Professional Documents
Culture Documents
(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF
(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF
(Studies in Natural Language and Linguistic Theory) William D. Davies, Stanley Dubinsky - New Horizons in The Analysis of Control and Raising - Springer (2007) PDF
VOLUME 71
Managing Editors
Editorial Board
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.
Published by Springer,
P.O. Box 17, 3300 AA Dordrecht, The Netherlands.
www.springer.com
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
Raising in DP Revisited 15
IVY SICHEL
vii
viii CONTENTS
Bibliography 327
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.
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.
6 WILLIAM D. DAVIES AND STANLEY DUBINSKY
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).
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.
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
LOOKING OUT OVER THE HORIZON 7
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).
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.
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.
LOOKING OUT OVER THE HORIZON 9
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).
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).
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
LOOKING OUT OVER THE HORIZON 11
(10) a. What kind of bee does Mary know how to defend oneself against?
b. *Where did John wonder who to introduce himself to?
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
12 WILLIAM D. DAVIES AND STANLEY DUBINSKY
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
RAISING IN DP REVISITED
15
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 15–34.
© 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):
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
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
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:
genitive DPs. For convenience, the underlying structure is given under the Roman
numeral heading each diagnostic4:
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):
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
RAISING IN DP REVISITED 21
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.
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):
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):
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):
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):
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
RAISING IN DP REVISITED 25
b. ha-simla Sela
the-dress of-her
‘her dress’
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):
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.
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:
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
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.
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.
RAISING IN DP REVISITED 29
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):
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:
The contrast between raising and control surfaces also in DP, control DPs in (40),
and raising DPs in (41):
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:
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-
RAISING IN DP REVISITED 31
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:
(45) a. John’s mother was in favor of the refusal [PRO to vindicate himself/oneself/herself]
(46) a. People who know John often discuss [his / *ec working too hard]
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
(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]
5. CONCLUSIONS
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
RAISING IN DP REVISITED 33
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:
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
35
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 35–70.
© 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.
(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
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
38 CHRISTOPHER HIRSCH AND KEN WEXLER
they use (with phasal v) is slightly different than the adult representation. On the
other hand, consider the analysis of (4a), the raised construction:
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
THE LATE DEVELOPMENT OF RAISING 39
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
THE LATE DEVELOPMENT OF RAISING 41
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.
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.
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
THE LATE DEVELOPMENT OF RAISING 45
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.
THE LATE DEVELOPMENT OF RAISING 47
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
THE LATE DEVELOPMENT OF RAISING 49
125%
100%
Accuracy
75% MR-foil
ER-foil
50% DR-foil
25%
0%
3 4 5 6 7 8 9
Age Group (Years Old)
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
50 CHRISTOPHER HIRSCH AND KEN WEXLER
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).
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.
THE LATE DEVELOPMENT OF RAISING 51
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
52 CHRISTOPHER HIRSCH AND KEN WEXLER
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.
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
THE LATE DEVELOPMENT OF RAISING 53
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.
54 CHRISTOPHER HIRSCH AND KEN WEXLER
100%
75%
Accuracy
Raising
Passive
50%
25%
3 4 5 6 7 8 9
Age Group (Years Old)
10
# Individuals AC (of 10)
6 Raising
4 Passive
0
3 4 5 6 7 8 9
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
56 CHRISTOPHER HIRSCH AND KEN WEXLER
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
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.
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
58 CHRISTOPHER HIRSCH AND KEN WEXLER
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
THE LATE DEVELOPMENT OF RAISING 59
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
60 CHRISTOPHER HIRSCH AND KEN WEXLER
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:
Yet structures like (12) are quite grammatical in other languages, where think can
take a nonfinite complement:
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.
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).
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
62 CHRISTOPHER HIRSCH AND KEN WEXLER
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)):
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
64 CHRISTOPHER HIRSCH AND KEN WEXLER
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
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
THE LATE DEVELOPMENT OF RAISING 65
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
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.
THE LATE DEVELOPMENT OF RAISING 67
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.
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)
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.
70 CHRISTOPHER HIRSCH AND KEN WEXLER
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
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.
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
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 71–107.
© 2007 Springer.
72 JAMES H. YOON
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.
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.
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).
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.
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.
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).
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
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.
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).
(15) ….
DP’i V’/VP
XP V
: movement
DPi ZP : coindexing
: predication/theta-role assignment
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
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 79
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:
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:
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.
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).
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.
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
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 83
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
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)).
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.
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 85
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.
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):
(28) …… v’
VP v
WP V’
ZP V
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):
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
(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.’
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).
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.
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 89
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
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
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.
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.
(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).
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
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.
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
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 93
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.
(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
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.
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 95
Adjuncts do not undergo Passive even when they are nominal, as is well known:
Neither does the P(roleptic) Object in English. Passive of (44a) is bad, with or
without the preposition accompanying the P-Object.
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
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
VP VP
?? V’ DP V’
DP
ZP V [e] XP V
Opi Z’ ZP
DP
Opi Z’
ei
ei
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 97
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.
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)
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
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.
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.
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
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
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.
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:
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:
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
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 103
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.
MAJOR ARGUMENTS IN KOREAN (AND JAPANESE) 105
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:
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.
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).
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:
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.
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
1. INTRODUCTION
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).
111
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 111–131.
© 2007 Springer.
112 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
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):
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
(6) . . . [vP perimenan [DP ton kapetanio]i [CP proi na ferthi me aksioprepia ]]
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
114 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
their subjects. In other words, the absence of finiteness signals the appearance of
RtO/ECM as a last resort, case-marking mechanism:
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).
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.
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):
NOT REALLY ECM, NOT EXACTLY CONTROL 115
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:
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:
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):
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.
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:
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).
NOT REALLY ECM, NOT EXACTLY CONTROL 117
*
(16b) evala ton kosta na tighanize psaria
made-1SG the Kostas-ACC SUBJ fried-3SG fish-ACC
‘I made Kostas fry fish.’
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.
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
118 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
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:
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
120 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
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.
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:
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.
NOT REALLY ECM, NOT EXACTLY CONTROL 121
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:
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:
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:
The above constructions, with the verbs perimeno ‘expect’ and pitho ‘persuade’,
are grammatical if the clitic doubles the accusative DP:
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.
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,
124 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
(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:
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’.
5.2.1 Passivization
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:
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:
NOT REALLY ECM, NOT EXACTLY CONTROL 127
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:
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
128 GEORGE KOTZOGLOU AND DIMITRA PAPANGELI
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’).
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
NOT REALLY ECM, NOT EXACTLY CONTROL 129
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:
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).
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:
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
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
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 133–157.
© 2007 Springer.
134 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
(2b) ‘Degrees of Agreement’: Overt agreement morphology and abstract Agr are related via
individual phi-features, not the phi-complex as a whole.
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.
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.
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):
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,
CONTROL IN MODERN GREEK 137
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.
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
138 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
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.
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
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:
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
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.
140 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
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.
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:
(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.
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.
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
142 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
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):
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).
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
CONTROL IN MODERN GREEK 143
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:
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).
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?
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.
CONTROL IN MODERN GREEK 145
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?
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.
The relevant derivation is given in (28) – recall that derivations are strictly bottom-up.
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):
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):
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.
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
CONTROL IN MODERN GREEK 147
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
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.
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.
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
148 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
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).
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.
This section briefly considers certain questions for and implications of our analysis,
pointing the way toward areas of potentially fruitful further research.36
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
(43) MP
na NegP
min TP
Spec T⬘
T AgrP
V Spec Agr⬘
Agr VP
V Spec V⬘
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
CONTROL IN MODERN GREEK 151
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).
(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):
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
152 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
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).
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.
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.
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
154 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
* 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:
CONTROL IN MODERN GREEK 155
‘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.
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.
156 KONSTANTIA KAPETANGIANNI AND T. DANIEL SEELY
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.
CONTROL IN MODERN GREEK 157
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
1. INTRODUCTION
159
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 159–183.
© 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).
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 ]]]]]]
The subjunctive in Greek can be found both in main and embedded (complement
and adjunct) clauses:8
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:
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’
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
FINITENESS AND CONTROL IN GREEK 163
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:
3. CONTROL IN GREEK
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:
However, the control pattern attested in these cases does not have the properties
of exhaustive control.14 First, partial control is permitted (12):
Second, the controlled null subject can take split antecedents (13):
Moreover, Control Shift is also attested. Example (13) involves a violation of the
Minimal Distance Principle on control (Rosenbaum 1967):
FINITENESS AND CONTROL IN GREEK 165
(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’
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.
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)
166 VASSILIOS SPYROPOULOS
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.
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:
FINITENESS AND CONTROL IN GREEK 169
(24) DS = PC
(25) AS = EC
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
In control subjunctives nominal predicate modifiers that modify the null subject
always appear in nominative case.
FINITENESS AND CONTROL IN GREEK 171
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.
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):
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.
FINITENESS AND CONTROL IN GREEK 173
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:
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):
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:
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
FINITENESS AND CONTROL IN GREEK 175
controlled pro in Greek? (b) How is control derived, since pro by itself is not refer-
entially restricted?
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:
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’
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.
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):
b. Dependent subjunctive
[CP C0[iTense] [MP M0 [TP T0max[T0, Agr] … ]]]
c. Anaphoric subjunctive
[CP C0 [MP M0 [TP T0max[T0, Agr] … ]]]
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
178 VASSILIOS SPYROPOULOS
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
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
180 VASSILIOS SPYROPOULOS
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):
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
FINITENESS AND CONTROL IN GREEK 181
(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’:
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:
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).
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:
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
1. INTRODUCTION
187
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 187–211.
© 2007 Springer.
188 GABRIELA ALBOIU
(2) a. Mrs. Dallowayi wanted [PROi+ to meet at 9] (but it was too early)
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
(5) [TP John T [vP John tried [TP to [vP John v [VP read the new Chomsky]]]]]
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.
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
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
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):
b.
[ VictorD încearca [θ1 <încearca> [sa cînte θ2 la trombonD <cînte> θ3 ]]]
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.
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.
(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)]
‘Victor is trying to play the trombone.’
MOVING FORWARD WITH ROMANIAN BACKWARD CONTROL 195
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.
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
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).
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.
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
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
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).
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.
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.
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.
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.’
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
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.
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:
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.
MOVING FORWARD WITH ROMANIAN BACKWARD CONTROL 203
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.
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]]]
‘Victor is trying to play the trombone.’
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
(35)
v VP
u[θvm]
V CLOWP
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.
MOVING FORWARD WITH ROMANIAN BACKWARD CONTROL 205
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
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
(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).
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]]]
‘Victor / Someone is trying to play the 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
MOVING FORWARD WITH ROMANIAN BACKWARD CONTROL 207
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.
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.
6. CONCLUSIONS
*
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.
MOVING FORWARD WITH ROMANIAN BACKWARD CONTROL 211
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).
Affects truth-
A-bar chain [Foc] formal Prosodic functional
effects feature marking values of S
Contrastive focus + + + +
Rhematic focus − − − −
CILENE RODRIGUES
1. INTRODUCTION
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.
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
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.
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) ).
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.
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.
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
218 CILENE RODRIGUES
(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
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
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.
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
220 CILENE RODRIGUES
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.
(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
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.
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
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
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).
(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.
AGREEMENT AND FLOTATION IN PARTIAL AND INVERSE 223
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
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.
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
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
(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]]]]]
AGREEMENT AND FLOTATION IN PARTIAL AND INVERSE 225
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).
(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.
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
226 CILENE RODRIGUES
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).
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.
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
(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
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.
AGREEMENT AND FLOTATION IN PARTIAL AND INVERSE 229
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.
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
231
W.D. Davies and S. Dubinsky(eds.), New Horizons in the Analysis of Control and Raising, 231–248.
© 2007 Springer.
232 MARCELLO MODESTO
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.’
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
(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.
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.
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
234 MARCELLO MODESTO
(9) a. Num = {T2, F1, o1, João1, disse1, que1, comprou1, um1, carro1 }
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
[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).
(13) a. Num = {T2, o1, João1, disse1, que1, ele1, comprou1, um1, carro1 }
c. [TP ele2 Case,φ [T’ [comprou1φdef, D + Tφ.EPP] [VP t2 [V’ t1 [DP um carro]]]]]]
(14) TP
[ 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).
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.’
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
236 MARCELLO MODESTO
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]
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.
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.
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
238 MARCELLO MODESTO
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.’
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.’
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.
(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.’
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
240 MARCELLO MODESTO
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):
(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.
(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.
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.
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:
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.
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.’
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
NULL SUBJECTS IN BRAZILIAN PORTUGUESE AND FINNISH 243
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
244 MARCELLO MODESTO
(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
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
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
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
NULL SUBJECTS IN BRAZILIAN PORTUGUESE AND FINNISH 245
c. [TopP o Feco2 [FP t2 [TP a Dani1 convenceu t2 [CP que [FP e2 [TP t2 pode se
the Feco the Dani convinced that can self
eleger]]]]]]
to.elect
d. [TopP o Feco2 [FP a Dani1 [TP t1 convenceu e2 [CP que [FP e1 [TP t1 pode se
the Feco the Dani convinced that can self
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.
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
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
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.
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.
NULL SUBJECTS IN BRAZILIAN PORTUGUESE AND FINNISH 247
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:
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
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.
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
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 251–262.
© 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?
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).
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
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:
254 CEDRIC BOECKX AND NORBERT HORNSTEIN
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.
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).
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)?
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
ON (NON-)OBLIGATORY CONTROL 255
(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
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
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
256 CEDRIC BOECKX AND NORBERT HORNSTEIN
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).
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).
ON (NON-)OBLIGATORY CONTROL 257
5. EXTENSIONS
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.
258 CEDRIC BOECKX AND NORBERT HORNSTEIN
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.
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).
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
ON (NON-)OBLIGATORY CONTROL 259
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
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
260 CEDRIC BOECKX AND NORBERT HORNSTEIN
* 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
(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:
(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.
ON (NON-)OBLIGATORY CONTROL 261
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.
262 CEDRIC BOECKX AND NORBERT HORNSTEIN
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
1. INTRODUCTION
This paper discusses control into wh-infinitivals as illustrated in the following examples.
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.
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
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 263–279.
© 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.
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.
a. implicative
b. aspectual
c. factive
d. propositional
e. desiderative
Note that forget and remember can also take wh-infinitives, as the following
example shows.
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
Here, we argue that control into wh-infinitivals can be either obligatory or generic.
Consider, first, the following sentences.
(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):
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.
(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.
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.
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.
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).
(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.
Note that the sentences in (10–12) all are incompatible with progressive aspect,
thus supporting an analysis of generic control for these sentences.
(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.
CONTROL AND WH-INFINITIVALS 269
(36) Which shares can John’s new program figure out when to buy?
(38) What kind of bee does Mary know how to defend oneself against?
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:
(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.
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:
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.
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
CONTROL AND WH-INFINITIVALS 271
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.
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
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:
(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:
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.
CONTROL AND WH-INFINITIVALS 273
(61) Which shares can John’s new program figure out when to buy?
Thus, (61) has an embedded reading on the DP which shares, while (62) must have
a matrix reading on why.
(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.
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.
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
(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.
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).
CONTROL AND WH-INFINITIVALS 275
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:
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:
Multiple sluicing is acceptable in both exhaustive control and partial control environ-
ments. Let us examine the results with wh-infinitivals:
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
CONTROL AND WH-INFINITIVALS 277
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.
The core proposal here is that the right edge of CP is structured as follows:
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.
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]]].
(88) Whatj does Johni know [FinP ti [WhP when [IP ti wash tj ]]].
(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).
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.
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.
CONTROL AND WH-INFINITIVALS 279
4. CONCLUSIONS
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:
b. Evak said that Kimi forced/advised Suej [PRO*i/j/*i+j/*k to take the lead]
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
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):
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.
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.
b. Kim promised/offered/asked Sue (for) the apple. (transfer after reference time)
(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
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)
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.
(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.
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).
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.
CONTROL VIA SELECTION 287
b. subevent structure [Tr e1* act (Kim, (Sue, leave)) {en+1} leave (Sue) ]Tr
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.
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].
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.
c. Plain English: At the event time e1*, Kim expresses volition with respect to a future
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.
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 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.
CONTROL VIA SELECTION 289
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.
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
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
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:
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).
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.
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.
CONTROL VIA SELECTION 291
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
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
W.D. Davies and S. Dubinsky (eds.), New Horizons in the Analysis of Control and Raising, 293–325.
© 2007 Springer.
294 IDAN LANDAU
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.
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’.
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
(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.
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.
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 297
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.
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.
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).
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).
This observation is enough to rule out (12a), on a par with (11b). Furthermore, a
commitative paraphrase is not necessary for PC.
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.
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 301
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.
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.
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.
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 303
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.
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.
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 305
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
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
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).
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.
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.
Control
DP1.DAT …V … [ PRO1 … Vpsych …]
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
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
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 309
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.
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).
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.
MOVEMENT-RESISTANT ASPECTS OF CONTROL 311
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).
8. FINITE CONTROL
Hornstein 2003 (Sect. 1.8.2) describes OC into finite clauses in Brazilian Portuguese
(BP), as below:
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.
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 313
9. PRO-GATE
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.
(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).
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.
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:
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,
MOVEMENT-RESISTANT ASPECTS OF CONTROL 315
(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).
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.
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.
(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
(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.
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.
MOVEMENT-RESISTANT ASPECTS OF CONTROL 317
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
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).
However, the two constructions cannot be combined – compare (57a) and (57b).
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
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).
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 319
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.
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
Landau 2003 notes that the option of ‘sideward movement’ – used by Hornstein
to explain OC into adjuncts – overgenerates examples like (64).
(65) a. *John’s1 examination of 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.
(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.
(68) a. *Lots of things can be the matter with your transmission without being the
matter with mine.
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.
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
MOVEMENT-RESISTANT ASPECTS OF CONTROL 323
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].
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.
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.
Amsterdam: John Benjamins.
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
Papers in Linguistics 15:57–83.
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.
Amsterdam: John Benjamins.
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
Linguistics 5:44–53.
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
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
Linguistic Theory 22:811–877.
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:
Oxford University Press.
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
Roussou, Anna. 2000. On the left periphery: Modal particles and complementizers. Journal of Greek
Linguistics 1:65–94.
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:
Oxford University Press.
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
Linguistics 1:144–63.
340 BIBLIOGRAPHY
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
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
Studies in Natural Language and Linguistic Theory
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
Studies in Natural Language and Linguistic Theory
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