features of the matrix subject. Thus, since PRO agrees with
T-Agr it will inherit all the φ-features of its controller, but it may not match its
controller for the semantic number feature. Therefore, PRO can be semantically
plural, while the controller is singular.
(22) a. [TP T [VP DP [V’ V [CP [IP PRO T-Agr [VP tpro • • •]]]] (cf. Landau 2000:8)
Agree II Agree III Agree I
b. [ TP T [ VP DP [V’ V [CP T-Agr [IP PRO tT-Agr [VP tpro • • •]]]]

Agree II Agree III Agree I
The clearest argument for (22) comes from sentences like (21), which Landau takes
to be evidence that partially controlled clauses, but not exhaustively controlled
clauses, are tensed. There is, however, no independent reason to posit the structure
in (22b). In fact, partial control may not even be conditioned by the presence of an
embedded tensed infinitival. Consider (23). Given the presence of a plural predicate,
we can safely conclude that (23) involves a partial control interpretation. (23a) is
an acceptable sentence even though the predicate embedded under the modal aux-
iliary must coincide with the modal in tense information, as shown by the ungram-
maticality of (23b).
(23) a. I can’t meet tomorrow. My daughter is getting married.
b. *Yesterday I couldn’t meet tomorrow. My daughter will be getting married.
Landau (2000:37–38) lists modals as predicates that license exhaustive control.

But as (23) shows, empirical facts point toward the opposite conclusion. What
licenses partial control here is the presence of a modal. In fact, a verb like try, a
prototypical exhaustive control verb, allows partial control, as long as it is embedded
under a modal auxiliary (cf. (24)). This is quite unexpected under any theory of
control that postulates that the difference between exhaustive and partial control
stems from tense properties of the infinitival T.12
(24) a. *I tried to meet yesterday, but I couldn’t guarantee that I would be there.
b. I can try to meet today, but I can’t guarantee that I’ll be there.
Arguably the linguistic requirement on partial control is that the null associative
plural pronoun must occur within the scope of a modal.13 This conclusion accords
with Wurmbrand’s (to appear) observation that partially controlled infinitival
clauses, such as (17), are not [+tense] clauses, being rather clauses whose T is
headed by a modal expressing future. Thus, in (17), we might say that the null asso-
ciative pronoun remains obligatorily stranded in the embedded [spec VP] because its
semantics requires it to stay under the scope of the future modal.
The existence of languages like British English, which allow a syntactic
number feature mismatch between the matrix and embedded clause, might
be evidence for the movement plus stranding analysis proposed here. Heim
et al. (1991) argue that reciprocal expressions are composed by a distributor
(each) and a reciprocator (other). At LF, the distributor adjoins to a plural DP
that is then interpreted as its antecedent. Assuming so, we might say that in
(18a) the distributor each adjoins to the copy of the DP [DP pro [DP the chair]]
that is in the embedded [spec VP]. Hornstein’s (2001) movement theory of
binding might also work here. For Hornstein, local antecedent–reciprocal relation-
ships result from movement. The antecedent and the reciprocal are merged
together, forming a unit, but the antecedent moves leftwards, leaving the
reciprocal behind. Under this analysis, (18a) involves a derivational step in
which the DP [DP pro [DP the chair]] is merged with each other, before moving
to the embedded [specVP].14
5. INVERSE PARTIAL CONTROL
Torrego (1996) observes that Spanish allows a full DP to float inside an infinitival
embedded clause, as in (25). This floating DP is semantically related to the matrix
subject in that it is obligatorily interpreted as a subset of the set denoted by the
matrix subject.
(25) no sabemos si firmar los lingüistas la carta

not know-1PL whether sign-INF the linguists the letter
‘We don’t know whether the linguists among us should sign the letter’
Thus, we may understand the relation between the floating DP and its antecedent as
an inverse partial control relation, where the set denoted by the controller is a superset
of the set denoted by the floating DP. Notice in sentences like (25), if there is no DP
floating, then an exhaustive control interpretation is obligatory. Therefore, these are
control structures.15
(26) (nosotros) no sabemos si firmar la carta

we not know-1PL whether sign-INF the letter
‘We don’t whether we should sign the letter’
It is important to note that the floating DP cannot occur in non-obligatory-

controlled clauses. Witness the data in (27).
(27) a. No sabemos si firmar la carta es una buena idea

not know-1PL whether sign-INF the letter is-3SG a good idea
‘We don’t know whether signing the letter is a good idea.’
b. *No sabemos si firmar los lingüistas la carta es una buena idea

not know-1PL whether sign-INF the linguists the letter is-3SG a good idea
‘We don’t know whether the linguists (among us) signing the letter is a good idea.’
The data in (25) and (27) are on a par with the Korean data in (28), which shows
that classifier stranding is possible in OC, but impossible in NOC.16
(28) a. Chingwutul-i [New York-ey sey-myeng ka-leyko] sitohay-ss-ta

friends-NOM New York-to 3-CLASSIFIER go-C try-PAST-DEC
‘Three friends tried to go to N.Y.’
b. *Chingwutul-i [New York-ey sey-myeng kanunkes-i caymi-iss-ul-kes-ila-ko]

friends-NOM New York-to 3-CLASSIFIER going-NOM fun-be-RC-thing-COPULAR-C
malhay-ss-ta
said-PAST-DEC
‘Friends said that 3 (friends) going to New York would be fun’
This suggests that only obligatory-controlled PRO is able to support stranded

material. This is the first time this restriction on stranding has been observed,
and I believe it should be tied to the Case Filter, which states that phonologi-
cally overt NPs must check Case. Any stranded nominal expression that has
phonological features is expected to require Case. Thus, assuming that the
subject position of an infinitival clause is not a Case position, it follows that
non-obligatory-controlled PRO does not license stranded material. In OC
configurations, on the other hand, stranding is possible because the controller,
which moves to a Case position (the subject of the finite matrix clause), forms
a chain with the DP containing the stranded material. If it is the case that
once one member of a given chain is valued for Case, all the other members
of that chain are also valued, then it follows that in OC the movement of the
controller to a Case position is what licenses stranded overt DPs inside the
embedded clause.
Having said that, I will now present empirical arguments for assuming that the
inverse partial control in (25) involves movement of controller plus stranding of
the floating DP. In other words, I will demonstrate that the underlying structure
of (25) is (29), and not (30):
(29) [TP pro2 sabemos [CP si [TP t2 [T’ ir1 [VP [XPlos lingüistas t2] t1]]]]]
(30) [TP pro sabemos [CP si [TP PRO2 [T ir1 [VP [DP los lingüistas t2] t1]]]]]
5.1 Evidence for a movement plus stranding analysis of Inverse Partial Control
As the data in (31) shows, the DP-floating phenomenon under analysis is not
restricted to infinitival embedded clauses. It also happens in matrix clauses and in
clitic-doubling structures (cf. Torrego 1996).
(31) a. Fuimos los lingüistas

went-1PL the linguists
‘The linguists among us went.’
b. Nos vio a los lingüistas

to-us saw-3SG to the linguists
‘She saw the linguists among us’
However, as Torrego (1996) demonstrates, there are restrictions on the licensing of

floating DPs. First, the antecedent of a floating DP cannot be an overt pronoun
in an A-position. Example (31) conforms to this restriction. In (31a) the anteced-
ent is in an A-position (specTP), but is a null pronoun. In (31b) the clitic nos is
arguably in an A’-position. Example (32), on the other hand, is out because the
antecedent is an overt pronoun, nosotros, in an A-position.
(32) *Nosotros fuimos los lingüistas

we went-1PL the linguists
‘The linguists among us went…’
(33) shows that this restriction on the antecedent is also observed in inverse partial
control configurations, in which the controller cannot be an overt pronoun.
(33) (*Nosotros) no sabemos si ir los lingüistas.

we not know-1PL whether go-inf the linguists
‘We do not know whether the linguists among us should go’
If the putative structure of (32) were (30), it would be hard to explain why the
controller in (33) must be null. According to (30), the local antecedent of the float-
ing DP los lingüistas is PRO, which is a null pronoun. Therefore, given the analysis
in (30), there should be no such requirement on the controller, which is not syn-
tactically related to the floating DP. Thus, the fact that controller is subject to the
restriction under discussion strongly suggests that the underlying structure of (33)
is (29), not (30). The syntactic antecedent of los lingüistas is the controller, and
that is why the controller must be null if it is a pronoun in an A-position.
Additional evidence for a movement analysis comes from constraints on
extraction out of subject in Spanish. In this grammar, extraction out of left
branch DPs is possible only when DP is in [spec VP] (34b) or in [spec CP] (35), but
not when in [spec TP] (34a) (cf. Torrego 1984; Uriagereka 1988).
(34) a. *[de qué conferenciantes]1 te parece que [las propuestas t1] me van impresionar
of what speaker to-you seems that the proposals me will impress-INF
b. ? [de qué conferenciantes]1 te parece que me van a impresionar [las propuestas t1]
of what speaker to-you seems that me will impress-INF the proposals
‘Of which speakers does it seem to you that his proposals will impress me’
(35) [De qué autora]1 no sabes [ [qué traduciones t1]2 t2 han ganado premios internacionales]
of what author no know-2SG what translations have won awards international
‘Of which author do you not know which of his translations have won international
awards’
Inverse partial control structures present a similar constraint: the embedded float-
ing DP either follows the verb or is in [spec CP] (Torrego 1996).
(36) a. *No sabemos [si los lingüistas asistir]

not know-1PL whether the linguists help-INF
b. No sabemos [si firmar los lingüistas la carta

not know-1PL whether sign-INF the linguists the letter
‘We do not know whether the linguists among us should sign the letter.’
c. Algunos no sabemos [cuántos firmar la carta]

some not know-1PL how many sign-INF the letter
‘Some of us do not know how many of us should sign the letter.’
This paradigm is consistent with a movement analysis. (36a) is ungrammatical because

it involves left branch extraction out of a DP in [spec TP]. Since V moves to T in
Spanish, the order in (36a) suggests that the DP formed by los lingüistas and pro moves
to [spec TP], and then pro is extracted and moved to the matrix clause, thus violating
a restriction on left branch extraction. (36b) and (36c) are parallel to (34b) and (35).
(36b) is a case of extraction out of a DP in [spec VP], and (36c) involves extraction
out of a DP in [spec CP].17
A third and final piece of evidence in favor of a movement analysis of inverse
partial control is the fact that in these structures, the controller controls the gender
features of the floating DP:
(37) a. Muchos (de nosotros) no sabemos [[cuántos] firmar la carta

many-MASC (of us-MASC) not know-1PL how many-MASC sign-1PL the letter
‘Many of us do not how many should sign the letter.’
b. *Muchos (de nosotros) no sabemos [[cuántas] firmar la carta.

many-MASC (of us-MASC) not know-1PL many-FEM sign-INF the letter
‘Many of us do not how many of us should sign the letter.’
(Torrego 1996)
(38) a. Muchas (de nosotros) no sabemos [[cuántas] firmar la carta

many-FEM (of us-MASC) not know-1PL how many-FEM sign-1PL the letter
b. *Muchas (de nosotros) no sabemos [[cuántos] firmar la carta.

many-FEM (of us-MASC) not know-1PL many-MASC sign-INF the letter
(Torrego 1996)
To the extent that obligatory gender agreement with the controller is an argument
for a movement analysis of OC (cf. section 3), the data above strengthen the suggestion
that inverse partial control is also an instance of movement plus stranding.
5.2 The syntax of Inverse Partial Control
Inverse partial control differs semantically from partial control in that the con-
troller in inverse partial control is not understood as one of the participants of
the event denoted by the embedded predicate. Consider for instance, the sen-
tence in (39). As the English translation shows, the controller in this construc-
tion is understood only as restricting the referent denoted by the floating DP.
Given this, I propose the controller in (39) starts the derivation as the comple-
ment of the NP lingüistas, and then moves upward, as shown by the derivation
sketched in (40).18
(39) No sabemos si ir los lingüistas

not know-1STPL whether go-INF the linguists
‘We don’t know whether the linguists among us should go’
(40)
[TP pro1 [T’ no sabemos1 [CP si [TP t1 [T’ ir2 [VP [DP los [NP lingüistas t1]] t2]]]]]
A potential objection to this analysis is that the movement of pro in (39) is too long,
crossing over the DP and the embedded CP. Clearly, an answer to this depends
on a more fine-grained analysis of the structure of nominal expressions. Dikken
(1998), for instance, suggests that the surface order in English Possessive DPs like
John’s dog involves movement of the possessor to a functional projection (FP)
that is the complement of D. If there is such a functional projection, then in (40)
it might serve as an escape hatch for pro. Within the DP, pro moves to [spec FP];
from there it moves to the embedded [spec TP], and then to the matrix clause. As
for availability of movement over the infinitival embedded CP, I follow Chomsky
2001b, in assuming that infinitival CPs are φ-incomplete, being, therefore, unable
to define a spell-out domain. For this reason, the embedded CP in (40) is porous
for extraction of the controller.
6. CONCLUSIONS
In this paper I examined agreement on past participles and adjectives in non-

finite control complements in Romance and discussed its theoretical implica-
tions for analyses of control. I proposed that the embedded subject of a plural
predicate such as meet may be a complex DP consisting of a DP and a
null pronoun with properties similar to associative morphemes found overtly
in other languages. When the DP moves to matrix position, the result is partial
control: the matrix controller is understood as a subset of the agents of the

embedded predicate. When the null pronoun moves to matrix position, the result
is what I dubbed inverse partial control, in which a DP like los lingüistas is left
‘floating’ in the embedded clause. The agreement data support an analysis in
which matrix controllers in partial control originate in the embedded clause and
move upward.
1
The Portuguese data reported throughout the paper are from European and Brazilian Portuguese.
Thus, when not explicitly differentiated, the term Portuguese refers to European and Brazilian
Portuguese.
2
For space reasons, I will not discuss the fact that gender agreement is also observed in adjunct
controlled clauses (cf. Rodrigues 2004).
3
All the Italian speakers I tested did not accept partial control. Thus, partial control in general seems
to be blocked in Italian.
4
The datum reported in (9c) is from European Portuguese. In Brazilian Portuguese, gender agree-
ment is obligatory. Rodrigues 2004 shows that third person referential null subjects in Brazilian
Portuguese are obligatory-controlled elements.
5
Notice that this analysis does not explain how agreement in φ-features ensures identity in reference.
The fact that PRO and its controller indirectly agree in φ-features does not tell us why these lexical
items refer to the same entity.
6
Cf. Menuzzi 2000 for evidence that the expression a gente is pronominal in Portuguese.
7
Assuming that features of a pronoun are subject to an internal hierarchy (cf. Harley and Ritter 2002),
Costa and Pereira (2003) argue that semantic and grammatical features are separate features and are
checked in different syntactic domains, namely phases. In (14a), the semantic plural number of a
gente first agrees with the adjective cansados ‘tired’. As a result the secondary predicate records plural
agreement features. After that, the grammatical features of the pronouns agree with T, which is thus
valued as singular.
8
In some varieties of Brazilian Portuguese, the pronominal expression a gente does not trigger plural
agreement, but it triggers gender agreement. Thus, assuming that in these varieties only semantic
gender features are able to trigger syntactic agreement, it is still unclear why PRO in (13) does not
trigger gender agreement on the adjective.
9
For space reasons, I will omit the vP layer in syntactic representations, and represent unpronounced
copies as traces.
10
This analysis has roots in Kayne 2002 where it is proposed that obligatory control is an instance of
clitic doubling. However, I am proposing here that only partial control configurations involve building
a complex DP, followed by a stranding process. Exhaustive control involves movement, but not strand-
ing. This difference is responsible for the fact that in exhaustive control the set of referents denoted
by the controller is identical to the set of referent denoted by the controllee, whereas in partial control
the set of referents denoted by the controller is a strict subset of the set denoted by the controllee.
11
This is suggested in Hornstein 2003 (fn. 77), who argues that partial control is derived by movement
of the controller plus an LF process that adjoins a null associative plural to the embedded copy of
the controller. Barrie and Pittman (2004) also argue that partial control is formed by movement.
However, for them it involves movement of controller plus an LF chain-splitting mechanism which
inserts a PRO as the subject of the controlled clause.
12
In fact, speakers report that (24a) is even more natural when the embedded clause is gerundive.
This preference might be triggered by fact that gerundive clauses are truncated clauses which may
involve nominalization (cf. Chierchia 1984; Abney 1987). Thus, it is plausible that (i) contains only
one clausal domain. In that case the partial controlled element (i.e. the external argument of meet-
ing) and the modal are inside the same clausal domain. There might be a preference for having both
the partially controlled empty category and the modal within the same clause domain. At any rate,
whatever the explanation of this preference turns out to be, our point here is constant. Sentences
like (24a) and (i) show that partial control is dependent on the presence of a modal.
(i) I can try meeting tomorrow morning, but in the afternoon I really can’t.
13
I will not develop an analysis for this restriction here. See Rodrigues (forthcoming) for an analysis.
14
Neither Landau’s analysis nor the analysis suggested here explains why (18a) is restricted to British
English. Interestingly, however, as pointed by Landau, in British English semantically plural DPs
such as committee and team can license syntactically plural anaphors.
(i) The team/the committee invited each other for the party.
Landau takes this to suggest that in this language, partially controlled PRO also licenses syntacti-
cally plural anaphors because it has a semantically plural number feature. Note, however that
Landau does not present any independent evidence that semantically plural pronouns in British
English also license syntactically plural anaphors. This is a relevant observation since languages do
present a partition between full DPs and pronouns with respect to matching between syntactic and
semantic number features. In Portuguese and Italian, for example, pronouns with a semantic plural
feature do license syntactic plural number features (cf. (14) ), but semantically plural DPs do
not (cf. footnote 8). Thus, the movement plus stranding analysis has the advantage of explaining
the number mismatch observed in (18a). It is due to the presence of a plural DP in the subject of
the embedded predicate. Hence, under the present analysis, we do not need to presuppose (without
evidence) that semantically plural pronouns (at least PRO) can license syntactically plural
anaphors in British English.
15
Some speakers of Spanish allow an exhaustive control interpretation for (25). So, for them (25) can
also mean we don’t know whether we, the linguists, should sign the letter. This interpretation is also
compatible with the movement plus stranding analysis I am suggesting here. Los lingüistas starts
the derivation as a modifier of the pro, but gets stranded as pro moves leftwards, towards the matrix
clause.
16
See Ko 2005a, b for arguments that Korean caseless floating numeral quantifiers form a constituent
with their associate NPs in the underlying structure, while other types of floating quantifiers may
not do so.
17
Esther Torrego pointed out to me that inversion in inverse partial control might not be evidence for
movement since inversion is not obligatory in raising predicates (i). A complete answer to this issue
depends on a detailed analysis of raising complements. However, it is plausible that the infinitival
clause in (i) is a truncated clause headed by a VP. Spanish speakers judge sentences like (ii) as
degraded. Example (ii) differs from (i) in that it has a negation inserted with the embedded clause.
This might reflect the fact that the embedded clause (i) is pruned, having no structure above VP. In
addition, note that Spanish speakers consider clitic climbing with raising predicates to be possible,
though in sentences like (iii), they have strong preference for reading the clitic as an experiencer,
instead of being the theme of odiar ‘to hate’. This supports an analysis of (i) as a restructured
domain. See also Wurmbrand (to appear) for evidence that raising verbs are restructuring verbs.
(i) Parecemos los linguistas haber firmado la carta.

seem-1PL the linguists have-INF signed the letter
‘It seems that the linguists among us have signed the letter.’
(ii) ?? Parecemos los linguistas no haber firmado la carta.

seem-1PL the linguists no have-INF signed the letter
‘It seems that the linguists among us haven’t signed the letter.’
(iii) Juan le parece odiar

Juan her seem-3SG hate-INF
‘Juan seems to hate her.’
18
I am putting aside questions about the position of the negation.
MARCELLO MODESTO
NULL SUBJECTS IN BRAZILIAN PORTUGUESE

AND FINNISH: THEY ARE NOT DERIVED
BY MOVEMENT*
1. INTRODUCTION
It is well known that some languages, despite permitting null subjects, do not
display the full array of characteristics usually associated with the Null Sub-
ject Parameter as proposed by Rizzi (1982, 1986). In the 1980s, there was much
debate about languages such as Chinese, which display no overt verbal agree-
ment and still allow for the subject position to be null. More recently, languages
like Brazilian Portuguese (BP) and (colloquial) Finnish have been discussed. In
those languages, third person null referential subjects are not allowed in matrix
contexts, which has been related to the fact that (at least in BP) verbal agreement
is ‘poor’.1
However, third person null referential subjects are productive in embedded
contexts in both languages, which is problematic for theories which try to explain
the Null Subject Parameter.
In a series of works (Modesto 2000a, b, 2004, 2007), I have demonstrated
that embedded null subjects in BP show all the properties that are characteris-
tic of obligatorily controlled subjects of nonfinite clauses. Rodrigues (2004)
has demonstrated that the same also applies to Finnish. The properties are
as follows: they must have an antecedent (1); the antecedent has to c-command the
subject empty category (2); the antecedent has to be local (3); the antecedent may
not be split (4); in VP ellipsis contexts, only a sloppy reading is possible (5); and
only a covariant interpretation is possible with ‘only NP’ antecedents (6):
(1) a. *e telefonou. BP
called
b. *Parece que e telefonou. BP

seems that called
c. *e oli soittanut. Finnish

called
d. *Vaikuttaa siltä että e oli soittanut. Finnish

seemsit it that called
231
W.D. Davies and S. Dubinsky(eds.), New Horizons in the Analysis of Control and Raising, 231–248.
© 2007 Springer.
232 MARCELLO MODESTO
(2) a. [ O amigo do Feco2 ]1 disse que e1/ 2/ 3 ganhou a competição. BP

* *
the friend of Feco said that won the championship
‘Feco’s friend said that he had won the championship.’
b. [ Veljeni2 vaimo]1 oli niin iloinen, ettei e1/ 2/ 3 voinut nukkua. Finnish
* *
Brother-GEN wife was so happy that-not could sleep.
‘My brother’s wife was so happy that she could not sleep.’
(3) a. O Feco1 disse que a Dani2 acha que e 1/2 ganhou na loto. BP
*
the Feco said that the Dani thinks that won in.the lottery
‘Feco said that Dani thinks that she won the lottery.’
b. Jukka1 sanoi että Liisa2 ajattelee että e 1/2 oli voittanut arpajaisissa. Finnish
*
Jukka said that Liisa thinks that had won lottery
‘Jukka said that Liisa thinks that she won the lottery.’
(4) a. *O Feco1 disse que a Dani2 acredita que e1+2 vão morar juntos.
the Feco said that the Dani thinks that will live together
‘Feco said that Dani thinks that they will live together.’
b. *Jukka1 kysyi vaimoltaan2 e1+2 voivatko mennä Espanjaan Finnish

Jukka asked his.wife can go to.Spain
lomalle.
for.vacation
‘Jukka asked his wife if they can go to Spain for vacation.’
(5) a. O Pedro1 acha que e1 é inteligente e o Paulo também. BP

the Pedro thinks that is intelligent and the Paulo too
‘Pedro thinks that he is intelligent and Paulo does too.’
b. Jukka1 sanoi että e1 oli voittanut arpajaisissa, ja niin Pekkakin. Finnish

Jukka said that had won the.lottery and so Pekka.also
‘Jukka said that he had won the lottery and Pekka did too.’
(6) a. Só o Maluf acha que e vai ganhar as eleições. BP

only the Maluf thinks that will win the elections
‘Only Maluf thinks that he will win the elections.’
b. Vain Jukka ajatteli että e oli voittanut arpajaisissa. Finnish

only Jukka thought that had won the.lottery
‘Only Jukka thought that he had won the lottery.’
The fact that null subjects in BP and Finnish present obligatory control-like char-
acteristics led Rodrigues (2004) to propose a movement analysis of null subjects
in those languages following a trend initiated by Hornstein’s (1999) analysis of con-
trol. In this paper, I will show that a movement analysis of null subjects of finite
clauses in BP and Finnish is not a good analysis for two reasons: (i) it leaves some of
the data unaccounted for, and (ii) it is not the simplest analysis. After proving that
this is truly the case, although the data to be presented here is not actually related to
the control vs. raising issue, I will speculate if the movement analysis of control may
also be mistaken. In turn, I will propose, following Modesto (2007), a modification
NULL SUBJECTS IN BRAZILIAN PORTUGUESE AND FINNISH 233
of Modesto’s (2000a, b) analyses in which embedded null subjects are taken to be

φPs, in the sense of Holmberg (2005), which are A’-bound by a topic.
2. THE MOVEMENT ANALYSIS OF NULL SUBJECTS IN BP

AND FINNISH: RODRIGUES 2004
Before I start, it should be noted that I will review Rodrigues’ analysis as

an example, noting that any other implementation of a movement analy-
sis of the facts presented here would deal with the same type of problems.
So I am not arguing against any specific implementation, but against the
idea that null subjects of finite clauses may be derived by movement out of a
Case domain.2
The first obvious problem that any movement analysis of null subjects of
finite clauses must have to deal with is the fact that A-movement takes place out
of a phase: a Case-marked domain. To deal with that problem, Rodrigues makes
the assumptions in (7)3:
(7) a. in null subject languages (with strong Agr systems), verbal affixes are listed in the
lexicon as separate lexical items and carry a D-feature, φ-features and possibly a Case
feature, but in non-null subject languages (with weak Agr systems), verbal affixes
are not independent units, entering the derivation already attached to their hosts.
b. in BP and Finnish, verbal agreement became φ-defective (i.e. weak), so a φ-complete

item must be inserted in the complement domain of T, such that the θ-roles of the
verb and the φ-features of T can be checked (by the operation Agree).
c. in BP and Finnish, verbal affixes still carry a D-feature, so verb movement to T can
check T’s EPP feature and Spec TP need not be projected.
d. a structural Case feature is only checked in a spec-head relation, not by Agree (contra
Chomsky 2000 and thereafter).
e. movement may be greedy, i.e. it may happen to satisfy the requirements of the moved
element alone.
f. DPs are moved to [Spec FP] before moving out of a phase.
g. The phase-impenetrability condition (PIC) is defined in a way such that inside the
domain of a strong phase (HP), only sub-domains that are themselves phases are not
accessible to operations outside HP. That means that only TP is spelled out when C is
reached but intermediate projections which would constitute an expanded CP domain
are still accessible to the derivation.
To give a concrete example, take the sentence in (8), with numeration (9a). The
first relevant step in the derivation is after (9b) is formed. T has its EPP feature
checked by the verbal affix and its φ-features by the DP o João after the operation
Agree takes place.4
(8) O João1 disse que e1 comprou um carro.

the João said that bought a car
‘João said that he bought a car.’
(9) a. Num = {T2, F1, o1, João1, disse1, que1, comprou1, um1, carro1 }
b. [TP [T comprou1φdef,D + T−φEPP] [VP [DP o João]Case,φ [V’ t1 [DP um carro]]]]
c. [FP [DP o João]2 Case,φ [F’ [TP [comprou1φdef,D + T−φEEP ] [VP t2 [V’ t1 [DP um carro]]]]]]]
At this point of the derivation, Rodrigues claims that the system can either move o
João to [Spec TP] or continue the derivation by merging the next item in the numera-
tion. Movement is allowed since T and o João have agreed in φ-features. However,
since the probe has already checked its EPP feature, I would assume that Merge over
Move requires that the derivation continue by merging the next item in the numera-
tion. So F is inserted and o João moves to its specifier position, as in (9c).5 Rodrigues
does not discuss what licenses such movement, but if it does not take place, the DP
will not be able to get out of its phase. The head C is then inserted and the deriva-
tion proceeds with movement of o João to matrix vP, as in (10).6 The derivation then
proceeds as shown in (11).
(10) [VP [DP o João] 2 Case, φ [V’ disseφ,def, D [CP que [FP t2 [F’ [TP [comprou 1φ,def, D + T−φEEP] [VP t2 [V’ t1
[DP um carro]]]]]]]]]
(11) TP
[DP o João] 2 Case,φ T’
[disse3 φ-def,D +Tφ, φ,EPP] [VP t2 [V’ t3 [CP que [FP t2 [F’ [TP [comprou1φ-def,D+T-φEEP]
[VP t2 [V’ t1 [DP um carro]]]]]]]]]
Consider now sentence (12) with the numeration in (13a). The first relevant step of
the derivation is the one following the formation of (13b). T has already checked
its EPP and φ-features. In this case, the pronoun does move to Spec TP to check its
own Case feature, as shown in (13c), although T has no feature to check. In other
words, according to Rodrigues (2004), besides the probe/goal/Agree system, move-
ment may happen for completely selfish reasons of the moved element. Movement
of the DP to Spec TP (over Merge of the next item in the numeration), in this case,
is licensed by the fact that if it did not happen at this point, the derivation would
not converge: the Case feature of ele would remain unchecked since at the matrix
level the DP o João is inserted as the subject. It is important to keep in mind that
movement of the embedded subject to Spec TP is only allowed as a last resort.
The derivation then continues as shown in (14).
(12) O João1 disse que ele1 comprou um carro.

the João said that he bought a car
‘João said that he bought a car.’
(13) a. Num = {T2, o1, João1, disse1, que1, ele1, comprou1, um1, carro1 }
b. [TP [comprou11φdef, D + TφEPP] [VP ele Case,φ [V’ t1 [DP um carro]]]]
c. [TP ele2 Case,φ [T’ [comprou1φdef, D + Tφ.EPP] [VP t2 [V’ t1 [DP um carro]]]]]]
(14) TP
[DP o João]4 Case,φ T’
[ disse3 φ-def,D +Tφ,EPP] [VP t4 [V’t3 [CP que [TP ele2Case,φ [T’ [comprou1φ-def,D +Tφ,EPP ]
[VP t2 [V’ t1 [DP um carro]]]]]]]
This very brief presentation of this analysis is intended to show that Rodrigues
uses the assumptions in (7) to account for the fact that the subject DP sometimes
checks case in the embedded clause (when it is overt) and sometimes not (when it
is null). What is left to be done, although I will not do it here, is to check if there
is any independent support for the assumptions in (7).
We now turn to some of the problems confronting this analysis. As noted by
Modesto (2000a, b), contrary to what movement analyses predict, the choice
of the antecedent of the null subject in BP is not consistent with the Minimal
Distance Principle. For verbs that take a direct object plus a sentential argument
with a null subject, it is always the matrix subject, never the object, that is the
antecedent of the null embedded subject, as seen in (15). The same is true in
Finnish, as seen in (16).
(15) a. A Dani1 convenceu o Feco2 que e1/ 2 pode se eleger. BP

*
the Dani convinced the Feco that can self to.elect
‘Dani convinced Feco that she can get elected.’
b. A Dani1 avisou o Feco2 que e1/ 2 precisa trabalhar até mais tarde.
*
the Dani warned the Feco that has to.work until more late
‘Dani warned Feco that she has to work until late.’
(16) a. Liisa1 vakuutti Jussille2 että e1/ 2 voi tulla valituksi. Finnish
*
Liisa assured to.Jussi that could become elected
‘Liisa assured Jussi that she can get elected.’
b. Liisa1 takasi Jussille2 että e1/ 2 saa ylennyksen.

*
Liisa guaranteed to.Jussi that will.get promotion
‘Liisa guaranteed Jussi that she would get the promotion.’
Other BP verbs exhibiting the exact same behavior include: informar ‘to inform’,
alertar ‘to alert’, prevenir ‘to forewarn’ and instruir ‘to instruct’, among others
which take an indirect object plus a sentential complement. While the examples
that follow are all based on the verb convencer ‘to convince’ (15a), all the facts are
reproducible with each of the other verbs in this class.
To solve the problem at hand, Rodrigues assumes that the sentential argument
of verbs of the convencer class is not a complement, but an adjunct to vP; and that
movement of the subject out of the adjunct is an instance of sideward movement.7

In that way, as seen in (17), movement of the embedded subject to the matrix
object position is blocked by Merge over Move, and movement of the embedded
subject to matrix subject position respects the MLC, since the matrix object does
not c-command the embedded subject. In this case, we can assume that movement
to Spec FP does not take place, since there is no need to escape the phase through
its edge in sideward movement.
TP
(17)
[a Dani] T’
[convenceu] vP
vP CP
t v’
[CP que [TP [T pode] [vP t [v’t se eleger]]]]]
t VP
t [o Feco]
To back up her assumption that the sentential argument is an adjunct, Rodrigues

evokes Larson’s (1991) VP shell analysis of the ‘promise’ class of control verbs where
the subject position is dethematized and the Case of the object is absorbed, much as
in passives (although she adjoins the sentential complement much higher than Larson
does). The problem with this is that Larson proposed such a structure for the ‘promise’
class because he wanted to account for several facts which indicate that promise-NP-
infinitive constructions are analogous to double object structures, in clear opposition
to verbs like ‘persuade’ and ‘force’. Verbs like convencer in BP, however, clearly pattern
with ‘persuade’ and ‘force’, not with ‘promise’, as seen in (18), (19), and (20).
(18) a. ??Who do you think John promised to leave?

b. ??Who do you think John promised a sports car?
c. Who do you think John persuaded to leave?
d. Quem você acha que a Dani convenceu a ir embora?
who you think that the Dani convinced to to.go away
‘Who do you think that Dani convinced to leave?’
(19) a. What did John promise Mary? (answer: to leave at 5 o’clock)

b. *What did John persuade Mary? (answer: to leave at 5 o’clock)
c. *O que a Dani convenceu o Feco?
what the Dani convinced the Feco?
‘What did Dani convince Feco?’
(20) a. John promised Ø to leave.

b. ??John promised Mary Ø.
c. *John persuaded Ø to leave.
d. John persuaded Mary Ø.
e. *A Dani convenceu Ø a sair.

the Dani convinced Ø to to.leave
‘*Dani convinced to leave.’
f. A Dani convenceu o Feco Ø.

the Dani convinced the Feco Ø
‘Dani convinced Feco.’
Based on such data, it can easily be concluded that convencer is not a double object
verb and, therefore, the structure proposed by Larson should not be applied to it.
It seems, then, that there are few facts supporting such a structure for the convencer
class of verbs other than as a means of accounting for the data in (15).
Another reason leading Larson to propose that structure for ‘promise’ was to
account for the fact that ‘promise’ is a subject control verb. As shown in (21), con-
vencer is an object control verb. If the sentential complement of verbs like convencer
are adjuncts and that explains why the subject of embedded finite clauses cannot move
to the matrix object position, then one would have to say that finite sentential comple-
ments of convencer are adjuncts but nonfinite sentential complements are not.
(21) A Dani1 convenceu o Feco2 a PRO*1/2 sair.

the Dani convinced the Feco to to.leave
‘Dani convinced Feco to leave.’
The only two arguments Rodrigues gives to assume that those complements are
adjuncts are the fact that they resist extraction of nonargument wh-phrases and the
subject of the clause embedded under convencer can be an epithet referring back to
the matrix object. As seen in (22a), it is indeed a fact that the sentence cannot be inter-
preted as asking when or why Dani traveled. However, it seems to be a characteristic
of nonargument wh-phrases in BP to attach as close as possible and not a peculiarity
of the convencer class. It is also impossible for (22b) to be interpreted as asking when
or why Dani traveled. In spite of that, one would hardly suppose that the sentential
argument of dizer ‘to say’, is an adjunct.
(22) a. *?Quando/por que a Dani convenceu o Feco que e viajou twh ?

when /why the Dani convinced the Feco that traveled
(When/why did Dani convince Feco that she traveled?)
b. *?Quando/por que a Dani disse que e viajou twh ?

when /why the Dani said that traveled
(When/why did Dani say that she traveled?)
So, the sole remaining argument supporting Rodrigues’ structure for the convencer
class verbs is that they allow an epithet in embedded subject position referring to
the matrix object. In fact, sentence (23a), from Rodrigues (2004), is possible. How-
ever, the impossibility of taking the matrix subject as the antecedent of the epithet
may be just a consequence of pragmatics. In (23b), where the pragmatics favor the
interpretation of the matrix subject as antecedent, that interpretation is indeed
possible. In any case, if (23a) was possible due to lack of c-command between the
matrix object position and the embedded subject position, sentence (24a), where
a pronoun in object position is coreferent with the embedded subject, should be
possible, in view of the fact that (24b) is possible. But (24a) is clearly ungrammati-
cal, which shows that there is c-command between those two positions.
(23) a. O Ira1 convenceu o Diogo2 que o bobão*1/2/3 não deveria comprar o carro.
the Ira convinced the Diogo that the silly not should to.buy the car
‘Ira convinced Diogo that the fool should not buy the car.’
b. O Maluf1 convenceu o Diogo2 que o desgraçado1/??2/3 era o melhor candidato.

the Maluf convinced the Diogo that the bastard was the best candidate
‘Maluf convinced Diogo that the bastard was the best candidate.’
(24) a. *A Dani1 convenceu ele2 que o Feco2 está errado.

the Dani convinced him that the Feco is wrong
(*Dani convinced him that Feco is wrong.)
b. O Ira1 acredita que ele1/2 é um gênio mais fervorosamente que a mãe

the Ira believes that he is a genius more fervently that the mother
do Feco2 (acredita).
of.the Feco (believes)
‘Ira believes that he is a genius more fervently than Feco’s mother does.’
We must then conclude that there are no arguments in favor of saying that the senten-
tial complement of the convencer class verbs is an adjunct and that the choice of the
null subject’s antecedent is not consistent with the MDP. An argument that the matrix
object does in fact c-command the sentential argument is the fact that a quantifier in
object position may bind into the embedded subject position, as shown in (25).
(25) A Dani convenceu cada homem1 que sua1 mulher era fiel.
the Dani convinced each man that his wife was faithful
‘Dani convinced each man that his wife was faithful.’
3. MOVEMENT AND THE NULL SUBJECT ANTECEDENT
Another fact first discussed by Modesto (2000a, b), which is problematic for any
movement analysis, is that movement of the matrix object alters the interpretative
possibilities of the null embedded subject. In (26a, b) and (27a, b) we see that an
object that has been wh-moved or relativized becomes a possible antecedent for
the null subject. Importantly, the matrix subject ceases to be a possible anteced-
ent.8 In sentences where the object is topicalized (26c) and (27c), however, both the
object and matrix subject are possible antecedents.
(26) a. Quem2 que a Dani1 convenceu t2 que e*1/2 pode se eleger? BP

Who that the Dani convinced that can self to.elect
b. O cara2 que a Dani1 convenceu t2 que e*1/2 pode se eleger já

the guy that the Dani convinced that can self elect already
chegou.
arrived
‘The guy who Dani convinced that he can get elected has already arrived.’
c. O Feco2, a Dani1 convenceu t2 que e1/2 pode se eleger.

the Feco the Dani convinced that can self to.elect
‘(Speaking of) Feco, Dani convinced (her) that s/he can get elected.’
(27) a. Quem2 que a Dani1 avisou t2 que e*1/2 precisa trabalhar até mais tarde? BP
who did the Dani warned that has to.work until more late
‘Who did Dani warn that s/he has to work till late?’
b. O cara2 que a Dani1 avisou t2 que e*1/2 precisa trabalhar até mais tarde
the guy that the Dani warned that has to.work until more late
já chegou.
already arrived
‘The guy who Dani warned that he has to work till late has already arrived.’
c. O Feco2, a Dani1 avisou t2 que e1/2 precisa trabalhar até mais tarde.
the Feco the Dani warned that has to.work until more late
‘(Speaking of) Feco, Dani warned (him) that s/he has to work till late.’
Rodrigues claims that the derivation of (26a) is analogous to the derivation of

parasitic gaps in the analysis of Hornstein 2001. In the derivation in (28), the
wh-phrase checks Case downstairs and moves to Spec CP. A variable is then created
in Spec TP, for that copy has checked its Case and is bound by an operator. From
Spec CP, the wh-phrase moves sidewards to the object position of the matrix verb.
Merge over Move is violated here for convergence: if the wh-phrase moves through
the matrix subject position, the variable downstairs gets bound, violating principle C.
Remember now that in sideward movement, a DP does not need to use the edge of
the phase as an escape hatch, as seen in (17). Thus, it seems that the sole motivation
for the movement of the wh-phrase to Spec CP in (28) is to explain the violation of
Merge over Move. Rodrigues (2004) gives the following as the structure of (26a):
CP
(28)
[quem]2 C’
C TP
[a Dani]1 T’
[convenceu] vP
vP CP
t1 v’
[t2 que [TP t2 [T’ [T pode] [vP t2 [v’ t se eleger]]]]]
t VP
t t2
Note also that the wh-phrase checks Case twice. This requires the additional
assumption that a Case feature becomes reactivated when a copy leaves a phase.
This assumption seems to be not only unsupported by data but also at odds with
current Minimalist spirit. Additionally, it necessitates the assumption that phrases
come into the derivation with a nonspecified Case feature, since the wh checks
Nominative in the adjunct and Accusative in the matrix clause.
The observant reader may have noticed that movement to check Case in Spec TP
of the embedded clause was only licensed for convergence in (13c). In (28), however,
the wh-phrase moves to Spec TP even though there is a convergent derivation begin-
ning with the same numeration in which Merge over Move is respected. Instead
of moving the wh-phrase to Spec TP, the DP a Dani could be inserted as the object
of the matrix verb and the wh-phrase could move to matrix subject position (just as
in (17) above), deriving sentence (29), as shown in (30). Note that in (17), the moved
DP does not raise to Spec FP or any other position before moving to matrix subject
position. Therefore, the derivation of (29) should block the derivation of (26a):
(29) Quem1 t1 convenceu o Feco2 que e1/*2 pode se eleger ?

who convinced the Feco that can self to.elect
‘Who convinced Feco that s/he can get elected?’
(30) CP
[quem] C’
C TP
t T’
[convenceu] vP
vP CP
t v’
[que [TP [T pode] [vP t [v’ t se eleger]]]]]
t VP
t o Feco
Lastly, note that there is nothing in Rodrigues’ system that would prevent generating
sentence (31), which is the same as (26a) but with the matrix subject interpreted as the
antecedent of the null embedded subject. The derivation would proceed as shown in
(32). That interpretation, however, does not exist. In other words, the movement analysis
cannot explain why movement of the matrix object makes it the sole antecedent of the
null embedded subject.
(31) *Quem2 que a Dani1 convenceu t2 que e1 pode se eleger?

who that the Dani convinced that can self to.elect
‘Who did Dani convince that [s/he] can get elected?’
(32) CP
[quem]2 C’
C TP
[a Dani]1 T’
[convenceu] vP
vP CP
t1 v’
[ que [TP [T pode] [vP t1 [v’ t se eleger]]]]]
t VP
t t2
Summarizing what we have seen so far, the data in (26) shows, in contrast to the
data in (15), that there is a correlation between moving an object and taking
that object as the antecedent of the null subject. This remains unexplained in
Rodrigues’ analysis. It is unclear to me how any movement analysis of BP null
subjects would capture that correlation in a simple manner. The greatest appeal of
Hornstein’s analysis of control is its simplicity. That simplicity, however, does not
appear to carry over to movement analyses of null subjects.
4. THE RELATION BETWEEN MOVEMENT AND BEING

THE ANTECEDENT OF A NULL SUBJECT
Consider once again the sentences in (26) and (27). The ambiguity of the ‘c’ sen-
tences seems to correlate with the possibility of base-generation. In (33), we see that
a topic may be generated in its dislocated position, while a wh-phrase or a null oper-
ator must be moved. We can then hypothesize that the ‘c’ sentences are ambiguous
exactly because the topic may be taken to be moved or base-generated. If moved,
it seems natural to suppose that the sentence has only the interpretation where the
topic is the antecedent of the null subject, on a par with (26a, b) and (27a, b). How-
ever, when base-generated, the matrix subject is the antecedent.
(33) a. *Quem1 que a Dani conhece o jornalista que entrevistou e1?

Who that the Dani knows the journalist that interviewed
‘*Who does Dani know the journalist who interviewed?’
b. *O cara1 que a Dani conhece o jornalista que entrevistou e1 já chegou

the guy that the Dani knows the journalist that interviewed already arrived
‘*The guy that Dani knows the journalist who interviewed has arrived.’
c. O Feco1, a Dani conhece o jornalista que entrevistou e1.

the Feco the Dani knows the journalist that interviewed
‘Feco, Dani knows the journalist who interviewed her.’
That relation between movement and antecedence is confirmed by (34) and (35),
where the topic and the wh-phrase have not been moved and therefore cannot
antecede the null subject:
(34) O Feco2, a Dani1 convenceu ele2 que e1/*2 pode se eleger.

the Feco the Dani convinced her that can self to.elect
‘(Speaking of) Feco, Dani convinced him that she can get elected.’
(35) A Dani1 convenceu quem2 que e1/*2 pode se eleger?

The Dani convinced who that can self to.elect
The same correlation holds in Finnish, as shown in (36). Remember from (16)
that an object is not a possible antecedent for the null subject in Finnish. When
the object is wh-moved or relativized, however, it becomes a possible antecedent.
Unlike in BP, movement of the matrix object gives rise to ambiguity. However,
Finnish speakers show a clear preference for the matrix subject as antecedent.
(36) a. Kenelle2 Liisa1 vakuutti t2 että e1/?2 voi tulla valituksi?

to.whom Liisa assured that could become elected
‘Who did Liisa assure that s/he can get elected?’
b. Henkilö2 jolle Liisa1 vakuutti t2 että e1/?2 voi tulla valituksi saapui jo.
the person to.whom Liisa assured that could become elected arrived already
‘The person who Liisa assured that s/he can get elected has arrived.’
5. AN ALTERNATIVE ANALYSIS: THE TOPIC-PROMINENCE

PARAMETER
Having demonstrated that movement analyses of null subjects cannot explain the
data, I will briefly discuss my own analysis, referring the reader to Modesto (2007)
for a full account. First, I assume that BP and Finnish are topic-prominent lan-
guages, something already argued by Holmberg and Nikanne (2002) for Finnish
and Negrão and Viotti (2000) for BP, among others. I also assume that in topic-
prominent languages a functional head F is always generated in every clause and
it always carries an OCC (=EPP) feature (see Chomsky 2004), meaning that
something will have to be moved to or merged in its specifier position. In other
words, topic-prominent languages are characterized by having a second level of
predication (besides the subject–predicate level) where the element in Spec FP is
predicated to the rest of the sentence. This second level of predication is usually
referred to as topic-comment. All languages may in fact present topic-comment
structures; what defines a language as topic-prominent is the obligatoriness of
such structures, caused by the OCC feature of F0. The choice of what moves to
Spec FP is free since any phrase can satisfy the OCC feature of F0, although, as
any movement, movement to Spec FP is regulated by locality constraints (the
MLC of Chomsky 1995, 2000, which is subsumed by the probe/goal architecture
of the Agree operation of Chomsky 2001a). Therefore, for locality reasons, the
subject (being the closest goal able to satisfy the probe) is usually moved to Spec
FP in every clause and it will be interpreted as a ‘grammatical’ topic. The claim
is, then, that null embedded subjects are possible in topic-prominent languages
due to the fact that subjects occupy a higher (A’) position and so matrix subjects
are able to identify null embedded subjects by binding them. The derivation of a
sentence like (15a) would then be (37):
(37) [FP a Dani1 [TP t1 convenceu o Feco2 [CP que [FP e1 [TP t1 pode se eleger ]]]]]
Dani convinced Feco that she can get elected
Following Holmberg 2005, I assume that null subjects are non-referential sets
of φ-features, φP(hrases) in his terminology. In ‘rich’ agreement languages,
referentiality is given by verbal agreement, making φPs behave like overt pronouns.
In the languages discussed here, since agreement cannot provide a reference to
(or identify) the φP, the only way to interpret it will be taking it to be a variable
at LF. In other words, φPs will only be possible in weak agreement languages
when A’-bound. Specifically, in (37), the φP gets bound by the higher subject
that has been moved to Spec FP, to check its OCC feature. The chain formed by
the two topics is nothing more than the application of the regular chain forma-
tion operation that applies between copies (cf. Nunes 1995). In this case, it may
apply to distinct elements since they have the same set of φ-features.9 An overt
pronoun could be merged in the embedded subject position as well, in BP or
Finnish. In that case, coreference with the higher subject would be accidental
and non-coreference would be possible. The φP, however, does not refer by itself
and can only be interpreted when bound. It is, therefore, the fact that those
languages are topic-prominent (and so subjects are moved to an A-bar position,
i.e. Spec FP) that allows φPs in the embedded subject position. That explains
why English-type languages do not show null embedded subjects of the kind
described here: subjects in English-type languages remain in Spec TP and so
cannot variable-bind an embedded subject.
Note that the characteristics of null subjects in Finnish and BP listed in the
introduction of this chapter, exemplified in (1)–(6), are readily explained by the
proposed analysis. The antecedent requirement is due to the non-referentiality
of φPs; c-command and locality are requirements of the chain formation opera-
tion; split antecedents are banned since tripartite chains cannot be formed; sloppy
readings and covariant interpretations are due to the fact that the φPs are inter-
preted as variables (they get their reference by being bound).
Subject orientation, shown in (15) and (16) is also explained since objects do
not (usually) occupy the Spec FP position.
The notion of ‘grammatical’ topic, mentioned above, needs some clarification.
Movement of subjects to Spec FP is driven solely by the necessity of the probe
(F0) to have some phrase in its specifier position (which is what the OCC feature
means) and not by any feature or semantic property of the subject. Therefore, the
moved subject in Spec FP is interpreted as a grammatical topic (as opposed to a
semantic one) and need not have any semantic property of a topic. This position
(Spec FP) is, then, the position of unmarked topics advocated by Martins and
Nunes (2005), which allows weak pronouns and non-referential phrases. This is
different from the marked topic position in the left-periphery of the sentence (Spec
TopP), reserved for constituents marked with a topic feature, therefore exclud-
ing weak pronouns and non-referential phrases. We, then, explain the remark in
Holmberg (in press) that ‘the subject may check the EPP even if it is not a referen-
tial category, for example a quantified NP, but nonsubjects have to be referential
and interpretable as topics to check the EPP’ in Finnish.
The fact that a matrix object becomes the only possible antecedent for the null
subject when the object is moved (in BP) can be explained by Minimality (cf. Rizzi
1990). If the matrix subject is moved to Spec FP and the object is moved over the
grammatical topic, Minimality is violated. In (38a), (39a), and (40a) we see that, in
fact, in BP, a wh-phrase, a null operator and a topic may not be moved over another
topic. This indicates that the derivations of the sentences in (26) must be (38c) and
(39c) and not (38b) and (39b). In (40), two structures are possible inasmuch as the
(marked) topic may be moved to Spec TopP or base-generated in that position.10
(38) a. *Quem1, esses livros2, t1 leu t2 ?

who these books read
b. *[CP quem2 que [FP a Dani1 [TP t1 convenceu t2 [CP que [FP e1 [TP t1 pode se
who that the Dani convinced that can self
eleger]]]]]]
to.elect
c. [CP quem2 que [FP t2 [TP a Dani1 convenceu t2 [CP que [FP e2 [TP t2 pode se
who that the Dani convinced that can self
eleger ]]]]]]
to.elect
(39) a. *O cara1 que, esses livros2, t1 leu t2 já chegou.

the guy that, these books, read has arrived
b. *[o cara2 [CP Op2 que [FP a Dani1 [TP t1 convenceu t2 [CP que [FP e1 [TP t1 pode se
the guy that the Dani convinced that can self
eleger]]]]]]]…
to.elect
c. [o cara2 [CP Op2 que [FP t2 [TP a Dani1 convenceu t2 [CP que [FP e2 [TP t2 pode se
the guy that the Dani convinced that can self
eleger]]]]]]]
to.elect
(40) a. *A Dani1, esses livros2, t1 já leu t2

Dani, these books, already read
Cf. A Dani1, esses livros2, ele1 já leu t2.
b. *[TopP o Feco2 [FP a Dani1 [TP t1 convenceu t2 [CP que [FP e2 [TP t2 pode se
the Feco the Dani convinced that can self
eleger]]]]]]
to.elect
c. [TopP o Feco2 [FP t2 [TP a Dani1 convenceu t2 [CP que [FP e2 [TP t2 pode se
eleger]]]]]]
to.elect
d. [TopP o Feco2 [FP a Dani1 [TP t1 convenceu e2 [CP que [FP e1 [TP t1 pode se
eleger]]]]]]
to.elect
As seen in (41a) and (42a), Finnish, on the other hand, does not show Minimal-
ity effects: both a wh-phrase and a null operator may be moved over a topic
without challenging the grammaticality of the sentence. Therefore, due to the
absence of Minimality effects, locality is always respected in Finnish and the sub-
ject always moves to Spec FP.
(41) a. Kuka tämän kirjan on kirjoittanut?

who this book has written
b. [CP Kenelle2 [FP Liisa1 vakuutti [TP t1 tv t2 [CP että [FP e1 [TP t1 voit tulla
to.whom Liisa assured that could become
valituksi]]]]]]
elected
(42) a. henkilö joka tämän kirjan on kirjoittanut saapui jo.

the person who this book has written arrived already
b. [ henkilö2 [CP jolle2 [FP Liisa1 vakuutti [TP t1 tv t2 [CP että [FP e1 [TP t1 voi tulla
person to.whomLiisa assured that could become
valituksi ]]]]]]
elected
Confirmation of the analysis presented here comes from intervention effects in

both languages. Consider (43). Although Finnish does not present Minimality
effects with wh-phrases and null operators, one topic does intervene over another
topic. Therefore, when the phrase ‘assignment’ occupies the Spec TopP position
in (43b), a topic chain cannot be formed between the embedded subject and the
matrix topic. This is either (i) because the marked embedded topic was moved
through Spec FP or (ii) because, being base-generated, it prevents application of
the chain formation operation since another A’-element intervenes between the
two unmarked topics, and the sentence gets a generic interpretation only11:
(43) a. Oppilas1 tietää ettei e1 pysty ratkaisemaan tehtävää.

student knows that.not can solve assignment
‘The student knows that he cannot solve the assignment.’
b. Oppilas1 tietää ettei tehtävää e*1 pysty ratkaisemaan.

student knows that.not assignment can solve
‘The student knows that the assignment cannot be solved.’
The sentences in (44a, b) show that exactly the same intervention effect happens
in BP. A marked topic in the embedded clause, moved through Spec FP to Spec
TopP or base-generated in the latter position, prevents a chain from being formed
between the two subjects, which makes (44b) necessarily interpreted as generic.
Sentence (44c) shows that not every phrase intervenes between the two subjects
though, only referential expressions do, which is straightforwardly explained in
this analysis: not being referential, the adverbial phrase todo dia ‘every day’ does
not qualify as a possible occupant of Spec TopP, so it must be a clausal adjunct.
Being an adjunct, it does not intervene between the two unmarked topics and the
coreferential reading is possible.
Sentence (44d), however, is the most interesting case. If the locative na praia is
taken to qualify the matrix event, it can be merged in the higher Spec TopP and the
sentence can be interpreted as saying that ‘Feco told me that he sells hot dogs (for
a living, when we were at the beach)’, since the matrix subject is moved to Spec FP.
On the other hand, if the locative qualifies the embedded event, it presumably has
to be moved from the embedded clause. Moving the locative over the matrix subject
in Spec FP would cause a Minimality effect. So it has to be moved to matrix Spec
FP before moving to Spec TopP. In that case, the matrix subject has to remain in
Spec TP and no topic chain can be formed, since the two phrases in Spec FP do
not share the same features and reference, so the embedded clause has the generic
reading only. It is important to note that in this case an adverb in sentence-initial
position prevents the null embedded subject from being interpreted as coreferential
to the matrix subject. That this very surprising state of affairs is accounted for and
explained by the analysis presented here provides strong support for it.
(44) a. O Feco1 me falou que e1 vende cachorro quente na praia.

the Feco to.me said that sells dog hot on.the beach
‘Feco told me that he sells hot dogs on the beach.’
b. O Feco1 me falou que na praia e*1 vende cachorro quente.

the Feco to.me said that on.the beach sells dog hot
‘Feco told me that hot dogs are sold at the beach.’
c. O Feco1 me falou que todo dia e1 pega o metrô.

the Feco to.me said that every day takes the subway
‘Feco told me that he takes the subway every day.’
d. Na praia, o Feco1 me falou que e*1 vende cachorro quente.

on.the beach, the Feco to.me said that sells dog hot
‘Feco told me that hot dogs are sold at the beach.’ or
‘Feco told me at the beach that he sells hot dogs.’
6. CONCLUSIONS
In this work, I have shown that, although null subjects of finite embedded clauses
in BP and Finnish have properties which resemble those of controlled subjects (they
must have a close (local) c-commanding antecedent, etc.), they should not be ana-
lyzed as such. Taking those subjects to be controlled or derived by movement (which
are equivalent in the present context) cannot explain several facts in these languages.
For instance, it cannot explain why a matrix object may not normally be interpreted
as the antecedent of a null subject. It also does not explain the fact that moving an
object to the CP domain makes it possible for the object to be interpreted as the
antecedent. In order to explain such facts, movement analyses would have to resort
to ad hoc stipulations and unfounded assumptions. The analysis presented in sec-
tion 4, however, accounts for those facts in a straightforward manner.
Data presented here do not bear directly on control nor on whether control
should be analyzed on a par with raising. But a parallel can be made. An analysis
of control involving movement seems to be very simple and elegant, but leaves
much data unaccounted for (especially with respect to non-obligatory control)
and it makes wrong predictions (with respect to implicit controllers, for instance).
Trying to account for that data or remedy such predictions would probably turn the
simple analysis into one as complicated and inelegant as any of its predecessors.
* I would like to thank Angela Bartens, Arja Häkkinen, Auli Hakulinen, Helena Halmari, Paivi
Koskinen, Minna Niskanen, Anne Vainikka and Maria Vilkuna for their native judgments and to Hannu
Reime for all his generosity in helping me with data. Obviously, all errors remaining in here are my own.
1
The definition of ‘rich’ and ‘poor’ inflection is debatable. Here, I take ‘rich’ agreement to mean
loosely a system that has sufficient morphology to provide information on person, number (and
maybe gender) of the subject in a non-ambiguous fashion (cf. Speas 1994). Finnish is not usually
considered a poor agreement language, although there is no morphological difference between
singular and plural in the third person. In BP, there is no difference between the second and third
person both in the singular and in the plural, and the system is usually considered poor.
2
There is one other movement analysis of null subjects in BP that I am aware of: Ferreira 2000.
Ferreira’s analysis will not be reviewed here because it is not as elaborated as Rodrigues’ and because
it does not consider the most crucial facts presented in Modesto 2000a, b, which are reproduced in
(26–27) below. The interested reader can readily confirm that Ferreira’s analysis leaves such data
unaccounted for, in addition to all the data that Rodrigues 2004 also does not account for.
3
Obviously, besides the assumptions in (7), Rodrigues incorporates all the assumptions made in
Hornstein (1999) to make a movement analysis of control work, such as treating θ-roles as features
and allowing DPs to check, or have, several θ-roles.
4
Here and below, traces stand for copies deleted at PF; irrelevant details omitted throughout.
5
Besides mentioning that Spec FP is the position of preverbal subjects in Romance, as argued by
Raposo and Uriagereka (1995), Rodrigues does not discuss what this F projection is or what its
(semantic or syntactic) role in the derivation is. It seems to be there only to make movement out
of a phase to be in accordance with some version of the PIC. In Modesto (2007) and below, I
argue that there is in fact an F projection above TP which hosts (grammatical) topics in Romance
languages, explaining the topic-like characteristics of preverbal subjects in those languages (cf.
Barbosa 1995; Cardinaletti 1997; Alexiadou and Anagnostopoulou 1998; among many others).
6
Such a movement appears to be an instance of improper movement. It may be that such movement
is possible because the DP has not checked its Case yet, a point not discussed in Rodrigues 2004
although it is unclear why this should be so.
7
Although Rodrigues discusses only convencer, I will assume that her arguments apply to the whole
class of verbs that behave like convencer.
8
Rodrigues (2004) agrees that the wh-phrase becomes the only possible antecedent in sentences like (26a).
However, she says that ‘there might be nothing within the grammar preventing the matrix subject . . . to
be the antecedent’ in virtue of the fact that most speakers take the subject as the most likely antecedent
when confronted with the sentence in (i), where that reading is the most plausible pragmatically:
(i) Quem2 que a Maria1 convenceu t2 que e1/2 estava grávida?

who that the Maria convinced that was pregnant
‘Who did Maria convince that she was pregnant?’
This can be viewed differently, however. The fact that (26a), a pragmatically neutral sentence, is
interpreted as taking the wh-phrase as the only possible antecedent of the null subject shows that
this is a grammatical fact that needs to be explained. The fact that pragmatics may override grammar,
as seen in (i), does not show that the grammatical fact seen in (26a) is not a fact; it just shows that
pragmatics may interfere with grammar, which is widely known. Thus, for instance, examples such
as ‘John1 only loves JOHN1’ do not necessitate the conclusion that there is no grammatical reality
in principle C of the Binding theory.
9
In other words, it could be assumed that chain formation is possible here because a φP is nothing
more than a set of φ-features and, since the φ-features of the embedded and the higher subject are
the same, the φP is completely contained by its antecedent and, therefore, it is indistinguishable
from it.
10
Note that I assume with Chomsky (2001a) that all evaluation with respect to locality of movement
is done at the phase level. Therefore, an object may move to Spec FP on its way to a higher posi-
tion since any other locality abiding derivation would violate Minimality, which is a condition on
movement itself.
11
According to Holmberg (2005), the generic reading arises when a φP is not bound and, therefore,
cannot be interpreted referentially; or when it is bound by an abstract generic operator.
V
EXTENSIONS AND ALTERNATIVES

TO THE MTC
CEDRIC BOECKX AND NORBERT HORNSTEIN
ON (NON-)OBLIGATORY CONTROL*
1. THE ISSUE
For the purposes of the present paper, we will assume that a movement
approach to obligatory control (OC) is essentially correct. Although this view
is not shared by all linguists working within the generative tradition, we think
that the movement theory of control (MTC) has many desirable conceptual
and empirical virtues, and has done well in the face of what some viewed as
lethal problems (see Boeckx and Hornstein 2003, 2004, in press; and Boeckx
et al., forthcoming, for extensive discussion). Within the MTC, it is fair to
say that non-obligatory control (NOC) has been pushed to the side, with the
focus of inquiry resting on OC. There are good reasons for this. As OC is
mediated by movement (OC PRO being an A-trace), its properties reflect core
features of UG. NOC in contrast is the elsewhere case and is mediated by an
empty pronominal category, pro, in the subject position of nonfinite clauses
(see Hornstein 2001, 2003; Boeckx and Hornstein 2003, 2004.) Nonetheless,
the MTC is incomplete without an account of pro’s distribution, the cynosure
of what follows.
The present paper focuses on the following question: even assuming that PRO
reduces to a copy/trace left by movement and that movement is preferred
to pronominalization (as Hornstein 2001, 2003 has argued), what prevents a DP
that cannot licitly antecede PRO (i.e. a copy/trace of A-movement) from binding
a pro (i.e. a null pronoun) in the same position?1
Some concrete examples will make the problem clear.
(1) John persuaded Mary to leave
(2) John1 persuaded Mary2 [PRO*1/2 to leave]
(3) John1 persuaded Mary2 [pro1/*2 to leave]
Sentence (1) has the structure (2). Given (2) and the assumption that PRO=copy/
trace of A-movement we can explain why Mary is the antecedent and John
cannot be: For John to be the antecedent requires that it move over Mary on
its way to Spec, vP. As this violates minimality, it cannot be the antecedent.
In contrast, movement of Mary from the embedded clause to the object of persuade
is licit and does not violate minimality (if we assume, as the MTC does, that
movement into theta positions is legitimate). So, given (2), we explain why
251
© 2007 Springer.
252 CEDRIC BOECKX AND NORBERT HORNSTEIN
Mary is and John is not an antecedent of PRO in (1). However, why can’t
(1) be analyzed as (3)? Here we have pro in place of PRO and the relation
of John to pro should be fine. Moreover, in this structure, Mary should not
be a possible antecedent on the assumption that movement trumps pronomi-
nalization (cf. note 1). Thus, we can explain why (3) with pro indexed 2 is out
(because, (2) with index 2 is good). But why (3) with pro bound by John is illicit
remains unexplained.
To phrase the problem differently, we assume that a coupling between an
antecedent and a pronoun is licit just in case movement cannot establish the
same relation. Thus, if one can move from one position to another, a DP in the
‘target’ cannot bind a pronoun in the ‘launch’ site, i.e. position of the trace.
This is how we understand the assumption that movement is cheaper than
pronominalization. However, this also implies that if movement is not possible
between positions A and B, then binding should be. What we see in (3) is a concrete
example of this option. However, we also see that it is impossible; (1) cannot
be interpreted with John as the leaver. The problem for the MTC is why (1)
cannot be interpreted as (3) given the basics of the MTC?
2. MORE EVIDENCE FOR PRO
Before attempting an answer, let us consider another piece of evidence that favors
relating the availability of pro (and hence NOC) to the impossibility of movement:
NOC is always licensed inside islands. Thus, for example, in (4) the subject of the
gerund can be bound by John and this binding has all the hallmarks of NOC (see
Hornstein 1999, 2001 for discussion).
(4) John said that [pro washing himself delighted Mary]
Interestingly, we can also get OC here.
(5) John said that [PRO washing herself delighted Mary]
The relation between John and pro in (4) cannot be formed by movement as extrac-
tion from a subject gerund (to the matrix subject position) is illicit. Thus, pro is
allowed to mediate this relation. However, in (5) movement from the subject of the
gerund to where Mary sits is licit, so PRO is required here and pro relating to Mary
is not.2 If this is correct, then the position occupied by PRO can also be occupied
by pro.3 What cannot occur is the following: both a structure in which DP binds
pro is licit and in which movement from the position of pro to that of the binder is
also licit (i.e. the structure in which DP binds PRO in place of pro). If this is cor-
rect, then it follows that structures are never classified as OC or NOC (e.g. there is
never selection for an OC complement). Rather, it is relations that are OC or NOC.
Furthermore, as NOC is only licit where OC is not and given that OC is formed by
movement, it follows that NOC will occur where movement is prohibited, i.e. inside
islands. However, as should be clear, this does not help us with (3) above as here the
movement is illicit yet the structure is not available.
ON (NON-)OBLIGATORY CONTROL 253
That the MTC treats OC and NOC as relations is an important point that is worth
emphasizing. In the best case, any approach to control (including the MTC) should
refrain from cataloguing sentence types as OC or NOC. These two descriptive predi-
cates should not type to clauses because they are more analogous to ‘bound’ and
‘free’ than to ‘interrogative’ and ‘declarative’, i.e. OC and NOC describe relations
between nominal expressions, not selection/subcategorization relations between
predicates and types of clausal complements. As grammatical theory does not dis-
tinguish clauses as ‘reflexive’ or ‘pronominal’ by whether they contain anaphors or
pronouns, it should not, by parity of reasoning, identify sentences as ‘(obligatory/
non-obligatory) control’ clauses.
To put this another way: selection/subcategorization accounts are rather stipu-
lative. Thus to the degree that one needs to resort to such to cover some set
of data, one surrenders explanatory ambitions. Consequently, these descriptive
resources should be employed very sparingly.
3. A PROPOSAL
Consider another problematic case:
(6) John kissed Mary without [empty category] getting embarrassed
(7) John1 kissed Mary2 without PRO1/*2 getting embarrassed
(8) John1 kissed Mary2 without pro*1/2 getting embarrassed
(9) John1 kissed Mary2 without him*1/her2 getting embarrassed
Here is a case of adjunct control. If the structure of (6) is (7), then we can account
for why John must be the antecedent of PRO (it is OC and formed by sideward move-
ment, cf. Hornstein 2001 for details). This also explains why (8) with pro anteceded
by John is out and why (9) with him bound by John is out: again, move trumps pro-
nominalize. However, why is (8) (where pro and Mary are co-indexed) not an option?
Note, if anything, this is even more troublesome than (3) above as (9) indicates that
an overt pronoun bound by Mary is perfectly acceptable.4 So why is a null pronoun
with the same reading unacceptable? This fact suggests that the problem noted here is
not a purely grammatical one as pronouns bound by subjects are fine in this position,
though null pronouns are not. Why?5
If we insist that the problems in (3) and (8) get a unified approach (not an obvi-
ous requirement, but not a bad one either), then the facts in (9) indicate that more
than grammatical requirements are at issue. What else could be at stake? We would
like to suggest a parsing-based approach. More particularly, the grammar does not
block the structure in (3) with John as antecedent nor (8) with Mary binding pro.
Both are grammatically fine. The problem is that neither would ever be accepted by
a well-behaved parser.6
Let us make the following (as far as we can tell, fairly standard) assumptions:
a. Parsers move from left to right and project structure rapidly and deterministically on the
basis of local information.
b. Parsers are transparent with respect to grammars. So, if grammars encode a condition,
parsers respect it.7
Given assumption (b), we expect parsers to prefer traces to pronouns (if gram-
mars prefer movement to pronominalization) and consequently that parsers will
treat gaps as copies/traces in preference to analyzing them as null pronominal pros
(ceteris paribus). In addition, we expect parsers to be sensitive to earlier informa-
tion. So, as a parser builds structure left to right, it will prefer to treat a potential
gap as a copy/trace (rather than a pro) if it can ceteris paribus (there is one impor-
tant ceteris suggested below). Given this, let us return to the NOC cases above.
4. BACK TO THE PROBLEM
Consider (1) again, repeated here as (10).
(10) John1 persuaded Mary2 to leave
As the sentence is parsed we arrive at to and the parser realizes that it must assign
a subject to the embedded clause. Moreover, the parser ‘sees’ that the subject is
a null category, either a pro or PRO (copy/trace). As the parser incorporates the
principles of the grammar, it prefers to ‘drop’ a trace here if it can. Recall, that
grammars ‘prefer’ movement to pronominalization, so given that parsers are
transparent with respect to grammars (i.e. parsers employ the same principles as
grammars), the parser ‘prefers’ to drop a trace here if it can. As it can, it does.
Thus we get (11).
(11) John1 persuaded Mary2 [PRO to leave]
Furthermore, this copy/trace must have Mary as antecedent via minimality and so
(10) gets the parse in (11) which requires that Mary antecede the PRO.
Let us put this another way. What is wrong with (12)?
(12) John1 persuaded Mary2 [PRO1 to leave]
It would require that at to the parser drop a pro in this position, for the only empty
category that could take John as antecedent is a null pronoun. However, to drop a
pro requires ignoring the parsers (built-in) preference for a copy/trace over a pro-
noun, all things being equal; a preference the parser has in virtue of being structur-
ally transparent to the grammar which prefers movement over pronominalization.
This makes (12) computationally unavailable and this accounts for the lack of the
indicated interpretation.
One point is worth emphasizing here. The ‘preference’ the parser displays arises
as a design feature of a parser that conforms to Transparency (a very good condition
(perhaps an optimal one) for regulating the relation between grammars and pars-
ers). It is often assumed that parsing strictures can be overridden given greater
resources. So, for example, center embedding structures can be parsed given more
memory ‘space.’ The suggestion above, however, cannot be so easily ameliorated.
The problem is not one where extra resources would help. If parsing principles
must respect grammatical ones (i.e. if Transparency holds), then a parser cannot
circumvent these principles by using additional memory or attention resources.8
The parse is simply not available.9
The same account extends to (8), repeated here as (13).10
(13) John1 kissed Mary2 without pro*2 getting embarrassed
This too is parsing inadmissible. When the parser gets to the gerund and needs to
drop an ec, it must drop a copy/trace if it can. Thus it prefers a PRO to a pro. As a
PRO can be licitly dropped here, it must be. If it is, however, then John must be the
antecedent. Thus, if there is an empty category in the adjunct, it will be analyzed
as a PRO (copy/trace) by the parser and so the indicated reading in (13) will be
unavailable.
Observe that this account turns on there being an empty category in the adjunct.
By ‘seeing’ nothing there, the parse must ‘decide’ what sort of empty category to
drop into the subject position. As it prefers dropping traces if it can, it drops a
trace and not a null pronoun. However, if there is an overt pronoun occupying the
same position, the parser is not faced with any choice as to what it must do and
as pronouns are grammatically licit here, we can derive sentences like (14) with
an overt pronoun anteceded by Mary.11
(14) John1 kissed Mary2 without her2 getting embarrassed
Consider now the last set of cases. We noted that examples like (15) are fine with
the indicated interpretation.
(15) John believes that [pro washing himself] would delight Mary
Here the parser gets to the subject gerund and ‘encounters’ an empty category.12 It
can treat it as a copy/trace or a pro. Note, however, that it is inside an island and if
it wants to link John to this element, it must treat it as a pro. Observe that if it were
a copy/trace this relationship would be illicit as it would require movement from
an island. As a PRO is not allowed here (traces are never allowed within islands),
a pro is licensed by the grammar. However, this does not end matters. Note that
(16) is also acceptable.
(16) John believes that [PRO washing herself] would delight Mary
The PRO is a copy/trace residue of movement.13 So, it seems that the parser can drop a
trace here. Why then doesn’t this prevent dropping a pro in (15)? The answer is that the
parser here must weigh a competing parsing demand. It is known that parsers like to
assign interpretations to empty categories (and dependent elements in general) very
quickly.14 Thus, it is generally the case that pronouns greedily appropriate suitable
interpretive antecedents (referential anchors) very rapidly online. If we add to our
previous two assumptions, the further assumption that parsers are interpretively
greedy, then in cases such as (15) and (16), the parser has competing preferences:
it would both like to assign an interpretation to the empty category (at this point
in the parse) and it would prefer to treat the empty category as a trace rather than
a pronoun. In contexts like (15) and (16) these desiderata pull in opposite direc-
tions: if the empty category is understood as a pro it can be related to John and so
can rapidly be provided with an interpretation at this point. However, this will also
require overriding its preference for traces over pronouns. On the other hand, if it
drops a trace here, then though it cannot resolve the interpretation of the empty
category at this point (as there is no antecedent yet available for the PRO (copy/
trace)) it can adhere to its preference for traces over pronouns (i.e. PRO over pro).
Recall, this is a case where the antecedent will only become visible downstream. In
short, as both options have their virtues, we suggest that both parses are available.15
It is instructive to compare (15) and (16) with (12) and (13) above. In the lat-
ter two cases, the two parsing demands coincide. The antecedent for the empty
category is to the left of the empty category in both cases. Thus, whether the empty
category is analyzed as a trace or a null pronoun, the empty category can be inter-
preted. Thus, the parser’s desire to interpret the empty category quickly does not
compete with the parser’s desire to drop a trace rather than a pro. Consequently,
the trace (PRO) is dropped. The only relevant cases, then, will be those in which
the demands compete and these arise just in case the antecedent for the empty
category is downstream from the empty category.
If this analysis is roughly on the right track, then some predictions follow. Con-
sider a sentence like (17) uttered discourse initially.
(17) Having to wash behind the ears made Mary angry at Bill
Here, there is no parsing advantage to interpreting the empty category as a null pronoun
(there is nothing to link the empty category to so that it can be quickly interpreted). As
such, we would expect the parser to drop a PRO here, giving us a structure like (18).
(18) [PRO having to wash behind the ears] …..
Thus, the parser will analyze the empty category as a residue of A-movement. The
antecedent of this will then necessarily be Mary.16 Note that were there a pro here
it should be able to have Bill as antecedent. However, this reading seems unavail-
able in (17). If we substitute his for the in (17), Mary is doing the washing behind
Bill’s ears! Note, however, that an overt pronoun can have Bill as antecedent.
(19) Him having to wash behind the ears made Mary angry at Bill
The reason is that a pronoun is grammatically permitted and the parser does
nothing more than put what it hears where it hears it. Thus, what cannot occur
here, because of parsing preferences, is a null pronoun, viz. pro. Overt pronouns
are fine, as are traces due to sideward A-movement (viz. OC PROs).
5. EXTENSIONS
The interpretive properties of NOC constructions are explained once we iden-

tify NOC PRO with pro and deduce its distribution. The proposal here is that its
distribution is a dual function of the grammar and the parser. We have relied on the
assumption, common in the parsing literature, that parsers use the same sorts of
principles and target the same sorts of entities as grammars do, i.e. they are trans-
parent to grammars. We have also assumed that parsers like to resolve the inter-
pretation of dependent elements (e.g. empty categories, anaphors, pronouns) very
quickly. In combination, these two assumptions can account for the interpretation
of NOC constructions by predicting where and when pros will be available.
The transparency conception of the relationship between parsers and grammars
can be also used to explain other anomalous data. Let us briefly outline two cases.
Recently, several authors have investigated grammars which countenance side-
ward or inter-arboreal movement (see Nunes 1995, 2004; Bobaljik and Brown
1997, Uriagereka 1998; Hornstein 2001; Kayne 2002; Boeckx 2003b; Agbayani
and Zoerner 2004, Ferreira 2004; Kiguchi 2004; Rodrigues 2004; Boeckx and
Hornstein 2005; Cheng 2005). The argument for allowing such operations is
that they follow seamlessly from three standard assumptions: (i) that movement
reduces to more primitive operations Copy and Merge, (ii) that Extension holds
and so derivations must involve multiple unconnected sub-trees, and (iii) that deri-
vations are bottom-up. A consequence of these three assumptions is sideward
movement (SWM): the situation in which a copy from one sub-tree is merged to
another unconnected sub-tree.17
The one major empirical problem for SWM theories that we are aware of is the
following. Assume for the sake of argument that OC PRO and local reflexives are
the residues of overt A-movement. Then it should be possible to license configura-
tions like (20) via derivations like (21).18
(20) a. *John’s mother washed himself
b. *John’s1 mother washed PRO1
(21) a. Merge wash, John: [wash John]
b. Copy John: [wash John], John
c. Merge John, Mother: [John mother], [wash John]
d. Merge [John mother], [wash John]: [TP [DP John mother] [VP washed John]]
e. Either spell out the lower John as a reflexive or delete it and get PRO19
This derivation seems unexceptionable and so the sentences in (20) should be fully
acceptable, a sadly incorrect consequence. Curiously, the transparency assump-
tion can be used to explain the unacceptability of these sentences without denying
their generability. Consider the details.
Assume that the grammar obeys the A-over-A (A/A) principle. Hornstein
(2005a, b) has proposed reducing this to minimality in the following way.20, 21
Distance between points in a phrase marker is measured by the path traversed.
A path consists of the union of the MaxPs dominating the target and the source
of ‘movement.’ To illustrate, the path from PRO in (20b) to John is {DP, VP,
TP} given the structure in (21d). Paths measure distance. Minimality (of which
the A/A is a special case) requires that path length be minimized. This is a well-
attested principle of grammar. Given transparency, it must also be a principle of
parsing. With this in mind, consider the path from John to PRO/himself in (20a,
b) and compare it with the path from John’s mother to PRO/himself. The path of
the latter must be shorter by at least one MaxP for the DP John’s mother domi-
nates John but not the DP itself, assuming, as is natural, that Domination is not
reflexive. If both are potential antecedents for the reflexive, then the fact that the
distance from the reflexive to the containing DP is shorter than the one to John
blocks the parser from relating John to the reflexive because there is a shorter
path that suffices, viz. the path of the containing DP John’s mother to the PRO/
reflexive.22 Thus, though generable, as we saw, the structure with the indicated
relation is not parsable by transparent procedures and so is unacceptable.
The logic here is the same as it was above regarding the placement of pro. The
grammar has minimality as a core requirement. Thus, the parser must as well. How-
ever, this principle when used in evaluating the sentence from left to right cannot relate
the DP in Spec D to the reflexive as the containing DP blocks this via minimality/
A-over-A. If acceptability reflects both generability and parsability then being
unparsable in principle suffices to account for the unacceptability of sentences like
(20) even assuming that SWM is a fully acceptable grammatical operation.
There is an additional empirical reason for taking this approach to the exam-
ples in (20). There are languages in which this sort of anaphor binding seems to be
perfectly acceptable. It is well known that in some of the East Asian languages, an
antecedent need not c-command a local anaphor in order to bind it. Examples
like (22) are perfectly acceptable.
(22) Zhangsan de guiji hai-le ta ziji/??ta

Zhangsan DE trick harm-PERF he-self/he
Zhangsan’s tricks harmed himself.
Note that here the local reflexive taziji is in complementary distribution with the pro-
noun ta, as happens in the more standard c-command cases in (23).
(23) Zhangsan hai-le ta ziji /*ta

Zhangsan harm-PERF he-self/ he
Zhangsan harmed himself.
We take this complementarity with pronouns to indicate that the reflexives in both
(22) and (23) are locally bound by Zhangsan. Importantly, not all cases of sub-com-
mand binding are felicitous. Thus, if the Zhangsan is buried inside a human-headed
nominal, e.g. Zhangsan de mama (Zhangsan’s mother), then it is no longer a potential
antecedent for the local reflexive. The reason is that Chinese reflexives require
human antecedents. Once this is factored into the equation, the possibility of
sub-command is restricted by the A-over-A in those cases where both are poten-
tial antecedents of the reflexive. Where the container is not a licit antecedent,
sub-command obtains.23
(24) Zhangsan1 de mama hai-le *ta ziji1

Zhangsan’s DE mother harm-PERF he-self
Zhangsan’s mother harmed himself.
Note that English differs from Chinese in not requiring that the antecedent of a
reflexive be human, e.g. This argument speaks for itself. As such, the A/A reason-
ing will apply quite generally and antecedents that sub-command will be prohib-
ited; e.g. *John’s argument impressed himself.24
Consider now a second instance of the same logic. It is well known that pars-
ing is sensitive to islandhood. For example, it is well attested that the filled-gap
effect does not occur within islands.25 This effect is generally interpreted as try-
ing to insert the antecedent into the relevant position. So, for example, if there
is a WH in peripheral position and the parse comes to a DP position that could
have hosted the WH there will be a slowdown at this point and this slowdown is
understood as resulting from trying to interpret the WH in this position. For
convenience we will describe this slowdown as the result of trying to ‘drop a
trace/copy.’ Thus, filled-gap effects are the result of trying to insert the antecedent
into the relevant position for interpretive reasons. What is interesting is that pars-
ers do not show filled-gap effects within islands. In our terms, they cannot drop
traces/copies inside islands. If we assume that such copies are required to license
reconstruction effects, then we should not find reconstruction effects into islands
even if movement is grammatically possible from islands (see Boeckx 2003a for a
defense of the latter claim). Thus, even if movement from within an island is licit,
we have independent evidence that the parser obeys island conditions in the sense
that it cannot drop traces/copies within them and this suffices to block reconstruc-
tion into islands.
We mention this because Aoun and Li 2003 have provided interesting evidence
from Lebanese Arabic that such reconstruction into islands is indeed impossible
though relations of WHs across islands display Superiority Effects, which can be
explained (via minimality) were movement from the island permitted.26 Here is
not the place to rehearse the details. But it is interesting to note that the same logic
deployed above extends to these cases as well once the parsers ‘respect’ for islands (as
gauged by filled-gap effects) is recognized.27
6. CONCLUSIONS
To sum up, it is reasonable to assume that parsers and grammars are closely
related. One way of ensuring this is to assume that parsers are transparent
with respect to grammatical principles and categories as urged by Berwick and
Weinberg 1984. We have argued that this assumption allows for an account of the
distribution of pro in NOC constructions. The assumption further cleans up
other problems concerning movement in the context of minimalist accounts based

on the copy theory. If Transparency is elevated to a regulative principle it might
place some interesting restrictions on candidate principles of UG: e.g. they must be
directionally invariant. What we mean is that they should be applicable regardless
of whether deployed in a bottom-to-top, top-to-bottom, or even inside-to-outside
flow of information. In this sense the principles should be directionally invariant.
Most principles that have been proposed have this nice feature. It would be worth
considering what principles that fail to have it might look like. We reserve this for
future work.
* This paper represents a portion of the talk we delivered at the raising and control work-
shop organized by Bill Davies and Stan Dubinsky during the LSA Summer Institute 2006 at
Harvard/MIT. We wish to thank the organizers for the invitation and the audience for their
questions and comments.
1
Hornstein 2001 argues that a preference for movement over pronominal binding underlies the com-
plementary distribution of anaphors and bound pronouns (i.e. Principles A and B of the Binding
Theory) as well as the complementarity of OC and NOC PRO.
2
Kiguchi and Hornstein (2002) argue that the relevant movement proceeds sideways. It is also con-
sistent with what follows if we assumed a derivation reminiscent of psych verbs in Belletti and
Rizzi 1988. In the latter case, the gerund would begin its derivational life within the VP before
moving to the matrix subject position. As the surface object of the psych verb is base generated
higher than the derived subject, movement from within the adjunct to this position should be pos-
sible, thus licensing an OC PRO under movement. This would parallel the derivation of reflexives
constructions like Pictures of himself delighted Bill, but with movement rather than binding being
the relevant operation. As which of these derivations is correct is irrelevant here, we leave the spe-
cifics to the reader.
3
It can also be occupied by an overt pronoun:
(i) John1 believed that him1 washing behind his ears delighted Mary
As expected, moreover, an overt pronoun with Mary as antecedent is illicit.
(ii) John believed that *her1/PRO1 washing behind her ears delighted Mary1
The latter is expected if Move blocks pronominalization. The PRO here is a residue movement.
4
Another example with complements illustrates the same point:
(i) John1 would prefer *him/PRO1 leaving early
(ii) Mary1 said that John would prefer pro*1/her1 leaving early
In (i), John cannot antecede him because movement (i.e. PRO) is licit here. In (ii), Mary can ante-
cede her but one cannot have the same binding with a null pro. These data parallel those with
adjuncts above: why isn’t (ii) acceptable given that gerunds can have pro subjects and movement of
Mary from the subject of the gerund to the matrix TP is prohibited?
5
Note that it is not open to us to say that pro is not licensed here. We have assumed that pro is licensed
in the subject of gerunds to account for standard cases of NOC like (4). If pro is permitted here, then
it should be allowed in gerundive subjects in general. Many readers will appreciate that this requires
a reanalysis of the ‘pro’-drop parameter, which we think is necessary anyway, given the nonuniform
nature of phenomena that fall within the pro-drop rubric. Our hunch is that pro can freely occur in
non-case positions. We suspect that this is related to its being phonetically unspecified.
6
If we also assume that producers and parsers meet similar constraints, then this would not be pro-
duced either. Such an assumption is natural in any kind of analysis-by-synthesis model.
7
This does not imply that grammars are identical to parsers (Phillips 1996) – a position which we
think is untenable (Phillips 2004 appears to agree on this). Our assumption only implies that
the parser respects the design features of grammars. For discussion of the transparency relation
between grammars and parsers see Berwick and Weinberg 1984.
8
This does not mean that a pro can never be placed where a PRO can be. See below for a case where a
pro can be posited in a place where a PRO is licit in order to advance another parsing desideratum.
9
If this is correct a question that is often raised may prove to be without much content. It is often
asked if a sentence is unacceptable because it is ungrammatical or because it is unparsable. In the
above cases, this is a distinction without a difference: it violates the very same principles when pars-
ing as when grammatically generating.
10
The subject of the adjunct cannot be a PRO (trace) left by movement of Mary as this violates
economy (e.g. Merge-over-Move). See Hornstein 2001 for a derivation of the fact that adjunct
control is restricted to subjects.
11
The same account extends to the cases in note 4 which are left as an exercise for the reader.
12
The parser ‘knows’ that this is a subject because it follows that and because it knows that believe
does not take gerundive complements. This kind of structural and lexical information is standardly
assumed to be available online to the parser. Thus, at the point where the gerund is encountered,
the relevant information that the gerund is a subject (and hence an island) is known.
13
Either sideward movement or movement as in a psych verb construction. See note 2 above.
14
See, for example, Nicol and Swinney 1989, Osterhout and Mobley 1995, and Badecker and Straub
2002, for discussion. Thanks to Nina Kazanina for very helpful discussion and references.
15
It is possible that different speakers weigh these options differently. Nina Kazanina (personal
communication) has found many speakers for whom sentences like (15) with John as anteced-
ent become very odd when Mary is encountered. This suggests that these speakers value trans-
parency more than reference resolution. One of us (NH) gets similar effects with sentences
like (i):
(i) John said that after washing himself Mary danced with Dave
(ii) John said that Mary danced with Dave after washing himself
It goes without saying that the proposal above is not a fully worked-out account and that much
more detailed work needs to be done to flesh it out.
16
For details, see Kiguchi and Hornstein 2002. This paper provides further evidence that the empty
category here is an OC PRO and not a pro.
17
It is worth emphasizing that sideward movement follows from these basic assumptions and is not
something that must be specifically added. Rather, blocking it requires special stipulations. In
this sense, sideward movement is an interesting consequence of the most general, least stipulative
assumptions. This does not mean that sideward movement ever correctly describes any linguistic
phenomena, but it does suggest that the burden of proof is on those that wish to exclude it a priori.
Conceptually, the operation is on solid ground.
18
This assumes that inherent reflexives like John washed involve a copy of John in object position. For
discussion see Hornstein 2001, Boeckx et al. forthcoming.
19
The details of when each occurs are discussed in Hornstein 2001. The problem discussed here is
articulated in Landau 2003.
20
See Boeckx 2003a and Fukui 1999 for previous attempts to relate minimality and the A/A condition.
21
Details will be spare here. The interested reader is referred to Hornstein 2005a, b.
22
Kayne 2002 also proposes treating the unacceptability of (20) as an A/A violation.
23
If reflexives are formed by first merging the reflexive morpheme and its antecedent (e.g. [John-self])
and then moving the latter out, then the restriction that Chinese antecedents must be human is easy
to state. There are several versions of the movement theory of reflexivization, see Hornstein 2001
and references cited there.
24
It is worth observing that if this is correct this provides an independent argument in favor of side-
ward movement, though examples like (20) are generally taken to be lethal counterexamples to the
proposal that SWM exists.
25
For discussion see Phillips 2004 and references therein. The filled-gap effect is an attested slowdown
in reading time in sentences such as Who did John persuade Harry that Bill likes when the parser reaches
Harry and Bill. The interpretation is that the parser tries to integrate Who into the structure as
quickly as possible. When it hits persuade it attempts to do this only to find when it gets to Harry
that the position is filled. Thus the parser must rescind the attempted integration and this causes a
reading slowdown. Interestingly, these slow downs do not occur within islands (or more accurately,
within islands that the parser has reason to believe are islands at the time of the attempted integra-
tion). The interpretation is that the parser ‘knows’ that movement from islands is illicit and so does
not attempt to drop a trace within one.
26
See Aoun and Li 2003 for original discussion of the data, and Boeckx and Hornstein 2006 for
a reanalysis. The proposal here is different from the one made there concerning the absence of
reconstruction effects.
27
This suggests that islands are best understood not as constraints on movement but as restricting gap
creation (chopping) as Ross 1967 originally proposed, and as Boeckx 2003a has articulated in a
minimalist setting (see also Boeckx and Lasnik 2006; Hornstein et al. to appear). It is worth noting
that this understanding of lack of reconstruction has many parallels with the proposal in Aoun
and Li, where both the relation of the gap to the antecedent and the antecedent to the gap are
subject to minimality constraints. What we are proposing is that this duality is not redundant as it
follows from a principle of parser/grammar transparency. Joseph Aoun (personal communication)
informs us that he and Li are currently interpreting their earlier results along similar lines.
MICHAEL BARRIE
CONTROL AND WH-INFINITIVALS†
1. INTRODUCTION
This paper discusses control into wh-infinitivals as illustrated in the following examples.
(1) a. John knows when to wash the dishes.
b. Mary learned how to fly a 747.
c. Alexis told Janice where to meet before the play.
d. Alex told Jack where to get good cheese.
I show that there are two types of control into wh-infinitivals, specifically obliga-
tory control (OC) (represented by (1a) and (1c)) and non-obligatory control (NOC)
(represented by (1b) and (1d)). I further argue that NOC into wh-infinitivals is
instantiated as generic control. I present data from multiple sluicing and extrac-
tion across weak islands that distinguish these two types of control. Further, the
data will be shown to support an analysis in which OC into wh-infinitivals arises
by movement of the controller from the embedded clause into the matrix clause
(in the sense of Hornstein 1999, 2001), while NOC entails no such movement. I
argue that the embedded subject is represented by pro, which is clause-bound by
a generic operator. In short, I argue for the following representations for (1a) and
(1b), respectively.
(2) a. Johni knows [when ti to wash the dishes]
b. Mary learned [how pro to fly a 747]
The analysis proceeds as follows. The multiple sluicing data crucially rely on Fox
and Pesetsky’s (2005) proposal that linearization proceeds on a phase-by-phase
basis. In multiple sluicing environments, the two surviving elements must be able
to be linearized. This is only possible if the two elements were originally merged
in the same phase (for reasons to be made clear below). If one of the two sluiced
elements is in a higher clause in the surface structure, and multiple sluicing pro-
duces a grammatical result, then the higher element must have originated in the
same phase as the other sluiced element. In the case of control constructions,
the evidence from multiple sluicing suggests that the controller originates in the
embedded clause.
263
© 2007 Springer.
264 MICHAEL BARRIE
The section on extraction from wh-infinitival clauses revisits some of the origi-
nal data presented in Huang 1982 and discussed in Manzini 1992. I present evi-
dence that the data has been mischaracterized as involving an argument/adjunct
asymmetry. Instead, I show that extraction across wh-infinitivals is sensitive to the
distinction between OC and NOC. In particular, OC environments do not allow
extraction out of a wh-infinitival, whereas NOC environments do allow extrac-
tion. I will argue that in OC the controller raises from the embedded clause to the
matrix clause, thus using up the single escape hatch in the CP domain and prevent-
ing the wh-phrase from extracting out. In NOC, there is no controller that raises
from the embedded to the matrix clause. Thus, the wh-phrase is free to extract to
the matrix clause.
This paper is organized as follows. Section 2 presents the patterns of control
into wh-infinitivals, establishes a contrast between OC and NOC, and presents the
data on multiple sluicing and extraction from weak islands. Section 3 develops an
analysis of this type of control. First we establish the structure of the left periph-
ery of wh-infinitivals and then present an analysis in which the controller raises
from the embedded clause to the matrix clause in OC, but not in NOC. Section 4
presents a brief conclusion.
2. PATTERNS OF CONTROL INTO WH-INFINITIVALS
This section introduces the basic empirical facts on control into wh-infinitivals in Eng-
lish. We begin by discussing the types of predicates in which wh-infinitivals are
found and the types of wh-phrases that are found in these constructions. Next, a
distinction is made between OC and generic control, and it is shown that control
into wh-infinitivals can be of both types. The last two subsections present two
asymmetries (extraction across weak islands and multiple sluicing) that highlight
the difference between these two types of controls.
2.1 Predicate types
Landau (1999) presents a typology of control predicates consisting of implica-

tives, aspectuals, factives, propositionals, desideratives, and interrogatives. The
data below show that wh-infinitives appear only with desiderative and interroga-
tive predicates. Note that I define interrogative predicates as those that obligato-
rily appear with a wh-phrase in the left edge of the complement.
(3) Typology of control predicates
a. implicative
i. John managed to eat a coconut.
ii. * John managed how to eat a coconut.

CONTROL AND WH-INFINITIVALS 265
b. aspectual
i. Mary began to write a novel
ii. * Mary began when to write a novel
c. factive
i. Alice hates/likes to eat in the basement
ii. * Alice hates/likes where to eat.
d. propositional
i. Bill claimed to be fluent in 6 languages.
ii. * Bill claimed how many languages to be fluent in.
e. desiderative
i. Peter knows to arrive by 7 pm.
ii. Peter knows when to arrive by.
iii. Susan told John where to eat.
iv. Alice said when to pick up the dry-cleaning
f. interrogative (:= must have wh-XP)
i. Susan wondered where to eat.
ii. * Susan wondered to eat in the kitchen.
iii. Peter mapped out how to drive to Montreal.
iv. * Peter mapped out to drive to Montreal through Vermont.
Note that forget and remember can also take wh-infinitives, as the following
example shows.
(4) John remembered/forgot when to wash the dishes.
In this case, I argue that forget and remember are desideratives, not factives, since
they no longer imply the truth or falsity of the embedded clause. Thus, (5) is
contradictory while (6) is perfectly acceptable.
(5) #John remembered to wash the dishes, but he didn’t wash them.
(6) John remembered when to wash the dishes, but he didn’t wash them.
266 MICHAEL BARRIE
2.2 Types of wh-phrases
Control infinitives can appear with any wh-element, except why:
(7) Johni doesn’t know who ei to visit twh

Johni doesn’t know what ei to eat twh
Johni doesn’t know where ei to go twh
Johni doesn’t know when ei to leave twh
Johni doesn’t know how ei to sing twh
*Johni doesn’t know why ei to wait twh
Full XP’s are also available:
(8) John wondered which book to buy.
(9) John wondered what book to buy.
2.3 Two types of control
Here, we argue that control into wh-infinitivals can be either obligatory or generic.
Consider, first, the following sentences.
(10) The sign says where earb not to smoke.
(11) John’s new stock market analysis program can figure out when earb to buy Microsoft
shares.
(12) Mary knows how earb to defend oneself against killer bees.
Most traditional analyses (Bresnan 1982; Chomsky 1981; Manzini and Roussou
2000) assume that wh-control is arbitrary control as shown by the element earb in
(10–12). Landau (1999, 2000) argues that control into wh-infinitivals is not arbi-
trary control, but is rather OC. This conclusion is based convincingly on evidence
such as the following (Landau 1999:52):
(13) Johni wondered who ei to introduce himselfi to.
In (13), the controlled element (be it PRO or a trace) must be coreferential with
the controller rather than take generic or arbitrary reference, or the anaphor
would not be licensed. Although I agree with Landau that (13) is an instance of
OC, I argue that the sentences in (10–12) are examples of generic control. Before
considering the arguments in favor of generic control (as opposed to arbitrary
control or OC), observe that the sentences in (10–12) can be paraphrased with a
generic one.
(14) The sign says where one should not smoke.

(15) John’s new stock market analysis program can figure out when one should buy Micro-
soft shares.
(16) Mary knows how one should defend oneself against killer bees.
Note furthermore, that the other examples of control into wh-infinitivals cannot
be paraphrased with generic one, but with a coreferential pronoun.
(17) John wondered which book to buy. =John wondered which book he should buy.
(18) John knows where to pick up the opera tickets for tonight. =John knows where he
must pick up the opera tickets for tonight.
The standard argument for arbitrary control into wh-infinitivals comes from the
presence of oneself anaphors in these constructions.
(19) Mary wasn’t sure how to introduce oneself to the Queen.
Since Mary is not an appropriate binder for oneself, the claim is that there is a
PROarb in the subject position of the embedded clause which binds the anaphor.
Landau (1999) claims that the oneself-test has been misused to diagnose arbitrary
control, however. He offers the following pair of sentences as evidence. The unac-
ceptability of oneself in (21) is unexpected.
(20) Mary wasn’t sure when to introduce oneself to John.
(21) *Maryi wasn’t sure when to introduce oneself to heri.
Consider though, the following context: Mary has just been appointed to a high-rank-
ing diplomatic position in a foreign country whose customs she is unfamiliar with.
She knows that there will much formality involved in introductions. The fol-
lowing sentence becomes a bit better, but is still somewhat degraded.
(22) ?Maryi wasn’t sure how to introduce oneself to heri.
Since I argue, however, that these sentences are generic, not arbitrary, I attribute the
degraded status of (21) and (22) to the presence of the pronoun which has a definite
referent. Generic statements do not felicitously contain definite referents. If we replace
the definite referent with a generic DP, the sentence improves considerably, as shown
in (23).
(23) A princessj shouldn’t have to explain how to introduce oneself to herj.
Landau maintains that in examples (10–12), there is an implicit controller as in

the following example:
(24) The flight attendant said (to the passengersi) where ei to sit.
268 MICHAEL BARRIE
Such an analysis is difficult to maintain in examples (10) and (11) because in these
two examples habitual aspect is used, which does not facilitate a pragmatically
felicitous implicit recipient.
(25) ?#The sign says (to everyone who happens to read it) where not to smoke.
I argue here that the proper characterization of NOC into wh-infinitivals is that of
generic control using various diagnostics found in Krifka et al. 1995. Krifka et al.
give three such diagnostics for generic (or what they call characterizing) sentences.
The first test is straightforward. Generic sentences are incompatible with progres-
sive aspect.
(26) John knows where to be standing at 6 pm. (OC)
(27) * John knows where to be getting good cheese. (generic control)
Note that the sentences in (10–12) all are incompatible with progressive aspect,
thus supporting an analysis of generic control for these sentences.
(28) *The sign says where not to be smoking.
(29) *John’s new stock market analysis program can figure out when to be buying
Microsoft shares.
(30) *Mary knows how to be defending oneself against killer bees.
They also observe that generic sentences can be modified by the adverb usually.
(31) John usually knows where to get good opera tickets cheap. (generic control)
(32) *John usually knows where to meet before the opera tonight. (OC)
Again, in (10–12), the adverb usually is licit, thus supporting an analysis in which
these sentences can be analyzed as generic control.
(33) The sign usually says where not to smoke (but in this airport, it doesn’t give any indication).
(34) John’s new stock market analysis program can usually figure out when to buy
Microsoft shares.
(35) Mary usually knows how to defend oneself against killer bees.
Their third diagnostic for generic sentences is that in generics, the property
described by the predicate is an essential property of some entity mentioned in the
sentence. Thus, in (31), it is an essential property of John that he knows where to
get good opera tickets cheap. In the sentence John knows where to meet before
the opera tonight, however, it is not an essential property of John that he knows
the information contained in the infinitival predicate. Likewise, in the sentences in
(10–12), the properties described by the embedded infinitival clauses are essential
properties of the sign, John’s computer program and Mary, respectively.
Finally, one of the prototypical properties of genericity is that generics allow

for exceptions. So for example, it is still true that John knows where to get good
opera tickets cheap, even if on one particular occasion, there were no good, cheap
opera tickets at one of the places he suggested. However, it is only true that that
John knows where to meet before the opera tonight if, in fact, he absolutely knows
this information. This fact about generics is attested by the use of the adverb
usually as described above.
I conclude on the basis of the above diagnostics, then, that there are two types
of control into wh-infinitivals: OC and NOC or, more specifically, generic control.
Next, I discuss two asymmetries with respect to this distinction. The first is extrac-
tion across weak islands, followed by multiple sluicing.
2.4 Extraction possibilities
In this section, I demonstrate that generic wh-infinitivals and obligatorily control-

led wh-infinitivals differ in their extraction abilities. As illustrated in the data
below, wh-movement is permitted across a wh-island with a generic, but not with
OC. Thus, (36–38), which contain embedded generic infinitivals, are licit; whereas
(39–41), which exhibit OC, are ruled out.
(36) Which shares can John’s new program figure out when to buy?
(37) What kind of plane does John know how to fly?
(38) What kind of bee does Mary know how to defend oneself against?
(39) *Which restaurant does John know when to meet at?
(40) *Whose car does John know where to park?
(41) *Where did John wonder who to introduce himself to?
Consider, now, the following sentence.
(42) Mary told John where to buy champagne.
This is ambiguous between a generic reading and an OC reading. (The OC read-

ing is facilitated by making the event more specific: Mary told John where to buy
the champagne for tonight.) This sentence could be followed up thus:
(43) a. at any good wine shop. (generic reading)
b. at the store around the corner; it’s already been ordered. (OC reading)
If we extract from the embedded clause, only the generic reading is available:
(44) What did Mary tell John where to buy?

270 MICHAEL BARRIE
The lack of ambiguity is highlighted by the following ungrammatical example:
(45) *What did Mary tell John where to buy for the party tonight?
The paradigms in (7) and (36–45) will help us construct a movement-based analy-
sis of OC into wh-infinitivals.
To sum up our findings so far, we have seen that wh-infinitives appear only
with desideratives and interrogatives and that why is not permitted in infinitival
clauses. We have also seen that there are two types of control into wh-infinitivals
– obligatory and generic. We have also seen that wh-movement across wh-islands
is available only with generic control.
2.5 Multiple sluicing
In this section, I show that OC into wh-infinitivals can undergo multiple sluicing,
while generic control cannot. Consider the following two pairs of examples:
(46) a. A certain traveler decided where to eat in a certain city.

I forget which traveler in which city.
b. A certain traveler knows where to eat in a certain city.

*I forget which traveler in which city. (*on generic control reading)
(47) a. A certain dignitary knows how to introduce himself to a certain monarch.

I forget which dignitary to which monarch.
b. A certain dignitary knows how to introduce oneself to a certain monarch.

*I forget which dignitary to which monarch.
Examples (46a) and (47a) illustrate OC and license multiple sluicing of the matrix
subject and the embedded object. By way of contrast, examples (46b) and (47b)
illustrate generic control as argued above and do not allow multiple sluicing. The
data are introduced here to underscore the difference between OC and generic con-
trol into wh-infinitivals. We will account for this difference in the next section. Before
ending this section, let us convince ourselves that (46b) and (47b) constitute generic
control by applying the same diagnostics as above. Here are the test sentences.
(48) a. *A certain traveler knows where to be eating in a certain city.
b. *A certain dignitary knows how to be introducing oneself to a certain monarch.
(49) a. A certain traveler usually knows where to eat in a certain city.
b. A certain dignitary usually knows how to introduce oneself to a certain monarch.
These sentences are clearly incompatible with the progressive aspect and can
appear with the adverb usually indicating that they allow exceptions. For exam-
ple, the traveler usually knows where to eat in whatever city, but perhaps his last
recommendation was a flop. Also, the dignitary might know how people should
introduce themselves to a given monarch, but perhaps he forgets the protocol for
how an archduke introduces himself to that monarch.
The tests for lack of genericity are a bit less straightforward and require some
explanation. Here, again, are the test sentences.
(50) a. A certain traveler decided where to be eating in a certain city.
b. ?A certain dignitary knows how to be introducing himself to a certain monarch.
(51) a. *A certain traveler usually decided where to eat in a certain city.
b. *A certain dignitary usually knows how to introduce himself to a certain monarch.
Both of these sentences are unacceptable with the adverb usually, suggesting that
they are not generic. The test with progressive aspect is less than clear, however.
The first sentence refers to the complete event of eating dinner. It is odd for some-
one to decide where an ongoing activity such as eating dinner should take place.
Rather, one decides where the complete event takes place. Consider the following
scenario, however. There is a certain scene in a movie where one of the characters
must be eating dinner when another character, say Sam, walks in and they meet.
The screen writer did not specify the exact location, and the director is unavailable
to make the decision. In this case, we could utter the following:
(52) The actor decided where to be eating when Sam walks in.
A similar argument can be made for the second sentence with the dignitary and
the monarch.
2.6 Conclusions
I have argued that control into wh-infinitivals in English can be instantiated as either
OC or generic control (but not arbitrary control). Furthermore, I presented two
asymmetries with respect to this contrast involving extraction from weak islands
and multiple sluicing. In particular, generic control allows extraction from a weak
island where OC does not. OC, however, permits multiple sluicing of the matrix
subject and the embedded object where generic control does not. The next section
provides an analysis for the structure of these two types of control structures into
wh-infinitivals, explaining the source of the two asymmetries discussed above.
3. ANALYSIS
3.1 Wh-infinitivals do not involve full CP
Following earlier work, I assume that why is merged higher than other wh-phrases
in the CP layer (Collins 1991; Ko 2005b; Lin 1992; Rizzi 1990, 1999). If infinitivals
generally have a reduced CP layer (or lack it altogether), this explains why
272 MICHAEL BARRIE
why is absent in infinitival clauses. This begs the question as to where other wh-ele-
ments appear in infinitival clauses. Consider an expanded CP (Rizzi 1997, 1999):
(53) ForceP > (TopP) > IntP > (TopP) > FocP > (TopP) > WhP > FinP
WhP is used as a label of convenience both here and in Rizzi 1999 to show that
wh-elements can appear in more than one position, depending on various factors.
Rizzi suggests that wh-phrases appear in SpecIntP in matrix clauses and SpecWhP
in embedded clauses. Ko (2005b), on the other hand, argues that the higher projec-
tion is reserved for why and its equivalents and that the other wh-phrases are merged
lower in the CP layer. What is important for us is that why is merged higher in the
CP layer than other wh-phrases. The wh-infinitival phrase must contain minimally
a FinP (finite/nonfinite distinction) and a WhP (to host wh-phrases), but nothing
more. I suggest, in fact, that such phrases cannot contain any higher projections
based on the following observation. Note that topics and focused elements are not
available in embedded infinitival clauses, as shown in the following examples:
(54) John wants to give a book to this student.
(55) EVEN THIS STUDENT, John wants to give a book to.
(56) *John wants EVEN THIS STUDENT to give a book to.
(57) This student, John wants to give a book to.
(58) *John wants this student, to give a book to.
(55) and (57) show that topics and focused phrases are available in the matrix clause.
But (56) and (58) show that there is no position in the left periphery of the embed-
ded infinitival clause to host topic and focus. If why is merged high (at IntP) and
infinitival clauses contain only the lower (right) portion of the expanded CP, then
this explains why why is not licit in infinitival clauses – there is no position for it.
There is a position for other wh-elements, though, which is WhP. In sum, then, we
conclude that FinP selects WhP, leaving WhP at the right edge of the CP domain.
This approach makes interesting predictions concerning where why-phrases
can be interpreted. Consider the following contrast:
(59) When did John decide to buy a car? (ambiguous)
(60) Why did John decide to buy a car? (unambiguous)
In (59), when can be interpreted in either the matrix or the embedded clause, while
in (60), why can be interpreted only in the matrix clause. This follows from the fact
that why could not have originated in the embedded clause, since there is no place
for it to merge, while when can merge in SpecWhP in either the matrix or embed-
ded clause. Recall also from above that for generic control, wh-extraction may
take place across a weak island (61). Again, however, if the element in the matrix
SpecCP is why, it cannot be interpreted in the embedded clause.
(61) Which shares can John’s new program figure out when to buy?
(62) Why did John decide which car to buy?
Thus, (61) has an embedded reading on the DP which shares, while (62) must have
a matrix reading on why.
3.2 Wh-infinitivals constitute a phase
We have seen that wh-infinitival clauses contain a significantly reduced CP layer,

thus we must determine whether these constructions constitute a phase or not.
This, of course, is necessary to formulate an analysis of control into wh-infiniti-
vals and to explain the asymmetry with respect to extraction possibilities between
OC and NOC constructions. Recall that extraction out of a wh-infinitival is possible
only with NOC and not with OC. In this section, we will see that the left edge of
wh-infinitivals does indeed constitute a phase.
To test for phasehood, Legate (2003) creates diagnostics in which a putative
phase edge must be able to be targeted as a site for potential reconstruction. The
logic of this test proceeds as follows:
(63) Schematic Test for Phasehood

Op … X … • … Y (• is test site for phasehood)
Show a configuration in which the following scope ordering is possible:
X > Op > Y
Consider the following sentence:
(64) Bill told Mary1 in which of Picasso2’s galleries to send him2 every portrait of herself1.
Here, an inverse scope reading is available between the wh-phrase and the QP, and
the anaphor is still bound by Bill. Note that the wh-phrase cannot reconstruct to
get this order, or a Principle C violation would result.
(65) Bill > every portrait of himself > which gallery
Thus, there must be a landing site at the left edge of the infinitival clause and a
fortiori a phase edge. Since wh-infinitivals constitute a phase, we must seek an
explanation for the lack of island effects in generic control constructions and the
availability of multiple sluicing in OC constructions.
3.3 Generic control involves progen
Hornstein (1999, 2001) suggests that instances of NOC involve pro. This was met
with immediate resistance, since pro is not generally available in English as English
is not a pro-drop language. Hornstein’s suggestion might be tenable, if we assume
that in languages such as English, pro is restricted to generic and arbitrary (and
274 MICHAEL BARRIE
non-referential) readings. Thus, example (10) has the representation shown in

(66), which will be modified subsequently.
(66) The sign says [WhP where [FinP [IP progen not to smoke]]].
Example (1) has the following representation, which will also be modified.
(67) Johni knows [WhP when [FinP [IP ti to wash the dishes]]].
Crucially, the difference here is that the subject has raised from the embedded
clause in (67) to the matrix clause, but in (66), the base position of the subject is
in the matrix clause, while the subject of the embedded clause is a generic pro. If
this analysis is on the right track, then the facts related to multiple sluicing and
extraction fall into place. We discuss these next.
3.4 Multiple sluicing
In section 2, we uncovered an asymmetry with respect to multiple sluicing with con-

trolled wh-infinitivals. Crucially, multiple sluicing is available with OC, but not with
generic control. We account for this asymmetry here within the framework of cyclic
linearization (Fox and Pesetsky 2005), assuming the structures in (66) and (67) above.
Recall that multiply sluiced elements must be clause-mates:
(68) Multiple Sluicing
a. A certain boy talked to a certain girl

I forget which boy to which girl.
b. Fred told a certain boy to talk to a certain girl

c. A certain boy told Fred to talk to a certain girl.

*I forget which boy to which girl.
Following Fox and Pesetsky, the two multiply sluiced elements must also be phase-
mates, where co-phasehood includes material up to and including the left phase edge
for overtly moved elements and elements that have undergone initial Merge, but
not covertly moved elements. So, in (68a), the matrix subject starts in SpecvP of the
embedded clause, where it becomes linearized with the DP a certain girl. Under Fox
and Pesetsky’s theory, the grammar records the ordering statement < a certain boy, to
a certain girl > , which survives upon sluicing since both of these XPs are present in
the sluice. These two XPs can then be ordered with respect to each other. Looking at
(68c), now, we see that at no time are which boy and to which girl in the same lineariza-
tion domain (again, roughly a phase). Thus, there is no ordering statement of the type
< a certain boy, to a certain girl > that could order these two XPs upon sluicing, thus
giving rise to an ungrammatical result (see Barrie, to appear, for additional details).
Adopting this approach to linearization makes certain predictions regard-

ing raising and control. Looking at (69), the matrix subject DP a certain boy
can appear in a multiple sluicing construction with the embedded object. This
is unsurprising since this is a raising construction, where the matrix subject is
assumed to originate in the embedded clause and thus satisfies the co-phasehood
constraint for multiply sluiced elements. Considering (70) now, we observe that
multiple sluicing is available with the control predicate decide. This indicates
that the matrix subject must also originate in the embedded clause and raises to
its surface position.
(69) A certain boy appears e to have talked to a certain girl.

(70) A certain boy decided e to talk to a certain girl.

One may object to this explanation since an intervening coreferential pronoun

(though not anaphor) can salvage a multiply sluiced construction:
(71) A certain boy1 said he1 would talk to a certain girl.

(72) A certain boy told himself to talk to a certain girl.

*I forget which boy to which girl.
Since a pronoun can salvage this construction, one might ask why PRO cannot do
the same. Although I have no coherent explanation for the facts in (71) and (72), PRO
has been argued to be anaphoric and anaphors do not salvage multiple sluicing.
This is a good point at which to summarize the various proposals for the fol-
lowing three sentences:
(73) Three (putative) types of control into wh-infinitives
a. John1 decided where e to eat.
b. John1 decided where e to meet.
c. John1 knows how e to fly a 747.
Table 1 summarizes the identity of e in each of these sentences following Landau’s

approach, a traditional approach, and the current proposal.
Table 1. Types of Null Subjects in Control Infinitives

Type Landau Traditional Current proposal
(73) a. PRO1 PROarb tJohn
(73) b. PRO1+ PROarb tJohn
(73) c. PRO1+ PROarb progen
276 MICHAEL BARRIE
Thus, I argue for a movement approach to OC, where I follow Landau and
assume that OC includes both exhaustive control and partial control. First, let us
convince ourselves that multiple sluicing is permitted in uncontroversial partial
control environments:
(74) A certain traveler wanted to eat in a certain restaurant.

I forget which traveler in which restaurant.
(75) A certain traveler wanted to meet in a certain restaurant.

I forget which traveler in which restaurant.
Multiple sluicing is acceptable in both exhaustive control and partial control environ-
ments. Let us examine the results with wh-infinitivals:
(76) A certain traveler decided where to eat in a certain city.

(77) A certain traveler decided where to meet in a certain city.

Again, both are fine. But, consider the following:
(78) A certain dignitary knows how to introduce himself to a certain monarch.

I forget which dignitary to which monarch.
(79) A certain dignitary knows how to introduce oneself to a certain monarch.

*I forget which dignitary to which monarch.
Here, we see clear evidence that the matrix subject is not the controller as the
matrix subject is not a suitable referent for the anaphor oneself. It is also not the
controller in a partial control construction (as argued in Landau 1999, 2000) since
partial control is able to license multiply sluiced constructions and these construc-
tions are not. We see the same effect in the following pairs of sentences, though
the contrast is slightly less robust:
(80) A certain caterer knows where to pick up the cheese for a certain party.
I forget which caterer for which party.
(81) A certain connoisseur knows where to get good cheese in a certain city.
*I forget which connoisseur in which city.
Thus, we see that exhaustive control and partial control constructions support
multiple sluicing, but that constructions with generic control do not. This asym-
metry is correctly predicted if we assume that exhaustive control and partial
control involve movement of the controller from the embedded clause to the
superordinate clause (whether in standard control sentences or in wh-control) and
the generic control does not involve movement, but rather a progen as the subject of
the embedded wh-infinitival. In sum, then, generic control involves progen; there is
no movement of the matrix subject with generic control; and with OC, the matrix
subject moves up from embedded clause.
This proposal makes a strong prediction regarding constructions with multiple
generic arguments. If we assume that progen is bound by a generic operator higher up
in the same clause (Krifka et al. 1995), then adjuncts that also contain a progen are also
bound by the same operator. Thus, we must understand the null subjects of both clauses
to be coreferential. This appears to be the case. Consider the following example:
(82) John figured out how progen1 to eat sushi without progen1/*2 using a fork.
Here, the eater of the sushi must also be the nonuser of the fork. Furthermore, if
the adjunct is adjoined to the matrix clause, then progen cannot be bound by the
generic operator in the embedded infinitival. In such a situation, we should not be
able to understand progen to be coreferential with the null subject of the wh-infini-
tival. Again, this prediction is borne out.
(83) John1 figured out how progen2 to eat sushi without e1/*2/*3 asking for help.
In (83), the empty subject in the adjunct is coreferential with John and not with
progen in the wh-infinitival clause.
This subsection has accounted for the asymmetry between OC and NOC into
wh-infinitivals with respect to multiple sluicing. The analysis crucially relied on
certain assumptions regarding control. Specifically, I assumed that OC arises via
movement of the controller from the infinitival clause to the superordinate clause
and that empty the subject in NOC constructions is pro. Next, we account for the
asymmetry between OC and NOC with respect to island violations. In order to do
this, we first need to sharpen our view of the right edge of CP.
3.5 On the right edge of CP
The core proposal here is that the right edge of CP is structured as follows:
(84) FinP > WhP
This simplifies the selectional restrictions of the matrix verb. The matrix verb always
selects FinP with Fin0 [-Finite], regardless of whether there is an embedded wh-element
or not. The embedded infinitival in (85) and (86) is headed by FinP in both cases.
(85) Mary told John to wash the dishes.
(86) Mary told John when to wash the dishes.
Now, for OC into infinitival clauses, the subject raises from the embedded clause
to the matrix clause, using [Spec,FinP] as an intermediate landing site. The pro-
posed structure for (1), then, is as follows:
278 MICHAEL BARRIE
(87) Johni knows [FinP ti [WhP when [IP ti to wash the dishes]]].
We now have an account of the asymmetry in wh-raising out of wh-infinitivals.

Wh-raising out of wh-infinitivals with OC is not a possibility, since the [Spec,FinP]
escape hatch has already been used by the matrix subject.
(88) Whatj does Johni know [FinP ti [WhP when [IP ti wash tj ]]].
Wh-raising out of wh-infinitivals with generic control is possible, since the

[Spec,FinP] escape hatch has not been used.
(89) What kind of carj does John know [FinP tj [WhP how [IP progen to park tj]]].
The trace of what kind of car is now free to raise through the intermediate
[Spec,FinP] on its way up to the matrix clause.
This analysis may seem problematic for sentences with overt prepositional
complementizers and wh-infinitivals as shown in (90) (marginal for some speak-
ers, but available for others).
(90) %Jason decided where for everyone to sit.
Given the structure in (84), we would expect the prepositional complementizer for
to precede the wh-phrase, contrary to fact. I suggest that for is actually the K0 of a
KP, which is part of the extended DP domain. Evidence for this claim comes from
the fact that the sequence for DP behaves as a constituent in several regards.
Consider the following data.
(91) a. For John and for Mary to get along would be a miracle.
b. I would like very much for Susan and for Jack to wait outside.
c. For Fred – but not for Baxter – to lose the race would be surprising.
These data strongly suggest that the sequence for DP is a constituent. I suggest,
then, that the embedded subject in an infinitival should be represented as in (92).
The traditional analysis of for in Fin0 does not capture the constituency facts.
(92) [KP for [DP subject ]].
Thus, the word order facts in (90) are easily accounted for if we assume that for is
not a Fin0, but rather a K0 that selects its subject directly.
In sum, then, we conclude that FinP selects WhP, leaving FinP at the left edge
of the CP domain. We also conclude that a wh-element can use [Spec,FinP] as an
escape hatch, unless the matrix subject has already used it.
4. CONCLUSIONS
I have argued that a distinction must be maintained between OC and generic

control into wh-infinitivals based on various diagnostics for genericity and on
two asymmetries for these two types of control in these constructions: extrac-
tion across weak islands and multiple sluicing. I argued that the controller in OC
constructions raises from the embedded clause to the matrix clause. For generic
control, however, I suggested that the embedded null subject is represented by
progen and that the matrix subject originates in the matrix clause. The generaliza-
tions that were accounted for were as follows. First, we offered an account of why
why is not found in embedded infinitivals (7). Furthermore, we explained why
wh-island ‘violations’ are permitted with generic control (36–38), but not with
OC (39–41). Wh-infinitivals are headed by a reduced CP consisting only of FinP
> WhP. In OC, the subject raises from the embedded clause, using [Spec,FinP] as
an escape hatch thereby eliminating the use of this position as an escape hatch
for subsequent wh-movement out of embedded infinitival. In generic control, the
subject does not raise from the embedded clause. Thus, [Spec,FinP] is free to be
used as an escape hatch for wh-movement from the embedded clause to the matrix
clause.
JOHAN ROORYCK
CONTROL VIA SELECTION
1. INTRODUCTION
In the last five years, a vigorous debate has unfolded around control theory,
focusing on the movement theory of control (MTC) (Hornstein 1999, 2001;
Culicover and Jackendoff 2001; Landau 2003; Boeckx and Hornstein 2003;
Davies and Dubinsky 2004). The MTC invokes theoretical parsimony to reduce
control to raising, thereby eliminating the theoretically problematic empty
pronoun PRO (but see fn. 3 infra). The opponents of the MTC justifiably argue
that its partisans sacrifice empirical adequacy to theoretical elegance.
The purpose of this paper will be twofold. First, in line with the critics of the
MTC, I will show that there are four empirical generalizations, which cannot be
explained by the MTC. Second, I will briefly recapitulate the analysis of control
proposed in Rooryck 2001, which offers an explanation of the full variety of control
facts in infinitival complementation by proposing an analysis in terms of selection.
This analysis provides an explanation for the semantically fine-grained differences
between control verbs (promise vs. ask vs. offer), which have long been observed
(see Abraham 1982, 1983; Rudanko 1985, 1989; Ružička 1983a, b, Siebert-Ott
1983, 1985; Wegener 1989), but ignored in the majority of the generative work on
control – a gap that stretches back in time to Rosenbaum 1967.1
In this paper, I will limit myself to data involving subject and object control
of infinitival complements selected by the matrix control verb as in (1), where the
controller of PRO is located in the immediately superordinate clause:
(1) a. Evak said that Kimi remembered [PROi/*k to leave]
b. Evak said that Kimi forced/advised Suej [PRO*i/j/*i+j/*k to take the lead]
For an analysis of control in adjuncts within the framework developed here,

I refer the reader to Rooryck 2001 (Chap. 3).
2. FOUR EMPIRICAL GENERALIZATIONS
2.1 Generalization 1: Variable but local control verbs (VLC verbs)
There is a substantial set of verbs, illustrated in (2–4), that allow either their subject
or object to function as the controller, but not a superordinate argument.
281
W.D. Davies and S. Dubinsky(eds.), New Horizons in the Analysis of Control and Raising, 281–292.
© 2007 Springer.
282 JOHAN ROORYCK
(2) Subject-or-object control:

Evak said that Kimi offered Suej [PROi/j/i+j/*k to leave]
(3) Subject-to-Object control shift

a. Evak said that Kimi promised Suej [PROi/*j/i+j/*k to leave]
b. Evak said that Kimi promised Suej [PRO*i/j/*i+j/*kto be allowed to leave]
(4) Object-to-Subject control shift

a. Evak said that Kimi asked Suej [PRO*i/j/ i+j/*k to leave]
b. Evak said that Kimi asked Suej [PROi/*j/*i+j/*k to be allowed to leave]
It is important to emphasize the locality of the domain for control in these cases.
The domain where the controllers dwell is the next clause up, but no further
(Manzini 1983). As a result, these cases can simply not be reduced to non-oblig-
atory control (NOC), as NOC strictly involves cases of control which do not
obey such locality. Pace the OC/NOC distinction (see Vanden Wyngaerd 1994:
Chap. 8; O’Neill 1997), the locality of the controller domain is similar to that of
anaphoric contexts such as (5):
(5) a. Theyk said that Kimi talked to Suej about herselfi/j/*k
b. Theyk said that Kimi talked to Suej about themselvesi+j/*k
A case like (5b), where both Kim and Sue are the (joint) local antecedents of the
anaphor themselves, can be usefully compared to (2), where Kim and Sue are the
(joint) local antecedents of PRO.
2.2 Generalization 2: The semantic coherence of variable control verbs
The VLC verbs in (2–4) all belong to a coherent semantic class, namely verbs
which express a commitment to, or a request for, the temporally unrealized, future
transfer of a Theme.2 In both (6a) and (6b), the DP the apple is transferred from
a Source to a Goal. With a verb of giving, the actual transfer corresponds to the
reference time. In the case of VLC verbs, the transfer is delayed to an unspeci-
fied moment after the reference time of promising, offering or asking.
(6) a. Kim gave/granted Sue the apple (transfer at reference time)
b. Kim promised/offered/asked Sue (for) the apple. (transfer after reference time)
In control contexts, the DP Theme argument of VLC verbs is realized as an

infinitival complement, cf. (2–4). The distinction between ‘actual’ transfer (6a) and
‘delayed’ transfer is relevant for control properties in the following way. ‘Actual’
transfer verbs involve only object control, while ‘delayed’ transfer verbs display
variable but local control as already exemplified in (2–4):
CONTROL VIA SELECTION 283
(7) a. ‘Persephonei had granted himj PRO*i/j to keep his/*her wits in Hades’
adapted from http://homepage.mac.com/cparada/GML/Odysseus.html
b. ‘Mais [le temps et l’expérience]i luij ont donné d’ PRO*i/j avoir

but the time and the-experience her.DAT have given C.INF-have
confiance en elle, (…)’
confidence in her
‘But time and experience have given her (lit. to have) confidence in herself’
[French] www.koraawards.co.za/web/finalists.asp?c=10&l=fr&y=2000
c. (…) ils [les dieux] m’avaient accordé de vivre à une époque où la tâche qui
they [the gods] me-had granted COMP live in a time where the task REL
m’était échue (…)
me-was entrusted
‘The gods had granted it to me to live in a time where the task entrusted to me …’
(Marguerite Yourcenar, Mémoires d’Hadrien (1951), Gallimard Folio, p. 126)
An empirically adequate theory of control should be able to explain this correla-

tion between fine-grained semantic distinctions and control properties.
2.3 Generalization 3: The semantic coherence of each subtype of variable

control verbs
The observation made above in 2.2 can be refined. In fact, each of the control
properties illustrated in (2–4) (i.e. subject-or-object control, subject-to-object con-
trol shift, object-to-subject control shift), correlates with a semantically specific
and coherent subtype of transfer.
• Subject-or-object control involves verbs of ‘tentative transfer’:

These include propose, offer, suggest.
(Abraham 1982, 1983; Ružička 1983a, b; Siebert-Ott 1983, 1985 for German).
• Subject-to-object control shift verbs involve verbs referring to a commitment
to future transfer. These include promise, threaten, guarantee.
• Object-to-subject control shift verbs involve verbs of request. These include
ask, beg, request, and the like.
• Control shifts in (2) and (3) occur from Source to Goal in the context of
passives and modals (Ružička 1983a, b; Bennis and Hoekstra 1989:246).
These facts are consistently found in a range of languages including French,

Dutch, and German (cf. (8–10) ).
(8) a. Paulk a dit que Pierrei a offert/proposé à Jeanj de PROi/j/i+j/*k partir (S-or-O Fr.)
b. Paulk zei dat Pieti Janj had voorgesteld/aangeboden om PROi/j/i+j/*k weg te gaan
(S-or-O Du.)
Paul said that Pierre offered/proposed to Pierre to leave
(9) a. Paulk a dit que Pierrei a promis à Jeanj de PROi/i+j/*j/*k pouvoir partir (S-to-O Fr.)
Paulk zei dat Pieti Janj beloofd had om PROi/i+j/*j/*k weg te mogen gaan (S-to-O Du.)
Paul said that Pierre promised Pierre to leave
284 JOHAN ROORYCK
b. Paulk a dit que Pierrei a promis à Jeanj de PRO*i/j/*i+j/*k pouvoir partir (S-to-O Fr.)
Paulk zei dat Pieti Janj beloofd had PRO*i/j/*i+j/*k weg te mogen gaan (S-to-O Du.)
Paul said that Pierre promised Pierre to be allowed to leave
(10) a. Paulk a dit que Pierrei a demandé à Jeanj de PRO*i/j/i+j/*k partir (O-to-S Fr.)
Paulk zei dat Pieti Janj vroeg om PRO*i/j/i+j/*k weg te gaan (O-to-S Du.)
Paul said that Pierre asked Pierre to leave
b. Paulk a dit que Pierrei a demandé à Jeanj de PROi/*i+j/*j/*k pouvoir partir (O-to-S Fr.)
Paulk zei dat Pieti Janj vroeg om PROi/*i+j/*j/*k weg te mogen gaan (O-to-S Du.)
Paul said that Pierre asked Pierre to be allowed to leave
It has sometimes been suggested that these facts are not stable, and that
speakers exhibit variation with respect to the control properties of these
verbs. Some people exclusively or preferably have object control for a verb
such as propose, for instance. (See also Siebert-Ott 1983, 1985 and Wegener
1989 for similar subtle differences between German anbieten ‘propose’ and
vorschlagen ‘offer’.)
Some at the LSA Workshop attempted to downplay the relevance of these
facts, invoking this purported instability of the data to dismiss them as irrel-
evant for a syntactic analysis of control. However, the variation in judgments
does not make the generalization go away. Why would it be the case that pre-
cisely this semantically coherent set of verbs (i.e. verbs of ‘delayed’ transfer)
displays variation among speakers? Bona fide object control verbs of the force
type (force, induce, coerce, but also advise, tell) never display such variation and
are consistently object control for all speakers. An empirically adequate theory
of control should have an explanation for, as opposed to a dismissal of, this
generalization.
2.4 Generalization 4: Split control and progressive Aktionsart
All VLC verbs in (2–4) exhibit split control. In addition, a subset of object control
verbs expressing aspectually ‘progressive’ coercion (cf. (11) ) also display split con-
trol, cf. (12).
(11) a. Kim (*?slowly) forced/coerced Suej [PROj to accompany her]
b. Kimi (slowly) convinced/persuaded Suej [PROj to accompany her]
(12) a. *? Kim forced/coerced Suej [PROi+j to go to dinner together]
b. Kimi convinced/persuaded Suej [PROi+j to go to dinner together]
Once again, an adequate theory of control should be able to provide an explanation

for such fine-grained data.
2.5 Discussion in the light of the movement theory of control (MTC)
If MTC-proponents take generalizations 2.1–2.4 seriously, a number of rather

insuperable challenges arise. Regarding generalizations 2.1, 2.2, and 2.4, the
MTC has to explain how the semantics of a matrix verb is able to restrict land-
ing sites for (A-) movement. Regarding generalization 2.3, the MTC would have
to assume that passives and modals in the infinitival complement conspire with
the semantics of promise and ask type verbs to influence the landing sites for
(A-) movement.
One way to treat these data under the MTC would be to claim that the facts
of variable control are not part of Core Syntax, but belong to the poorly under-
stood conditions governing NOC. Only control conforming to all requirements
of obligatory control (OC) (Locality, Uniqueness of Antecedent, c-command,
precedence, cf. Williams 1980:209) is part of Narrow Syntax and subject to MTC.
The problem with this line of defense is that not all requirements on OC have the
same weight. Uniqueness of the antecedent is not even observed by local ana-
phors (cf. (5) ). And certainly the controllers of VLC verbs conform to the impor-
tant requirements of locality, c-command, and precedence. In fact, during the
LSA Workshop, Boeckx and Hornstein did adopt an NOC approach to VLC
verbs. The trouble for such an approach is that it predicts VLC verbs should be
able to pick nonlocal antecedents, i.e. antecedents beyond the immediately super-
ordinate clause, contrary to fact. Moreover, generalization 2.4 is left unexplained
under such an approach. ‘Nonprogressive’ force/coerce are OC, while ‘progressive’
convince/persuade are NOC. Sadly, the only way out for MTC proponents is to
discredit the data presented here as subject to idiosyncratic variation, and hence
irrelevant.
3. CONTROL VIA SELECTION (CVS)
The challenge posed by the generalizations 2.1–2.4 can be formulated as follows.

How can a fine-grained semantics of control verbs be related to the controllers
available in the matrix clause? I argue that the control properties of control verbs
can be derived via the mechanism of selection. In order to do so, I will make explicit
my assumptions about selection in general, and about selection of infinitival com-
plements in particular. I will show that the temporal selection of the infinitival
complement is crucial for identifying the set of possible controllers in the matrix
verb.
I take (s-)selection to be a Minimalist head–head relation (Chomsky 1995:173),
involving the identification of relevant features of the selecting head with similar
properties of the selected head. In (13), eat and apple share a set of semantic
features, while eat and beauty do not.
(13) John eats an apple/*?beauty

286 JOHAN ROORYCK
In the context of the selection of infinitival complements, the (s-)selection of

infinitives by a control verb determines their temporal interpretation (cf. also
Stowell 1982). More specifically, I assume that the temporal feature of infinitival
(C-)T is [unrealized]. The mechanism of s-selection ensures that the [unrealized] (C-)T
feature of the infinitive is identified by similar [unrealized] temporal properties
in the subevent structure of the control verb. The assumption of selection via
identification of identical features now forces me to make explicit what can be
considered an [unrealized] subevent in the subevent structure of the matrix verb,
such that it can be identified with the [unrealized] (C-)T feature of the infinitive.
The definition of [unrealized] subevent is twofold:
(14) Definition of temporally [unrealized] subevents

A subevent is temporally [unrealized] iff (a) or (b):
a. The (sub)event is not linked to a specific point on the time axis representing the
temporal development of the verb. Such a subevent can be temporally undefined
with respect to the past and future of the specific event time e1*, which is linked to
the temporal morphemes in T°. This type of subevent is represented as en.
b. The (sub)event includes either a point or an extended time period on the time axis
situated in the future with respect to the specific event time e1* linked to the temporal
morphemes in T°; the (sub)event refers to a ‘possible future’ or an indefinitely extended
future period, and is represented as en + 1.
The reason for this double definition of ‘unrealized subevent’ is related to the nature
of the notion ‘unrealized’ itself. It has often been recognized that infinitives have
both an atemporal and a modal ‘future/irrealis’ character. Bresnan (1972) notes that
infinitival complements refer to ‘something hypothetical or unrealized’. Guillaume
(1929) already defined the infinitival tense as a tense ‘in posse’: it expresses ‘potential’
time, or eventuality, which is opposed to tense ‘in esse’, a ‘real’, or finite time
reference linked to the time axis. Stowell (1982) makes a similar observation stating
that the tense of infinitives must be semantically interpreted as unrealized or as a
‘possible future’.
The first part of the definition, (14a) appeals to the atemporal nature of the
infinitive, unlinked to the time-axis. The second part of the definition, (14b) is
related to the ‘future/irrealis’ modality inherent in infinitives. As both aspects
are crucial for the semantics of the tense of infinitives, it is only proper that an
explicit theory of temporal selection of infinitives take these into account.
The semantic analysis of control verbs in terms of subevents can be rep-
resented using Pustejovsky’s (1988) representation for event structure. A verb
such as force then involves the subevent structure of a Transition (Tr), a cover
term for Vendler’s (1967) Accomplishments and Achievements. As illustrated in
(15b), this Transition involves two successive subevents, one for the initiation of
the force event and one for its resulting subevent. The initiating subevent e1 is
punctual, and corresponds to the reference time of force, indicated by an aster-
isk in (15b). The resulting subevent, represented here as en + 1, is [unrealized] in
the sense of (14b): it takes place at an undetermined moment after the initiating
subevent e1.
(15) a. Kim forced Sue to leave
b. subevent structure [Tr e1* act (Kim, (Sue, leave)) {en+1} leave (Sue) ]Tr
c. syntactic structure [V force] [CP (C-)T (…) [V leave] ]

| |
{en+1} {Tunrealized}
d. Plain English: At the event time e1*, Kim undertakes action with respect to Sue’s
leaving, resulting in a subevent at an undetermined moment after e1*, at which Sue
leaves.
The syntactic representation in (15c) illustrates selectional identification of the

unrealized subevent of the control verb {en + 1} and the infinitival {Tunrealized}. This
identification ensures that the infinitive is temporally interpreted as occurring in
the future with respect to subevent e1*, i.e. the time when force is applied.
At this point in the analysis, one last assumption is necessary to derive the strict
object control properties of force. I assume that identification of the infinitival
[unrealized] (C-)T feature with the [unrealized] subevents of the control verb makes
only the arguments included in that unrealized subevent available as controllers.
In other words, the infinitival (C-)T can only ‘see’ those arguments which are part
of the subevent it is identified with.
This assumption now derives object control for force: the unrealized subevent
{en + 1} with which the infinitival (C-)T is identified only contains the argument Sue,
so only this argument is available for control. The initiating, punctual subevent e1
contains both the Agent and Patient arguments Kim and Sue. However, since this
subevent is not [unrealized] in the sense of (14), the infinitival (C-)T is not identified
with it, and therefore the arguments of this subevent are unavailable for control.
Identification of the unrealized (C-)T and the unrealized subevent does not
in and of itself result in control. I adopt a version of Borer’s (1989) anaphoric
AGR3: infinitival [unrealized] (C-)T has anaphoric phi-features, which do the job
of PRO. Identification of [unrealized] (C-)T with the [unrealized] subevent entails
identification of (C-)T’s anaphoric phi-features with the phi-features of all and
only those argument(s) contained in the [unrealized] subevent. Control thus rides
piggyback on the temporal identification of the infinitive by the matrix verb. The
question is now how this perspective can give a full account of the intriguing
control properties of VLC verbs (2.1–2.3) and split control verbs (2.4).
3.1 Deriving the control properties of VLC verbs
3.1.1 The subevent structure of LC verbs
Recall from generalization 2.2 that VLC verbs are ‘delayed’ transfer verbs,
and from generalization 2.3 that every control pattern corresponds to a coherent
semantic class. This ‘delayed’ transfer can be represented in a Pustejovskian
representation as a future Transition, embedded as the second subevent of a
288 JOHAN ROORYCK
Transition. The subevent structures of resp. promise, ask, and offer all involve
a Transition headed by a punctual subevent related to the moment of promising,
asking, or offering, and a second Transition subevent expressing the temporally
undefined transfer. All subevents within this second Transition subevent are
temporally undefined and therefore [unrealized].
(16) a. Kim promised Sue an apple
b. Subevent structure (also threaten, guarantee in the relevant dialects/languages)

[Tre1* vol (Kim, Tr2) [Tr2 en+1 act (Kim) & move (Kim, apple)
& at (Sue, apple) ]Tr2 ]Tr
c. Plain English: At the event time e1*, Kim expresses volition with respect to a future
transfer, which is to be executed by Kim at an undetermined moment after e1*, at
which Sue comes into possession of the transferred ‘Theme’ at the final moment in
the development of the event.
(17) a. Kim asked Sue for an apple
b. Subevent structure (also beg, request, implore in the relevant dialects/languages)

[Tr e1* vol (Kim, Tr2) [Tr2 en+1 act (Sue) & move (Sue, apple)
& at (Kim, apple) ]Tr2 ]Tr
transfer, which is to be executed by Sue at an undetermined moment after e1*, at
which Kim comes into possession of the transferred ‘Theme’ at the final moment
in the development of the event.
(18) a. Kim offered Sue an apple
b. Subevent structure (also for propose, suggest in the relevant dialects/languages)

[Tr e1* vol (Kim, Tr2) [Tr2 en+1 act (Kim) & move (Kim, apple)
en+2 act (Sue, en+1) & at (Sue, apple) ]Tr2 ]Tr
transfer, which is to be executed by Kim at an undetermined moment after e1*, at
which Sue may act on whether Kim’s transfer comes about, and thus may come into
possession of the transferred ‘Theme’ at the final moment in the development of
the event.
Since the second ‘delayed’ Transition subevent involves both the Source and Goal
arguments of VLC control verbs, both these arguments occur in the [unrealized]
subevents of the second Transition. The assumption that identification of the
infinitival (C-)T with an unrealized subevent limits the set of controllers to those
embedded in that subevent now has as a consequence that both subject and object
of VLC verbs are potential controllers for the embedded infinitives. Identification
of [unrealized] (C-)T can take place with any of the subevents of the [unrealized]
Transition. This in turn will entail identification of (C-)T’s anaphoric phi-features
with only the argument(s) contained in the [unrealized] Transition. As a result,
control by either subject or object, or both (split control) is possible.
3.1.2 Control shifts
Note that this analysis only makes subject and object available as controllers.
It does not explain why the three verb classes behave differently with respect
to the direction of control shifts (subject-to-object, object-to-subject), or the
absence of control shifts with verbs of the offer class. There is however a further
generalization to make about control shifts. It appears that the differences among
VLC verbs involve a Thematic hierarchy. Among potential controllers, the most
Agentive argument with respect to the transfer is the ‘default’ controller. In the
case of promise and ask, the most Agentive argument is the Source initiating the
transfer. In the case of offer, the Source and Goal arguments can be viewed as
equally Agentive with respect to the transfer; that is, although the Source argu-
ment initiates the transfer, it may well be ‘resisted’ by the Goal. As a result, there
is no ‘default’ controller, and controller choice is a matter of context.
Control shifts occur when the implicit agent of the passive infinitive, or the instance
behind the permission/obligation inherent in the modal, is identified with the Agen-
tive argument of the control verb. The agentive argument then becomes unavailable
for control, and control shifts to the Goal argument of promise and ask.
Note that this analysis also offers some insight into the variation among
speakers with respect to the control properties of these verbs. The fact that for
some people a verb, such as suggest preferably has object control, can be related
to the fact that controller choice with VLC verbs depends on a rather subtle
thematic hierarchy, which may be influenced by individual factors of interpreta-
tion. The important thing is that both arguments of VLC verbs are available for
control for a majority of speakers, as has been repeatedly attested for various
languages in a regrettably ignored literature.
3.1.3 ‘Delayed’/unrealized transfers (offer/promise/ask) vs. actual

transfers (give, grant, confer)
Summarizing, all VLC verbs involve ‘delayed’/unrealized transfers, with both Source
and Goal situated in [unrealized] subevents, enabling these arguments to function
as controllers. Recall now that ‘actual’ transfer verbs involve strictly object control
(cf. (7) ), repeated here.
(7) a. ‘Persephonei had granted himj PRO*i/jto keep his/*her wits in Hades’
adapted from http://homepage.mac.com/cparada/GML/Odysseus.html
b. ‘Mais [le temps et l’ expérience]i luij ont donné d’ PRO*i/j avoir

but the time and the-experience her.DAT have given C.INF-have
confiance en elle, (..)’
confidence in her
‘But time and experience have given her (lit. to have) confidence in herself’
[French] www.koraawards.co.za/web/finalists.asp?c=10&l=fr&y=2000
Semantically, verbs of the give, grant, confer type only differ from VLC verbs in
that the actual transfer corresponds to the reference time (cf. (6) ), repeated here.
290 JOHAN ROORYCK
(6) a. Kim gave/granted Sue the apple (transfer at reference time)
b. Kim promised/offered/asked Sue (for) the apple. (transfer after reference time)
In the analysis presented here, the correlation between actual transfer and object
control is derived from the following representation:
(19) a. Mary gave the book to Bill
b. Subevent structure of give (also grant, confer)

[Tr e1* act (Mary) & move (Mary, book) eat (book, Bill)]Tr (Pustejovsky 1988:(2a) )
The subevent eat in (19b) should be viewed as [unrealized] in the sense of (14b),
i.e. a future subevent occurring after e1* that is temporally indefinitely extended.
In the event structure of ‘actual’ transfer verbs, only the Goal argument occurs
in an [unrealized] subevent, so only this argument is available as a controller, just
like for force in (15).
3.1.4 Split control and progressive Aktionsart
Under generalization 2.4, I mentioned that all VLC verbs in (2–4) exhibit split
control. In terms of the analysis presented here, this is easily understood, as both
the Source and Goals arguments of these verbs are embedded in an [unrealized] sub-
event with which the infinitival (C-)T is identified. As a result, these arguments can
also jointly function as controllers, i.e. antecedents of the anaphoric phi-features of
the infinitival (C-)T.
In addition, a subset of object control verbs expressing aspectually
‘progressive’ coercion also display split control, cf. (11–12), repeated here. Other
verbs of ‘progressive’ coercion include cajole, accustom, nag, browbeat.
(11) a. Kim (*?slowly) forced/coerced Suej [PROj to accompany her]
b. Kimi (slowly) convinced/persuaded Suej [PROj to accompany her]
(12) a *? Kim forced/coerced Suej [PROi+j to go to dinner together]
The analysis of control via selection provides an answer for this problem.
Verbs such as convince and persuade can be represented as Transitions, just
like strictly object control verbs such as force. The difference between ‘strictly
object control’ verbs of the force type and ‘object control plus split control’
verbs such as convince and persuade lies in the representation of the first,
initiating subevent. The ‘progressive’ nature of the coercion expressed in
these verbs is represented in (20c) by a Process subevent. By contrast, the
initiating subevent of force is represented as a punctual subevent as in (15b),
repeated here.
(20) a. Kimi (slowly) convinced/persuaded Suej [PROj to accompany her]
c. Event structure for convince/persuade (with subevent Pr for ‘Process’ subevent)

[Tr [Pr e1 … en ]Pr act (Kim, (Sue, leave) ) en+1 leave (Sue) ]Tr
(15) a. Kim forced Sue to leave
b. subevent structure [Tr e1* act (Kim, (Sue, leave) ) en+1 leave (Sue) ]Tr
The representation in (20c) can now be related to the explanation for control
proposed here. Process subevents can be viewed as [unrealized] in the sense
defined in (14a): their subparts are not linked to a specific, punctual point on
the time axis representing the temporal development of the verb. As a result, the
infinitival (C-)T can be identified with such a Process subevent, making both
the arguments contained in that subevent jointly available for control.
4. CONCLUSIONS
In this paper, I have presented four generalizations about control that are deeply
problematic for the MTC. I have proposed an alternative to the MTC, which
derives control via selection, arguing that a fine-grained semantics of the event
structure of VLC verbs can be related to their specific control properties. The
analysis makes use of no more than three theoretical assumptions regarding
selection, which are entirely compatible with Minimalist assumptions, and
allow control to be derived in infinitival complements via selection. Within this
analysis, the four empirical generalizations mentioned are derived without fur-
ther ado. The derivation of control via selection makes both a control module
and PRO superfluous. In this way, control via selection achieves results similar
to the MTC, but without the drawbacks and stipulations.
1
The neglect with which these studies have been treated does not augur well for the reception of my
own work on this issue. Such neglect should not be surprising. Sociologically, the horizon for the
debate on raising and control has been limited to New England ever since Rosenbaum 1967. The
domain of control is of course hardly unique in this regard. What is unique, however, is the curi-
ously stubborn persistence with which relevant empirical facts and generalizations have been sys-
tematically downplayed as marginal exceptions, and alternative analyses ignored. I will, however,
leave the study of the sociological order of the field to future historiographers of linguistics.
2
For reasons of expository simplicity, I will limit the discussion to verbs where the transfer takes place
at an undefined moment after the moment of promising, offering, or asking, i.e. verbs where the
event expressed by the infinitive takes place in an undefined future. It is important to point out that
VLC-verbs are in no way limited to verbs expressing an undefined future transfer. A verb like thank
for instance (and more generally the semantic set of verbs expressing thanks) functions like ask in
that they involve object-to-subject control shift (Hei thanked herj for PRO*i/j doing that/for PROi*j
being allowed to do that/for PROi*j being included). Thank differs from ask in that the transfer takes
place in a temporally undefined past moment before the thanking. As the analysis will show, it is
this lack of temporal anchoring that is crucial for the determination of the set of controllers.
292 JOHAN ROORYCK
3
In this context, it is interesting to observe that the elimination of PRO is touted with some fanfare
as one of the great achievements of the MTC. Proponents of the MTC fail to acknowledge that
the elimination of PRO was already a direct consequence of Borer’s (1989) analysis: if AGR
(alternatively the phi-features of T, in current Minimalist terms) is anaphoric, the empty subject of
the infinitive might just as well be pro.
IDAN LANDAU
MOVEMENT-RESISTANT ASPECTS OF CONTROL
1. INTRODUCTION
This paper is a contribution to the ongoing debate about the nature of obliga-
tory control (OC) in recent syntactic theorizing.* Although the debate has seen
many participants and approaches, I focus here on two opposing views in par-
ticular: The view represented in Landau 1999, 2000, 2003, 2004a, 2006, and the
view represented in Hornstein 1999, 2001, 2003 and Boeckx and Hornstein 2004.
According to the former, OC is formed by an abstract Agree relation, mediated by
functional heads. On the latter view (the movement theory of control (MTC)), OC
is formed by A-movement, an instance of raising.
Landau 2003 provides numerous independent arguments showing that the
MTC as developed in Hornstein 1999 fails to account for the most significant
generalizations about OC and non-obligatory control (NOC). In response,
Hornstein (2003) addresses a variety of empirical problems and offers novel,
sometimes ingenious, analyses for them. It also criticizes key aspects of the
proposal in Landau 1999. Boeckx and Hornstein 2004, in turn, claims that most
of the arguments in Landau 2003 against the MTC do not survive upon closer
scrutiny.
These are welcome developments. As the debate proceeds, theoretical
positions are continually sharpened, bringing to light more and more empirical
consequences.
The purpose of the present paper is threefold. It first examines Hornstein’s
(2003) treatment of various empirical challenges to the MTC and evaluates
how well his solutions meet these challenges. It next clarifies and defends cer-
tain aspects of Landau 1999 that are criticized (and misdescribed) in Horn-
stein 2003. Lastly, it shows why the reply in Boeckx and Hornstein 2004 fails
to address the essence of the critique expressed in Landau 2003. The general
organization of this paper more or less follows the discussion in Hornstein
2003 and then turns to some issues exclusively treated in Boeckx and Horn-
stein 2004.
One cautionary note to the reader: Because this is an ongoing debate with a
considerable history, it becomes increasingly cumbersome to elaborate this history
on each new installment. Therefore, much material – both data and arguments – that
is already well-represented in the previous stages of the debate has been omitted.
The discussion to follow thus presupposes some familiarity with the relevant
literature. It is my hope that the arguments below will be appreciated and judged
against the background of this knowledge.
293
© 2007 Springer.
294 IDAN LANDAU
2. THE MINIMAL DISTANCE PRINCIPLE (MDP)
An immediate consequence of reducing OC to movement is the emergence of locality

constraints. In particular, a very strict constraint – the minimal link condition
(MLC) – which restricts A-movement, is now taken to restrict OC. Hornstein
(1999) claims this to be the source of the observation that most transitive control
verbs display object control (the minimal distance principle (MDP)). The famous
case of promise is treated as a marked exception.
Landau 1999, 2000 (Chap. 5) and 2003 extensively discusses the serious
shortcomings of the MDP.1 There it is shown, most notably, that the strictly
configurational character of the MLC makes it look like an accident that the
same semantic verb classes, again and again across languages, ‘violate’ the
MDP. Rather than repeat those arguments here (see Jackendoff and Culicover
2003 for pertinent discussion), I will instead address some novel points made in
Hornstein 2003.
As evidence for the workings of the MDP/MLC, Hornstein mentions
Rosenbaum’s observation that object control verbs, when optionally lacking an
object, shift to subject control: ‘[W]hen the object is not generated (at least in overt
syntax), the subject can (and must) be the controller’(Hornstein 2003:29).
(1) a. John1 asked/begged/got Mary2 PRO*1/2 to leave.
b. John1 asked/begged/got PRO1 to leave.
This description is misleadingly selective. In fact, many verbs retain their object
control interpretation even in the absence of an overt object. English exhibits this
pattern with communication verbs, other languages (like Hebrew below) extend it
to many other instances.
(2) a. Mary1 said/shouted to John2 PRO*1/2 to leave.

b. Mary1 said/shouted PRO*1 to leave.
(3) a. Gil1 hirša le-Rina2 PRO*1/2 la’azov.

Gil permitted to-Rina PRO to-leave
‘Gil1 permitted Rina2 PRO*1/2 to leave’.
b. Gil1 hirša PRO*1 la’azov.

Gil permitted PRO to-leave
‘Gil1 permitted PRO*1 to leave’.
Consider now how an MLC-approach might account for the difference between
ask and say. It could be argued that the null object of ask is not syntactically
represented, whereas that of say is present in the syntax, i.e. a small pro. The
latter then imposes object control but the former, being invisible to the MLC,
does not. This account, again, does not speak to the obvious question, which
is why things are this way and not any other way. It also approaches circularity:
there is no independent evidence for the presence/absence of a pro object, other
MOVEMENT-RESISTANT ASPECTS OF CONTROL 295
than the absence/presence of control shift.2 Internally to the MTC, it is also not
clear how English could license the Last Resort element pro in contexts that risk
no grammatical violation.
By contrast, if controller choice is assigned to lexical and contextual factors
to begin with, the finding that different verbs show different patterns of control
shift is anything but surprising. Under this view, neither ask nor say licenses a pro
object; the difference is located in their fine-grained lexical structure.
How is this view more explanatory than the MLC view? First, it opens up the
way for a close investigation of the lexical semantics involved in such matters.
I will not attempt here any elaborate investigation, but simply sketch the outlines
of one possible approach. Notice that (1b) has the following paraphrase, with the
italicized material tacitly ‘filled in’.
(4) John1 asked X for [X’s permission to him1 PRO1 to leave].
In this semantic representation, X controls the external argument of permission

and John the internal one. Permission being an object control predicate, PRO gets
to be controlled by the implicit him, identical to the matrix subject.
What would be the analogue for (2b)? Conceivably, the following paraphrase.
(5) Mary1 said to X [her1 instruction to X*1 PRO*1 to leave].
Here, the understood noun, recovered from the meaning of say, is instruction (or
perhaps order). Interestingly, the external argument of this noun is controlled by
the matrix subject Mary, not by the matrix implicit goal X, which controls the
internal argument of instruction. For this reason, the latter cannot also be
co-indexed with Mary or else condition B will be violated. Since the internal
argument of the noun controls PRO, we obtain the result that PRO must be disjoint
from the matrix subject.
The above reasoning is nothing more than intuition spelled out loud, and a
formal account of these data must go much beyond such intuitions.3 Nevertheless,
the purpose of this exercise is simply to draw attention to the kind of considerations
that are likely to provide fruitful insights in the study of control shift. In this regard,
the MLC approach is rather limited. Notice that even if implicit arguments are
pro-like elements (but see the next point), the MLC fails to explain their effect on
control. Whereas (4) conforms to the MLC, (5) does not – Mary controls the NP
subject across the matrix goal.
In Boeckx and Hornstein 2004 it is conceded that implicit control reduces, in
their theory, to movement of a null pro. The alternative, they maintain, ‘would
force us to say that lexical control exists along with syntactic control . . . deciding
which control takes over when is no trivial matter, and certainly weakens the
theory . . . so our account leads to a more restrictive theory’ (Boeckx and
Hornstein 2004:439). This position indicates a willingness to sacrifice well-known
facts on the altar of ‘restrictiveness’. It is a fact that control shift is lexically
governed; it is a fact that the control properties of ask and say differ when their
object is unexpressed (‘no trivial matter’, and thus, one would think, not a matter
296 IDAN LANDAU
to be ignored); and it is a fact that a uniform pro-analysis of implicit control fails

to explain this contrast. All these facts, apparently, do not suffice to compromise
the a priori imperative of restrictiveness.
The second aspect in which the lexical/contextual account of control shift is
superior to the MLC is more straightforward. Recall that Hornstein must posit
a pro object in (2b) to explain the blocking of subject control. Null pronouns,
unlike implicit arguments, are visible to condition A; in particular, they can bind
anaphors (Rizzi 1986). We may ask whether the null object of say exhibits this
property. The answer is no; unlike an overt pronoun, the null goal argument of
say cannot bind an anaphor.
(6) a. We all heard the amazing story about [Bill and Kevin]1.
John said to them1 at each other’s1 parties to take off their clothes.
b. We all heard the amazing story about [Bill and Kevin]1.

*John said to pro1 at each other’s1 parties to take off their clothes.
This contrast casts further doubt on the idea that object control in (2b)/(3b) is mediated
via a syntactic object pro. But then, if something other than the MLC explains
object control in (2b)/(3b) (in fact, the MLC wrongly predicts subject control), that
something might as well explain object control in the canonical examples (1a)/(2a).
Thus, neither canonical nor shifted control shows evidence for the MLC.
Boeckx and Hornstein (2004) argue that Rizzi’s binding/control asymmetry can
be explained without sacrificing the assumption that implicit control is mediated
by pro. Following Hornstein (2001), they assume that binding itself is subsumed
under movement, the anaphor being a ‘lexicalized copy’of its antecedent. The
following condition is then proposed to explain why pro can control but not bind:
‘An anaphor cannot be lexicalized if its antecedent is not’(Boeckx and Hornstein
2004:439). Hence, a null element (like pro) can antecede other null elements
(unpronounced subject copies in OC) but not lexical ones (reflexives in binding).
Unfortunately, the proposed condition is disconfirmed in NOC environments,
where PRO (or pro, according to Boeckx and Hornstein) happily binds a reflexive.
Note that PRO in (7a, b) has no antecedent in its clause.
(7) a. [PRO praising oneself/myself] wouldn’t be polite.
b. John was furious. [PRO to get himself a new CD-player like the one stolen]
would cost a fortune.
The combination of (6b) and (7) presents a paradox to Boeckx and Hornstein;
conceivably, they might introduce a novel distinction between pros that can, and
pros that cannot bind, restricting the latter to nonsubject positions. This,
however, would merely replicate the already-existing distinction between implicit
arguments (available in both internal and external positions) and PRO (restricted
to subject positions).4
More generally, it appears that that the postulated pro controller in implicit
control constructions creates more problems than it solves: It does not block control
shift where it should, and it does not bind reflexives where it could. Its motivation
is purely theory-internal – to secure the claim that control is movement.
Finally, a word about the appeal to ‘markedness’ in Hornstein 1999, 2003
to explain the exceptional behavior of promise, vow, commit, etc. Hornstein
emphasizes that this view makes sense of the late acquisition of such construc-
tions by children, first documented in Chomsky 1969. This may be so. What is
not warranted by the acquisition evidence is the following statement: ‘[A]ny view
that “regularizes” the properties of promise, for example, by claiming that it falls
under a broader generalization in terms of which its behavior is grammatically
impeccable, cannot account for why it is acquired late’(Hornstein 2003:34).
First, note that this reasoning is dubious on general grounds. There is no
obvious entailment from late acquisition to markedness. A host of grammatical
constructions emerge relatively late in acquisition – passive, purpose clauses,
parasitic gaps, etc. – none of which is marked in the relevant sense (i.e. an
exception to a UG principle). Markedness is but one, probably minor, factor in
fixing the time course of grammatical development.
Second, one cannot argue for a particular syntactic analysis on the basis of an
isolated developmental fact. The rich literature on the acquisition of control has
uncovered many facts that do not follow from the MTC. For example, there is a
systematic delay in the acquisition of OC into adjuncts compared to the acquisi-
tion of OC into complements (McDaniel and Cairns 1990; McDaniel et al. 1991;
Cairns et al. 1994). Applying the same logic, one could argue that this delay ‘can-
not’ be explained by any analysis that reduces both types of OC to the same mecha-
nism (e.g. A-movement). Clearly, the conclusion is unwarranted; and so is Hornstein’s
conclusion with respect to potential non-markedness accounts of the promise-delay.
Other developmental facts are equally puzzling. Young children (3-year-olds)
err in control assignment with object control verbs as frequently as they succeed
in subject control verbs. Thus, although incorrect object control assignment to
promise persists to age 7 or so, tell and remind elicit considerable incorrect subject
control at age 3, which later dies out (Sherman and Lust 1993). This progression
from more to fewer MDP-violations is anything but expected on the markedness
theory. Finally, parallels between children’s preferences for controllers of PRO
and antecedents of pronouns in comparable structures suggest that non-movement
generalizations are operative in this domain (Cairns et al. 1994). The place and
relative weight of the promise-delay effect within this rich empirical array still
waits to be determined.5
3. SPLIT CONTROL
Hornstein (1999) takes the ban on split control to be an exceptionless criterion for
OC. This was challenged in Landau 1999, 2000, but in Hornstein 2003: (fn. 13) this
position is reiterated, pointing to the controversial status of the English examples.
There is little doubt that English is pretty resistant to split control in OC – most
OC verbs do not allow split control. But some do, as (8a) from Koster and May
(1982:96) shows. The German example (8b) is from Wurmbrand 2001 and the
298 IDAN LANDAU
Hebrew example (8c) is my own. Note that the latter involves a verb of ‘strong’
influence.
(8) a. John proposed to Mary to help each other.
b. Ich habe ihm angeboten einander zu helfen.

I have him offered each-other to help
‘I offered him to help each other’.
c. Gil šixnea et Rina le’hacig et acmam bli buša.

Gil persuaded ACC Rina to-present ACC themselves without shame
‘Gil persuaded Rina to present themselves without shame’.
Even if some (or most) speakers reject split control in some (or most) contexts,
the fact that the possibility is real for others is a cause for worry. In fact, the very
variability observed is a cause for worry for theories in which controller choice
is wired into the syntax. In particular, it is virtually impossible to imagine how
split control can arise through raising, given that two distinct DP chains cannot
share their tail position. My own Agree-based analysis does not fare much better.6
Rather, the point here is that split control is a problem for everybody, and cannot
be simply wished away.
4. PARTIAL CONTROL (PC)
Section 1.6.4 in Hornstein 2003 is dedicated to a close critique of the analysis of

OC proposed in Landau 1999, mostly directed at the technical implementation
of the analysis. Unfortunately, these criticisms overlook significant revisions
made in Landau 2000, revisions which resolve many of the issues raised. Even so,
Hornstein misinterprets several aspects of Landau 1999, so some clarifications
are in order.
4.1 Semantic plurality
Consider the structure of a PC configuration (F = matrix T/v).
(9) a. The chair1 preferred [PRO1+ to gather in the conference room].

b. [CP DP .. F .. [CP [T-Agr+C0 T] [TP PRO [I’ tT-Agr [VP tPRO ..]]]]]
Agree2 Agree3 Move Agree1
Agree1 establishes embedded agreement (followed by raising of PRO to [Spec,TP])

and Agree2 matrix (subject or object) agreement. A PC infinitive is tensed, hence
headed by C0T with an uninterpretable Tense feature. The embedded T-Agr raises
to check off this feature, thereby reaching an edge position in which it is visible to
matrix operations. Agree3 applies to establish the link generating OC.7
Crucially, all valuation of normal φ-features flows from the single non-anaphoric
occurrence of these features – the controller DP. Landau 2000 assumes that the
special feature [SP] (semantic plurality), on which DP and PRO may differ, is
inherently specified on both; just like lexical nouns are inherently specified as [+SP]
(committee) or [−SP] (chair), so can PRO be specified for either value. This departs
from Landau (1999), where the [SP] value of PRO is contextually acquired.
At once, the issues of Inclusiveness and the alleged ‘copying’ capacity of Agree
raised by Hornstein (2003:39) disappear. Note that fn. 71 repeats the allegation that
my conception of Agree is non-standard in that it involves assignment, not check-
ing. Even supposing this were a terrible flaw (which I doubt), the simple fact is that
my conception of Agree is directly drawn from Chomsky 2000; namely, matching,
valuation, and deletion. To the extent that valuation does not violate Inclusiveness,
nothing in (9b) does either.
Further comments on the interaction of syntactic and semantic plurality in PC
reveal a misunderstanding. For example, Hornstein writes that ‘in PC cases, there is
evidence both that the PRO is semantically plural and that it is syntactically singular.
This is a problem for Landau’s (1999) proposal’ (2003:44). In fact, not only is this not
a problem – it is precisely the PC-effect that I had discovered! The key point about
PRO in PC is that it behaves like a group name (e.g. committee), which is semantically
plural but syntactically singular. Thus, the contrast in (10), in American English,
parallels the one in (11).
(10) a. The committee gathered before the vote.
b. *The committee consulted each other before the vote.
(11) a. The chair preferred to gather before the vote.
b. *The chair preferred to consult each other before the vote.
Furthermore, in British English, where (10b) is fine – so is (11b). This is because

the reciprocal phrase is licensed by syntactic plurality in American English,
but merely semantic plurality in British English. Such variation is found across
languages, dialects and even idiolects. This point is worth emphasizing – and it fact
it was: ‘When testing the predictions of [the PC-generalization] via the licensing
of “plural” morphemes in PC-complements, one should be careful to establish
independently the specific agreement requirement of every relevant morpheme’
(Landau 1999:64; 2000:50).
Despite this careful statement, Hornstein (2003:fn. 78) challenges my description
of the facts on the basis of the failure of PC in (12a).
(12) a. *John wants/decided to be similar/sing alike/be mutually supporting.
b. *John is similar/sings alike/is mutually supporting with Bill.
Hornstein notes that all these predicates also resist the commitative construction
(12b), suggesting that ‘the relevant generalization behind PC is that certain verbs
can select embedded commitatives.’
300 IDAN LANDAU
Hornstein failed to test whether the predicates in (12) are licensed by semantic
or syntactic plurality. In fact, they require the latter (in American English).
(13) a. The members are mutually supporting.
b. *The committee is/are mutually supporting.
(14) a. They sing alike.
b. *This team sing(s) alike.
(15) a. John and Mary are similar.
b. *This couple is/are similar.
This observation is enough to rule out (12a), on a par with (11b). Furthermore, a
commitative paraphrase is not necessary for PC.
(16) a. The chair voted/decided to disperse until next week.
b. *The chair dispersed with the rest of us.
I conclude that none of the facts presented by Hornstein pose any challenge to the
original statement of the PC-generalization. In fact, they are fully expected once
the precise predictions of the generalization are understood.8
4.2 Minimality
Next, Hornstein argues that (9b) violates minimality – F cannot Agree with T-Agr
(or with PRO, in EC) across the closer DP, namely the controller. This type of
Agree is thus ‘specially suited to the phenomenon at hand’. Strikingly, Horn-
stein fails to mention the explicit answer to this problem given in Landau 1999
(84–86); 2000 (70–72), where the structure of the MLC-violation in (9b) is shown
to be formally equivalent to the structure of analogous violations of superior-
ity in multiple-wh languages, as well as other dependencies discussed in Richards
1997: An attracting head first Agrees with a close target, then with a remote one.
To accommodate these phenomena, Richards developed his Principle of Mini-
mal Compliance (PMC), which I adopted for the OC dependencies. Whether or
not one agrees with this account, it should at least be addressed.
4.3 Asymmetry of Agree
The next point Hornstein makes is that the ‘chaining’ of Agree relations depicted
in (9b) does not guarantee control. There seem to be two objections here, the first of
which rests, again, on misattribution. For PRO to be controlled by DP, Hornstein
argues, it is not enough that Agree is a transitive relation: ‘Landau must also be
assuming that AGREE is symmetric’ (pp. 39–40). However, Agree is asymmetric, in
virtue of the distinction between probe and goal. ‘[I]f AGREE is not symmetric, then
we cannot establish a control relation through the series of AGREE operations….
The reason is that we cannot deduce that DP agrees with PRO from the fact that
other agreement operations took place’ (p. 40).
This objection rests on Hornstein’s own equivocation between the terms AGREE
and Agree, revealed in the above quote; it finds no support in Landau 1999. As
discussed above, I adopted the Chomskyan view of Agree (=AGREE), based on
valuation, hence intrinsically asymmetric. The error in the objection rests on the
assumption that it is AGREE itself that must be symmetric (to guarantee control),
rather than its output – agreement. Indeed, Agree (T, DP) is asymmetric (DP values
T’s φ-features, T values DP’s case feature), but crucially, its output is symmetric: We
say that T and DP agree with each other, simply because they come to share their
features. When interested in agreement (as opposed to Agree), we care about a
symmetric outcome of an asymmetric operation.
Consider an analogy from reference assignment in discourse.
(17) John1 entered the room. He1 looked around. He1 sat down. Then he1 poured himself1
a glass of beer.
Analyzing the process by which reference is assigned in this discourse, we want to

say that John fixes the reference of the first he, which fixes the reference of the
second he, which fixes the reference of the third he, which fixes the reference of
himself. These are all asymmetric relations. By contrast, focusing on the resulting
interpretation, we may simply say that the five nominal expressions are co-indexed,
a symmetric relation. I see no problem in this description, here or in (9b).
The chain of Agree operations in (9b) achieves two goals. First, it matches the
φ-features of PRO with those of the controller DP. Second, it co-indexes those two
elements, establishing variable binding. Hornstein states that co-indexing can only
be achieved by Move, not by Agree, which appears to be a stipulation. Furthermore,
the claim (in fn. 72) that co-indexing does not capture the tight referential depend-
ency in OC (presumably, de se interpretation) is a double-edged sword; as previously
shown in Landau 1999 (41–42); 2000 (29–30); 2003, neither does Hornstein’s ‘complex
monadic predicate’ distinguish variable binding from de se.9
4.4 Null case
Hornstein points out that my theory of OC is not obviously consistent with the
theory of null case (Martin 1996, 2001). He then attempts to reconcile the two
theories, again attributing to me unfounded stipulations. In fact, Landau 1999, 2000
says nothing about the distribution of PRO. Save for one unfortunate typo, the posi-
tion is summarized at the outset: ‘Most probably, PRO occurs only in the subject
position of nonfinite clauses. This study has virtually nothing to add to this
observation… perhaps the distribution of PRO is an irreducible fact of UG’
(Landau 1999:11; 2000:2).
I never advocated the null case theory, and I think that there are overwhelming
reasons to reject it. Hornstein (2003) discusses several good arguments, and others
302 IDAN LANDAU
can be adduced (see Baltin and Barrett 2002). Most notably, Martin’s notion of
[±tense] is both semantically incoherent and fails to demarcate control from
raising complements. My own view is that PRO is case-marked just like any other
DP (see sections 6 and 8 below, and Landau 2006). Ample evidence for this comes
from languages exhibiting subject-oriented case concord in controlled infinitives
(Russian, Icelandic) or subjunctives (Greek, Romanian, Hebrew, Persian, etc.).
These issues are extensively discussed in Landau 2004a, 2006, where I also provide
an alternative account of PRO’s distribution, divorced from case theory altogether.
4.5 Gerunds and tense
Next, Hornstein (example (78) ) notes that gerundive complements tolerate PC, a
‘problem’ for my analysis. The facts are well known, though; I have noted similar
examples in the past (Landau 1999:58; 2000:45). So what is the problem? First,
Stowell (1982) argues against gerunds being [+tense] (the condition for PC).
Second, gerunds are generally assumed to be TPs, not CPs, so they should not
afford the PC mechanism, which is crucially linked to C.
As to the first point, we have to conclude that Stowell was wrong. While some
gerunds are untensed, others are tensed; cf. the following contrast.
(18) a. *Yesterday, John avoided leaving tomorrow.
b. Yesterday, John preferred leaving tomorrow.
Stowell argued that it is precisely this variability in temporal interpretation that

attests to the lack of intrinsic tense in gerunds; supposedly, their tense is specified
by the matrix verb. The problem is that the same logic can be applied to minimal
infinitival pairs, potentially voiding Stowell’s claim that they do contain tense.
(19) a. *Yesterday, John managed to solve the problem tomorrow.
b. Yesterday, John hoped to solve the problem tomorrow.
The alternative (see Landau 1999, 2000) is very simple. Taking the temporal
mismatch as a reliable diagnostic for semantic tense, we conclude that the
complements in (18a) and (19a) are untensed whereas those in (18b) and (19b)
are tensed. Notice that this distinction cuts across the gerund/infinitive categories.
This is probably correct. Crosslinguistically, gerunds are significantly rarer than
infinitives. Many languages, in fact, express (18a, b) with infinitival complements;
crucially, their tense properties remain constant. Conversely, complements to
English verbs that select either form (prefer, hate, start, etc.) are consistently tensed
or untensed, whether realized as a gerund or as an infinitive.
As to the second point – is it a real problem to assume that gerunds are CPs?
This is less clear today than it was 20 years ago. First, the lack of interrogative
gerunds could be handled at the featural level, not necessarily the category level.
For-infinitives are similarly never interrogative, yet we do not conclude that they are
bare TPs. Second, it is quite possible that some gerunds do contain complementiz-
ers. Landau 2002 argues that from is a negative complementizer heading gerundive
complements of negative verbs (refrain, prevent, etc.). All in all, the evidence against
gerunds being CPs is very weak, at best. But if gerunds may be tensed CPs, nothing
blocks them from licensing PC.
4.6 T-to-C
Hornstein suggests (p. 42) that the mechanism of raising T-Agr to C, posited
in PC complements, may overgenerate to EC-complements. The reason is that
EC-complements display strict tense dependency (in fact, identity) between
the matrix and the embedded tenses. Presumably T would have to raise to C to
impose this selectional restriction (without violating locality of selection). But
then the contrast with PC complements is lost.
The point is correct, but harmless. Indeed, Landau 1999, 2000 left the realization
of tense selection unstated. This lacuna is filled in Landau 2004a, where a fully
explicit theory of clausal complementation is laid out. Still, the problem could be
easily solved within the framework of Landau 1999. EC-verbs could select the feature
[–Tense] on the C head of their complements, which in turn is matched against the
embedded T. Alternatively, the lack of [Tense] on C in these complements could
allow for a direct selectional relation between the matrix verb and the embedded T,
the idea being that selection, like checking, is obstructed only by potential interven-
ers. Either way, T-to-C is not needed in EC-complements, and the contrast with PC-
complements is preserved. That said, Landau (2004a) recasts the EC/PC distinction
in terms of pure Agree, without movement, although the key contrast – the nature
of [Tense] on C – is retained.
Lastly, Hornstein argues (p. 42) that there is little evidence for overt embedded
T-to-C in English, so my analysis of PC in (9b) is dubious. It is worth pointing
out that there is likewise little evidence for overt NP-raising in John likes
himself, even though NP-raising from the object position allegedly occurs in this
sentence, according to Hornstein (2001). More seriously, few would endorse such a
literal, ‘phonetic’ view of ‘evidence for movement’. Both my analysis and the
MTC are couched in a richly theoretical framework, where quite a few degrees
of freedom (e.g. null morphemes, vacuous movement, late insertion) separate
phonetic strings from abstract syntactic trees. Evidence for movement could consist
in phonetic rearrangement of terminals, but surely does not have to.
4.7 A meaning postulate for PC
In place of my syntactic account of PC, Hornstein proposes a lexical account. A meaning

postulate could be built in the lexical entry of PC verbs (like want), specifying that
the subject of their nonfinite complement may refer to either the controller, or a
group containing the controller and some contextually specified others.
Hornstein admits that this is ‘a very uninteresting account’, but one that works.
Yet it is important to realize what is being missed by this account. The meaning
304 IDAN LANDAU
postulate approach leaves three major puzzles unanswered. First, why is it that
the embedded tense is relevant at all to PC – rather than, say, the transitivity of the
embedded predicate, the animacy of its object, the social status of the subject, and
so on? Second, why is it that presence, rather than absence, of tense, licenses PC? In
other words, why does decide license PC and manage force EC, and not the other
way round? Third, why is it that no raising predicates license PC? Evidently, mean-
ing postulates are too unrestricted; although they can construct any list we design
them to, they provide virtually no insight into the underlying generalizations. By
contrast, the syntactic account crucially identifies the presence of embedded tense
(and PRO) as the vehicle of PC, making sense of the observed generalization.10
Hornstein notes that meaning postulates can only be imposed on arguments,
predicting lack of PC in adjuncts. The observation is correct, as can be seen in
right-adjoined nonfinite adjuncts headed by before/after/while/without. Hornstein
also includes rationale clauses in this set, unfortunately so, since these form a
category of their own. Despite the OC characteristics displayed in Hornstein’s
example (85), rationale clauses also show NOC characteristics, a fact known
since Williams’ (1974) famous example Grass is green (in order) to promote
photosynthesis. I refer the reader to Landau 1999 (206–211); 2000 (179–183) where
the complexities of rationale clauses are addressed.
The lack of PC in temporal adjuncts, which are clearly tensed, does not jeopard-
ize the PC-generalization. To begin with, Landau 1999, 2000 set adjunct control
apart from complement control. Only the latter is mediated by Agree, an opera-
tion which cannot penetrate islands (like adjuncts). My own view is that right-
adjoined adjuncts are construed as predicates, following Williams 1992. Being
directly predicated of a singular subject, they can support PC no more than any
secondary predicate can.
(20) a. *John called Mary before meeting in the restaurant.
b. *John called Mary together/while together/as a team/extremely polarized.
Thus, both the OC character of right-adjoined temporal adjuncts and the lack
of PC follow straightforwardly from predication.
5. NON-OBLIGATORY CONTROL
The previous section used the careful label ‘right-adjoined temporal adjuncts’, rather
than just adjuncts, in order to emphasize that not all adjuncts are alike. Rationale
clauses, as mentioned above, do not obviously show OC. Moreover, left-adjoined
temporal adjuncts support NOC (with subtle differences between participial and
gerundive adjuncts), a fact documented at length in Bresnan 1982; Williams 1992;
Kawasaki 1993; Landau 1999, 2000; and Lyngfelt 1999. These facts were brought
up again in Landau’s 2003 critique of Hornstein 1999.
The one concession Hornstein makes is to accept that PRO in NOC may bear
more resemblance to a logophor than to a pronoun. He notes (p. 51) that the
implications of this distinction are not evident, however, in light of the uncertainty
surrounding the syntax and semantics of logophors. Those comments are well taken.
I also agree with fn. 98, where it is said that Landau 1999 was not explicit about
the ‘competition’ between OC and NOC. In particular, the question why OC holds
whenever it can was not addressed. I agree that some economy metric is needed: Try
to establish control syntactically (by Agree/Move) before you resort to pragmatics
(logophoric/pronominal coreference).
6. CASE PERCOLATION
Landau 2003 pointed out that languages with case concord reveal a crucial
contrast between raising and control constructions. When the embedded
predicate assigns inherent (quirky) case to its subject, this case shows up on
the matrix subject in raising, but not on the matrix controller in control. The
latter bears the regular local case (e.g. nominative), while PRO bears the quirky
case (revealed on an agreeing element, like a secondary predicate, or floating
quantifier). Icelandic, for example, displays the following pattern (Sigurðsson
1991), illustrated below with dative quirky case (FQ = floating quantifier).
(21) Icelandic
a. Raising
DP1.DAT …V … [ t1 … V … FQ.DAT]
b. Control
DP1.NOM …V … [ PRO1 … V … FQ.DAT]
Icelandic is not unique; Russian, Hungarian, and Greek behave similarly (see
Landau 2004a and the references therein).11 This pattern strongly suggests that
one chain is involved in raising and two in control, contra the reductionist
analysis.
Hornstein (2003:(41b) ) brings up interesting data that challenge this picture.
In Chilean Spanish, the quirky dative case licensed by the embedded (psychologi-
cal) predicate may show up on the controller.12
(22) a. Marta le quiere gustar a Juan.

Marta CL.DAT wants to-please to Juan
‘Marta wants for Juan to like her’ (Marta wants to be liked by Juan).
(22) b. A Juan le quiere gustar Marta.

to Juan CL.DAT wants to-please Marta
‘Juan wants to like Marta’
In (22b), a Juan receives the two experiencer θ-roles, but crucially, its dative case
is assigned by the embedded verb gustar (querer does not assign dative case).
Hornstein notes that this pattern is the expected one under the ‘control = raising’
view, admitting that the Icelandic pattern remains a problem (fn. 32).
306 IDAN LANDAU
The actual force of this (admittedly intriguing) datum, however, is questionable.

Gonzalez (1990) makes clear that this type of ‘unusual inversion’ is severely
restricted, ‘occurring only with the Equi-Clause-Reduction predicates querer
‘want’ and tratar ‘try’ (p. 101). In fact, it is most common with querer, while some
speakers also accept it with tratar. Other predicates disallow ‘dative percolation’.
(23) a. A Marta le trataron/*desearon/*intentaron de gustar los gatos,

to Marta CL.DAT tried/ *desired/ *tried of to-please the cats
pero le produjeron alergia.
but CL.DAT produced allergy
‘Marta tried/*desired/*tried to like cats, but they produced allergy on her’
b. A Juan le quiere gustar Marta.

to Juan CL.DAT wants to-please Marta
‘Juan wants to like Marta’.
Furthermore, Bošković (1994:fn. 35) notes that the status of the judgments in (22)
is quite contentious; some Chilean Spanish speakers reject (22b), while others find
no semantic difference between (22a,b). Indeed, Gonzalez notes that when the dative
argument is a clitic, the sentence becomes ambiguous. Given that Spanish allows
postverbal subjects, it is not clear why the first reading of (24) is missing from (22b).
(24) Te quiero gustar.

CL.DAT want.1SG to-please
‘I want you to like me’.
‘You want to like me’.
The puzzle for Hornstein’s analysis is why this construction is so limited in

distribution (at most two restructuring verbs, for some speakers in some dialects).
Given the purely syntactic mechanism of A-movement, nothing seems to block
the raising of a quirky dative DP to the external argument position of any matrix
verb. This option is not even expected to be restricted to restructuring verbs, as all
OC verbs trigger raising in the MTC.
A further puzzle, already recognized by Gonzalez (1990:fn. 7), is the non-iterative
nature of dative percolation.
(25) *A Marta le quieren tratar de gustar los gatos.

to Marta CL.DAT wanted to-try of to-please the cats
‘Marta wants to try to like cats’.
As the English translation indicates, OC dependencies can be chained; so can

A-movements. If A-movement derives (22b), then the fact that it cannot apply
successive-cyclically to derive (25) is unexpected.
These observations suffice to undermine the claim that (22b) demonstrates
movement into a θ-position. Still, we may ask how such examples are derived.
I will offer two possibilities, leaving the choice between them open.
A first possibility is to treat querer as a modal of sorts, along the lines suggested
in Wurmbrand 2001 for wollen ‘want’ in German. Wurmbrand observes that the
verb want in many languages exhibits ‘quasi-modal’ properties. Modals are raising
predicates, that assign no θ-role to their subject. The semantic import of the modal
– obligation, necessity, permission, etc. – is rooted in lexical entailments interact-
ing with context, not in θ-theory. If indeed querer is a modal element, the alterna-
tion in (22) is of a familiar sort, on a par with other modal alternations.
(26) a. John may visit Mary.
b. Mary may be visited by John.
c. There may be no more than one visit per week.
On that account, the sense that Marta is the wanter in (22a) and Juan is in (22b)
is not due to a different distribution of θ-roles, but rather due to the association
of the matrix subject with the lexical entailments of the modal (desire in (22),
permission in (26) ).13 (22b) is monoclausal; a Juan raises to a nonthematic sub-
ject position, preserving its inherent case, the standard behavior of (derived)
quirky subjects. Since a clause may contain at most a single modal, example (25)
is impossible.
A second possibility, perhaps more daring, is to acknowledge that querer is a
standard subject control verb, which assigns its own external θ-role. The challenge,
on this account, is to explain how the external argument of querer comes to bear
the dative case associated with the embedded verb.
(27) Chilean Spanish
Control
DP1.DAT …V … [ PRO1 … Vpsych …]
Consider the Agree-based mechanism illustrated in (9b). Establishing control

requires checking (valuation) of φ-features between the matrix v/T and the
embedded T-Agr. To the extent that structural case is a reflex of φ-features, it
is potentially checked in this process. As discussed in section 4.4, there is every
reason to believe that PRO bears case. The question then is – how is the case of
PRO acquired? More technically, how is the case feature of PRO valued?
Several options suggest themselves. The normal way would be parallel to case
checking in finite clauses. The infinitival T-Agr contains a valued case feature,
and by Agreeing with PRO, values the latter’s case. The actual value could vary
across languages: in Icelandic/Greek it is nominative, in Russian/Hungarian it
is dative. On this scenario, the matrix functional head effecting control (say, T)
only values the φ-features of PRO, not its case. Below I illustrate the relevant
valuations with their direction.14
(28) Case Independence
Subj. T-Agr … [CP T-Agr+C [TP tT-Agr [VP PRO …]]]
← Case f-set→ Case→

f-set→
308 IDAN LANDAU
This is what happens in the normal case, even if no direct evidence for the actual
case value of PRO is available in the language; the only special feature of languages
like Russian and Icelandic in this regard is the availability of case concord, which
reveals the case of PRO. Scenario (28) also applies in Icelandic when the embedded
predicate assigns quirky case, the only difference being that the case feature of
PRO is valued by the lexical predicate, not by T-Agr.
Consider now the peculiar case of Chilean Spanish (27), where the embedded
predicate appears to value the case feature of the controller. I suggest that this
may come about whenever the case feature of the matrix functional head (here,
T-Agr) is optionally unvalued. In this situation, the value of PRO’s case feature
(determined by the embedded predicate) may actually ‘percolate up’ to the controller
via the matrix T-Agr, since the latter cannot provide its own value. The valuation
scheme is the following.15
(29) Case Percolation

Subj. T-Agr…[CP T-Agr+C [TP tT-Agr [VP PRO … V …]]]
← Q.Case ← Q.Case ← Q.Case ← Q.Case

f-set→ f-set→
The difference between Icelandic and Chilean Spanish, then, boils down to this: In
Icelandic, finite T-Agr obligatorily bears a valued nominative case feature. In Chilean
Spanish, it is either valued nominative or unvalued. Notice that the locus of this
difference – a property of a functional head – makes it a natural parameter.
The logic of this system predicts another possibility: The case feature of
the embedded T-Agr is unvalued, and as a result PRO inherits its case from the
controller (more precisely, from the matrix functional head that values the case of
the controller). Indeed, as Cecchetto and Oniga (2004) report, this is the situation
in Latin: Subject controlled PRO bears nominative case, object controlled PRO
bears accusative case. Latin is the mirror image of Chilean Spanish, as can be
seen below.16
(30) Case transmission

Subj. T-Agr … [CP T-Agr+C [TP tT-Agr [VP PRO … V …]]]
← Case Case → Case →

f-set→ f-set→
Putting all these data together, we can appreciate the typological space afforded by
the Agree-based approach to OC. Depending on which, if any, of the functional
heads Agreeing with the controller and with PRO are unvalued for case, the
case of PRO is independent from, transmitted from, or percolates to the case of
the controller. In contrast, we predict that in genuine raising, the trace position
would be either caseless, or marked with quirky case. This would look like case
transmission or percolation, respectively. Technically, though, the raising chain

is case valued only once, whereas the OC dependency is valued twice. What is
excluded in principle is case-independence in a raising chain; this follows from the
traditional assumption that A-chains bear a unique case.
Under the reductionist view, however, the last prediction is lost. As far as their
case patterns are concerned, raising and OC should not differ, since both involve an
A-chain with a single case value. For this reason, the Icelandic/Russian/Hungarian
pattern is problematic for the MTC. By contrast, The Chilean Spanish facts, which
are extremely idiosyncratic to begin with, are harmless to the Agree-based theory,
and in fact follow from its design.17
7. BACKWARD CONTROL
The claim that backward control exists in natural language is perhaps the most
interesting contribution of the reductionist camp to the debate on the nature of
OC. According to this camp, backward control is nothing but covert movement of
the ‘controller’ DP to its matrix thematic position; if OC is A-movement and A-movement
can be covert, then backward control is an inevitable possibility. PRO-based
approaches, in contrast, cannot explain how PRO can be licensed and interpreted in a
position higher than the controller’s.
In this section I will not attempt any analysis of backward control. Instead,
I will lay out some skeptical thoughts about the force of the conclusions war-
ranted by the evidence that is currently available. I believe that this skepticism is in
place, and I hope that it will prompt further research into this important topic.
Hornstein cites two studies arguing for backward control: Farrell 1995 and
Polinsky and Potsdam 2002. Let us consider them in turn.
Farrell (1995) discusses causative complements to fazer (‘make’) and mandar
(‘have’; literally, ‘send’) in Brazilian Portuguese (BP), illustrated below.
(31) a. A mulher fez o nenê dormir.

‘The woman made the baby sleep’.
b. Eu mandei o sapateiro concertar esse sapato.

‘I had the cobbler fix these shoes’.
These infinitival complements, according to Farrell, have an ECM syntax and

an object control semantics. Like object control complements, they do not show
voice transparency under passivization of the embedded verb; the causee is
subject to selectional restrictions – it cannot be a clause or an expletive; furthermore,
with mandar, it must be animate.
At the same time, the overt causee seems to occupy the embedded subject position
(the one reserved for PRO under normal object control). It can be realized as a subject
(1st person) pronoun (33), which is otherwise excluded from object positions, and
can follow the embedded verb when the latter is unaccusative, a standard property of
subjects in BP.
310 IDAN LANDAU
(33) A professora mandou/fez eu apagar o quadro.

the teacher had/ made I.NOM erase the board.
‘The teacher had/made me erase the board’.
Hornstein concludes from these facts that the BP periphrastic causative

construction displays backward control: The causee raises covertly to the
thematic position of the matrix object, establishing object control. It is
curious, though, that Farrell himself, although using the same terminology
of ‘backward control’, draws different conclusions from his own data. Farrell
argues that the matrix object position is in fact never realized in the syntax.
Following Jackendoff 1990, he views Lexical Conceptual Structure as the level
where control is established. In fact, he takes his data to show that any syntactic
approach to control is misguided.
Let me first say that I find the BP data fascinating, although more puzzling
than acknowledged by others. First, notice that the causee can be realized as an
object clitic on the main verb.
(34) O professor os fez estudar mais.

The professor CL-3.MASC.PL made study more
‘The professor made them study more’.
Farrell explains that the causee may either get nominative case from the embedded
Infl, as in (33), or accusative case from the matrix verb, as in (34). He points
out that unlike standard object control verbs (35b), fazer (or mandar) cannot be
passivized in the causative construction (35a).
(35) a. *O nenê foi feito dormir.

(The baby was made sleep)
b. Os alunos foram forçados a estudarem mais.

‘The students were forced to study more’.
Farrell and Hornstein take this fact to show that the causee does not occupy a
matrix object position. But does it really show that? Recall that (34) has shown
that an ECM analysis is possible for these constructions. If so, it is far from clear
what blocks (35a). If the causee is an embedded ECM subject, matrix passivization
should withdraw its accusative case and allow raising to the matrix subject position.
Notice that this should be possible whether the causee stops at the matrix object
position, to pick up a θ-role (as Hornstein would have it), or not (as Farrell would).
In fact, passivization of causatives is unavailable in many languages ((35a) is
impossible in English and French too). Little follows from this poorly understood
observation about the surface position of the causee.
In sum, contrary to Hornstein, I do not consider these facts as ‘evidence that
the thematic complement is not a syntactic object of the matrix in overt syntax’
(p. 57). In fact, they strike me as a real puzzle for everyone.
A much more persuasive case for backward control is made in Polinsky and
Potsdam 2002. Through a series of tests P&P show that the ergative argument in
constructions like (37) in Tsez occupies the embedded subject position, although
associated with both the lower and the higher subject θ-roles (II/III are noun class
agreement markers).
(37) ∆1 [kid-ba1 ziya b-išra] y-oq-si.

II.ABS girl.II-ERG cow.III.ABS III-feed.INF II-begin-PAST.EVID
‘The girl began to feed the cow’.
As I said, I find most of P&P’s arguments compelling, so I will only comment on

the weaker aspects of their analysis.
First, perhaps the most striking feature of backward control is its rarity. In Tsez,
only two verbs display it; the numbers hardly exceed five in other languages. Most com-
monly, the backward control verbs are aspectuals (begin, continue, stop, etc.), which
also have a standard raising guise. This suspicious overlap is recognized by P&P, and is
attributed to the tendency for such verbs to undergo V-to-T raising. In their analysis, V
in T may satisfy the EPP, allowing the ‘controller’ to remain in its embedded position.
Yet the ability of a verb to satisfy EPP is a stipulated lexical property – a [+D] feature.
Another puzzling issue is the status of the matrix null controller (∆ in (37)) with
respect to case. On the one hand, P&P (p. 258) explain the apparently exceptional
agreement of -oqa ‘begin’ with the ergative argument (agreement in noun class
is always with the absolutive argument) by positing a co-indexed null matrix
absolutive argument which is the true trigger of agreement. On the other hand,
the actual analysis P&P end up with (see their example 65) deprives the control
verb -oqa ‘begin’ of its case assigning capacity. The lack of absolutive case in the
matrix clause is in fact crucial: It explains the very effect of OC. If an independent
DP were merged as the external argument of –oqa, its case feature would remain
unchecked. Thus, the only DP that can check the matrix θ-role is the embedded
subject, which has already checked its ergative case feature downstairs. This is
done by covert raising of this DP to the matrix clause.
It seems to me that P&P run into a contradiction here. To explain the
agreement facts, they must allow absolutive case in the matrix clause. To explain
OC, they must exclude it. Conceivably, they could acknowledge that the LF A-
chain of ‘the girl’ in (37) bears two distinct cases, not an unprecedented situation
(see Bejar and Massam 1999). This, however, would rob them of the account
for why a second DP cannot be merged in the matrix clause (e.g. *The boy.ABS
[the girl.ERG the cow.ABS to-feed] began). That the very effect of OC hangs on
a problematic technical assumption, rather than on a natural semantic analysis
of begin, is a testimony, in my view, to the limitations inherent to the movement
approach to OC.
Having said that, I wish to stress that the evidence for backward control (more
so in Tsez than in Brazilian Portuguese) is a real challenge to standard theories of
OC, perhaps the most challenging phenomenon presented so far by proponents of
the movement approach.
312 IDAN LANDAU
8. FINITE CONTROL
Hornstein 2003 (Sect. 1.8.2) describes OC into finite clauses in Brazilian Portuguese
(BP), as below:
(38) O Joao1 disse [que ec1 comprou um caro novo].

‘John1 said that he1 bought a new car’.
BP has lost referential pro-drop, and appears to allow referential null subjects in
finite clauses only under OC. Hornstein suggests that here too, A-movement applies
to produce OC. Importantly, he assumes that case marking of the embedded
subject is optional (a fact related to the simplification of the verbal paradigm).
When this position is caseless, the embedded subject may raise to the matrix
clause, where it checks its case (and an additional θ-feature). BP also allows
raising out of finite complements.18
Finite control is in fact more common than normally assumed. It is found
throughout the Balkan family, where the infinitive was lost and replaced by the
subjunctive. Other languages that exhibit finite control are Hebrew, Kannada,
Persian, and Dogrib (see Landau 2004a for a comprehensive description and
analysis). Ironically, the BP construction may be a poor choice to illustrate
this phenomenon. According to Modesto (this volume) constructions like (38)
involve embedded topic drop, which is contingent on the presence of a coreferent
matrix topic. Modesto shows that this analysis fares better than the OC analysis
in explaining some peculiar interactions with Ā-movement.
At any rate, even if the controlled empty category in BP finite complements
were the classical OC PRO (or for Hornstein, an A-trace), there would be no
support whatsoever for the alleged lack of case for this element. The BP data do
not speak to this issue; languages with case concord, in contrast, make it clear that
OC PRO in finite clauses bears the standard case that any lexical subject would
bear in that position.
The examples below illustrate low case marking of OC PRO in finite complements.
(39) Greek (Philippaki-Warburton and Catsimali 1999)

a. Anangasan tin Eleni [PRO na milisi afti i idhja].
forced.3PL the Eleni.ACC PRO.NOM PRT speak.3sg she herself.NOM
‘They forced Helen to speak herself’.
Hungarian (J. Horvath, personal communication)

b. János bíztatta Marit [hogy PRO maga/ *magát
John.NOM urged.3sg.DEF Mary.ACC that PRO.NOM herself.NOM/*ACC
beszéljen az orvossal].
talk.SUBJ.3sg the doctor-with
‘John urged Mary to talk to the doctor herself’.
Notice that the emphatic reflexives bear nominative case. Such emphatic elements
display case-concord – they agree in case with the DP they modify. Thus, we
can conclude that PRO bears nominative case in (39) – despite the fact that the
controller is marked accusative. This situation is parallel to what we have seen in

Icelandic and Russian OC infinitives: The controller and PRO bear distinct cases
(see section 6). On a theory that divorces the distribution of PRO from case, as
in Landau 2004a, 2006, this phenomenon presents no mystery. If, however, lack
of case at the base position is a pre-requisite for A-movement, and OC is but
A-movement, then data such as (39) are extremely puzzling.
9. PRO-GATE
The last section in Hornstein 2003 argues that analysing OC as A-movement

provides an elegant account for an old chestnut in the study of control – the
PRO-gate. This is the weakest part of the paper, containing very few empirically
valid claims. Therefore, it is important to clarify the factual nature of PRO-gate in
particular and subject gerunds in general.
The PRO-gate effect is illustrated in (40); a PRO subject in a gerundive subject
clause allows a pronoun to be Ā-bound even to the left of a variable (the wh-trace),
in contrast to a lexical subject in the same position, which gives rise to the familiar
WCO violation.
(40) a. Who1 did [PRO1 kissing his1 mother] upset t1?
b. *Who1 did [Mary’s/his1/him1 kissing his1 mother] upset t1?
Following Kiguchi 2000 and Kiguchi and Hornstein 2001, Hornstein makes the
following argument.
(41) a. A-movement circumvents WCO (i.e. a pronoun bound by an A-trace will not
violate WCO).
b. PRO in OC is an A-trace.
c. PRO in subject gerunds displays OC.
d. Therefore, PRO in subject gerunds circumvents WCO (= PRO-gate).
(41a) is well established (e.g. Who1 t1 seemed to his1 wife t1 to be clever?). (41b)
was precisely the issue under debate in Landau (2003); for the sake of the present
argument, let us grant it. The novel claim, then, is (41c). If true, then (41) tracks a
valid deduction, and PRO-gate is indeed explained.
However, (41c) is false. In what follows I show why the claim that PRO in
subject gerunds falls under OC is unsupported.
To show that PRO in subject gerunds requires an antecedent, Hornstein cites (42).
(42) a. *Shaving himself impressed Mary.
b. *Shaving himself made it seem cold outside.

314 IDAN LANDAU
These examples show little out of context. It has been widely documented (e.g.
Grinder 1970; Kuno 1975; Williams 1992; Bresnan 1982; Manzini 1983;
Lebeaux 1984; Kawasaki 1993; Landau 1999, 2000) that a local object DP is
not an obligatory controller for PRO in a subject gerund. Indeed, Kiguchi and
Hornstein (2001) acknowledge that previous discourse may allow extrasentential
reference for such PRO, citing (43a). Furthermore, (42a) is possible in a strongly
deictic context, where the speaker points to (say) John, saying ‘Shaving himself…’.
Compare (43b), where the 1st/2nd person PRO is easily accessible in any context.
(43) a. John1 put on a bushy hat. PRO1 having shaved himself1 earlier made it seem
very cold.
b. PRO1 shaving myself1/yourself1 impressed Mary.
So the first property of OC already fails to hold of PRO in subject gerunds:

It does not require a local grammatical antecedent.
Next, Hornstein claims that split control is blocked.
(44) *John1 said that PRO1+2 shaving themselves upset everyone2.
Again, the example is misleading. Control into subject gerunds is subject to logo-
phoricity constraints, as recognized in the studies cited above. Those constraints are
quite complex, involving both syntactic prominence, animacy, discourse salience and
mental perspective. Provided that both controllers qualify as logophoric antecedents,
control can be split between them. Example (45a) is provided by Grinder (1970);
(45b) is my own minimal modification of (44), which makes it grammatical.
(45) a. That [PRO1+2 covering themselves with mud] disturbed Spiro1 amused Dick2.
b. John1 told Bill2 that PRO1+2 shaving themselves would upset everyone.
Unlike OC environments, where split control is indeed uncommon (see section 3),
NOC environments allow it rather liberally, contra Hornstein’s description.
Hornstein also claims that PRO in subject gerunds only permits a de se
interpretation and a sloppy reading when the controller is modified by only
(Hornstein 2003:(134), (135) ). However, the judgments on such examples are
very subtle, and are easily reversed with contextual manipulation. Consider the
following scenario.
(46) Fred has recently barely survived a ferocious assault by an alligator that was videotaped
live. Fred lost his memory of this entire event, and he now watches the videotape.
Describing Fred’s reaction to the videotape, we say:
a. Being assaulted by an alligator amused Fred.

b. Fred was amused to have been assaulted by an alligator.
Speakers judge that in this scenario (46a) makes sense (even if it is morally
regrettable) but (46b) must imply that Fred is a suicidal freak. In other words,
(a) is compatible with Fred’s not being aware that he himself is the victim of the
assault, but (b) is not. The implication is that PRO in subject gerunds can be
interpreted de re, unlike PRO in OC complements, which is necessarily interpreted
de se. Again, subject gerunds behave like NOC.
As to only-sentences, the strict reading is possible given the right context. (47a)
can imply that only Bill was amused by his flirting around (his wife did not find it all
that amusing); the same reading is available in the stripping constructions (47b, c).
(47) a. Flirting around amused only Bill.
b. Flirting around amused Bill, but not his wife.
c. Flirting around amused Bill, but offended his wife.
I should mention that there is some speaker variation here, but clearly the majority
allows the strict reading. Nothing like that is observed in classical OC (into
complements).
The fourth argument in Hornstein 2003 in favor of (41c) is that the antecedent
of PRO must be the most prominent DP available and cannot be buried inside
another DP.
(48) a. *PRO1 shaving himself made Mary believe John1.
b. *PRO1 shaving himself upset John’s1 mother.
The examples are again partial. Notice that Mary in (48a) is a potential logophoric
antecedent (subject of a mental attitude) while John is not. If NOC PRO is a
logophor, this fact alone – and not locality – would rule out control by John.
Whenever two DPs follow the gerund, it is indeed easier to take the closer one as
a controller. This is no doubt due to the fact that this DP will be more prominent
on the logophoric scale (its own perspective ‘embedding’ that of the more
distant DP). Processing limitations (favoring shorter dependencies) might also
be at work. But this is just a strong tendency, not a grammatical constraint.
Example (49), where long-distance control skips a potential controller, is provided
by Richardson (1986).
(49) [PRO1 storming out of the room that way after losing the game] convinced everyone
that John1 is very immature.
Another way of showing that proximity in NOC is not a syntactic constraint

is by using previous discourse to highlight the lower DP, thus making it the
prominent logophoric center (see Kawasaki 1993 for evidence that ‘topicality’
affects NOC). Compare (48a), where locality supposedly rules out control by
John – with the following.
(50) a. John1 finally got what he wanted. PRO1 shaving himself made Mary believe him1.
b. John1 knew that [PRO1 shaving himself] made Mary believe him1.
316 IDAN LANDAU
Furthermore, a local DP can perfectly be skipped if it is an expletive.
(51) a. [PRO1 getting himself a new pair of trekking shoes] made it look like John1 was
about to leave on a journey.
b. *[PRO1 getting himself a new pair of trekking shoes] made Mary realize John1
was about to leave on a journey.
This non-syntactic contrast is mysterious if the controller is moved from the

position of PRO by sideward movement. For (51a and b), the decision whether
sideward movement is allowed into the most embedded subject position would
crucially depend on the identity of DPs merging after that point, in a higher cycle
(namely, it or Mary). This look-ahead is precisely what is banned in a strongly
derivational theory like the MTC.
(48b) does not represent a general fact either, and the claim associated with it
ignores familiar cases (discussed in Chomsky 1981, 1986, and in Landau 1999, 2000)
where a controller can be embedded inside a DP. (52a) is from Chomsky 1981.
(52) a. PRO1 finishing his work on time is important to John’s1 development/*friends.
b. [PRO1 perjuring himself] tarnished John’s1 reputation.
c. [PRO1 reciting these silly lines] insulted John’s1 intelligence/*sister.
The contrast between the good versions of (52a, c) and the bad ones (including
(48b)) is that the controller in the former is embedded inside DPs which are not
in themselves potential controllers. Moreover, nouns like reputation, development,
intelligence, etc. all designate some attribute of the personality of their posses-
sor; Landau 1999, 2000 dubs them ‘logophoric extensions’. It is a fact – and
crucially, not a syntactic fact – that these nouns are ‘transparent’ to NOC, whereas
individual-referring nouns are not. I cannot see how the (sideward) movement
approach can make sense of the data in (49)–(52).
Thus, on every single criterion PRO in subject gerunds displays NOC.19
Hornstein points out that whenever sideward movement is blocked – e.g. when
the gerund is inside an island – NOC should emerge and with it, WCO effects (see his
examples 138, 139). The fact that NOC ‘emerges’ is of course not surprising – it
emerges already in the basic paradigm, as shown above, when the gerund is not
inside an island. As to the WCO effects – the judgments are not clear-cut at all.
Native speakers find the following example acceptable (i.e. PRO-gate applies even
to gerunds inside islands).
(53) The fact that PRO1 losing his1 life is a distinct possibility frightens every soldier1.
Finally – perhaps the deadliest blow to this analysis – PRO-gate effects are attested
even in environments of arbitrary control (J. Bobaljik, personal communication).
The judgments below were confirmed with several native speakers.
(54) a. PROarb calling him1 an idiot would upset any/?every professor1.
b. PROarb overcooking his1 lunch would make any/?every pupil1 cry.
Observe that the deduction in (41) presupposes that for PRO to act as a gate,
it must be locally controlled by the operator that binds the pronoun; only if
it is so controlled can it be even contemplated that PRO is a residue of prior
A-movement of the operator. But clearly, arbitrary PRO is not controlled by
any local DP, hence could not be an A-trace. The fact that it still displays the
PRO-gate effect eliminates the entire motivation for the analysis.
To summarize, the argument that PRO-gate can be explained by sideward
A-movement is seriously flawed. First, PRO in subject gerunds systematically
displays NOC, as a careful consideration of the data reveals. On Hornstein’s
analysis, NOC PRO is a small pro, hence it is incorrectly predicted to trigger WCO
violations. But even if PRO in subject gerunds had been a genuine OC PRO, the
fact is that OC is not a necessary condition for PRO-gate. The PRO-gate puzzle
appears more damaging than corroborating to the movement approach to control.
10. WH-INFINITIVES
Another case of OC being misclassified as NOC involves infinitival wh-complements

like (55a).
(55) a. John wondered what to do.
b. ? What did John say how to cook?
Hornstein 1999, 2001, following the mainstream tradition, asserts that these fall
under NOC and derives this effect from the islandhood of wh-complements.
Specifically, the claim is that since movement cannot cross a wh-island, a
‘last-resort’ pro is inserted in the embedded subject position, yielding NOC.
Landau 2000, 2003 shows that the common conception is false – wh-complements
display OC, specifically, of the partial control type. Thus, they resist long-distance
control and truly arbitrary control, force sloppy readings under VP-ellipsis, etc.
Landau 2003 concludes that ‘Hornstein must either deny the islandhood of
interrogative complements or abandon the link between islandhood and NOC’
(p. 483).
Not surprisingly, Boeckx and Hornstein 2004 chooses the former option,
claiming that wh-infinitives are ‘very weak islands’– see (55b) (their judgment).
Furthermore, they block wh-movement due to Relativized Minimality, but
whether or not they should block A-movement (which supposedly underlies OC)
is unclear.
The inconsistent claims surrounding these data are somewhat elusive, and the
position taken in Boeckx and Hornstein is very hard to test, especially since the status
of examples like (55b) is debateable. Still, one can put to test the speculation that
318 IDAN LANDAU
infinitival wh-islands are invisible to A-movement. Notice that the verb inquire is
compatible with a wh-complement, and with the expletive there (in the passive voice).
(56) a. Someone inquired how to welcome Malay visitors.
b. There were Malay visitors welcomed.
However, the two constructions cannot be combined – compare (57a) and (57b).
(57) a. *There were inquired how to be Malay visitors welcomed.
b. There were likely to be Malay visitors welcomed.
The plausible derivation for (57a, b) would have the expletive merge in the embedded
subject position and raise to the matrix one. Then the matrix T establishes Agree with
the embedded associate (witness the plural agreement). Alternatively, the expletive
may directly merge in the matrix subject position, following long-distance Agree.
Crucially, whatever the cross-clausal link is – Agree or Move – it is of the
A-type. And also crucially, this link is allowed to cross a raising infinitive
(57b) but not a wh-infinitive (57a). Thus, the evidence at hand suggests that
wh-infinitives are islands to A-movement (and strong ones – (57a) is completely
impossible). Boeckx and Hornstein’s speculation cannot be maintained, leaving
the initial puzzle unanswered: How can movement (of the subject) escape a wh-
infinitive for the purposes of OC, but no other movement type may do so?20
Note, in comparison, that the Agree-based approach to OC (and specifically,
to PC) faces no such difficulties. Under this approach, Agree targets the C head of
the wh-complement, which is accessible to the higher phase by assumption (see the
Phase Impenetrability Condition of Chomsky 2000). The fact that the specifier of
the complement is filled by a wh-phrase is immaterial, since the OC dependency
consists of φ-features. Nothing else needs to be said about these constructions.21
11. CONTROL ACROSS PASSIVE
Landau 2003 observes that Hornstein 1999 provides no principled way to rule
out (58a), given that both (58b) and (58c) are possible. The same type of A-
movement allegedly involved in the latter should also be able to apply in the
former. Landau (2003) also shows the solution offered in Hornstein (2000) to
be empirically inadequate – predicting, in fact, (58b) to be as bad as (58a).
(58) a. *John was hoped to leave.
b. John was persuaded to leave.
c. John was expected to leave.
Boeckx and Hornstein (2004) return to the issue, proposing a new solution. They
claim that the problem with (58a) is the general inability of hope to passivize with
a (non-expletive) subject. Thus, only verbs that license a DP object can passivize;
(58a) is ungrammatical because (60a) is, and (58b) is grammatical because (60b) is.
(60) a. *A victory was hoped.
b. John was persuaded.
Boeckx and Hornstein’s condition may be a necessary condition on English

passive formation, but its relevance to the point at stake is unclear. Importantly,
it misses the generalization (‘Visser’s Generalization’) that (58a) only illustrates:
No subject control verb may promote the embedded subject to the matrix subject
position via passivization – not even verbs that do passivize their DP objects.
(61) a. The decision was regretted/hated.
b. They regretted/hated to have passed the decision.
c. *They were regretted/hated to have passed the decision.
This generalization is obscured, in English, by the availability of ‘passive ECM’

variants of subject control verbs. Thus, it might appear that the pair in (62)
confirms Boeckx and Hornstein’s hypothesis.
(62) a. The program was decided.
b. John was decided to be our representative.
However, (62b) is a passive ECM, not a passive control construction.22 This is

made evident by its tense restrictions, typical of ECM and absent in control.
(63) a. John decided to apologize to Mary tomorrow.
b. *John was decided to apologize to Mary tomorrow.
Most languages lack this option, and do not even allow (62b). Furthermore, in
languages where unergative verbs form impersonal passives (e.g. Dutch), it seems
that Case-licensing a DP object is not a necessary condition on passive; nonetheless,
sentences like (58a)/(63b) are consistently ruled out. As far as we know – this is
a universal phenomenon. The solution in Boeckx and Hornstein 2004 (like its
predecessor in Hornstein 2000) fails to address this general problem.23
12. SIDEWARD MOVEMENT FROM COMPLEMENTS
Landau 2003 notes that the option of ‘sideward movement’ – used by Hornstein
to explain OC into adjuncts – overgenerates examples like (64).
(64) *John’s1 friends prefer [t1 to behave himself].

320 IDAN LANDAU
If OC is movement, and sideward movement is allowed, nothing seems to block

movement of the embedded subject into [Spec,DP] of a matrix argument,
generating nonexisting interpretations.
Boeckx and Hornstein 2004 proposes two possible ways to deal with this
problem: (i) The DP John’s friends is derived from a small clause headed by John,
which functions as a predicate. Semantically, John cannot be both a predicate
(of friends) and an argument (of behave himself), hence (64) is uninterpretable;
(ii) Possessive DPs are adjuncts, at least in cases like John’s friends. Movement to
adjoined positions is blocked (violating Greed), hence (64) is underivable.
Boeckx and Hornstein’s solution is too narrow, confusing one example (which
may be solved locally) with a general pattern (which calls for a deeper revision).
Both options (i) and (ii) above can be ruled out if we make sure that the ‘raised’
genitive phrase is a genuine argument of its head noun. Such is the case in (65).
(65) a. *John’s1 examination of the patient convinced Mary [t1 to applaud himself].
b. *John’s1 examining the patient convinced Mary [t1 to applaud himself].
Examination and examining in (65) are complex event nominals in the sense
of Grimshaw (1990), taking an internal agent, namely John. There is no reason
whatsoever to analyse John here either as a predicate or as an adjunct. The fact
that the ungrammatical pattern is constant across (64) and (65) indicates that
Boeckx and Hornstein’s proposal does not go to the heart of the problem.
13. NOVEL RAISING-CONTROL CONTRASTS
While the debate about the nature of OC is largely informed by well-known

facts, one should not assume that all the relevant facts have been uncovered,
even in this well-trodden terrain. The empirical adequacy of the competing views
should be continually measured against new discoveries.24 In this section I will
briefly discuss two such discoveries, made by Postal (2004:Chap. 2).25
Postal discusses predicates like be the matter/wrong with, whose subject position
is antipronominal.
(66) a. Something1 is the matter with my transmission, but that sort of thing/*it1 is not the
matter with his.
b. *He said something1 was wrong with her values, and it1 was wrong with them.
Interestingly, these two predicates may occur inside raising complements, but not
inside controlled clauses.
(67) a. Lots of things seem to be the matter with your transmission.
b. Such a thing is bound to be wrong with someone’s liver.

(68) a. *Lots of things can be the matter with your transmission without being the
matter with mine.
b. *That can be detectable without being wrong with your liver.
The contrast is explained if PRO is a null pronominal (or reflexive) of sorts,

whereas the trace of a raised DP is simply a silent copy of this DP, retaining
all its distributional properties. Notice that according to the MTC, OC into
adjuncts like (68) is achieved via NP movement, hence the contrast with (67)
is unexplained.
A second observation pointing to the pronominal nature of PRO involves
antecedence relations between certain DPs and their metonyms. Observe first that
pronominal antecedence tolerates certain metonymous shifts but not others.
(69) a. I am parked on 26th Street (= my car is parked on 26th Street).
b. Microsoft went up (= Microsoft’s stock’s price went up).
(70) a. John1 claimed that he1 was parked on 26th Street.
b. *Microsoft1 claimed that it1 would go up.
Thus, for whatever reason, a pronoun may be metonymous to its antecedent in

the case of a car-possessor relation, but not in the case of a company-stock’s
price relation.26 Crucially, now, the same contrast is preserved in control, but
not in raising.
(71) a. John plans to be parked on 26th Street.
b. *Microsoft1 plans to go up.
(72) a. John seems/is likely to be parked on 26th Street.
b. Microsoft1 seems/is likely to go up.
The fact that (71b) patterns with (70b) and not with (72b) strongly suggests that
the null subject of the control complement is more akin to a pronoun than to a
copy of NP movement. Postal further shows that a ‘stock price’ PRO can be
controlled, provided its antecedent is construed as a stock price as well.
(73) Microsoft went up today after going down yesterday.
Therefore, the problem with (71b) is specifically the metonymous shift, which
is independently shown to be restricted for pronouns (as well as reflexives, e.g.
*Microsoft believes itself to have gone up). A raising analysis of OC will be hard-
pressed to make sense of this pattern of data.
322 IDAN LANDAU
14. CONCLUSIONS
This paper has attempted to assess the current status of two accounts of control:
The movement analysis and the Agree-based analysis. Although situated within
the same framework, each account comes with its own baggage of auxiliary
assumptions; quite often, the empirical success of the accounts relies on those
assumptions, rather than on first principles. It is now worth considering where
each account stands.
Consider first the issues raised in Hornstein 2003. As argued above, few of the empir-
ical challenges to the movement analysis are genuinely solved. MDP-violations, partial
control, the proper delineation of NOC and case-independence in OC reveal
serious shortcomings of the analysis. The PRO-gate phenomenon lies entirely
beyond the explanatory capacity of the MTC (in its present form). Moreover,
the responses in Boeckx and Hornstein 2004 to the issues raised in Landau 2003
leave much to be desired. On many points – blocking passivization of subject
control verbs, blocking sideward movement from complements, the treatment of
implicit control and of OC into wh-infinitives – their solutions are extremely narrow
in scope, failing to address the deeper generalizations.
Finally, novel facts that came to light recently point to a referential distinction
between PRO and NP-trace, which is also congenial to the traditional view but
extremely puzzling under the raising analysis.
Remarking on the current success of movement-based analyses, Hornstein
(2003) writes that ‘what I find interesting is not that they indeed have empirical
weaknesses, but that there are not more of them’ (p. 63). That, however, is a question
of perspective; in my view, the above list of problems is quite worrisome.
Compare now the Agree-based approach. First, this analysis is free of many
of the problems attributed to it in Hornstein 2003; in some cases (i.e. semantic
plurality in PC) the objections have no force; in others (e.g. T-to-C) minimal
modifications, of the sort developed in Landau 2004a, remove the difficulties; and
yet in others (e.g. finite control), a closer consideration of the full picture reveals a
clear advantage for the Agree-based analysis over the movement analysis.
The Agree-based analysis of OC is of course not without problems. Two such
problems emerge from the debate. First, there is the existence of split control.
This, I think, is an outstanding problem for both sides of the debate, whether
acknowledged as such or not. Second, there is the phenomenon of backward
control. If not misanalysed (see the skeptical comments in section 7), then
backward control represents a challenge to any nonmovement approach to OC.
Hopefully, more research into this phenomenon will be available to inform future
theoretical attempts to deal with it.
Thus, on the empirical front, which I believe is the decisive one, it seems
fair to say that the movement analysis of OC still has a long way to go to reach
the empirical adequacy of its rivals, the Agree-based analysis included. The
conclusions of Landau 2003 do not call for revision.
On the meta-theoretical front, our perspective now is clearer than it was a few
years ago. The theoretical apparatus constructed in Hornstein 1999 was indeed
very tight and elegant. The only problem was – it could not cope with the wealth
of facts involved in control. To their credit, Hornstein 2003 and Boeckx and
Hornstein 2004 are much more empirically minded, addressing a variety of factual
issues. The inevitable price is, however, a considerable enrichment of the sparse
theoretical toolbox afforded by Hornstein 1999. Novel mechanisms abound,
none of which follows from the bare foundations of the movement analysis: Log-
ophoric PRO is allowed in NOC, a new spellout condition is imposed on binders,
a meaning postulate for PC is proposed, stipulations about case in finite control
are introduced, speculations about wh-islands are entertained, etc.
All of these are, of course, natural developments. It is difficult to see how any
theory can account for the facts of control under the a priori strictures imposed in
Hornstein 1999. But now, I think, the claim for theoretical elegance and economy,
repeated in Hornstein 2003 and Boeckx and Hornstein 2004, is far less convinc-
ing. Supplemented by quite a number of auxiliary assumptions, the movement
analysis of OC does not seem obviously simpler (more elegant, less redundant,
etc.) than its rivals. The fact is – no one can tell for sure. Meta-theoretical com-
parisons, so revered in certain linguistic circles, can be easily pulled toward what-
ever direction suits one’s taste; worse, they can never be objectively settled, which
leaves us with the good old adjudicator – empirical adequacy. And its verdict, I
believe, is quite clear in this case.
* This paper has benefited from the careful reading and comments of Paul Postal, Eric Potsdam, and
Susi Wurmbrand, for which I am grateful. I am the only one to blame for any remaining shortcomings.
1
Incidentally, the fact that the MDP is a misanalysis of a complex of semantic factors was already
perceived quite clearly in Postal 1970, the first serious generative study of control following Rosen-
baum 1967. Postal observed that the reference of PRO in infinitival complements mimics the
reference of a subject pronoun in the corresponding finite complements, which contain a modal.
He also documented the now-familiar classes of counterexamples to the MDP – the promise/vow
class, split control, subject control across an object with ask and the lack of control shift in com-
munication verbs (see (2) below).
2
See Landau 2003 for similar comments on the Hornstein 1999 treatment of the distinction between
reflexive and non-reflexive verbs (wash vs. hit).
3
For several attempts in this spirit, see Farkas 1988, Sag and Pollard 1991, and Petter 1998.
4
Notice that implicit arguments must exist on anyone’s theory of syntax, which frugally reserves
pro as a last resort device. Otherwise, pros will proliferate without limit (in the sentence The letter
made mother angry, does letter contain a Goal pro, mother a ‘son/daughter’ pro, and angry a
Subject-Matter pro?). If so, a theory which posits just one type of pro, always visible to binding, is
to be preferred over more complex alternatives.
5
Hornstein proposes that the peculiarity of promise might reside in the presence of a null (dative)
preposition introducing the goal argument. Thus, the goal DP does not c-command PRO, and the
MLC selects the subject as the closest controller. It is unclear how this proposal is consistent with
the well-known observation – acknowledged by Hornstein himself (example 69) – that prepositions
are invisible to c-command relations. Nor is it clear how the proposal could distinguish minimal
pairs like the following.
i. Bill1 vowed to Jane2 [PRO1 to marry her2/*him1].
ii. Bill1 appealed to Jane2 [PRO2 to marry him1/*her2].

6
Still, one can imagine that the infinitival Agr is targeted by both the matrix v and T. This would
no doubt be an exotic option, but one that does not run into tree-geometric contradictions, as the
raising analysis appears to.
324 IDAN LANDAU
7
Boeckx and Hornstein (2004, fn. 1) remark that since Agree is implicated in Move, ‘once control is
treated in terms of Agree, it becomes very hard to rule out movement within control structures in
a principled fashion’. In fact, movement is ruled out in a principled way if θ-roles are not features
and cannot drive movement. The problem is just the opposite: How to prevent the alleged θ-driven
movement from applying where it should not. This is the overgeneration problem discussed in
Landau (2003) (see sections 11, 12 below).
8
Boeckx and Hornstein 2004 downplays the significance of partial control by saying that ‘it is a
special lexical property of meet and a handful of other verbs that allows them to give rise to a
partial control reading’ (p. 449). This statement is based on their observation that given the right
context, even raising verbs support ‘partial readings’:
i. John seems to be meeting all the time.
First, note that the relevant lexical property is absolutely systematic All collective predicates participate
in PC. Second, any predicate may be turned into a collective predicate by adding together, giving rise
to productive PC constructions that cannot be relegated to idiomatic exceptions. While meet may be
such an exception, other collective predicates are not, pointing to a clear distinction between raising and
control environments:
We admitted that…
ii. *John seems to be working together/gathering all the time.
iii. John planned to work together/gather all the time.

9
Hornstein wonders, more than once, how Agree can establish antecedence – ‘especially when it is as round
about as in (76b)’ (=(9b) above). This is like wondering how co-indexing can establish antecedence between
John and himself in (17). Does the number of intermediate steps ‘weaken’ the dependency?
10
In another recent attempt at a semantic reduction of PC, Jackendoff and Culicover 2003 proposes
that PC is coerced through an abstract INTENTION predicate. This explains, they claim, why PC is
restricted to control verbs expressing intentions and to irrealis complements. The problem is that
neither claim is true. As shown at length in Landau 1999 (58–62); 2000 (45–48), PC occurs under
factive and propositional verbs, both allowing realis complements, and neither of which involves
intention; e.g. the Italian examples:
i. Maria pensava che Gianni si fosse pentito di essersi baciarti alla festa.
Mary thought that John had regretted to-be-SI kissed at-the party.
‘Mary thought that John had regretted to have kissed at the party’
ii. Il presidente crede di essersi riuniti inutilmente la notte scorsa.

the chair believes to-be-SI gathered in vain the night last.
‘The chair believes to have gathered in vain last night’
I realize that the temptation to explain PC in purely semantic terms is hard to resist, yet partial expla-
nations, unlike partial control, cannot be coerced.
11
I am abstracting away from the case split internal to Russian between subject control (without an overt
C), where PRO inherits nominative case from the controller, and the ‘elsewhere’ situation, where PRO
is locally assigned dative case. Only the latter option is analogous to Icelandic. For relevant discussion,
see Comrie 1974, Neidle 1988, Franks 1990, Franks and Hornstein 1992, and Babby 1998.
12
See Gonzalez 1988, 1990, cited in Bošković 1994.
13
Want normally disallows expletive subjects, but Wurmbrand 2001 shows that under certain cir-
cumstances this restriction is lifted. I thank Susi Wurmbrand for discussion of these issues.
14
The φ-features of PRO come to be valued after those of the embedded T-Agr are; I omit this part
from (28).
15
Notice that the case feature of the embedded T-Agr may be valued in (29), but still ‘overwritten’ by
the quirky case of PRO, as usual in such contexts.
16
Apparently Icelandic also exhibits this pattern, alongside case-independence (see Andrews 1976;
Thráinsson 1979).
17
If the Chilean Spanish facts reduce to an isolated property of querer, independent evidence will be
needed to exemplify the model of Case Percolation in (29). If none shows up, we may conclude that
UG blocks this option (say, by stipulating that the case feature of finite T-Agr is always valued).
18
Hornstein remarks that ‘it is interesting that these BP clauses tolerate both control and super-raising,
suggesting that both raising and control are reflexes of the same operation’(p. 60). However, nothing
of the sort is suggested. Finite clauses are likewise transparent to both wh-movement and pronominal
binding, yet no underlying common operation is suggested by that.
19
It should be stressed, again, that this conclusion is supported by a vast literature, cited above.
Boeckx and Hornstein 2004 discusses some peculiarities of the specific examples given in Landau
2003, failing to address the general issue. The complex, non-pronominal behavior of NOC PRO
has been documented in other languages as well (see Babby and Franks 1998 on Russian; Lyngfelt
1999 on Swedish).
20
Obviously, solutions can be devised; for example, it might be suggested that that θ-driven movement
is subject to different islands than other movements. This will correctly distinguish OC from both
expletive raising and wh-movement. But the issue is not whether solutions can be devised – they
always can – but rather whether they can be substantiated on the basis of independent evidence.
Many of the suggestions in Boeckx and Hornstein 2004 – e.g. regarding the spellout of binders, the
islandhood of wh-infinitives, the functional (?) status of raising verbs, the effect of θ-features on
equi-distance – are just that: suggestions, unaccompanied by arguments.
21
Barrie (this volume) challenges the uniform OC status of wh-complements, claiming that in generic
contexts the embedded null subject is progen, whereas in non-generic contexts it is a trace of the con-
troller. Landau 2000 argued that there are no genuine examples of NOC in wh-complements. Even
the oneself-test does not reliably diagnose NOC (at most, it permits partial control).
i. *Maryi wasn’t sure when to introduce oneself to heri.
Commenting on (i), Barrie claims that its ungrammatical status has nothing to do with control. Instead,
it is related to the fact that definite pronouns do not occur felicitously in generic statements. Anticipat-
ing and refuting this claim, however, Landau (2000:40) already observed that examples like (i) become
perfect when the complement clause, still generic, is made finite, with the indefinite subject one (ii). This
strongly suggests that the problem in (i) is unrelated to genericity; rather, the embedded PRO is partially
controlled by Mary, inducing a violation of condition B.
ii. Maryi wasn’t sure when one should introduce oneself to heri.
To substantiate his account of (i), Barrie argues that when the pronoun is construed generically, true
NOC is possible, citing (iii).
iii. A princessi shouldn’t have to explain how to introduce oneself to heri.
However, notice that explain, unlike sure, introduces a goal argument. It is this implicit argument,
generically bound in (iii), that controls PRO in this example, again an instance of OC. As far as I can
see, then, the OC status of wh-complements stands firm.
22
Why the active source of these passive ECM constructions is ungrammatical is an old puzzle (see
Postal 1974; Pesetsky 1995; Bošković 1997).
23
Landau 2003 suggested that (58a) is blocked by the Ban on Improper Movement. In recent
terms, this would be covered by the Phase Impenetrability Condition, which makes [Spec,TP]
inaccessible under a phasal CP. A simpler view, however, would be that a lexical subject (or its
copy) is not licensed to begin with in the infinitival complement of control verbs, because it
would fail to check off certain uninterpretable features that characterize T and C in these clauses
(see Landau 2004 for details).
24
Recently, findings from ERP studies revealed a systematic difference between the syntactic processing
of raising and control constructions (Featherston et al. 2000), in line with the conclusions of
Landau 2003 and this paper.
25
Postal argues for a third one, namely, that subjects of middle predicates can be raised, but not
controlled. However, the actual scope, and ultimate source, of this putative contrast are less
obvious to me.
26
Notice that there is nothing wrong about treating companies as agents or attitude-holders; compare
(70b) with Microsoft1 claimed that its1 stock price would go up.
BIBLIOGRAPHY
Abney, Steven. 1987. The English noun phrase in its sentential aspect. MIT dissertation.
Abraham, Werner. 1982. Zur kontroll im deutschen. GAGL1 112–167.
Abraham, Werner. 1983. The control relation in German. In Werner Abraham (ed.), On the formal
syntax of the Westgermania, 217–242. New York: John Benjamins.
Agbayani, Brian, and Ed Zoerner. 2004. Gapping, pseudogapping, and sideward movement. Studia
Linguistica 58:185–211.
Alboiu, Gabriela. 1999. (De)-Focusing and object raising in Romanian. Canadian Journal of Linguistics
44:1–22.
Alboiu, Gabriela. 2002. The features of movement in Romanian. Bucharest: The University of Bucharest
Press.
Alboiu, Gabriela. 2003. Operator asymmetries in Romanian: Syntax and/or phonology. In Ana
Teresa Pérez-Leroux and Yves Roberge (eds.), Romance linguistics: Theory and acquisition, 3–18.
Amsterdam: John Benjamins.
Alboiu, Gabriela. 2004a. Optionality at the interface: Triggering focus in Romanian. In Henk van
Riemsdijk and Anne Breitbarth (eds.), Triggers, 49–75. Berlin: Mouton de Gruyter.
Alboiu, Gabriela. 2004b. Shared arguments in control. Toronto Working Papers in Linguistics 22:53–74.
Alboiu, Gabriela. 2006. Are we in agreement? In Cedric Boeckx (ed.), Agreement systems, 13–39.
Alboiu, Gabriela, and Virginia Motapanyane. 2000. The generative approach to Romanian grammar:
An overview. In Virginia Motapanyane (ed.), Comparative studies in Romanian syntax, 1–48.
New York: Elsevier.
Alexiadou, Artemis, and Elena Anagnostopoulou. 1997. Notes on ECM, control, and raising. ZAS
Working Papers in Linguistics 8:17–27.
Alexiadou, Artemis, and Elena Anagnostopoulou. 1998. Parametrizing AGR: Word Order,
V-movement, and EPP-checking. Natural Language and Linguistic Theory 16:491–539.
Alexiadou, Artemis, and Elena Anagnostopoulou. 2002. Raising without infinitives and the role of
agreement. In Artemis Alexiadou, Elena Anagnostopoulou, Sjef Barbiers, and Hans-Martin
Gärtner (eds.), Dimensions of movement, 17–30. Amsterdam: John Benjamins.
Andrews, Avery D. 1971. Case agreement of predicate modifiers in Ancient Greek. Linguistic Inquiry
2:127–151.
Andrews, Avery D. 1976. The VP-complement analysis in Modern Icelandic. In Proceedings of NELS
6:1–21. Reprinted in Maling, Joan, and Annie Zaenen (eds.), Modern Icelandic syntax, Syntax and
Semantics 24, 165–185. New York: Academic Press.
Aoun, Joseph, and Y.-H. Audrey Li. 2003. Essays on the representational and derivational nature of
grammar: The diversity of Wh-constructions. Cambridge, MA: MIT Press.
Avrutin, Sergey, and Kenneth Wexler. 1999/2000. Children’s knowledge of subjunctive clauses:
Obviation, binding, and reference. Language Acquisition 8:69–102.
Babby, Leonard. 1998. Subject control as direct predication: Evidence from Russian. Proceedings of
the 6th Meeting of Formal Approaches to Slavic Linguistics, 17–37. Michigan University: Michigan
Slavic Publications.
Babby, Leonard H., and Steven Franks. 1998. The syntax of adverbial participles in Russian revisited.
Slavic and East European Journal 42:117–149.
Babyonyshev, Maria, Jennifer Ganger, David Pesetsky, and Kenneth Wexler. 2001. The maturation of
grammatical principles: Evidence from Russian unaccusatives. Linguistic Inquiry 32:1–44.
Babyonyshev, Maria, and Dina Brun. 2003. The role of aspect in the acquisition of passive constructions
in Russian. Paper presented at GALA, Utrecht, The Netherlands.
Badawi, El-Said, Mike G. Carter, and Adrian Gully. 2004. Modern written Arabic: A comprehensive
grammar. London: Routledge.
327
328 BIBLIOGRAPHY
Badecker, William, and Kathleen Straub. 2002. The processing role of structural constraints on the
interpretation of pronouns and anaphors. Journal of Experimental Psychology: Learning, Memory,
and Cognition 28:748–769.
Baltin, Mark, and Leslie Barrett. 2002. The null content of null case. Ms., NYU.
Barbosa, Pilar. 1995. Null subjects. MIT dissertation.
Barrie, Michael. 2006. Control as movement and cyclic linearization. Proceedings of North East Linguistic
Society (NELS) 36 (1):143–153. Amherst: GLSA.
Barrie, Michael, and Christine Pittman. 2004. Partial control and the movement towards movement.
Toronto Working Papers in Linguistics 22:75–92.
Basilico, David. 2003. The topic of small clauses. Linguistic Inquiry 34:1–35.
Becker, Misha. 2004. The search for structure: How learners distinguish raising from control. Handout
from talk given to the Linguistics Department at the University of Maryland.
Becker, Misha. 2005. Raising, control, and the subset principle. In Proceedings of the West Coast Conference
on Formal Linguistics (WCCFL), 24:52–60. Somerville, MA: Cascadilla Proceedings Project.
Becker, Misha. 2006. There began to be a learnability puzzle. In Linguistic Inquiry 37:441–456.
Bejar, Susana, and Diane Massam. 1999. Multiple case checking. Syntax 2:65–79.
Belletti, Adriana. 2004. Aspects of the low IP area. In Luigi Rizzi (ed.), The structure of CP and IP:
The cartography of syntactic structures, vol. 2, 16–52. Oxford: Oxford University Press.
Belletti, Adriana. 2005. Extended doubling and the VP periphery. Probus 17:1–35.
Belletti, Adriana, and Luigi Rizzi. 1988. Psych-verbs and theta-theory. Natural Language and Linguistic
Theory 6:291–352.
Bennis, Hans, and Teun Hoekstra. 1989. Generatieve grammatica. Dordrecht: Foris.
Berwick, Robert, and Amy Weinberg. 1984. The grammatical basis of linguistic performance.
Cambridge, MA: MIT Press.
Bever, Thomas. 1970. The cognitive basis for linguistic structures. In John R. Hayes (ed.), Cognition
and the development of language, 279–362. New York: Wiley.
Bishop, Dorothy, Caroline Adams, and Courtenay Norbury. 2006. Distinct genetic influences on
grammar and phonological short-term memory deficits: Evidence from 6-year-old twins. Genes,
Brain, and Behavior 5:158–169.
Bobaljik, Jonathan D., and Samuel Brown. 1997. Interarboreal operations: Head movement and the
extension requirement. Linguistic Inquiry 28:345–56.
Boeckx, Cedric. 2003a. Islands and chains. Amsterdam: John Benjamins.
Boeckx, Cedric. 2003b. Indirect binding. Syntax 6:213–236.
Boeckx, Cedric and Howard Lasnik. 2006. Intervention and repair. Linguistic Inquiry 37:150–155.
Boeckx, Cedric, and Norbert Hornstein. 2003. Reply to ‘Control is not movement’. Linguistic Inquiry
34:269–280.
Boeckx, Cedric, and Norbert Hornstein. 2004. Movement under control. Linguistic Inquiry 35:431–452.
Boeckx, Cedric, and Norbert Hornstein. 2005. The status of D-structure: The case of binominal each.
Syntax 8:23–43.
Boeckx, Cedric and Norbert Hornstein. 2006. Control in Icelandic and theories of control. Linguistic
Inquiry 37:591–606.
Boeckx, Cedric, and Norbert Hornstein. In press. Control in Icelandic and theories of control.
Linguistic Inquiry.
Boeckx, Cedric, and Norbert Hornstein. To appear. Superiority, reconstruction, and islands. In Robert
Freidin, Carlos P. Otero, and Maria Luisa Zubizaretta (eds.), Fundamental issues in linguistic
theory. Cambridge, MA: MIT Press.
Boeckx, Cedric, Norbert Hornstein, and Jairo Nunes. Forthcoming. Control as movement. Ms., Harvard
University, University of Maryland, and University of São Paulo.
Bolinger, Dwight L. 1961. Syntactic blends and other matters. Language 37:366–381.
Borer, Hagit. 1989. Anaphoric AGR. In Osvaldo Jaeggli and Kenneth J. Safir (eds.), The null subject
parameter, 69–109. Dordrecht: Kluwer.
Borer, Hagit, and Yosef Grodzinsky 1986. Syntactic vs. lexical cliticization: The case of Hebrew dative
clitics. In Hagit Borer (ed.), The syntax of pronominal clitics, Syntax and Semantics 19, 175–217.
San Diego, CA: Academic Press.
Borer, Hagit, and Kenneth Wexler. 1987. The maturation of syntax. In Thomas Roeper and Edwin
Williams (eds.), Parameter setting, 123–172. Dordrecht: Reidel.
Borer, Hagit, and Kenneth Wexler. 1992. Bi-unique relations and the maturation of grammatical
principles. Natural Language and Linguistic Theory 10:147–189.
Borsley, Robert. 2001. What do “prepositional complementizers” do? Probus 13:155–171.
Bošković, Željko. 1994. D-structure, theta-criterion, and movement into theta-positions. Linguistic
Analysis 24:247–286.
BIBLIOGRAPHY 329
Bošković, Željko. 1997. The syntax of nonfinite complementation: An economy approach. Cambridge,
MA: MIT Press.
Bouchard, Denis. 1984. On the content of empty categories. Dordrecht: Foris.
Bowers, John. 2001. Predication. In Mark Baltin and Chris Collins (eds.), Handbook of contemporary
syntactic theory, 299–333. Oxford: Blackwell Publishing.
Bresnan, Joan. 1972. Theory of complementation in English syntax. MIT dissertation.
Bresnan, Joan. 1982. Control and complementation. Linguistic Inquiry 13:343–434.
Bruening, Benjamin. 2001. Syntax at the edge: Cross-clausal phenomena and the syntax of Passamaquoddy.
MIT dissertation.
Burzio, Luigi. 1986. Italian syntax. Dordrecht: Kluwer.
Burzio, Luigi. 1991. The morphological basis of anaphora. Journal of Linguistics 27:81–105.
Byun, Hyuna, and Cho Yongjoon. 2005. The absence of PBC effects in raising to object constructions
in Korean. Paper presented in the LSA workshop on New Horizons in the Syntax of Raising and
Control. Harvard University.
Cairns, Helen S., Dana McDaniel, Jennifer Ryan Hsu, and Michelle Rapp. 1994. A longitudinal study
of principles of control and pronominal reference in child English. Language 70:260–288.
Cameron-Faulkner, Thea, Elena Lieven, and Michael Tomasello. 2003. A construction-based analysis
of child directed speech. Cognitive Science 27:843–873.
Cardinaletti, Anna. 1997. Subjects and clause structure. In Liliane Haegeman (ed.), The new comparative
syntax, 33–63. London and New York: Longman.
Cecchetto, Carlo, and Renato Oniga. 2004. A challenge to null-case theory. Linguistic Inquiry 35:141–
149.
Cheng, Lisa. 2005. Verb copying in Mandarin Chinese. Paper presented at the workshop on The Copy
Theory of Movement on the PF Side. University of Utrecht.
Chien, Yu-Chin, and Kenneth Wexler. 1990. Children’s knowledge of locality conditions in binding as
evidence for the modularity of syntax and pragmatics. Language Acquisition 1:225–295.
Chierchia, Gennaro. 1984. Topics in the syntax and semantics of infinitives and gerunds. University of
Massachusetts, Amherst, dissertation.
Chomsky, Carol. 1969. The acquisition of syntax in children from 5 to 10. Cambridge, MA: MIT Press.
Chomsky, Noam. 1970. Remarks on nominalization. In Roderick Jacobs and Peter Rosenbaum (eds.),
Readings in transformational grammar, 184–221. Waltham, MA: Ginn.
Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris.
Chomsky, Noam. 1986a. Knowledge of language: Its nature, origin, and use. New York: Praeger.
Chomsky, Noam. 1986b. Barriers. Cambridge, MA: MIT Press.
Chomsky, Noam. 1993. A minimalist program for linguistic theory. In Kenneth Hale and Samuel
J. Keyser (eds.), The view from building 20, 1–52. Cambridge, MA: MIT Press.
Chomsky, Noam. 1995. The minimalist program. Cambridge, MA: MIT Press.
Chomsky, Noam. 1998. Minimalist enquires: The framework. In R. Martin, D. Michaels, and
J. Uriagereka, eds, step by step: Essays on Minimalist Syntax in Honour of Howard Lasnik,
Cambridge, Mass: MIT Press.
Chomsky, Noam. 2000. Minimalist inquiries: The framework. In Roger Martin, David Michaels, and
Juan Uriagereka (eds.), Step by step: Essays on minimalist syntax in honor of Howard Lasnik,
89–155. Cambridge, MA: MIT Press.
Chomsky, Noam. 2001a. Derivation by phase. In Michaels Kenstowicz (ed.), Ken Hale: A life in
language, 1–52. Cambridge, MA: MIT Press.
Chomsky, Noam. 2001b. Beyond explanatory adequacy. MIT Occasional Papers in Linguistics 20:1–28.
Chomsky, Noam. 2004. Beyond explanatory adequacy. In Adrianna Belletti (ed.), Structures and
beyond – The cartography of syntactic structures, vol. 3, 104–131. Oxford: Oxford University Press.
Chomsky, Noam. 2005. On phases. Ms., MIT.
Chomsky, Noam. 2006. Approaching UG from below. Ms., MIT.
Chomsky, Noam, and Howard Lasnik. 1993. The theory of principles and parameters. In The
minimalist parameter, 13–128. Cambridge, MA: MIT Press.
Chung, Dae-Ho. 2004. What does bare – ko coordination say about postverbal morphology in Korean?
Lingua 115:549–568.
Cinque, Gugliemo. 1991. Types of A-bar dependencies. Cambridge, MA: MIT Press.
Cinque, Gugliemo. 2004. ‘Restructuring’ and functional structure. In Adriana Belletti (ed.),
Structures and beyond: The cartography of syntactic structures, vol. 3, 132–192. Oxford:
Oxford University Press.
Collins, Chris. 1991. Why and how come. In Lisa Cheng and Hamida Demirdache (eds.), More papers
on WH-Movement, MIT Working Papers in Linguistics 15:31–45.
Collins, Chris. 2005. A smuggling approach to raising in English. Linguistic Inquiry 36:289–298.
330 BIBLIOGRAPHY
Comrie, Bernard. 1974. The second dative: A transformational approach. In Richard Brecht and
Catherine Chvany (eds.), Slavic transformational syntax, 123–150. Michigan Slavic Materials No. 10.
Michigan University Press.
Cormack, Annabel, and Neil Smith. 2002. Compositionality, copy theory, and control. UCL Working
Papers in Linguistics 14:355–373.
Cormack, Annabel, and Neil Smith. 2003. Backward control in Korean and Japanese. UCL Working
Cornilescu, Alexandra. 2000a. The double-subject construction in Romanian. In Virginia Motapanyane
(ed.), Comparative studies in Romanian syntax, 83–134. Dordrecht: Elsevier.
Cornilescu, Alexandra. 2000b. Notes on the interpretation of the prepositional accusative in Romanian.
University of Bucharest Working Papers in Linguistics 2:91–107.
Costa, Jõao. 1999. Word order variation. A constraint based approach. The Hague: Holland Academic
Graphics.
Costa, Jõao, and Sandra Pereira. 2005. Phases and autonomous features: A case of mixed agreement
in European Portuguese. In Martha McGinnis and Norvin Richards (eds.), Perspectives on phases,
MIT Working Papers in Linguistics 49:115–124.
Crawford, Jean. 2005. The acquisition of the Sesotho passive: Reanalyzing a counterexample to
maturation. Ms., Boston University.
Culicover, Peter W., and Ray Jackendoff. 2001. Control is not movement. Linguistic Inquiry 32:493–512.
D’Alessandro, Roberta. 2004. Impersonal si constructions: Agreement and interpretation. Stuttgart
University dissertation.
Darzi, Ali. 1996. Word order, NP-movements, and opacity conditions in Persian. University of Illinois,
Urbana-Champaign dissertation.
Davies, William D. 2005. Madurese prolepsis and its implications for a typology of raising. Language
81:645–665.
Davies, William D., and Stanley Dubinsky. 2004. The grammar of raising and control. Oxford: Blackwell
Publishing.
Déchaine, Rose-Marie, and Martina Wiltschko. 2002. Decomposing pronouns. Linguistic Inquiry
33:409–442.
Deprez, Viviane. 1992. Raising constructions in Haitian Creole. Natural Language and Linguistic
Theory 10:191–232.
Diesing, Molly. 1992. Indefinites. Cambridge, MA: MIT Press.
Dikken, Marcel den. 1998. Predicate inversion in DP. In Artemis Alexiadou and Chris Wilder (eds.),
Possessors, predicates, and movement in the determiner phrase. Amsterdam: John Benjamins.
Dobrovie-Sorin, Carmen. 1994. The syntax of Romanian. Berlin: Mouton de Gruyter.
Dobrovie-Sorin, Carmen. 2001. Head-to-head merge in Balkan subjunctives and locality. In María-Luisa
Rivero and Angela Ralli (eds.), Comparative syntax of Balkan languages, 44–74. Oxford: Oxford
University Press.
Doron, Edit, and Caroline Heycock. 1999. Filling and licensing multiple specifiers. In David Adger,
Susan Pintzuk, Bernadette Plunkett, and George Tsoulas (eds.), Specifiers, 69–89. Oxford: Oxford
University Press.
Epstein, Samuel. 1999. Un-principled syntax and the derivation of syntactic relations. In Samuel Epstein
and Norbert Hornstein (eds.), Working Minimalism, 317–345. Cambridge, MA: MIT Press.
Epstein, Samuel. 2001. Deriving the proper binding constraint. Ms., University of Michigan.
Epstein, Samuel, Erich Groat, Ruriko Kawashima, and Hisatsugu Kitahara. 1998. A derivational
approach to syntactic relations. Oxford: Oxford University Press.
Epstein, Samuel, and T. Daniel Seely. 1999. SPEC-ifying the GG subject: Eliminating A-chains and
the EPP within a derivational model. Paper presented at The Role of Grammatical Functions in
Transformational Grammar (1999 LSA Linguistic Institute Workshop), University of Illinois.
Epstein, Samuel, and T. Daniel Seely. 2002. Derivation and explanation in the minimalist program.
Oxford: Blackwell Publishing.
Epstein, Samuel, and T. Daniel Seely. 2006. Derivations in minimalism. Cambridge: Cambridge Uni-
versity Press.
Erteschik-Shir, Nomi. 1997. The dynamics of focus structure. Cambridge: Cambridge University Press.
Fanselow, Gisbert. 2004. The MLC and derivational economy. In Arthur Stepanov, Gisbert Fanselow,
and Ralf Vogel (eds.), Minimality effects in syntax, 73–124. Berlin: Mouton de Gruyter.
Farkas, Donka. 1985. Intensional descriptions and the Romance subjunctive mood. New York: Garland
Publishing.
Farkas, Donka. 1988. On obligatory control. Linguistics and Philosophy 11:27–58.
Farkas, Donka. 1992. On the semantics of subjunctive complements. In Paul Hirschbühler and
Konrad Koerner (eds.), Papers from the 20th linguistic symposium on Romance languages, 69–105.
BIBLIOGRAPHY 331
Farrell, Patrick. 1995. Backward control in Brazilian Portuguese. In Janet Fuller, Ho Han, and David
Parkinson (eds.), Proceedings of Eastern States Conference on Linguistics (ESCOL 94), 116–127.
Ithaca, NY: Department of Linguistics, Cornell University.
Featherston, Samuel, Matthias Gross, Thomas F. Münte, and Harald Clahsen. 2000. Brain potentials
in the processing of complex sentences: An ERP study of control and raising constructions.
Journal of Psycholinguistic Research 29:141–154.
Ferreira, Madalena. 2000. Argumentos nulos em Português Brasileiro. MA thesis, Universidade
Estadual de Campinas.
Ferreira. Madalena. 2004. Hyperraising and null subjects in Brazilian Portuguese. Ms., MIT.
Fox, Danny, and Yosef Grodzinsky. 1998. Children’s passive: A view from the by-phrase. Linguistic
Inquiry 29:311–332.
Fox, Danny, and David Pesetsky. 2005. Cyclic linearization of syntactic structure. Theoretical
Linguistics 31:1–46.
Frampton, John, Sam Gutmann, Julie Legate, and Charles Yang. 2000. Remarks on ‘Derivation by
phase’: Feature valuation, agreement, and intervention. Ms., Northeastern University.
Franks, Steven. 1990. Case, configuration, and argumenthood: Reflections on the second dative.
Russian Linguistics 14:231–254.
Franks, Steven, and Norbert Hornstein. 1992. Secondary predication in Russian and proper
government of PRO. In Richard Larson, Sabine Iatridou, Utpal Lahiri, and James Higginbotham
(eds.), Control and grammar, 1–50. Dordrecht: Kluwer Academic Publishers.
Froud, Karen, Kenneth Wexler, and Vina Tsakali. In preparation. The development of raising.
Ms., MIT.
Fukui, Naoki. 1999. An A-over-A perspective on locality. In Enoch Iwamoto and Masatake Muraki
(eds.), Linguistics: In search of the human mind, 109–129. Tokyo: Kaitakusha.
Ganger, Jennifer, Sabrina Dunn, and Peter Gordon. 2004. Genes take over when the input fails: A twin
study of the passive. Poster presented at 28th Annual Boston University Conference on Language
Development (BUCLD 28).
George, Leland M., and Jaklin Kornfilt. 1981. Finiteness and boundedness in Turkish. In Frank Heny
(ed.), Binding and filtering, 105–127. London: Croom Helm.
Giannakidou, Anastasia. 1998. Polarity sensitivity as (non)veridical dependency. Amsterdam: John
Benjamins.
Gibson, Edward. 1998. Linguistic complexity: Locality of syntactic dependencies. Cognition 68:1–76.
Gonzalez, Nora. 1988. Object raising in Spanish. New York: Thomas Y. Crowell.
Gonzalez, Nora. 1990. Unusual inversion in Chilean Spanish. In Paul Postal and Brian Joseph (eds.),
Studies in relational grammar, 87–103. Chicago IL: University of Chicago Press.
Goodluck, Helen, Arhonto Terzi, and Gema Chocano Díaz. 2001. The acquisition of control
crosslinguistically: Structural and lexical factors in learning to license PRO. Journal of Child
Language 28:153–172.
Gordon, Peter, and Jill Chafetz. 1990. Verb-based versus class-based accounts of actionality effects in
children’s comprehension of passives. Cognition 36:227–254.
Gordon, Peter, Randall Hendrick, and Marcus Johnson. 2001. Memory interference during language
processing. Journal of Experimental Psychology: Learning, Memory and Cognition 27:1411–1423.
Grimshaw, Jane. 1990. Argument structure. Cambridge, MA: MIT Press.
Grinder, John T. 1970. Super equi-NP deletion. Papers from the Regional Meeting of the Chicago
Linguistic Society (CLS) 6:297–317.
Grosu, Alex, and Julia Horvath. 1987. On nonfiniteness in extraction constructions. Natural Language
and Linguistic Theory 5:181–196.
Guasti, Maria. 2002. Language acquisition: The growth of grammar. Cambridge, MA: MIT Press.
Guillaume, Gustave. 1929. Temps et verbe. Paris: Masson.
Hadjivassiliou, Angela, Irene Philippaki-Warburton, and Vassilios Spyropoulos. 2000. Greek
ECM constructions revisited. In Katerina Nicolaidis and Marina Mattheoudakis (eds.), 13th
International Symposium on Theoretical and Applied Linguistics, 70–80.
Haegeman, Liliane. 1990. Non-overt subjects in diary contexts. In Joan Mascaró and Marina Nespor
(eds.), Grammar in progress, 167–174. Dordrecht: Foris.
Haegeman, Liliane, and Henk van Riemsdijk. 1986. Verb projection raising, scope, and the typology
of rules affecting verbs. Linguistic Inquiry 17:417–466.
Hamburger, Henry, and Stephen Crain. 1982. Relative acquisition. In Stan A. Kuczaj (ed.), Language
development Vol. I: Syntax and semantics, 245–274. Hillsdale, NJ: Lawrence Erlbaum.
Han, Chung-Hye, and Jong-Bok Kim. 2004. Are there ‘double relative clauses’ in Korean? Linquistic
Inquiry 35:315–337.
Harley, Heidi, and Elizabeth Ritter. 2002. Person and number in pronouns: A feature geometric
analysis. Language 78:482–526.
332 BIBLIOGRAPHY
Heim, Irene, Howard Lasnik, and Robert May. 1991. Reciprocity and plurality. Linguistic Inquiry
22: 63–101.
Heycock, Caroline. 1993. Syntactic predication in Japanese. Journal of East Asian Linguistics 2:167–211.
Heycock, Carolilne, and Edit Doron. 2003. Categorical subjects. Gengo Kenkyuu 123: 95–135.
Higginbotham, James. 1989. Reference and control. Paper presented at the MIT Control Conference.
Hill, Virginia. 2002. Adhering focus. Linguistic Inquiry 33:164–172.
Hill, Virginia. 2003. Reduced CP fields and phases. University of Bucharest Working Papers in
Hiraiwa, Ken. 2002. Indeterminate agreement and raising in Japanese. Paper presented at the 21st
Annual Meeting of West Coast Conference on Formal Linguistics (WCCFL 21).
Hirsch, Christopher, and Jeremy Hartman. 2005. Some (wh) questions concerning passive interactions.
Paper presented at Generative Approaches to Language Acquisition 2005, Università degli Studi
di Siena.
Hirsch, Christopher, and Jeremy Hartman. 2006. Semantic constraints on passive comprehension:
Evidence for a theory of sequenced acquisition. Paper presented at Latsis Colloquium of the
University of Geneva 2006.
Hirsch, Christopher, Nadya Modyanova, and Kenneth Wexler. 2006. Predicting passive acquisition:
Environmental factors and biological maturation. Paper presented at the 22nd Scandinavian
Conference of Linguistics, Workshop in Language Acquisition, Aalborg, Denmark 2006.
Hirsch, Christopher, Robyn Orfitelli and Kenneth Wexler. 2006. When seem means think – The role
of the experiencer-phrase in children’s comprehension of raising. Paper presented at Generative
Approaches to Language Acquisition North America 2006, McGill University.
Hirsch, Christopher, and Kenneth Wexler. 2004a. Children’s passives and their resulting interpreta-
tion. Paper presented at Generative Approaches to Language Acquisition North America 2004,
University of Hawai’i at Manoa.
Hirsch, Christopher, and Kenneth Wexler. 2004b. ‘Who’ seems to rescue raising. Paper presented at
Generative Approaches to Language Acquisition North America 2004, University of Hawai’i at
Manoa.
Hirsch, Christopher, and Kenneth Wexler. 2005. By the way, children don’t know by. Paper presented
at the 29th Annual Boston University Conference on Language Development 2005.
Hirsch, Christopher, and Kenneth Wexler. 2006. Acquiring verbal passives: Evidence for a
maturational account. Paper presented at the annual meeting of the Linguistic Society of America,
Albuquerque.
Hoji, Hajime. 1985. Logical form constraints and configurational structures in Japanese. University of
Washington dissertation.
Hoji, Hajime. 1991. Raising-to-object, ECM, and the major object in Japanese. Paper presented at the
Japanese GrammarWorkshop. University of Rochester.
Hoji, Hajime. 2005. Major object analysis of the so-called raising-to-object constructions in Japanese.
Paper presented at New Horizons in the Grammar of Raising and Control Workshop. LSA Sum-
mer Institute, Harvard University.
Holmberg, Anders. 2005. Is there a little pro? Evidence from Finnish. Linguistic Inquiry 36:533–564.
Holmberg, Anders, and Urpo Nikanne. 2002. Expletives, subjects, and topics in Finnish. In Peter
Svenonius (ed.), Subjects, expletives, and the EPP, 71–105. Oxford: Oxford University Press.
Holmberg, Anders. To appear. The empty generic subject pronoun in Finnish. In Satu Manninen,
Diane Nelson, Katrin Hiietam, Elsi Kaiser, and Virve Vihman (eds.), Passives and impersonals in
European languages. Amsterdam: John Benjamins.
Holton, David, Peter Mackridge, and the Irene Philippaki-Warburton. 1997. Greek: A comprehensive
grammar of the modern language. London: Routledge.
Hong, Ki-Sun. 1990. Subject-to-object raising in Korean. In Katarzyna Dziwirek, Patrick Farrell,
and Errapel Mejías-Bikandi (eds.), Grammatical relations: A crosslinguistic perspective, 215–225.
Stanford: CSLI Publications.
Hong, Ki-Sun. 1997. Yenge-wa kwuke-uy insang kwumwnun pikyo pwunsek (Subject-to-object raising
constructions in English and Korean). Language Research 33:409–434.
Hong, Soo-Min. 2005. “Exceptional” case-marking and resultative constructions. University of
Maryland, College Park, dissertation.
Horgan, Dianne. 1978. The development of the full passive. Journal of Child Language 5:65–80.
Hornstein, Norbert. 1990. As time goes by. Cambridge, MA: MIT Press.
Hornstein, Norbert. 1995. Logical form. Oxford: Blackwell Publishing.
Hornstein, Norbert. 1998. Adjunct control and parasitic gaps. University of Maryland Working Papers
in Linguistics 6:102–121.
Hornstein, Norbert. 1999. Movement and control. Linguistic Inquiry 30:69–96.
BIBLIOGRAPHY 333
Hornstein, Norbert. 2000. On A-chains: A reply to Brody. Syntax 3:129–143.

Hornstein, Norbert. 2001: Move! A Minimalist theory of construal. Oxford: Blackwell Publishing.
Hornstein, Norbert. 2003. On control. In Hendrick, Randall (ed.), Minimalist syntax, 6–81. Oxford:
Blackwell Publishing.
Hornstein, Norbert. 2005a. Deriving c-command. Ms., University of Maryland, College Park.
Hornstein, Norbert. 2005b. What do labels do. Ms., University of Maryland, College Park.
Hornstein, Norbert. Forthcoming. Pronouns in a minimalist setting. University of Maryland Working
Papers in Linguistics.
Hornstein, Norbert, Howard Lasnik, and Juan Uriagereka. In press. The dynamics of islands: Specu-
lations on the locality of movement. Linguistic Analysis.
Householder, Fred W., Kostas Kazazis, and Andreas Koutsoudas 1964. Reference grammar of literary
Dhimotiki. Bloomington: Indiana University Press.
Huang, C.-T. James. 1982. Logical relations in Chinese and the theory of grammar. MIT dissertation.
Huang, Yan. 2000. Anaphora. Oxford: Oxford University Press.
Hyams, Nina, and William Snyder. 2005. Young children never smuggle: Reflexive clitics and the
universal freezing hypothesis. Paper presented at the 29th Annual Boston University Conference
on Language Development (BUCLD 29).
Hyams, Nina, Dimitrios Ntelitheos, and Cecile Manorohanta. 2006. The acquisition of the Malagasy
voicing system. Ms., UCLA.
Iatridou, Sabine. 1993. On nominative case assignment and a few related things. MIT Working Papers
in Linguistics 19:175–196.
Ingria, Robert. 1981. Sentential complementation in Modern Greek. MIT dissertation.
Isac, Dana. 2002. The force of negative mood. Paper presented at the Annual Meeting of the Linguistic
Society of America, San Francisco.
Ito, Masuyo, and Kenneth Wexler. 2002. The acquisition of Japanese unaccusatives. Ms., MIT.
Jackendoff, Ray, and Peter W. Culicover. 2003. The semantic basis of control in English. Language
79:517–556.
Jackendoff, Ray. 1990. Semantic structures. Cambridge, MA: MIT Press.
Jacobson, Pauline. 1990. Raising as function composition. Linguistics and Philosophy 13:423–475.
Joseph, Brian. 1976. Raising in Modern Greek: A copying process? In Judith Aissen and Jorge
Hankamer (eds.), Harvard studies in syntax and semantics 2, 241–78. Cambridge, MA: Harvard
University Press.
Joseph, Brian D. 1983. The synchrony and diachrony of the Balkan infinitive: A study in areal, general,
and historical linguistics. Cambridge: Cambridge University Press.
Joseph, Brian D. 1992. Diachronic perspectives on control. In Richard K. Larson, Sabine Iatridou,
Utpal Lahiri, and James Higginbotham (eds.), Control and grammar, 195–234. Dordrecht:
Kluwer.
Joseph, Brian D., and Irene Philippaki-Warburton. 1987. Modern Greek. London: Croom Helm.
Kakouriotis, Athanasios. 1980. Raising in Modern Greek. Lingua 52:157–177.
Kapetangianni, Konstantia. 2004. Subjunctive na clauses in Modern Greek: A challenge for
Minimalism, MA thesis. Eastern Michigan University.
Kapetangianni, Konstantia, and T. Daniel Seely. 2003. On control in Modern Greek: Explanation by
deduction: Agreement by degrees. Ms., University of Michigan and Eastern Michigan University.
Kawasaki, Noriko. 1993. Control and arbitrary interpretation in English. University of Massachusetts,
Amherst, dissertation.
Kayne, Richard. 1984. ECP extensions. Connectedness and binary branching. Dordrecht: Foris.
Kayne, Richard. 1989. Facets of Romance past participle agreement. In Paola Benincà (ed.), Dialect
variation in the theory of grammar, 85–104. Dordrecht: Foris.
Kayne, Richard. 1991. Romance clitics, verb movement, and PRO. Linguistic Inquiry 22:647–686.
Kayne, Richard. 1993. Towards a modular theory of auxiliary selection. Reprinted in Richard Kayne
(ed.), 2000. Parameters and universals. Oxford: Oxford University Press.
Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press.
Kayne, Richard. 1999. Prepositional complementizers as attractors. Probus 11(1):39–73.
Kayne, Richard. 2000. Parameter and universals. Oxford: Oxford University Press.
Kayne, Richard. 2002. Pronouns and their antecedents. In Samuel D. Epstein and T. Daniel Seely (eds.),
Derivation and explanation in the minimalist program, 133–166. Oxford: Blackwell Publishing.
Kayne, Richard. 2004. Prepositions as probes. In Adriana Belletti (ed.), Structures and beyond: The
cartography of syntactic structures, vol. 3, 192–212. Oxford: Oxford University Press.
Kempchinsky, Paula. 1986. Romance subjunctive clauses and logical form. UCLA dissertation.
Kiguchi, Hirohisa. 2000. PRO-gate and movement. Ms., University of Maryland, College Park.
Kiguchi, Hirohisa. 2004. Syntax unchained. University of Maryland, College Park, dissertation.
334 BIBLIOGRAPHY
Kiguchi, Hirohisa, and Norbert Hornstein. 2001. PRO-gate and sideward movement. Ms., University
of Maryland, College Park.
Kiguchi, Hirohisa, and Norbert Hornstein. 2002. PRO-gate. Ms., University of Maryland, College Park.
Kim, Soo-Won, and Joan Maling. 1993. Syntactic case and frequency adverbials. Harvard Studies in
Korean Linguistics 5:368–378.
Kim, Yeng-Hee. 1985. Cwu.e olliki [Subject raising]. Kwukehak [Korean Linguistics] 14:337–360.
Kim, Young-Joo. 1990. The syntax and semantics of Korean case: The interaction between lexical and
syntactic levels of representation. Harvard University dissertation.
Ko, Heejeong 2005a. Syntactic edges and linearization. MIT dissertation.
Ko, Heejeong. 2005b. Syntax of why-in situ: Merge into [Spec, CP] in the overt syntax. Natural
Language and Linguistic Theory 23:867–916.
Koizumi, Masatoshi. 1993. Object agreement phrases and the split VP hypothesis. MIT Working
Papers in Linguistics 18: 99–148.
Koster, Jan. 1984. On binding and control. Linguistic Inquiry 15:417–459.
Koster, Jan, and Robert May. 1982. On the constituency of infinitives. Language 58:117–143.
Kotzoglou, George. 2002. Greek ECM and how to control it. Reading Working Papers in Linguistics
6:39–56.
Krapova, Iliyana. 2001. Subjunctives in Bulgarian and Modern Greek. In María-Luisa Rivero and
Angela Ralli (eds.), Comparative syntax of the Balkan languages, 105–126. Oxford: Oxford
University Press.
Krapova, Iliyana, and Vassil Petkov. 1999. Subjunctive complements, null subjects and case-checking
in Bulgarian. In Proceedings of the Meeting of Formal Approaches to Slavic linguistics, 7:267–285.
Ann Arbor: University of Michigan.
Kratzer, Angelica. 1995. Stage-level and individual-level predicates. In Gregory N. Carlson and Francis
Jeffry Pelletier (eds.), The generic book, 125–175. Chicago, IL: University of Chicago Press.
Krifka, Manfred, Francis Jeffry Pelletier, Gregory N. Carlson, Alice Ter Meulen, Gennaro Chierchia,
and Godehard Link. 1995. Genericity: An introduction. In Gregory N. Carlson and Francis Jeffry
Pelletier (eds.), The generic book, 1–124. Chicago, IL: University of Chicago Press.
Kuno, Susumu. 1973. The structure of the Japanese language. Cambridge, MA: MIT Press.
Kuno, Susumu. 1975. Super equi-NP deletion is a pseudo-transformation. Proceedings of the North
East Linguistic Society (NELS) 5:29–44.
Kuno, Susumu. 1976. Subject raising. In Masayoshi Shibatani (ed.), Japanese generative grammar,
Syntax and Semantics 5, 17–29. New York: Academic Press.
Kuroda, Shige-Yuki. 1972. The categorical judgment and the thetic judgment. Foundations of Language
9:153–185.
Ladusaw, William. 1994. Thetic and categorical, stage and individual, weak and strong. In Proceedings
of SALT-IV, 220–229. Cornell University: CLC Publications.
Landau, Idan. 1998. Projection and bounding in possessor raising. In Kimary Shahin, Susan Blake,
and Eun-Sook Kim (eds.), Proceedings of the West Coast Conference on Formal Linguistics
(WCCFL) 17:405–419. Stanford: CSLI.
Landau, Idan. 1999. Elements of control. MIT dissertation.
Landau, Idan. 2000. Elements of control: Structure and meaning in infinitival constructions. Dordrecht:
Kluwer.
Landau, Idan. 2001. Control and extraposition: The case of super-equi. Natural Language and
Linguistic Theory 19:109–152.
Landau, Idan. 2002. The scale of finiteness and the calculus of control. Ms., Ben Gurion University.
Landau, Idan. 2003. Movement out of control. Linguistic Inquiry 34: 471–498.
Landau, Idan. 2004a. The scale of finiteness and the calculus of control. Natural Language and
Landau, Idan. 2004b. Movement resistant aspects of control. In http://www.bgu.ac.il/idanl/files/
Language.revised.mine.pdf
Landau, Idan. 2006. Severing the distribution of PRO from case. Syntax 9:153–170.
Larson, Richard. 1991. Promise and the theory of control. Linguistic Inquiry 22:103–139.
Lasnik, Howard. 1999. Minimalist analysis. Oxford: Blackwell Publishers.
Lebeaux, David. 1984. Anaphoric binding and the definition of PRO. Proceedings of the North East
Linguistic Society (NELS) 14:253–274.
Lee, Hyeonjin, and Kenneth Wexler. 2001. Nominative case omission and unaccusatives in Korean
acquisition. In Hee-Don Ahn and Namkil Kim (eds.), Selected Papers from the 12th International
Conference on Korean Linguistics, 263–279. Seoul: Kjungjin Publishing.
Lee, Jeong-Shik. 1992. Case alternations in Korean: Case minimality. University of Connecticut
dissertation.
BIBLIOGRAPHY 335
Lee, Pil-Young. 1992. Wanhyeng pomwun-eyse-uy cwu.e insang-ey tayha.ye [On subject raising from
closed complements]. Swulyen Emwun Noncip [Swulyen Studies in Language and Literature]
19:139–171.
Lee, Wooseung. 2003. Tense-shared coordination constructions in Korean: Implications for morphosyntax.
Harvard Studies in Korean Linguistics 10:609–620.
Legate, Julie. 2003. Some interface properties of the phase. Linguistic Inquiry 34:506–516.
Li, Y.-H. Audrey. 1999. Plurality in a classifier language. Journal of East Asian Linguistics 8:75–99.
Lin, Jo-Wang. 1992. The syntax of zenmeyang ‘how’ and weishenme ‘why’ in Mandarin Chinese.
Journal of East Asian Linguistics 1:293–331.
Lyngfelt, Benjamin. 1999. Optimal control: An OT perspective on the interpretation of PRO in
Swedish. Working Papers in Scandinavian Syntax 63:75–104.
McDaniel, Dana, Helen S. Cairns, and Jennifer Ryan Hsu. 1991. Control principles in the grammars
of young children. Language Acquisition 1:121–138.
MacWhinney, Brian. 2000. The CHILDES project: Tools for analyzing talk. Third edition. Mahwah,
NJ: Lawrence Erlbaum Associates.
Manzini, M. Rita. 1983. On control and control theory. Linguistic Inquiry 14:421–446.
Manzini, M. Rita. 1992. Locality: A theory and some of its empirical consequences. Cambridge,
MA: MIT Press.
Manzini, M. Rita, and Anna Roussou. 1999. A minimalist theory of A-movement and control. In
Corinne Iten and Ad Neeleman (eds.), UCL Working Papers in Linguistics 11:403–442.
Manzini, M. Rita, and Anna Roussou. 2000. A minimalist theory of A-movement and control. Lingua
110: 409–447.
Maratsos, Michael, and Rona Abramovitch. 1975. How children understand full, truncated, and
anomalous passives. Journal of Verbal Learning and Verbal Behavior 14:45–157.
Maratsos, Michael, Dana Fox, Judith Becker, and Mary Anne Chalkley. 1985. Semantic restrictions
on children’s passives. Cognition 19:167–191.
Maratsos, Michael, Stan Kuczaj, Dana Fox, and Mary Anne Chalkley. 1979. Some empirical studies in the
acquisition of transformational relations: Passives, negatives, and the past tense. In W. Andrew Collins
(ed.), Children’s language and communications, 1–45. Hillsdale, NJ: Lawrence Erlbaum Associates.
Martin, Roger A. 1996. A minimalist theory of PRO and control. University of Connecticut dissertation.
Martin, Roger A. 2001. Null case and the distribution of PRO. Linguistic Inquiry 32:141–166.
Martins, Anna Maria, and Jairo Nunes. 2005. Raising issues in Brazilian and European Portuguese.
Journal of Portuguese Linguistics 4:53–77.
Massam, Diane. 1985. Case theory and the projection principle. MIT dissertation.
May, Robert. 1977. The grammar of quantification. MIT dissertation.
McDaniel, Dana, and Helen S. Cairns. 1990. Processing and acquisition of control structures by young
children. In Lyn Frazier and Jill De Villiers (eds.), Language processing and language acquisition,
311–323. Dordrechet: Kluwer.
McGinnis, Martha. 1998. Locality in A-movement. MIT dissertation.
Menuzzi, Sergio. 2000. First plural anaphora in Brazilian Portuguese. Chains and constraint interaction
in binding. In Jõao Costa (ed.), Portuguese syntax: New comparative studies, 191–240. Oxford:
Miyagawa, Shigeru. 1989. Structure and case marking in Japanese. Syntax and Semantics 22. San
Diego, CA: Academic Press.
Miyagawa, Shigeru. 2001. The EPP, scrambling, and Wh-in situ. In Michael Kenstowicz (ed.), Ken
Hale: A life in language, 293–338. Cambridge, MA: MIT Press.
Miyamoto, Edson, Kenneth Wexler, Takako Aikawa, and Shigeru Miyagawa. 1999. Case-dropping
and unaccusatives in Japanese acquisition. Proceedings of the 23rd Annual Boston University
Conference on Language Development (BUCLD 23), 443–452. Somerville, MA: Cascadilla Press.
Modesto, Marcello. 2000a. On the identification of null arguments. University of Southern California
dissertation.
Modesto, Marcello. 2000b. Null subjects without ‘rich’ agreement. In Mary A. Kato and
Esmeralda Negrão (eds.), Brazilian Portuguese and the null subject parameter. Frankfurt:
Vervuert-Iberoamericana.
Modesto, Marcello. 2004. Sujeitos nulos em línguas de tópico proeminente. Revista da Abralin 3:121–148.
Modesto, Marcello. 2007. Topic prominence and null subjects. In Theresa Biberauer (ed.), The limits
of syntactic variation. Amsterdam: John Benjamins.
Moore, John. 1998. Turkish copy-raising and A-chain locality. Natural Language and Linguistic Theory
16:149–89.
Morrell, Kenneth S. 1989. Studies on the phrase structure of early Attic prose. Harvard University dis-
sertation.
336 BIBLIOGRAPHY
Motapanyane, Virginia. 1995. Theoretical implications of complementation in Romanian. Padova: Unipress.

Müller, Gereon. 2004. Phrase impenetrability and wh-intervention. In Arthur Stepanov, Gisbert
Fanselow, and Ralf Vogel (eds.), Minimality effects in syntax, 289–326. Berlin: Mouton de Gruyter.
Nakanishi, Kumiko, and Satoshi Tomioka. 2002. On Japanese associative plurals. In Proceedings of
the North East Linguistic Society (NELS) 32:423–439.
Negrão, Esmeralda V., and Evani Viotti. 2000. Brazilian Portuguese as a discourse-oriented language.
In Mary A. Kato and Esmeralda Negrão (eds.), Brazilian Portuguese and the null subject parameter,
105–125. Frankfurt: Vervuert-Iberoamericana.
Neidle, Carol Jan. 1988. The role of case in Russian syntax. Dordrecht: Kluwer.
Nicol, Janet, and David Swinney. 1989. The role of structure in coreference assignment during sentence
processing. Journal of Psycholinguistic Research 18:5–20.
Nunes, Jairo. 1995. The copy theory of movement and linerarization of chains in the minimalist program.
University of Maryland dissertation.
Nunes, Jairo. 1999. Linearization of chains and phonetic realization of chain links. In Samuel D. Epstein
and Norbert Hornstein (eds.), Working minimalism, 217–250. Cambridge, MA: MIT Press.
Nunes, Jairo. 2004. Linearization of chains and sideward movement. Cambridge, MA: MIT Press.
O’Grady, William. 1991. Categories and case: The sentence structure of Korean. Amsterdam: John Benjamins.
O’Neil, John. 1997. Means of control: Deriving the properties of PRO in the minimalist program.
Harvard University dissertation.
Oka, Toshifusa. 1988. Abstract case and empty pronouns. Tsukuba English Studies 7:187–227.
Ordóñez, Francisco. 1998. Post verbal asymmetries in Spanish. Natural Language and Linguistic
Theory 16:313–346.
Osterhout, Lee, and Linda A. Mobley. 1995. Event-related brain potentials elicited by failure to agree.
Journal of Memory and Language 34:739–773.
Papangeli, Dimitra. 2000. Clitic doubling in Modern Greek: A head-complement relation. UCL
Working Papers in Linguistics 12:473–499.
Papangeli, Dimitra. 2004. The morphosyntax of argument realization. LOT Dissertation series
86. Utrecht: Landelijke Onderzoek school Taalwetenschap (Netherlands Graduate School of
Linguistics).
Partee, Barbara H., and Vladimir Borschev. 2003. Genitives, relational nouns, and argument-modifier
ambiguity. In Ewald Lang, Claudia Maienborn, and Cathrine Fabricius-Hansen (eds.), Modifying
adjuncts, 67–112. New York: Mouton de Gruyter.
Pesetsky, D. 1995. Zero syntax. Cambridge, MA: MIT Press.
Pesetsky, David, and Esther Torrego. 2001. T-to-C movement: Causes and consequences. In Michael
Kenstowicz (ed.), Ken Hale: A life in language, 335–426. Cambridge, MA: MIT Press.
Pesetsky, David, and Esther Torrego. 2004a. The syntax of valuation and the interpretability of
features. Ms., MIT.
Pesetsky, David and Esther Torrego. 2004b. Tense, case, and the nature of syntactic categories. In
Jacqueline Gueron and Jacqueline Lacarme (eds.), The syntax of time, 495–539. Cambridge,
MA: MIT Press.
Petter, Marga. 1998. Getting PRO under control. The Hague: Holland Academic Graphics.
Philippaki-Warburton, Irene. 1987. The theory of empty categories and the pro-drop parameter in
Modern Greek. Journal of Linguistics 23:289–318.
Philippaki-Warburton, Irene. 1994. The subjunctive mood and the syntactic status of the particle na in
Modern Greek. Folia Linguistica 28:297–328.
Philippaki-Warburton, Irene. 1998. Functional categories and Modern Greek syntax. The Linguistic
Review 15:158–186.
Philippaki-Warburton, Irene, and Georgia Catsimali. 1999. On control in Greek. In Artemis Alexiadou,
Geoffrey Horrocks, and Melita Stavrou (eds.), Studies in Greek syntax, 153–68. Dordrecht: Kluwer.
Philippaki-Warburton, Irene, David Holton, and Peter Mackridge. 1997. Greek: A comprehensive
grammar of the modern language. London: Routledge.
Philippaki-Warburton, Irene, and Vassilios Spyropoulos. 1997. Provlimata ptosis sto plesio tis theorias
tu elachistu [Case problems in the framework of the minimalist program]. Studies in Greek
Philippaki-Warburton, Irene, and Vassilios Spyropoulos. 1999. On the boundaries of tection and
syntax. The Yearbook of Morphology 1998, 45–72. Dordrecht: Kluwer.
Philippaki-Warburton, Irene, and Vassilios Spyropoulos. 2002. Infinitely intriguing: The question
of finiteness of Greek subjunctives. Paper presented at the Workshop on Greek Syntax: The
Minimalist Seduction, University of Reading.
Philippaki-Warburton, Irene, and Vassilios Spyropoulos. 2004. A change of mood: The development
of the Greek mood system. Linguistics 42:791–817.
BIBLIOGRAPHY 337
Phillips Colin. 1996. Order and structure. MIT dissertation.

Phillips Colin. 2004. The real-time status of island constraints. Ms., University of Maryland, College Park.
Pinker, Steven. 1984. Language learnability and language development. Cambridge, MA: Harvard
University Press.
Pires, Acrisio. 2001. PRO, movement, and binding in Portuguese. Presented at the 30th Linguistic
Symposium on Romance Languages, University of Florida.
Pîrvulescu, Mihaela. 2001. Le concept de paradigme et la morphologie verbale. University of Toronto
dissertation.
Polinsky Maria, and Eric Potsdam. 2002. Backward control. Linguistic Inquiry 33:245–282.
Polinsky, Maria and Eric Potsdam. 2003. Backward control: Evidence from Malagasy. Cornell Work-
ing Papers in Linguistics 19:173–187.
Polinsky, Maria and Eric Potsdam. 2006. Expanding the scope of control and raising. Syntax 9: 171–
192.
Postal, Paul M. 1970. On coreferential complement subject deletion. Linguistic Inquiry 1:439–500.
Postal, Paul. 1974. On raising: One rule of English and its theoretical implications. Cambridge,
MA: MIT Press.
Postal, Paul. 2004. Skeptical linguistics essays. Oxford: Oxford University Press.
Potsdam, Eric, and Jeffrey. Runner. 2001. Richard returns: Copy raising and its implications. Papers
from the Regional Meeting of the Chicago Linguistic Society (CLS) 37:453–468.
Pustejovsky, James. 1988. The geometry of events. In Carol Tenny (ed.), Studies in generative approaches
to syntax. Lexicon Project Working Papers 24:19–40. Cambridge, MA: MIT Press.
Raposo, Eduardo. 1987. Case theory and Infl-to-Comp: The inflected infinitive in European
Portuguese. Linguistic Inquiry 18:85–109.
Raposo, Eduardo, and Juan Uriagereka. 1995. Two types of small clauses: Toward a syntax of theme/
rheme relations. In Anna Cardinaletti and Maria Theresa Guasti (eds.), Small clauses, 179–206.
New York: Academic Press.
Reinhart, Tanya. 2000. The theta system: Syntactic realization of verbal concepts. OTS Working
Papers in Linguistics [revised and published as Reinhart, 2002.]
Reinhart, Tanya. 2002. The theta system – an overview. Theoretical Linguistics 28:229–290.
Reihnart, Tanya, and Tal Siloni. 2005. The lexicon-syntax parameter: Reflexivization and other arity
operations. Linguistic Inquiry 36:389–436.
Reuland, Eric. 1983. Governing -ing. Linguistic Inquiry 14:101–136.
Richards, Norvin. 1997. What moves where when in which language. MIT dissertation.
Richards, Norvin. 1999. Featural cyclicity and the ordering of multiple specifiers. In Samuel D. Epstein
and Norbert Hornstein (eds.), Working minimalism, 127–157. Cambridge, MA: MIT Press.
Richardson, John F. 1986. Super-equi and anaphoric control. Papers from the Regional Meeting of the
Chicago Linguistic Society (CLS) 22:248–261.
Rivero, María-Luisa. 1994. The structure of the clause and V-movement in languages of the Balkans.
Natural Language and Linguistic Theory 12:63–120.
Rivero, María-Luisa, and Dana Geber. 2004. Raising in Romanian: Move and Agree. Paper presented
at the 34th Linguistic Symposium on Romance Languages (LSRL), Salt Lake City, March 2004.
Rizzi, Luigi. 1982. Issues in Italian syntax. Dordrecht: Foris.
Rizzi, Luigi. 1986. Null objects in Italian and the theory of pro. Linguistics Inquiry 17:501–557.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, MA: MIT Press.
Rizzi, Luigi. 1997. The fine structure of the left periphery. In Liliane Haegeman (ed.), Elements of
grammar: A handbook in generative syntax, 281–337. Dordrecht: Kluwer Academic Publishers.
Rizzi, Luigi. 1999. On the position of “int(errogative)” in the left periphery of the clause. Ms.,
Università di Siena.
Rizzi, Luigi. 2004. On the cartography of syntactic structures. In Luigi Rizzi (ed.), The structure of CP
and IP: The cartography of syntactic structures, vol. 2, 3–16. Oxford: Oxford University Press.
Roberts, Ian. 1997. Restructuring, head movement, and locality. Linguistic Inquiry 28: 423–460.
Rodrigues, Cilene. 2004. Impoverished morphology and movement out of case domains. University of
Maryland at College Park dissertation.
Rodrigues, Cilene. Forthcoming. Modality and partial control. Ms., University of Brasilia.
Rooryck, Johan. 2001. Configurations of sentential complementation: Perspectives from Romance
languages. London: Routledge.
Rosenbaum, Peter. 1967. The grammar of English predicate complement constructions. Cambridge,
MA: MIT Press.
Ross, John Robert. 1967. Constraints on variables in syntax. MIT dissertation. [Published 1986 as Infi-
nite syntax! Norwood, N.J.: Ablex.]
Roussou, Anna. 1994. The syntax of complementisers. University College of London dissertation.
338 BIBLIOGRAPHY
Roussou, Anna. 2000. On the left periphery: Modal particles and complementizers. Journal of Greek
Roussou, Anna. 2001. Control and raising in and out of subjunctive complemements. In María-Luisa
Rivero and Angela Ralli (eds.), Comparative syntax of the Balkan languages, 74–104. Oxford:
Roussou, Anna. 2004. More remarks on control. Paper presented at the Cambridge Linguistics Society
Seminar Series, University of Cambridge.
Roussou, Anna. 2005. Greek control revisited. Paper presented at the 7th International Conference on
Greek Linguistics, University of York.
Roussou, Anna. In preparation. In the mood of control. Ms., University of Patras.
Rudanko, Juhani. 1985. Towards classifying verbs governing object-controlled infinitival equi in
Modern English. Studia Neophilologica 57:145–155.
Rudanko, Juhani. 1989. Complementation and case grammar. Albany: SUNY Press.
Ružićka, Rudolf. 1983a. Autonomie und interaktion in syntax und semantik. In Rudolf Ružićka and
Wolfgang Motsch (eds.), Untersuchungen zur semantik, studia grammatica XXV:15–60. Berlin:
Akademie Verlag.
Ružićka, Rudolf. 1983b. Remarks on control. Linguistic Inquiry 14:309–324.
Sag, Ivan, and Carl Pollard. 1991. An integrated theory of complement control. Language 67:
63–113.
Saito, Mamoru. 1983. Comments on the papers in generative syntax. In Yukio Otsu, Henk van
Riemsdijk, Kazuko Inuoe, Akio Kamio, and Noriko Kawasaki (eds.), Studies in generative
grammar and language acquisition, 79–89. Tokyo: International Christian University.
Salmon, Nathan. 1986. Reflexivity. Notre Dame Journal of Formal Logic 27:401–429.
Schütze, Carson. 2001. On Korean ‘case stacking’: The varied functions of the particles ka and lul. The
Linguistic Review 18:193–232.
Schütze, Carson, and Kenneth Wexler. 1996. Subject case licensing and English root infinites.
Proceedings of the 20th Annual Boston University Conference on Language Development (BUCHD
20), 670–681. Somerville, MA: Cascadilla Press.
Seely, T. Daniel. 1988. Anaphoric relations, chains, and paths. University of Massachusetts, Amherst,
dissertation.
Seely, T. Daniel. 2002. On ‘construal as movement’. Ms., Eastern Michigan University.
Sells, Peter. 1990. Is there subject-to-object raising in Japanese? In Katarzyna Dziwirek, Patrick Farrell,
and Errapel Mejías-Bikandi (eds.), Grammatical relations: A cross-linguistic perspective, 445–457.
Stanford: CSLI Publications.
Sherman, Janet C. 1983. The acquisition of control in complement sentences: The role of structural and
lexical factors. Cornell University dissertation.
Sherman, Janet C., and Barbara Lust. 1993. Children are in control. Cognition 46:1–51.
Sichel, Ivy. 2003. Phrasal movement within DP. In Jacqueline Lecarme (ed.), Research in Afroasi-
atic grammar II: Selected papers from the 5th Conference on Afroasiatic Languages, Paris, 2000
447–449. Amsterdam: John Benjamins.
Sichel, Ivy. 2005. The non-thematic nature of possession. Handout of talk given at Generative
Linguists of the Old World, University of Geneva.
Sichel, Ivy. 2006. Control in DP. Handout of talk given at Ben Gurion University.
Siebert-Ott, Gesa. 1983. Kontroll-probleme in infiniten komplementkonstruktionen. Tübingen: Gunter
Narr Verlag.
Siebert-Ott, Gesa. 1985. Bemerkungen zu den elementen einer theorie der kontrolle. In Werner
Abraham (ed.), Erklärende syntax des deutschen, 255–270. Tübingen: Gunter Narr Verlag.
Sigurðsson, Halldór Á. 1991. Icelandic case-marked PRO and the licensing of lexical arguments.
Natural Language and Linguistic Theory 9:327–363.
Siloni, Tal. 1997. Noun phrases and nominalizations. Dordrecht: Kluwer.
Sitaridou, Ioanna. 2002. The synchrony and diachrony of Romance infinitives with nominative subjects.
University of Manchester dissertation.
Slobin, Dan I. 1966. Grammatical transformations and sentence comprehension in childhood and
adulthood. Journal of Verbal Learning and Verbal Behavior 5:219–227.
Solà, Jaume. 1992. Agreement and subjects. Universitat Autonoma de Barcelona dissertation.
Song, Jae-Gyun. 1994. Clause-embedding verbs and the interpretation of WH in situ. University of
Texas dissertation.
Speas, Margaret. 1994. Null arguments in a theory of economy of projection. In Elena Benedicto
and Jeffrey Runner (eds.), Functional projections, University of Massachusetts Occasional Papers
(UMOP) 17:179–208.
BIBLIOGRAPHY 339
Sportiche, Dominique. 1996. Clitic constructions. In Johan Rooryck and Laurie Zaring (eds.), Phrase
structure and the lexicon, 213–276. Dordrecht: Kluwer Academic Publishers.
Sportiche, Dominique. 1998. Partitions and atoms of clause structure. London: Routledge.
Spyropoulos, Vassilios. 1998. The structure of small clauses in Modern Greek. In Brian Joseph,
Geoffrey Horrocks, and Irene Philippaki-Warburton (eds.), Themes in Greek linguistics II,
169–196. Amsterdam: John Benjamins.
Spyropoulos, Vassilios. 1999. Agreement relations in Greek. University of Reading dissertation.
Spyropoulos, Vassilios. 2005a. Agreement and multiple-case licensing in Greek. In Melita Stavrou
and Arhonto Terzi (eds.), Advances in Greek generative grammar, 15–39. Amsterdam: John
Benjamins.
Spyropoulos, Vassilios. 2005b. The syntax of Classical Greek infinitive. In Katalin É. Kiss (ed.), Universal
grammar in the reconstruction of ancient languages, 295–337. Berlin: Mouton de Gruyter.
Spyropoulos, Vassilios, and Irene Philippaki-Warburton. 2001. Subjunctive, coreference and control
theory in Greek. Paper presented at the 5th International Conference on Greek Linguistics.
Spyropoulos, Vassilios, and Irene Philippaki-Warburton. 2002. Subject and EPP in Greek: The
discontinuous subject hypothesis. Journal of Greek Linguistics 2:149–186.
Stowell, Timothy. 1981. The origins of phrase structure. MIT dissertation.
Stowell, Timothy. 1982. The tense of infinitives. Linguistic Inquiry 13:561–570.
Stromswold, Karin. 1995. The acquisition of subject and object wh-questions. Language Acquisition
4:5–48.
Stromswold, Karin. 1996. Does the VP-internal subject stage really exist? Paper presented at the 21st
Annual Boston University Conference on Language Development (BUCLD 21).
Sudhalter, Vicki, and Martin D.S. Braine. 1985. How does comprehension of passive develop? A
comparison of actional and experiential verbs. Journal of Child Language 12:455–470.
Szabolcsi, Anna. 1983. The possessor that ran away from home. The Linguistic Review 3:89–102.
Szabolcsi, Anna. 1994. The noun phrase. In Ferenc Kiefer and Katalin É. Kiss (eds.), The syntactic
structure of Hungarian, Syntax and Semantics 27, 179–274. San Diego, CA: Academic Press.
Takano, Yuji. 2003. Nominative objects in Japanese complex predicate constructions: A prolepsis
analysis. Natural Language and Linguistic Theory 21:779–834.
Takano, Yuji. 2004. Coordination of verbs and two types of verbal inflection. Linquistic Inquiry
35:168–178.
Tanaka, Hidekazu. 2002. Raising to object out of CP. Linguistic Inquiry 33:637–652.
Taylor, Heather Lee. 2002. Tough-constructions involving degree ADVPs: A movement-as-construal
approach. Paper presented at the Michigan Linguistic Society, Eastern Michigan University.
Terzi, Arhonto. 1992. PRO in finite clauses: A study of the inflectional heads of the Balkan languages.
CUNY dissertation.
Terzi, Arhonto. 1997. PRO and null case in finite clauses. The Linguistic Review 14:335–60.
Terzi, Arhonto, and Kenneth Wexler. 2002. A-Chains and s-homophones in children’s grammar: Evidence
from Greek passives. Proceedings of the North East Linguistic Society (NELS) 32:519–537.
Theophanopoulou-Kontou, Dimitra, Georgia Catsimali, Amalia Moser, Vasiliki Nikiforidou, and Despina
Chila-Markopoulou (eds.), 1998. Themes in Modern Greek syntax (in Greek). University of Athens.
Thráinsson, Höskuldur. 1979. On complementation in Icelandic. New York: Garland Press.
Torrego, Esther. 1984. On inversion in Spanish and some of its effects. Linguistic Inquiry 15:103–129.
Torrego, Esther. 1996. On quantifier floating. In Katalin É. Kiss (ed.), Control clauses configurational
languages, 122–152. Oxford: Oxford University Press.
Torrego, Esther. 2002. A derivational approach to syntactic relations based on clitics. In Samuel
D. Epstein and T. Daniel Seely (eds.), Derivation and explanation in the minimalist program,
249–268. Oxford: Blackwell Publishing.
Turner, Elizabeth, and Ragnar Rommetveit. 1967. Experimental manipulation of the production of
active and passive voice in children. Language and Speech 10:169–180.
Ura, Hiroyuki. 2000. Checking theory and grammatical functions in universal grammar. Oxford: Oxford
University Press.
Uriagereka, Juan. 1988. On government. University of Connecticut dissertation.
Uriagereka, Juan. 1998. Rhyme and reason. Cambridge, MA: MIT Press.
Vallduví, Enric. 1995. Structural properties of information packaging in Catalan. In Katalin Kiss
(ed.), Discourse configurational languages, 122–152. Oxford: Oxford University Press.
Vanden Wyngaerd, Guido. 1994. PRO-legomena. Distribution and reference of infinitival subjects.
Berlin: Mouton de Gruyter.
Varlokosta, Spyridoula. 1993. Control in Modern Greek. University of Maryland Working Papers in
340 BIBLIOGRAPHY
Varlokosta, Spyridoula. 1994. Issues on Modern Greek sentential complementation. University of

Maryland dissertation.
Veloudis, Ioannis, and Irene Philippaki-Warburton. 1983. I ipotaktiki sta Nea Elinika (The subjunc-
tive in Modern Greek). Studies in Greek Linguistics 3:151–68.
Vendler, Zeno. 1967. Linguistics and philosophy. Ithaca: Cornell University Press.
Wegener, Heide. 1989. ‘Kontrolle’ semantisch gesehen. Zur Interpretation von Infinitivkomplementen
im deutschen. Deutsche Sprache 17:206–228.
Wellman, Henry M., David Cross, and Julanne Watson. 2001. Meta-analysis of theory-of-mind
development: The truth about false belief. Child Development 72:655–684.
Wellman, Henry M., Michelle Hollander and Carolyn A. Schult. 1996. Young children’s understanding
of thought-bubbles and of thought. Child Development 67:768–788.
Wexler, Kenneth. 1992. Some issues in the growth of control. In Richard K. Larson, Sabine Iatridou,
Utpal Lahiri, and James Higginbotham (eds.), Control and grammar, 253–295. Dordrecht:
Kluwer.
Wexler, Kenneth. 1993. Optional infinitives, head movement, and the economy of derivations. In
David Lightfoot and Norbert Hornstein (eds.), Verb movement, 305–382. Cambridge: Cambridge
University Press.
Wexler, Kenneth. 1998. Very early parameter setting and the unique checking constraint: A new
explanation of the optional infinitive stage. Lingua 106:23–79.
Wexler, Kenneth. 2002. Lenneberg’s dream: Learning, normal language development and specific
language impairment. In Yonata Levy and Jeannette Schaeffer (eds.), Language competence across
populations: Towards a definition of specific language impairment, 11–61. Mahwah, NJ: Lawrence
Erlbaum Associates.
Wexler, Kenneth. 2004. Theory of phasal development: Perfection in child grammar. MIT Working
Wexler, Kenneth, and Yu-Chin Chien. 1985. The development of lexical anaphors and pronouns.
Papers and Report on Child Language Development 24:138–149.
Wexler, Kenneth, and Peter W. Culicover. 1980. Formal principles of language acquisition. Cambridge,
MA: MIT Press.
Williams, Edwin. 1974. Rule ordering in syntax. MIT dissertation.
Williams, Edwin. 1980. Predication. Linguistic Inquiry 11:203–238.
Williams, Edwin. 1992. Adjunct control. In Richard K. Larson, Sabine Iatridou, Utpal Lahiri, and
James Higginbotham (eds.), Control and grammar, 297–322. Dordrecht: Kluwer Academic
Publishers.
Wurmbrand, Susi. 1998. Infinitives. MIT dissertation.
Wurmbrand, Susanne. 2001. Infinitives: Restructuring and clause structure. New York: Mouton de
Gruyter.
Wurmbrand, Susanne. 2004. Two types of restructuring – lexical vs. functional. Lingua 114:991–1014.
Yoon, James H. 1987. Some queries concerning the syntax of multiple nominative constructions in
Korean. In Harvard Studies in Korean Linguistics. 2:138–162. Seoul: Hanshin.
Yoon, James H. 1994. Korean verbal inflection and checking theory. In Morphology-syntax connection,
MIT Working Papers in Linguistics 22:251–270.
Yoon, James H. 1996. Ambiguity of government and the chain condition. Natural Language and
Yoon, James H. 2004a. The independence of grammatical case from interpretive factors. Paper
presented at the 2004 Linguistic Society of Korean International Conference, Seoul, Korea.
Yoon, James H. 2004b. Non-nominative (major) subjects and case-stacking in Korean. In Peri
Bhaskararao and Karumuri Venkata Subbarao (eds.), Non-nominative subjects, vol. 2, 275–324.
Berlin: Mouton de Gruyter.
Yoon, Jeong-Me. 1989. ECM and multiple subject constructions in Korean. In Harvard Studies in
Korean Linguistics 3:369–382. Seoul: Hanshin.
Yoon, Jeong-Me. 1991. The syntax of A-Chains: A typological study of ECM and scrambling. Cornell
University dissertation.
Zidani-Eroğlu, Leyla. 1997. Exceptionally case-marked NPs as matrix objects. Linguistic Inquiry
28:219–230.
NAME INDEX
A
Cheng, Lisa 257
Abney, Steven 15, 16, 228
Chien, Yu–Chin 70
Abraham, Werner 281, 283
Chierchia, Gennaro 228
Abramovitch, Rona 527
Cho, Yongjoon 130
Agbayani, Brian 257
Chomsky, Carol 297
Alexiadou, Artemis 113, 129, 150, 156, 157, 180, 181,
Chomsky, Noam 9, 15–17, 35–38, 70, 111, 114, 140,
209, 247
141, 152, 154–157, 159, 172, 176, 187–189, 195, 198,
Anagnostopoulou, Elena 113, 129, 150, 156, 157, 180,
199, 205, 209, 210, 214, 227, 233, 242, 243, 248, 266,
181, 209, 247
285, 297, 299, 301, 316, 318
Andrews, Avery 171, 324
Chung, Dae–Ho 105
Aoun, Joseph 259, 262
Cinque, Gugliemo 191, 208, 211
Avrutin, Sergey 70
Collins, Chris 59, 66, 154, 271
Comrie, Bernard 324
B Cormack, Annabel 181
Babby, Leonard 324, 325 Cornilescu, Alexandra 195, 196, 209
Babyonyshev, Maria 35, 38, 51, 53, 66 Costa, Jõao 211, 219, 228
Badawi, El–Said 182 Crain, Stephen 59
Badecker, William 261 Crawford, Jean 70
Baltin, Mark 302 Culicover, Peter 4, 6, 11, 17, 28, 62, 66, 179, 281, 294,
Barbosa, Pilar 124, 129, 247 324
Barrett, Leslie 302
Basilico, David 85–87, 105
D
Becker, Misha 7, 12, 39–41, 45, 49, 66, 67
D’Alessandro, Roberta 219
Bejar, Susana 199, 311
Deprez, Viviane 103
Belletti, Adriana 66, 211, 260
Diesing, Molly 68, 72, 83, 104
Bennis, Hans 283
Dikken, Marcel den 227
Berwick, Robert 259, 259
Dobrovie–Sorin, Carmen 179, 191, 192, 195–198, 209
Bever, Thomas 52
Doron, Edit 72, 85, 89
Bishop, Dorothy 66
Bobaljik, Jonathan 257, 316
Borer, Hagit 29, 30, 35–37, 51, 52, 59, 61, 66, E
189, 287, 292 Epstein, Samuel 133, 141, 142, 145, 154–157, 209
Borschev, Vladimir 33 Erteschik–Shir, Nomi 86
Bošković, Željko 17, 114, 176, 306, 324, 325
Bouchard, Denis 134, 154 F
Braine, Martin D.S. 52 Farkas, Donka 191, 196, 198, 323
Bresnan, Joan 266, 286, 304, 314 Farrell, Patrick 309, 310
Brown, Samuel 257 Featherston, Samuel 325
Bruening, Benjamin 103, 104, 123 Ferreira, Madalena 247, 257
Brun, Dina 53 Fox, Danny 52, 68, 69, 263, 274
Burzio, Luigi 29, 154 Frampton, John 218
Byun, Hyuna 130 Franks, Steven 324, 325
Froud, Karen 39, 66, 70
C Fukui, Naoki 261
Cairns, Helen 62, 297
Cameron–Faulkner, Thea 42 G
Cardinaletti, Anna 247 Ganger, Jennifer 66
Catsimali, Georgia 113, 114, 131, 150, 151, 160, 168, Geber, Dana 197
179, 180, 312 George, Leland 199
Cecchetto, Carl 308 Giannakidou, Anastasia 121
Chafetz, Jill 52, 68 Gibson, Edward 57
341
342 NAME INDEX
Gonzalez, Nora 306, 324 L

Goodluck, Helen 64, 65 Ladusaw, William 84
Gordon, Peter 52, 57, 68 Larson, Richard 70, 236, 237
Grimshaw, Jane 15, 17, 30, 320 Lasnik, Howard 114, 154, 159, 188, 262
Grinder, John 314 Lebeaux, David 314
Grodzinsky, Yosef 29, 30, 52, 68, 69 Lee, Hyeonjin 66
Grosu, Alex 197 Lee, Jeong–Shik 74, 83, 104
Guasti, Maria 62 Lee, Pil–Young 71, 74
Guillaume, Gustave 286 Lee, Wooseung 105
Legate, Julie 273
H Li, Y.–H. Audrey 221, 259, 262
Hadjivassiliou, Angela 126, 129 Lin, Jo–Wang 271
Haegeman, Liliane 189 Lyngfelt, Benjamin 304, 325
Hamburger, Henry 59
Han, Chung–Hye 105 M
Harley, Heidi 228 MacWhinney, Brian 42
Hartman, Jeremy 52, 53, 58, 66 Maling, Joan 95
Heim, Irene 223 Manorohanta, Cecile 52
Heycock, Caroline 72, 85, 89 Manzini, M. Rita 159, 188, 189, 192, 264, 266, 282, 314
Hill, Virginia 195, 196, 208–210 Maratsos, Michael 52, 54
Hiraiwa, Ken 71 Martin, Roger 17, 114, 159, 176, 301, 302
Hoekstra, Teun 283 Martins, Anna Maria 244
Hoji, Hajime 71, 94, 96–99, 103, 104, 130 Massam, Diane 103, 199, 311
Holmberg, Anders 233, 242–244, 248 May, Robert 93, 299
Holton, David 176, 179 McDaniel, Dana 62, 297
Hong, Ki–Sun 71, 73, 77, 94, 103, 104, 106 McGinnis, Martha 59
Hong, Soo–Min 71 Menuzzi, Sergio 228
Horvath, Julia 197, 312 Miyagawa, Shigeru 105
Householder, Fred 179 Miyamoto, Edson 66
Huang, C.–T. James 264 Mobley, Linda 261
Huang, Yan 178 Modyanova, Nadya 59, 66, 68, 70
Hyams, Nina 38, 52, 66 Moore, John 74, 94, 95, 106
Morrell, Kenneth 171
I Motapanyane, Virginia 191, 196, 209
Iatridou, Sabine 113, 114, 116, 117, 154, 159, 168,
169, 179 N
Ingria, Robert 7, 179 Nakanishi, Kumiko 221
Isac, Dana 196 Negrão, Esmeralda 242
Neidle, Carol 324
J Nicol, Janet 261
Jackendoff, Ray 4, 6, 11, 17, 28, 62, 179, 281, 294, Nikanne, Urpo 242
310, 324 Ntelitheos, Dimitrios 52
Jacobson, Pauline 33 Nunes, Jairo 17, 28, 156, 243, 244, 257
Johnson, Marcus 107
Joseph, Brian 7, 130, 160, 179, 262 O
O’Grady, William 75, 96
K O’Neil, John 188, 282
Kakouriotis, Athanasios 114 Oka, Toshifusa 71, 75
Kawasaki, Noriko 304, 314, 315 Oniga, Renato 308
Kayne, Richard 15, 16, 21, 33, 209, 214, 228, 257, 261 Orfitelli, Robyn 51, 66, 69
Kempchinsky, Paula 191 Ordóñez, Francisco 211
Kiguchi, Hirohisa 257, 260, 261, 313, 314 Osterhout, Lee 261
Kim, Jong–Bok 105
Kim, Soo–Won 95, 103 P
Kim, Yeng–Hee 97 Partee, Barbara 33
Ko, Heejeong 229, 271, 272 Pereira, Sandra 219, 228
Koizumi, Masatoshi 114, 220 Pesetsky, David 66, 127, 128, 154, 155, 176, 199, 203,
Kornfilt, Jaklin 114, 199 209, 218, 263, 274, 325
Koster, Jan 297 Petter, Marga 323
Krapova, Iliyana 179, 198 Philippaki–Warburton, Irene 113, 114, 119, 123, 129, 131,
Kratzer, Angelica 86 149, 150, 154–157, 160, 168, 174, 179–181, 312
Krifka, Manfred 268, 277 Phillips, Colin 261, 262
Kuno, Susumu 7, 71, 77, 80, 97, 314 Pinker, Steven 62, 70
Kuroda, Shige–Yuki 84 Pires, Acrisio 154
NAME INDEX 343
Pîrvulescu, Mihaela 196 Sudhalter, Vicki 52

Pittman, Christine 228 Swinney, David 261
Polinsky, Maria 3, 4, 8, 181, 188, 309, 311 Szabolcsi, Anna 21
Pollard, Carl 323
Postal, Paul 3, 16, 18, 111, 120, 125, 320, 321, 323, 325 T
Potsdam, Eric 3, 4, 8, 74, 94, 181, 188, 309, 311, 323 Takano, Yuji 71, 74, 75, 88, 94, 104
Pustejovsky, James 286, 290 Tanaka, Hidekazu 71, 97, 100, 104
Taylor, Heather 154
R Terzi, Arhonto 53, 114, 130, 154, 159, 168, 179–181,
Raposo, Eduardo 247 191, 196
Reinhart, Tanya 113, 117, 118, 119, 127, 128, 130, 131 Theophanopoulou–Kontou, Dimitra 131
Reuland, Eric 23 Thráinsson, Höskuldur 324
Richards, Norvin 300 Tomioka, Satoshi 221
Richardson, John 315 Torrego, Esther 59, 155, 176, 199, 203, 218, 223, 225,
Riemsdijk, Henk van 189 226, 229
Ritter, Elizabeth 228 Tsakali, Vina 66
Rivero, María–Luisa 196, 197 Turner, Elizabeth 52
Rizzi, Luigi 66, 86, 189, 196, 208, 231, 244, 260, 271,
272, 296 U
Roberts, Ian 210, Ura, Hiroyuki 103, 104, 106, 199
Rommetveit, Ragnar 52 Uriagereka, Juan 225, 247, 257
Rosenbaum, Peter 3, 164, 188, 281, 291, 294, 323
Ross, John Robert 262
V
Roussou, Anna 157, 159, 160, 175, 179, 180, 183, 188,
Vallduví, Enç 211
189, 192, 266
Vanden Wyngaerd, Guido 282
Rudanko, Juhani 281
Varlokosta, Spyridoula 113, 114, 134, 137, 138,
Runner, Jeffrey 4, 74, 94
147–150, 154, 156, 157, 159, 168, 169, 174–176,
Ružička, Rudolf 281, 283
178–180
Veloudis, Ioannis 179
S Vendler, Zeno 286
Sag, Ivan 323 Viotti, Evani 242
Saito, Mamoru 71
Salmon, Nathan 156
W
Schütze, Carson 65, 104
Wegener, Heide 281, 284
Sells, Peter 71, 103
Weinberg, Amy 259, 261
Sherman, Janet 62, 297
Wellman, Henry 43, 48
Siebert–Ott, Gesa 281, 283, 284
Williams, Edwin 30, 194, 285,
Sigurðsson, Halldór 179, 305
304, 314
Siloni, Tai 33, 113, 117–119, 130, 131
Wurmbrand, Susanne 17, 28, 189, 191, 208, 215, 223,
Sitaridou, Ioanna 199
229, 297, 306, 323, 324
Slobin, Dan 52
Smith, Neil 181
Snyder, William 52, 66 Y
Solà, Jaume 129 Yoon, Jeong–Me 5–8, 73, 74, 77, 79, 80, 88, 89, 92, 101,
Song, Jae–Gyun 75, 94, 96 103–105, 107
Sportiche, Dominique 214
Stowell, Timothy 17, 30, 176, 198, 286, 302 Z
Straub, Kathleen 261 Zidani–Eroǧlu, Leyla 120
Stromswold, Karin 36, 58 Zoerner, Ed 257
SUBJECT INDEX
A E
ACC-ing 22, 23, 33 EARH. See External Argument Requirement Hypothesis
A-chain 35–38, 63, 66, 70, 197–199, 201–203, 205, ECM. See Exceptional Case Marking
208, 210, 309 Ellipsis 137, 143, 174, 194, 231, 319
Adjunct control 253, 304 Epicene noun (epicene DP) 213–216
ϕ-agree. See Phi-agree EPP. See Extended Projection Principle
Aktionsart. See Transition Exceptional Case Marking (ECM) 15, 16, 23–26, 33,
Anaphoric tense 198, 203 65, 71, 104, 106, 107, 111–131, 151, 170, 179, 309,
Arabic 182, 259 310, 319, 325
Arbitrary control 4, 11, 266, 267, 271, 316, 317 Exhaustive control (EC) 163, 164, 167–169, 172–182,
190, 193, 194, 215, 217, 219, 221–223, 228, 229, 278,
B 302, 305, 306
Backward control 4, 5, 9, 165, 167, 169, 181, 188, 194, Expletive 6, 16–12, 28, 32, 33, 37, 39, 42, 43, 52, 69,
206, 309–311, 322 104, 311, 318, 320, 321, 326, 327
Bound variable 31, 32, 76, 92 Extended Projection Principle (EPP) 16, 31, 38, 87, 105,
Brazilian Portuguese (BP). See Portuguese 140, 156, 157, 181, 195, 208, 210, 220, 233–235, 242,
244, 311
C External Argument Requirement Hypothesis (EARH)
Case checking 10, 131, 138, 141, 144, 152, 154, 157, 38, 52, 61, 69
172, 178, 182, 307 Extraposition 17, 23, 26–28
Case concord 151, 302, 305, 308, 312
Case Filter 15, 62, 141, 224 F
Case percolation 5, 305, 308, 324 Factive 175, 176, 221, 264, 265, 324
Categorical judgment 84–87 ϕ-feature. See Phi-feature
Categorical subject 72, 78, 84, 85, 88, 90, 91, 93, 97, Finite complement 5, 7, 8, 11, 23, 50, 60, 70, 87, 303,
103–105 312, 323
CHILDES corpus 40, 42, 67, 70 Finiteness 4, 66, 114, 116, 159, 160, 168, 170, 172,
Chinese 221, 231, 258, 259, 261 176, 196
Clausal completeness 4 Finnish 231–233, 235, 242–248
Clitic climbing 191, 209, 229 Focus particle 17, 23, 24, 26, 27
Clitic doubling 8, 23, 25, 114, 118–120, 122, 123, 131, French 3, 11, 45, 59, 60, 118, 214, 283, 289, 310
225, 228
Clitic left dislocation (CLLD) 123, 124, 129
G
ϕ-complete. See Phi-complete
Generic control 263, 264, 266, 268–274, 276–279
Completeness. See Clausal completeness
German 11, 60, 107, 118, 283, 284, 297, 306
Construal-as-movement 141, 142, 146, 147, 153, 157
Government 7, 15, 17, 28, 32, 33, 138, 141, 156,
Control shift 164, 179, 284, 285, 291, 293,
180, 188
295, 296, 323
Government and Binding Theory (GB) 7, 15, 16, 133,
Control suspension 165, 166, 178–180
134, 137, 138, 140, 141, 154
Copy raising 4, 74, 94, 95, 106
Greek 5, 7–10, 53, 64, 65, 106, 111–131, 133, 150,
152–154, 156, 157, 159–161, 163, 166–168
D
(270 instances)
de dicto 93, 101, 102
de re 75, 90, 92, 93, 101, 102, 106, 174, 317
de se 31, 32, 75, 135, 143, 155, 301, 314, 315 H
ϕ-defective. See Phi-defective Hebrew 4, 6, 16–18, 21–26, 32, 33, 118, 296, 300,
Defective Agr 9, 10, 134, 140, 141, 143, 154, 156 304, 312
Defective v 7, 37, 38, 51, 64, 68, 70 Hungarian 305, 307, 309, 312
Desiderative 193, 197, 202, 221, 266, 267, 272
Dislocation 187, 195, 202–209, 211 I
DP-float (floating DP) 225 Icelandic 45, 59, 200, 201, 302, 305, 307–309, 313, 324
Dutch 11, 118, 283, 319 Idiom chunk 6, 16–21, 28, 32, 33, 74, 90, 94
345
346 SUBJECT INDEX
Implicative (implicative predicate) 188, 193, 198, 202, Phi-agree (ϕ-agree) 217, 218, 222
210, 221, 264 Phi-complete (ϕ-complete) 10, 140–143, 145, 147–149,
Implicit control 295, 296, 322 151–155, 157, 191, 195, 199, 204, 233
Improper movement 69, 247, 325 Phi-defective (ϕ-defective) 5, 9, 10, 134, 140, 141, 143,
Independent tense 157, 198, 203 144, 147, 150, 154, 156, 157, 233
Individual-level predicate 80, 83, 84, 104 Phi-feature (ϕ-feature) 10, 38, 134, 140, 141, 152, 153,
Inherent case 16, 101, 119, 180, 307 182, 202, 204, 205, 210, 213–215, 217, 218, 222, 228,
Interrogative 221, 253, 264, 265, 270, 302, 317 233, 234, 243, 248, 287, 288, 290, 292, 299, 301, 307,
Inverse partial control 213, 223–229 318, 324
Island 8, 11, 73, 74, 89, 105, 106, 130, 217, 218, 254, 257, PIC. See Phase Impenetrability Condition
259, 261, 263–266, 271–275, 279, 306, 318–320, 325 Portuguese 153, 211, 213, 215, 216, 219,
228, 229
J Brazilian Portuguese (BP) 5, 8, 9, 11, 228, 231–233,
Japanese 4, 5, 7, 66, 71, 72, 74–76, 78, 87–89, 93, 94 235–239, 241–244, 246, 247, 309–312, 325
PRO-gate 11, 313, 316, 317, 322
Prolepsis 5, 7, 8, 71, 94, 99, 100, 101, 106
K
Proleptic object 93, 94, 99, 106, 107
Korean 5, 7, 8, 66, 71–74, 76, 77, 78, 86–89, 91, 93–95,
Proper Binding Condition (PBC) 97–101, 107
97, 99, 102, 103, 224
Psychological predicate (psychological verb) 6, 52–54,
62, 305
L
Last Resort 105, 114, 156, 234, 295, 317, 323
Q
Long-distance Agree 71, 104, 318
Quantifier scope 75
Long-distance control 31, 32, 315, 317
R
M
Raising-to-object (RtO). See Subject-to-object raising
Madurese 8, 71, 89, 95, 106, 130
Reconstruction 76, 91, 259, 262, 273
Major Object (MOB) 77, 93–99, 101, 102
Relativized Minimality (minimality) 86, 103, 105, 106,
Major Subject (MS) 5, 7, 8, 72, 76–85, 87–94, 96, 97, 100,
251, 254, 258, 259, 261, 262, 300, 317
101, 103–105, 107
Resumptive pronoun 73, 89, 94, 100
Match 140, 155, 156, 177, 195, 205
Romanian 5, 8, 9, 118, 181, 187, 190–196, 198, 199, 201,
Minimal Distance Principle (MDP) 11, 164, 188, 235,
203, 205, 207–211, 302
238, 294, 296, 297, 322, 323
Russian 53, 66, 70, 302, 305, 307, 308, 309, 313,
Minimal Link Condition (MLC) 70, 236, 242, 294, 295,
324, 325
296, 300, 323
Minimality. See Relativized Minimality
MOB. See Major Object S
MoodP 155, 157 Secondary predicate 115, 116, 215, 219, 220, 228, 304,
MS. See Major Subject 305
Selectional restriction 6, 18, 20, 28, 277,
303, 309
N
Semantic selection (s-selection) 5, 10, 286
Negative concord 17, 23, 26
Sentential predicate 77–80, 84, 86, 88–91, 96, 97, 100,
Negative polarity item (NPI) 8, 120, 121, 196
104, 105
Nominalization 4, 6, 15, 21, 24, 25, 27, 28, 31, 32, 228,
Sideward movement (SWM) 11, 156, 236, 239, 253, 257,
251, 252, 254, 260
258, 261, 316, 319, 320, 322
Non-phasal v 37, 38, 52, 61, 63, 70, 187, 191, 194,
Sloppy identity 31, 32
196–200, 202, 208, 209
Sluicing 265, 266, 271–273, 275–279, 281
Nonsubject raising 73, 78, 79, 82, 83, 104
Small clause 43, 67, 320
NPI. See Negative polarity item
SOR. See Subject-to-object raising
Null case 154, 155, 172, 301
Spanish 59, 118, 211, 213, 215, 216, 223, 225, 226, 229,
305, 306, 308, 309, 324
O Split antecedent 134, 143, 146, 165, 167, 173, 179, 180,
Object raising 71, 81, 211 243
OCC (occurrence) 145, 187, 194, 199, 205–208, 210, 211, Split control 5, 11, 157, 284, 287, 288, 290, 297, 298, 314,
242, 243, 299 322, 323
s-selection. See Semantic selection
P SSR. See Subject-to-subject raising
Partial control 4, 5, 8, 9, 11, 157, 163, 164, 166, 167, 173, Stage-level predicate 68, 83–85, 105
179, 180, 188, 190, 192, 213, 215, 217–219, 221–229, Stranding 9, 213, 219–221, 223–225, 227–229
276, 317, 322, 324, 325 Strict reading 134, 155, 174, 194, 315
Past participle agreement 213–215, 218 Structural case 15, 33, 113, 117–119, 307
PBC. See Proper Binding Condition Subevent 286–291
Persian 302, 312 Subjacency 89, 105
Phase Impenetrability Condition (PIC) 36, 63, 64, 195, Subject-to-object raising (SOR) 71–74, 76–79, 81–91,
196, 233, 247, 318, 325 93–104, 106, 107
SUBJECT INDEX 347
Subject-to-subject raising (SSR) 35–37, 39, 45, 51, Transition (Aktionsart) 286–288
70, 179 Transparency 254, 255, 257, 258, 260, 261, 309
SWM. See Sideward movement Tsez 4, 311
T U
[Tense] feature 10, 176–178, 222, 303 Universal Phase Requirement (UPR) 36–38, 41, 44, 46,
Theme-Rheme 187, 205 47, 50–52, 56, 57, 59–66, 68–70, 142
Theta-chain 194, 197, 200, 201, 202 [unrealized] (C-)T feature 286, 287
Theta-checking 28, 144 [unrealized] subevent 286–290
Theta-criterion 20, 21, 118, 146, 188
Theta-licensed 18 V
Theta-role 7, 20, 21, 30, 52, 63, 68, 78, 94–97, 106, 113 Variable binding 90, 91, 301
(73 instances) Variable local control (VLC) 11, 281, 282, 284, 285,
Thetic judgment 84–87, 104 287–291
Thetic subject 85–88, 91, 105
Topic 3, 5, 12, 65, 70, 86, 103–105, 150, 157, W
187, 195, 202, 205, 207, 211, 233, 242–248, 274, 311, 314 wh-infinitives 11, 264, 265, 270, 275, 317,
Topic-prominent 242, 243 318, 322
Managing Editors
Liliane Haegeman, University of Geneva
Joan Maling, Brandeis University
James McCloskey, University of California, Santa Cruz
Publications
1. L. Burzio: Italian Syntax. A Government-binding Approach. 1986
ISBN Hb 90-277-2014-2; Pb 90-277-2015-0
2. W. D. Davies: Choctaw Verb Agreement and Universal Grammar. 1986
ISBN Hb 90-277-2065-7; Pb 90-277-2142-4
3. K. É. Kiss: Configurationality in Hungarian. 1987
ISBN Hb 90-277-1907-1; Pb 90-277-2456-3
4. D. Pulleyblank: Tone in Lexical Phonology. 1986
ISBN Hb 90-277-2123-8; Pb 90-277-2124-6
5. L. Hellan and K. K. Christensen: Topics in Scandinavian Syntax. 1986
ISBN Hb 90-277-2166-1; Pb 90-277-2167-X
6. K. P. Mohanan: The Theory of Lexical Phonology. 1986
ISBN Hb 90-277-2226-9; Pb 90-277-2227-7
7. J. L. Aissen: Tzotzil Clause Structure. 1987
ISBN Hb 90-277-2365-6; Pb 90-277-2441-5
8. T. Gunji: Japanese Phrase Structure Grammar. A Unification-based Approach.
1987 ISBN 1-55608-020-4
9. W. U. Wurzel: Inflectional Morphology and Naturalness. 1989
ISBN Hb 1-55608-025-5; Pb 1-55608-026-3
10. C. Neidle: The Role of Case in Russian Syntax. 1988
ISBN 1-55608-042-5
11. C. Lefebvre and P. Muysken: Mixed Categories. Nominalizations in Quechua.
1988 ISBN Hb 1-55608-050-6; Pb 1-55608-051-4
12. K. Michelson: A Comparative Study of Lake-Iroquoian Accent. 1988
ISBN 1-55608-054-9
13. K. Zagona: Verb Phrase Syntax. A Parametric Study of English and Spanish.
1988 ISBN Hb 1-55608-064-6; Pb 1-55608-065-4
14. R. Hendrick: Anaphora in Celtic and Universal Grammar. 1988
ISBN 1-55608-066-2
15. O. Jaeggli and K. J. Safir (eds.): The Null Subject Parameter. 1989
ISBN Hb 1-55608-086-7; Pb 1-55608-087-5
16. H. Lasnik: Essays on Anaphora. 1989
ISBN Hb 1-55608-090-5; Pb 1-55608-091-3
17. S. Steele: Agreement and Anti-Agreement. A Syntax of Luiseño. 1990
ISBN 0-7923-0260-5
18. E. Pearce: Parameters in Old French Syntax. Infinitival Complements. 1990

ISBN Hb 0-7923-0432-2; Pb 0-7923-0433-0
19. Y. A. Li: Order and Constituency in Mandarin Chinese. 1990
ISBN 0-7923-0500-0
20. H. Lasnik: Essays on Restrictiveness and Learnability. 1990
ISBN 0-7923-0628-7; Pb 0-7923-0629-5
21. M. J. Speas: Phrase Structure in Natural Language. 1990
ISBN 0-7923-0755-0; Pb 0-7923-0866-2
22. H. Haider and K. Netter (eds.): Representation and Derivation in the Theory
of Grammar. 1991 ISBN 0-7923-1150-7
23. J. Simpson: Warlpiri Morpho-Syntax. A Lexicalist Approach. 1991
ISBN 0-7923-1292-9
24. C. Georgopoulos: Syntactic Variables. Resumptive Pronouns and A’ Binding in
Palauan. 1991 ISBN 0-7923-1293-7
25. K. Leffel and D. Bouchard (eds.): Views on Phrase Structure. 1991
ISBN 0-7923-1295-3
26. C. Tellier: Licensing Theory and French Parasitic Gaps. 1991
ISBN 0-7923-1311-9; Pb 0-7923-1323-2
27. S.-Y. Kuroda: Japanese Syntax and Semantics. Collected Papers. 1992
ISBN 0-7923-1390-9; Pb 0-7923-1391-7
28. I. Roberts: Verbs and Diachronic Syntax. A Comparative History of English
and French. 1992 ISBN 0-7923-1705-X
29. A. Fassi Fehri: Issues in the Structure of Arabic Clauses and Words. 1993
ISBN 0-7923-2082-4
30. M. Bittner: Case, Scope, and Binding. 1994 ISBN 0-7923-2649-0
31. H. Haider, S. Olsen and S. Vikner (eds.): Studies in Comparative Germanic
Syntax. 1995 ISBN 0-7923-3280-6
32. N. Duffield: Particles and Projections in Irish Syntax. 1995
ISBN 0-7923-3550-3; Pb 0-7923-3674-7
33. J. Rooryck and L. Zaring (eds.): Phrase Structure and the Lexicon. 1996
ISBN 0-7923-3745-X
34. J. Bayer: Directionality and Logical Form. On the Scope of Focusing Particles
and Wh-in-situ. 1996 ISBN 0-7923-3752-2
35. R. Freidin (ed.): Current Issues in Comparative Grammar. 1996
ISBN 0-7923-3778-6; Pb 0-7923-3779-4
36. C.-T. J. Huang and Y.-H. A. Li (eds.): New Horizons in Chinese Linguistics.
1996 ISBN 0-7923-3867-7; Pb 0-7923-3868-5
37. A. Watanabe: Case Absorption and WH-Agreement. 1996
ISBN 0-7923-4203-8
38. H. Thráinsson, S. D. Epstein and S. Peter (eds.): Studies in Comparative
Germanic Syntax. Volume II. 1996 ISBN 0-7923-4215-1
39. C. J. W. Zwart: Morphosyntax of Verb Movement. A Minimalist Approach

to the Syntax of Dutch. 1997 ISBN 0-7923-4263-1; Pb 0-7923-4264-X
40. T. Siloni: Noun Phrases and Nominalizations. The Syntax of DPs. 1997
ISBN 0-7923-4608-4
41. B. S. Vance: Syntactic Change in Medieval French. 1997
ISBN 0-7923-4669-6
42. G. Müller: Incomplete Category Fronting. A Derivational Approach to
Remnant Movement in German. 1998 ISBN 0-7923-4837-0
43. A. Alexiadou, G. Horrocks and M. Stavrou (eds.): Studies in Greek Syntax.
1998 ISBN 0-7923-5290-4
44. R. Sybesma: The Mandarin VP. 1999 ISBN 0-7923-5462-1
45. K. Johnson and I. Roberts (eds.): Beyond Principles and Parameters. Essays
in Memory of Osvaldo Jaeggli. 1999 ISBN 0-7923-5501-6
46. R. M. Bhatt: Verb Movement and the Syntax of Kashmiri. 1999
ISBN 0-7923-6033-8
47. A. Neeleman and F. Weerman: Flexible Syntax. A Theory of Case and
Arguments. 1999 ISBN 0-7923-6058-3
48. C. Gerfen: Phonology and Phonetics in Coatzospan Mixtec. 1999
ISBN 0-7923-6034-6
49. I. Paul, V. Phillips and L. Travis (eds.): Formal Issues in Austronesian Linguistics.
2000 ISBN 0-7923-6068-0
50. M. Frascarelli: The Syntax-Phonology Interface in Focus and Topic Constructions
in Italian. 2000 ISBN 0-7923-6240-3
51. I. Landau: Elements of Control. Structure and Meaning in Infinitival
Constructions. 2000 ISBN 0-7923-6620-4
52. W. D. Davies and S. Dubinsky (eds.): Objects and other Subjects. Grammatical
Functions, Functional Categories and Configurationality. 2001
ISBN 1-4020-0064-2; Pb 1-4020-0065-0
53. J. Ouhalla and U. Shlonsky (eds.): Themes in Arabic and Hebrew Syntax. 2002
ISBN 1-4020-0536-9; Pb 1-4020-0537-7
54. E. Haeberli: Features, Categories and the Syntax of A-Positions. Cross-Linguistic
Variation in the Germanic Languages. 2002
ISBN 1-4020-0854-6; Pb 1-4020-0855-4
55. J. McDonough: The Navajo Sound System. 2003
ISBN 1-4020-1351-5; Pb 1-4020-1352-3
56. D. E. Holt (ed.): Optimality Theory and Language Change. 2003
ISBN 1-4020-1469-4; Pb 1-4020-1470-8
57. J. Camacho: The Structure of Coordination. Conjunction and Agreement
Phenomena in Spanish and Other Languages. 2003
ISBN 1-4020-1510-0; Pb 1-4020-1511-9
58. I. Toivonen: Non-Projecting Words. A Case Study of Swedish Particles. 2003
ISBN 1-4020-1531-3; Pb 1-4020-1532-1
59. D. Adger, C. de Cat and G. Tsoulas (eds). Peripheries. Syntactic Edges and
their Effects. 2004 ISBN 1-4020-1908-4; Pb 1-4020-1909-2
60. C. Goria: Subject Clitics in the Northern Italian Dialects. A Comparative
Study Based on the Minimalist Program and Optimality Theory. 2004
ISBN 1-4020-2736-2; Pb 1-4020-2737-0
61. V. Dayal and A. Mahajan (eds.): Clause Structure in South Asian Languages.
2004 ISBN 1-4020-2717-6; Pb 1-4020-2718-4
62. P. Kempchinsky and R. Slabakova (eds.): Aspectual Inquiries. 2005
ISBN 1-4020-3032-0; Pb 1-4020-3035-5
63. M. Arad: Roots and Patterns. Hebrew Morpho-syntax. 2005
ISBN 1-4020-3243-9; Pb 1-4020-3245-5
64. P. Štekauer and R. Lieber (eds.): Handbook of Word-Formation. 2005
ISBN 1-4020-3595-0; Pb 1-4020-3597-7
65. A. Johns, D. Massam and J. Ndayiragije (eds.): Ergativity. Emerging Issues.
2006 ISBN Hb 1-4020-4186-1; Pb 1-4020-4187-X
66. F. Lee: Remnant Raising and VSO Clausal Architecture. A Case Study from
San Lucas Quiavini Zapotec. 2006 ISBN 1-4020-4300-7
67. Olga M: Balkam Sprachbund Morpho-Syntactic Features. 2006
ISBN 978-1-4020-4487-8
68. Kiss, Katalin É. (eds.): Event Structure and the left Periphery. 2006
ISBN 978-1-4020-4753-4
69. Harbour, Daniel: Morphosemantic Number, from Kiowa Noun Classes to
UG Number Features. ISBN 978-1-4020-5037-4
70. Pereltsvaig, Asya: Copular Sentences in Russian A Theory of Intra-Clausal
Relations. ISBN-978-1-4020-5792-2
71. William D. Davies & Stanley Dubinsly (eds.): New Horizons in the Analy-
sis of Control and Raising ISBN-978-1-4020-6175-2
springer.com

(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF

Uploaded by

Copyright:

Available Formats

You might also like

(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF

Uploaded by

Copyright:

Available Formats

NEW HORIZONS IN THE ANALYSIS OF CONTROL AND RAISING

Studies in Natural Language and Linguistic Theory

Marcel den Dikken, City University of New York

Guglielmo Cinque, University of Venice

ISBN 978-1-4020-6175-2 (HB)

Printed on acid-free paper

All Rights Reserved

Part I: New Horizons in the Analysis of Control and Raising 1

Looking Out Over the Horizon 3

Part II: New Views of Raising 13

The Late Development of Raising: What Children Seem

Raising of Major Arguments in Korean (and Japanese) 71

Part III: Raising or Control in Greek 109

Not Really ECM, not Exactly Control: The ‘Quasi-ECM’

Control in Modern Greek: It’s Another Good Move 133

Finiteness and Control in Greek 159

Part IV: Control in Romance 185

Moving Forward with Romanian Backward Control and Raising 187

Agreement and Flotation in Partial and Inverse Partial

Null Subjects in Brazilian Portuguese and Finnish:

Part V: Extensions and Alternatives to the MTC 249

On (Non-)Obligatory Control 251

Control and Wh-infinitivals 263

Control via Selection 281

Movement-Resistant Aspects of Control 293

Name Index 341

Subject Index 345

Gabriela Alboiu Christopher Hirsch

Cedric A. Boeckx Konstantia Kapetangianni

Idan Landau T. Daniel Seely

NEW HORIZONS IN THE ANALYSIS

LOOKING OUT OVER THE HORIZON

1. THE GENESIS OF THIS BOOK

2. AN OVERVIEW OF THE CONTENTS

3. SURVEYING THE RESULTS OF THE RESEARCH

(1) a. The regents attempted to undermine the faculty.

(2) a. The candidates appear to be late.

(3) ha-sikuyim Sel rina [le-hagi’a ba-zman]

(4) Cheli-nun Yenghi-lul yenglihay-ss-ta-ko mitnun-ta.

(5) o petros ithele ti maria na traghudhai oli mera.

(6) A vítima quer se encontrar bêbada/* bêbadas.

(7) Quem2 que a Dani1 convenceu t2 que e*1/2 pode se eleger?

Modesto proposes an analysis in which pro must be A’-bound by an element in a

(8) John1 persuaded Mary2 [[empty category]2, 1 to leave].

Barrie’s chapter is that in (10), where extraction out of a wh-island is permitted

Barrie’s categorization of control in wh-infinitivals is distinct both from tradi-

4. ASSESSING THE RESULTS

NEW VIEWS OF RAISING

1. ON THE ABSENCE OF RAISING

The absence of raising in DP has figured prominently in syntactic theory ever

(1) a. John1 refused/promised [PRO1 to leave]

b. John1’s refusal/promise [PRO1 to leave]

(2) a. John1 appeared/was certain [ t1 to be on time]

b. *John1’s appearance/certainty [ t1 to be on time]

c. We wanted [Mary to arrive on time]

d. *Our desire [of Mary to arrive on time]

e. Our desire [for Mary to arrive on time]

The restriction against raising in DP has contributed to theories about various