Professional Documents
Culture Documents
Problems: Syntax
Problems: Syntax
Problems: Syntax
IN SYNTAX
STUDIES IN LANGUAGE
Editorial Board :
Noam Chomsky - Femand Vandamme
Liliane Tasmowski - Dominique Willems
PROBLEMS
IN SYNTAX
W. Abraham, Cl. Blanche-Benveniste, A. Colen, B. Comrie, M. Dascal
R. Declerck, S. Dik, M. Dominicy, G. Fauconnier, J. Koster, S. Kuno
J. McCawley, L. Melis, C. S. Smith, W. van Langendonck, A. Zaenen
Edited by
Liliane TASMOWSKI
Dominique WILLEMS
Plenum Press,
a Division of Plenum Publishing Corporation, 233 Spring Street, New York 10013
ISBN-13: 978-1-4612-9691-1 e-ISBN-13: 978-1-4613-2727-1
DOl: 10.1007/978-1-4613-2727-1
@ 1983. Communication and Cognition, Blandijnberg 2, B-9000 Ghent.
English edition first published in 1983 simultaneously by
a) Communication and Cognition, Ghent
b) Plenum Publishing Corporation.
Softcover reprint of the hardcover 1st edition 1983
Presentation ix
vii
viii CONTENTS
Marc DOMINICY, Time, tense and restriction (On the French peri-
phrasis 'venir de + infinitive') 325
References 377
PRESENT AnON
II. The themes developed by the contributors to this volume can be divided
into three main areas :
1. articles dealing with problems of definition or recognition of functions
and categories.
2. articles dealing with constructional properties of the sentence
3. articles dealing with tense and aspect.
Two further articles are exclusively concerned with the general question of the
organization of grammar.
1. What characterizes the first group of articles, concerned with the definition
and delimitation of certain complement-classes or functions, is the variety in the
range of the proposed arguments and the differences in hierarchy between them.
In the relation between syntactic and morphological arguments, for instance,
Werner Abraham and Bernard Comrie defend opposite points of view.
ix
x L. TASMOWSKI - D. WILLEMS
Despite the divergences, e.g. concerning the basic entities that are posited,
common properties emerge. Most of the proposed frameworks show little
interest in formalizing, do away with the notion transformation and are
concerned with a more surface oriented syntax - relational grammar being an
absolute exception. The tendency to be as general as possible and to integrate
a variety of facts within the same analysis is striking. Finally, the importance
given to lexical, semantic and pragmatic considerations in the argumentation
is to be underlined.
As for the research strategies used and the relation between theory and facts,
the contributions could be grouped into two main tendencies: those which,
starting from the facts, classical or new ones, tend. to evaluate the existing
theories and/or look for an adequate description which would account for the
facts (Declerck, Van Langendonck, Zaenen, Melis, Kuno) and those which,
starting from a general rule', look for adequacy throughout a variety of data
(Fauconnier, Dik).
RENAAT DECLERCK examines the various attempts made by trans-
formational grammars to establish a deep structure that could account for the
properties of English verbs of perception. He demonstrates the practical failure
of every such attempt, the most adequate seeming to be Postal's Raising analysis.
WILLY VAN LANGENDONCK also starts from the impossibility of a trans-
formational account for passive sentences in Dutch. He opts for dependency
grammar, which allows him to account for the variety of constructions available
in Dutch to mask the agent.
ANNIE ZAENEN, adducing material from Icelandic, shows how V/I is
triggered by a "topicalized" context. As Chomskyan scholars make stylistic
inversion dependent on WH-movement, they offer no general solution here. The
facts are, on the contrary, naturally fitted by lexical-functional grammar as
developed by Kaplan and Bresnan.
Without situating himself in a particular framework, LUDO MELIS analyses
sentences involving French transitive causatives of movement. He proposes
two different syntactic patterns for J'envoie Marie promener, "I-send-Mary-
walk", and J'envoie promener Marie, "I-send-walk-Mary". Treating as
complementary the figurative and the concrete use of the first pattern, he posits
a structure with a prepositional complement of locality for it.
GILLES F AUCONNIER treats together a whole range of facts formerly
considered to be quite distinct: French factitives, passives, predicative
constructions and tense. They are analysed as different manifestations of a single
principle, Union, the amalgamation of two predicates into one, which provokes
xii L. TASMOWSKI - D. WILLEMS
an intuitive way. They appear however with divergent status, as primitive entities
(in functional grammar), as semantic labelling of functions (in relational
grammar and traditional grammar) or as elements appearing in the description
of the lexicon (Chomskyan grammar). This signifies that the relevance of
semantics for syntax is at least implicitly recognized. A similar phenomenon
shows up in the lexical subclassification of the verbs, taken together under the
heading "verbs of movement" or "of causation" for instance, or in the sub-
classification of the nouns, such as "picture nouns", applied in many of the
contributions. In this connection it should be remembered that other trends,
not represented in this volume (cf. Gross), try to measure the impact of syntax
on semantics, precisely by defining and delimitating such lexical subclasses on
the basis of their syntactic behaviour. It may also be noticed that McCawley
more specifically calls attention to the general significance of the morphological
categories. Semantic correlates are thus ascribed to both functions and (sub)-
categories.
1.2. Another manifestation of the importance of semantics and pragmatics
for syntax is to be found in the explanatory principles admitted for the constraints
on particular constructions. It is for instance argued by Kuno that in reality
semantic and contextual factors, such as the degree of agentivity or of aware-
ness of the subject and the tension between topic and subject, are involved in
cases where explanations used to be sought in the purely syntactical Specified
Subject Condition.
The ever increasing importance of non-syntactic descriptions is also notice-
able in the way particular constructions have come to be handled, the passive
(here studied by Van Langendonck) being a case in point. It may be remembered
that, as they were fully aware of the differences in theme/rheme between an
active and a passive construction, most traditionally oriented European scholars
never accepted the idea of a transformational, purely syntactic relation. It
sufficed to falsify the active/passive synonymy to recognize the passive as a
construction in its own right. But the argument starts as to what should be
considered as such. As no single morphological device is unambiguous, semantics
and pragmatics come in again, finding, for instance, a partial explanation and
definition in the principle of the theme/subject coincidence. The range of
possible "passives" left open by such an approach induces one to look for the
semantic impact of every device to be found in the construction.
1.3. It is of course to be expected that the more deictic elements come in,
the less syntax will playa primary role, as is inaeed the case in the studies on
tense and aspect. Here, syntactic representation is only seen as a means of
PRESENTA nON xv
2.DEPE~ENCY· 5.~CALREP~ATION
"
~. LEXlCON~4.SJAX/
t 9. SEMANTIC FEATURES
In this scheme, functional grammar passes through all the stages, and this is in
general true of European traditional perspective. Dependency grammar brings
together 2, 4 and 6 and lexical functional grammar 3, 4 and 5. Relational
grammar situates itself at 1, 2, 3, and 4. Functional sentence perspective is
concerned with 4, 6 and 7. Problems of tense and aspect are approached through
5, 6 and 7. McCawley passes from 9 to 8 before attaining 4. Finally 4 is the
central concern of Koster's contribution.
IV. The articles irt this volume, by their variety and originality suggest some
directions for further research.
1. The broadening of the perspective calls into question the definition of the
object of investigation. For more than twenty years, language was generally
simply equated with the infinite set of grammatical sentences of that language,
a point of view that is challenged here by McCawley. But it is not clear whether
the notion "grammatical sentence" has been adequately defined. On the one
hand, since a presentential place is foreseen in which the topic is generated, one
wonders if oral utterances such as Cette villa, on ne peut pas dormir ("This
villa, one can't sleep", produced by Blanche-Benveniste), which does not fit
into any kind of left dislocation, must be accounted for. And the reasons why
a post-sentential place for some coda isn't provided for (cf. J'ai un bon conseil
a te donner, toi "I have some good advice to give you, you", or C'est sur qu'il
y ressemble a un cluImeau "Sure-that-he-resembles-it, -a-camel" as compared with
?A un chameau, c'est sur qu'il y ressemble) are not obvious. It seems that the
left and right boundaries of the sentence are not definitely fIXed.
If, on the other hand, one leaves autonomous syntax and admits the influence
of non syntactic factors, the notion "grammatical sentence" must become
undetermined. Indeed, the acceptability of an utterance is a function of prosodic
factors (La~me Jean veut\vs .. */ii pomme Je"iiiiVeilt "The apple John wants"),
of semantics (John said of Mary that Tom is boring is only acceptable if Tom
PRESENTATION xvii
several criteria are used, the necessity of hierarchy arises, which forms a new
source of divergences. Therefore, some linguists, such as Fauconnier, will take
subject as an undefined primitive. Dik considers it to be the product of a choice
at a certain stage of the generation, and for Koster it is probably configurational-
ly determined with respect to some head.
This situation certainly results from the impossibility of laying down a set
of syntactic postulates that might in some way be considered "principled".
Admitting that an adequate characterization of every language utterance would
be that it is physicalization of thought, and that this is necessarily so because
communication rests on perception, and supposing that predicativity is an
essential feature of communication, then a system taking that notion, or an
analysis of that notion, as a primitive term, would seem to rest on rather
plausible grounds. And indeed, if similarly conditioned primitives could be
established for each separate level of analysis, some comparison between the
explicit choices would become possible.
4. Another point that needs clarification is the extraordinary variety of
linguistic data presented here. First, the customary set of facts - the positive
ones and the complementary ones that have to be filtered out - is extended by
the addition of tokens from spoken language, dialect variants, historical data,
data from other languages, with or without historical links, and data from
language acquisition. Their complexity can furthermore be enhanced by the
recognition of various styles and registers of language, in fact implicitly present
in every grammaticality judgement, since saying that a particular construction
belongs to the language often implies that it is acceptable in a particular
situation, in accordance with the speaker's status, point of view, knowledge and
intentions. Variety being as essential a characteristic of language as regularity,
it certainly has to be taken into consideration. But if the primary data are a
heterogeneous collection, varying along the dimensions of space, time, speakers
etc., the rules themselves provide theoretical unity and delimit the object of
investigation, free to restrict or extend the range of facts that are considered.
Now, apart from logical questions of internal consistency and scientific require-
ments of explicitness, one evaluation criterion for a theory is its adequacy
with regard to the facts: the number of facts it is able to predict and the
adequacy of the prediction with regard to the known facts. In the actual absence
of a theory of variation and of an explicit stratification of linguistic facts
however, the criterion of descriptive adequacy seems quite arbitrary, not to say
useless.
PRESENT AnON xix
Liliane Tasmowski
Dominique Willems
SYNTACTIC-MORPHOLOGICAL DISCREPANCIES IN
MALTESE SENTENCE STRUCTURE*
Bernard Comrie
University of Southern California, Los Angeles
1. Introduction
3
4 B. COMRIE
The data used in the present paper are from Maltese, the dominant and
co-official (with English) language of the Republic of Malta in the Mediterranean,
some S8 miles south of Sicily. Maltese is, genetically, a variety of North African
vernacular Arabic. However, since the twelfth century Malta has been culturally
cut off from other forms of Arabic and subject to a strong Sicilian and, later,
standard Italian influence: this influence has been primarily on the lexicon
(including phraseology), and has also led to certain changes in morphology (in
the forms themselves, not in the range of categories distinguished). More recent-
ly, Maltese has borrowed a number of lexical items from English, though these
are not in general integrated into the overall system of the language, as the
Siculo-ltalian loans are. Maltese is treated here, however, purely from a
synchronic perspective, a part from some brief diachronic asides in section 7.
Although I have not systematically compared the aspects of Maltese syntax to
be discussed below with the corresponding constructions in other forms of
Arabic, unsystematic glances at what information is available on the syntax of
Arabic vernaculars, in particular those of North Africa, suggest that many of
these aspects may be quite widespread within vernacular Arabic (and thus not,
incidentally, attributable to Siculo-ltalian influence on Maltese). Development
of this theme, however, and also comparison with Classical Arabic and Modern
Written Arabic, must remain a task for future research. Maltese is here treated
simply as a language in its own right.
Some general features of Maltese sentence structure must be given before
the detailed presentation of the following sections. The basic word order is
Subject - Verb - Object, although other word orders, including Verb - Subject -
Object (the older Semitic order), are found; for further discussion, see Agius
& Kalmar (forthcoming). Most of the other features of canonical Verb - Object
order are present: in particular, adjectives and genitives follow their head noun,
auxiliaries precede the main verb, and Maltese has prepositions rather than post-
positions. These features are illustrated and elaborated in the discussion below.
SYNTACTIC-MORPHOLOGICAL DISCREPANCIES 5
2.1. Subjects}
Given that the verb shows the person and number (and gender) of the subject
to the same extent as do separate pronouns (except for the limited syncretism
of first and second person Singular in the Perfect, and of second person singular
and third person singular feminine in the Imperfect), subject pronouns are
usually omitted if unstressed; we may thus distinguish between unstressed and
stressed pronouns, a distinction that will be relevant also in the discussion of
direct objects. The following sentences illustrate these points :2
Other complex forms utilize a conjugated auxiliary verb. Thus, addition of the
Perfect of 'be' before a conjugated verb form throws the time reference of that
verb form further into the past: if the main verb is Imperfect, the sense is past
imperfective; if the main verb is Perfect, the sense is pluperfect:
Note that all the conjugated verb forms in such complexes must be of the same
person, number and gender: combinations like kon-t kiteb are ungrammatical
and meaningless; Maltese does not, incidentally, have an infinitive verb form
lacking person, number and gender (though there are nouns derived from verbs,
with clearly nominal rather than verbal form and function).
More complex forms also occur. For instance, the time reference can be
thrown into the future by using sa with the Imperfect of 'be' as an auxiliary,
or progressive qed can be added to complex forms whose first conjugated form
is Imperfect; these more complex forms, however, add nothing to the data
possibilities that are relevant to the present discussion.
In addition to verbal sentences, Maltese also has verbless (nominal) sentences,
namely with an adjectival, noun phrase, or adverbial predicate in the present
tense. (In other tenses, the verb kien 'be' is used.) Except in elliptical sentences,
the subject must be expressed even if an unstressed pronoun (given that there is
no possibility for Subject-Verb-Agreement). If the subject is a full noun phrase
then, under conditions which merit more detailed investigation, a coreferential
pronoun may be inserted between subject and predicate:
(8) Hu tabib.
he doctor
'He is a doctor. '
(9) Hu fqir.
he poor
'He is poor.'
(10) Hu hawn.
he here
'He is here.'
(11) Pawlu (hu) tabib.
Paul he doctor
'Paul is a doctor.'
(12) Marija (hi) fqir-a.
Mari she poor-F
'Mary is poor.'
(13) Pawlu hawn.
Paul here
'Paul is here'.
preposition iii (variant 'I), which otherwise indicates indirect objects (recipients),
some direct objects allow this preposition, while yet others cannot take lil.
The most detailed discussion to date is by Borg (1981: chapter 2) and can be
summarized as follows: lil is required for stressed pronouns and for proper
names of persons; lil is possible with other definite animate noun phrases, and
marginally possible with definite concrete inanimate noun phrases; otherwise
(with indefinite noun phrases, and with inanimate, especially abstract, noun
p!uases), it is impossible. By constructing paradigms of sentences including
noun phrases from these various sets, one can tell whether a given noun phrase
is a subject (never with iii), an indirect object (always with Ii!), or a direct object
(presence or absence of lil according to the above criteria). Illustrative examples
of direct objects are:
(Note that Maltese has no indefinite article; the definite article is (i)I-, with
morphophonemic alternants that are not relevant here.)
For unstressed personal pronouns, Maltese has a special set of verb suffixes,
as set out in Table 2. These suffixes are always distinct from the corresponding
Singular I qatl-it-ni
2 qatl-it-ek
3 masculine qatl-it-u
feminine qatl-it-ha
Plural I qatl-it-na
2 qatl-it-kom
3 qatl-it-hom
Note: After vowels, the third person singular masculine suffix takes the form -h.
SYNTACTIC --MORPHO LOG ICAL DISCREPANCIES 9
Note that topics which are not subjects do not trigger Verb-Subject-Agreement,
i.e. such agreement distinguishes between subjects and topics. It is also possible
to have, without Topicalization, cooccurrence of a full noun phrase direct
object and a coreferential object suffix:
The conditions under which sentences like (20) are possible, and the distinction
in force, if any, between (20) and (16), require further investigation.
As noted in section 1.2, Maltese has prepositions, and these combine with
a following full noun phrase without any further marking. Thus, with the
preposition gtumd 'at the place of (cf. French chez), we have ghand Pawlu
'at Paul's place' ,g~and ii-mara 'at the woman's place'. With unstressed pronouns,
however, pronominal suffixes are attached to the preposition, as indicated in
Table 3; thus 'at my place' isghand-i. These suffixes are the same as those for
direct objects of verbs, except in the first person singular (-i, or -ja after vowels,
10 B. COMRIE
Singular I gtiand-i
2 g"ttand-ek
3 masculine gnand-u
feminine gtiand-ha
Plural I g"ttand-na
2 g"ttand-kom
3 ghand-hom
Note: After vowels, the first person singular suffix has the form -ja, and the
third person singular masculine the form -h.
Finally in this section, we may note that Maltese also has indirect object
suffixes on verbs: these take the form of an / (etymologically cognate with the
preposition lil) followed by the prepositional object form of the pronoun
suffixes, e.g. kitb-it-/-i 'she wrote to me', kitb-it-hom-/-i 'she wrote them to me'.
Such forms playa marginal role in what follows (section 6.1).
3. Problematic constructions
Where English uses the verb haJle to express a possessive predicate, Maltese
uses a construction which can be described formally in the following terms.
SYNTACTIC-MORPHOLOGICAL DISCREPANCIES 11
The difference in word order (g'lia.nd typically before the item possessed, but
after the item located) could readily be attributed to the difference in topic-
comment structure. Thus an analysis close to the form of (21), equating
morphological and syntactic structure, would say that the possessed item is
subject of the sentence, and that gtuznd is the locative preposition, taking as
its object the possessor.
The alternative analysis, claiming a morphological-syntactic discrepancy,
would claim that in (21), the prepositional object suffix in fact encodes the
subject, i.e. the possessor is a morphologically irregular subject; the item g~nd
would be an irregular verb, agreeing with its subject as if that subject were a
prepositional object; while ktieb would (probably - see footnote 4) be
considered a direct object. Presented as baldly as this, the second analysis might
seem woefully inadequate, if not downright immoral: criticism of it would
center on the complete failure to consider the forms of Maltese in their own
right, attempting rather to impose on Maltese the alien syntactic structure of
the English construction I have a book. For now, I simply present the two
analyses, to return in sections 4-6 to demonstrate that the second analysis,
despite first appearances, is in fact the correct analysis of this construction.
3.2. Quasi-auxiliaries
group of forms with similar morphological and syntactic behavior, all of which
indicate various tense-aspect or mood features of the sentence. They are g~ad
'still', gliodd 'almost', donn 'seem', and qis 'be like, as if. Each of these items
functions as an invariable stem, to which can be attached direct object suffixes;
this form is then followed by a finite verb with a subject affix corresponding
to the direct object affix of the quasi-auxiliary:
The analysis following form would argue that in a form like gliad -ni the suffix
-ni correlates with a direct object, and indeed the form -ni is, qua form, unequi-
vocally the first person singular direct object suffix. The alternative analysis
would say that these suffixes correlate with the subject, i.e. the quasi-
auxiliaries are irregular verbs which agree with their subjects (as all Maltese
verbs must do), but by taking the morphology of direct object suffixation.
With quasi-auxiliaries, the first analysis, following form, can in fact be taken
a step further. The stems g~odd-, donn-, and qis- are etymologically the
imperative singulars of verbs meaning 'consider', 'consider', and 'measure',
respectively, and gfiodd and qis still function synchronically as imperatives of
these verbs 3 . Thus, for gfiodd and qis at least, one might maintain, in terms of
an analysis advocating one-one correspondence between form and function,
that these are synchronically second person singular imperatives, i.e. not only
does the suffix not refer to a subject (but to a direct object), but there is a
different subject, second person singular. Below, I will show that either variant
of the first analysis is incorrect. Incidentally, gfzad seems to have a different
origin: it still survives as an adverb, and can be used before an Imperfect verb
to indicate future time reference:
Note that in this use, it takes no suffix. However, it is not clear that the two
uses of g~ad can be given a uniform analysis, even by those who advocate
absolute correspondence of form and function, since the quasi-auxiliary has one
clear verb property that the adverb gtiad lacks: it takes the unequivocally
direct object suffix -ni (first person singular).
The discussion of sections 1-3 essentially recapitulates and reorganizes the
treatment of these various constructions given, with various degrees of explicit-
ness, in traditional grammars of Maltese. Even the competing analyses set out
above can be found, implicitly or explicitly, in such traditional grammars. The
original contribution of the present paper, however, contained in sections 4-6,
is to show that it is possible to evaluate between these competing analyses by
careful consideration of the properties of verbs and grammatical relations in
Maltese, without any appeal to aprioristic or aesthetic considerations. Although
some of the criteria used below have been alluded to occasionally in earlier
works -- for instance, Schabert (1976: 133-134) mentions negation as a test for
verbhood -, I am not aware of any attempt as detailed as this to marshal the
evidence in favor of one analysis or the other.
In the true locative construction, with the preposition g1iand in its literal
meaning 'at (the place ot)', the preposition governs a full noun phrase directly,
i.e. full noun phrase and prepositional object suffix are in complementary
distribution:
construction, g'liand must always have a suffix. If the possessor is a full noun
phrase, then it simply precedes gtzand, which takes the suffix of the same person,
number and gender as the possessor:
In the possessive sense, it would not be possible to say ghand Pawlu bajda or
gtumd Marija bajda, yet this is precisely what would be predicted by the analysis
which claims a one-one correspondence between morphology and sentence
structure.
One might ask how adherents of such an analysis might in fact deal with
the data of (30)-(31). The only obvious solution, maintaining other aspects of
their position, would be to say that some kind of Topicalization is involved, i.e.
Pawlu is topic, not subject, of (30) and, as usually in Maltese, the topicalized
nonsubject is taken up again by an unstressed pronoun. However, the absence of
any untopicalized version of this sentence (in particular *g1fand Pawlu bajda)
makes this dubious as a synchronic analysis. (For the diachronic possibilities,
see section 7.) In other words, attempting to maintain a neat one-one corres-
pondence between form and syntactic structure in one area leads only to
complication elsewhere, since the putative Topicalization would not have the
distinctive function it has elsewhere, where topicalized and nontopicalized
variants contrast.
The second analysis presented in section 3.1, however, actually predicts the
data of (30)-(31). Since the possessor, on this analysis, is a subject, it naturally
comes first in the sentence, preceding the verb. Since gTiand is analyzed as a
verb, it naturally follows the subject, and obligatorily agrees with the subject in
person, number and gender. The only irregularity is that gfiand agrees with its
subject by taking prepositional object suffixes rather than subject suffixes,
which is of course the hallmark of the analysis advocating morphological-
syntactic discrepancy.4
SYNTACTIC- MORPHO LOG ICAL DISCREPANCIES 15
It is not possible to construct the full noun phrase as a direct object of the quasi-
auxiliary:
If (33) means anything, it is 'consider the rat; he has fallen', i.e. with gtlOdd
taken as the imperative of 'consider'.
As with the possessive predicate construction, advocates of the 'one form -
one function' analysis would be forced to establish an obligatory instance of
synchronic Topicalization, lacking' a nontopicalized variant, to make any sense
at all out of sentences like (32). The alternative analysis, however, again actual-
ly predicts the observed pattern: in (32), ii-far is subject, therefore the verb
gflOdd must agree with it, the only irregularity being that this verb takes direct
object suffixes to show Subject-Verb-Agreement. Incidentally, exactly the
same pattern is evinced by ghad, without there being even a conceivable inter-
pretation for (34) :
5. Negation
The provisional generalization is thus: place the circumfix rna ... -x around the
first word of the verb complex.
The situation is complicated, however, by complexes where the first word
of the verb complex is not a verb, e.g. the particle sa indicating futurity. Here
SYNTACTIC-MORPHOLOGICAL DISCREPANCIES 17
basic Negative
Singular I jien(a) M'inie·x (majinie-x)
2 int(i) m'inti-x
3 masculine hu(wa) m'hu-x (m'huwie-x)
feminine hiGa) m'hi-x (m'hijie-x)
Plural anna m'atmie-x
2 intorn rn'intorn-x
3 huma m'humie-x
(42) Hutabib.
'He is a doctor.'
(Hu) m'hu-x tabib.
'He isn't a doctor.'
(43) Pawlu (hu) fqir.
'Paul is poor.'
18 B. COMRIE
(46) If the first word of the verb complex is a verb, prefix ma (m 'before a
vowel sound) and suffix -x to that first word; if the first word is not a
verb, insert before the verb complex a form consisting of the circum fix
m(a) ... -x on a pronoun agreeing with the subject.
(47) Ikteb!
'Write !'
(La) t-iktib-x !
'Don't write!'
With the possessive construction, however, this is impossible. Instead, the cir-
cumfix rna ... -x is added tog~and with its pronominal suffix:
Here, there seems to be no way in which the 'one form - one function'
approach can provide a motivated analysis of the negative construction: the
possibility of taking the circumfix rna ... -x is elsewhere clearly a criterion for
verbhood, and the negativ~ possessive predicate construction behaves radically
differently from the negative locative construction. The second analysis,
whereby the possessor is subject and ghand is an irregular verb, predicts the data
given above, most crucially in that g~and is negated like a verb.
For most of the quasi-auxiliaries, there is no doubt about their being verbs,
so the negation test, which primarily distinguishes verbs from nonverbs, is not
directly relevant: both the 'one form - one function' analysis and its rival main-
tain that the quaSi-auxiliaries are verbs (the 'one form - one function' analysis
on the basis that they take direct object suffixes). It should, however, be noted
that this test does encompass the quasi-auxiliary gfiad, which is not a verb in
origin :
6. Other tenses
All the examples of the possessive predicate so far have been present. In the
past and future, Maltese uses a construction which, in terms of its formal
relation to the present, involves irregularity by any account. In the past,gfiand-
is replaced by kell-; in the future, gfiand- is replaced by sa jkoll-. These forms
take the same pronominal suffixes as g'liand, and have exactly the same syntax,
for instance with regard to full noun phrases and negation (though sa jkoll-,
having sa as first word of its verb complex, falls under the second par.t of
generalization (46) concerning negation) :
These past and future forms are etymologically the third person singular
masculine form of the Perfect (kien) and Imperfect (j-kun) of 'be', respectively,
followed by an indirect object pronoun suffix, although it is doubtful whether
SYNT ACnC-MORPHOLOGICAL DISCREPANCIES 21
this analysis is still felt synchronically. One point that should, however, be
emphasized is that if the possessed item in the future is feminine (e.g. bajda
'egg'), the form of the possessive predicate remains jkoll-, even though the j
is etymologically a third person singular masculine prefix, and is not replaced
by the feminine form to give *t-koll- :
Note also that in (62), the negative m'hi-x agrees with the feminine possessor
Marija; such negative pronouns never agree with the object possessed.
Since the behavior of these other tenses exactly parallels that of g'liand in
the present, the set of predictions made correctly by the analysis advocated here,
and incorrectly by the 'one form - one meaning' approach, is precisely the
same in these other tenses.
In section 2.1, I observed that the time reference of a sentence can be pushed
into the past or future by use of the Perfect or sa plus the Imperfect of kien
'be', respectively, with the further proviso that the auxiliary 'be' and the main
verb must be of the same person, number and gender. We may now turn to such
constructions with the quasi-auxiliaries. In section 3.2 we had examples like
'the rat almost fell', so let us now push this sentence further into the past, as
'the rat had almost fallen'. The following examples give parallel versions of
simple Perfect (past time reference) and Perfect of kien with Perfect of main
verb (Pluperfect time reference) of examples with 'almost' :
In each such example, the aUxiliary kien agrees in person, number and gender
with the noun phrase expressed as direct object of g'liodd. Similar examples can
be constructed withgltzd, donn, and qis.
This is clearly not what is predicted by the 'one form - one function'
analysis, in any of its variants. If gtiodd is a singular imperative, then we would
expect no grammatical sentence with an auxiliary, since in Maltese the
imperative does not combine with a preceding form of kien; or perhaps a second
person singular auxiliary, but this is not possible (except, of course, irrelevant-
ly, if the sentence is 'you had almost...'). If gtiodd is first person singular (see
footnote 3), we would expect a first person singular auxiliary, but this again is
not possible (except, of course, irrelevantly, if the sentence is 'I had almost...').
If ghodd is some kind of invariant verb, one might expect an invariable third
person singular masculine auxiliary, the usual unmarked agreement form in
Maltese, but again this is impossible (except as per footnote 6, and in any event
the possibility of ')ther 'negative pronouns' contradicts the analysis of gfiodd
as an impersonal verb).9
Yet, once again, the paradigm of (64)-(67) is precisely what is predicted by
the alternative analysis rejecting the facile assumption of 'one form - one
function'. Despite the morphology, the verb forms kon-t and gTzodd-ni have the
same subject, first person singular, as do kien-et and gfzodd-ha, third person
singular feminine. Syntactically, then, these quasi-auxiliaries (and, of course,
any auxiliaries forming a verb complex with the quasi-auxiliary) must agree in
person, number, and gender with the subject of the main verb.
The crucial aspect of the discussion in section 6.2 was the requirement, in
the construction considered, that all verb forms have the same subject. There are
several other constructions in Maltese that have a similar requirement on verb
sequences. For instance, the verb sata' 'be able' requires that the following verb
have the same subject as itself. The restriction noted in section 3.2 that the
object suffix of a quasi-auxiliary must be coreferential with the subject of the
following verb is another similar example. (The statement of this last
generalization is neutral between the 'one form - one function' and the
alternative analyses; under the latter, however, it is simply a special case of the
like-subject constraint.) By creating chains of forms involving quasi-auxiliaries,
possessive predicates, etc., we can provide further evidence of the grammatical
relation subject advocated here: I 0
SYNTACTIC-MORPHOLOGICAL DISCREPANCIES 23
The most immediate aim of this paper has been to demonstrate, with a range
of independent arguments, one particularly clear instance of morphological-
syntactic discrepancy in a natural language, across a range of constructions.
In fact, with Maltese, it is possible to find even more instances of such dis-
crepancies, though to avoid charges of 'over-kill' I will restrict myself to
discussing just one further example. ii Maltese hemm, like English there, can be
used in both locative and existential constructions:
In terms of overt form, the main difference between (72) and (73) is in the word
order. Nonetheless, the structures of the two sentences are radically different.
Sentence (72) exhibits a nonverbal predicate, as can be seen from its negative
and past correspondents:
The form m'hemm-x in (76) shows that hemm is a verb, while the invariable
kien in (77) shows that bajda is not subject, otherwise, since bajda is a feminine
noun, one would have expected third person singular feminine kien-et, as in
(75).
How, more precisely, do all these considerations relate to the 'one form -
one function' controversy? Clearly, it has been shown that there is one pervasive
aspect of Maltese sentence structure, namely the identification of the verb and
of grammatical relations, that is largely independent of the morphology yet
still interacts with other factors in determining the grammaticality or ungramma-
ticality of sentences, i.e. morphology cannot be the whole of language structure.
However, an advocate of the 'one form - one function' approach might retreat
to a slightly weaker position, arguing that while the aspects of Maltese structure
discussed in this paper do not relate directly to morphological oppositions,
there are other aspects of structure that do relate directly to such morphological
categories. All I can reply to this is that I am not aware of any such struct"ural
aspects of Maltese : certainly, in the discussion of this paper, it has never been
necessary to refer to such aspects (other than strictly in the discussion of
morphology), while it has frequently been necessary, and not just advisable, to
refer to aspects of structure not directly encoded in the morphology, for
instance in order to know how to negate sentences, place them in other tenses,
and even to replace pronouns by full noun phrases. Thus the onus is clearly on
anyone who would maintain the 'one form - one function' approach to
demonstrate that there are aspects of Maltese sentence structure that either
require or prefer direct correlation with the morphology.
Does it then follow from all this that the relation between Maltese syntax
and morphology is arbitrary, with regard to those instances where a morpho-
logical-syntactic discrepancy has been demonstrated ? With regard to the strict
synchronic analysis of Maltese, I think the answer is 'Yes'. However, from a
broader perspective, an explanation, invoking diachronic and typological
considerations, can clearly be given. During the development of the cited
SYNTACTIC-MORPHOLOGICAL DISCREPANCIES 2S
(In Arabic, all prepositions govern the genitive case.) Moreover, we know that
shifts in subject properties often arise via Topicalization, Le. by a topic being
reanalyzed as a subject; this process has been documented for several
constructions in a variety of languages by Cole et al. (1980), possessive
predicates being one of the constructions most likely to undergo this kind of
reanalysis, no doubt because possessors, being high in animacy, are more likely
to be topic than are possessed items.
There are even traces of ongoing reanalysis in Maltese, in particular of the
reanalysis of nonverbs as verbs in the kinds of construction discussed in this
paper (though the reanalysis of grammatical relations seems complete). Although
I have analyzed the possessive predicate gtzand as a verb, it still takes the pre-
positional form of the first person object suffix -i, rather than the variant ani
expected with a verb. In addition to gfiand, possession can also be indicated,
though more rarely, by the preposition bi 'with'; as a preposition, bi gives
bi-ja 'with me', but when used in the possessive construction, it has assimilated
to verb form to the extent of taking the suffix ani, Le.bi-ni 'I have'. Similarly,
fi 'in' gives fi-ja 'in me', but as a verb fi-ni 'I have in me, I contain'. Although
gRad 'still' is not etymologically a verb, it invariably takes the first person
singular suffix in the form ani. Another item which fits into the set of quasi-
auxiliaries is it- 'ago'. This is not etymologically a verb, and in the first person
26 B. COMRIE
FOOTNOTES
*In the preparation of this paper, I have benefited immensely from previous
accounts of Maltese syntax, in particular Sutcliffe (1936), Aquilina (1965),
and Borg (1981). I am grateful to Albert J. Borg for detailed comments on an
earlier version of this paper, and also to the faculty and students who offered
comments on my presentation of this material at the Linguistic Circle of the
University of Southern California and in the Linguistics Program of the
University of Utah. Interpretations of data remain, of course, my own
responsibility.
I For a discussion of subjecthood in Maltese from a rather different perspective,
including relations to semantic roles and thematic structure, see Borg (1981 :
chapter 3).
2Maltese examples are presented in standard orthography, though with the
addition of relevant morpheme boundaries. The following abbreviations are used
in glosses: DO - Direct object; F - Feminine; M - Masculine; P - Plural;
PO - Prepositional object; S - Singular; SU - Subject. Each abbreviated gloss is
made up as follows: first, a letter indicating number, followed by a figure
indicating person, followed where relevant by a letter indicating gender; then,
separated from the first part of the gloss by an oblique stroke, a letter
combination indicating whether the morpheme encoded by that abbreviation has
the form of a subject, direct object, or prepositional object affix (note that these
glosses refer solely to the form of the affix, not necessarily to its syntactic
function). These appended glosses are added after the stem gloss for subjects of
verbs in the Perfect (including the third person singular masculine, morpho-
logically nUll), for direct objects, and for prepositional objects; they are added
before the stem gloss for subjects of verbs in the Imperfect (including the plural,
where morphologically person is indicated by a prefix and number by the plural
suffix ·u). Affixes and verb stems undergo a number of morphophonemic alter-
nations, which are commented. upon here only where problems might arise in
SYNTACTIC--MORPHOLOGICAL DISCREPANCIES 27
recogmzmg a given form. Note that Maltese has two genders, masculine and
feminine, though no gender distinction is made in the plural. Nonhuman nouns
are distributed essentially arbitrarily, from a semantic viewpoint, between the
two genders; nouns ending in -a in the singular are usually feminine (e.g. bajda
,egg', baqra 'cow'), others masculine (e.g. far 'rat', ktieb 'book'). Adjectives
agree in gender and number with their noun, the feminine suffix being -a.
3Bravmann (1953: 27) argues that the forms gflOdd and donn (and, presumably,
qis, though this is not mentioned explicitly) derive not from imperatives, but
rather from the first person singular Imperfect, which in Classical Arabic took a
prefix consisting of just a glottal stop where Maltese has no. While this derivation
is as plausible phonologically as that from the imperative, it fails to take into
account the fact that the replacement of the first person singular prefix by n-
must be very old in Maltese - this replacement is found throughout Western
Arabic vernaculars, and as such must antedate the final separation of Maltese
from other forms of Arabic at the end of the eleventh/beginning of the twelfth
century. Even if the derivation from first person singular is correct, however, the
arguments below still go through, mutatis mutandis: in the synchronic structure,
the subject is the noun phrase indicated by the direct object suffix on the quasi-
auxiliary, not first person singular (unless, of course, that subject noun phrase
happens to be first person singular).
4There is, further, evidence that the possessed noun phrase is a direct object,
given the occurrence of the preposition 7 (see section 2.2) in such examples as :
(i) Gnand-i '1 Marija.
at-Sl/PO Mary
'I have Mary.'
5With certain negative words, the suffix -x is not used; thus rna ... qatt means
'never'.
6Formally, the structure NP rna Pronoun-x- Predicate is the negation of the
structure NP Pronoun Predicate, cf. examples (11)-(12) above. This latter
structure presumably arose via Topicalization of the original subject, thereby
leading to a resumptive pronoun. Diachronically, it seems that the pronoun in
this structure has been reinterpreted as a (copular) verb - despite its clearly
nonverbal morphology -, thereby acquiring the negation possibility (rna ... -x)
of a verb. This would thus be a further example of syntactic-morphological
discrepancy. However, there is evidence that the diachronic development has
moved even beyond this stage, with loosening of the relation between the
'negative pronouns' and basic pronouns. First, 'negative pronouns' occur in
many environments where the basic pronouns are impossible (other than via
dislocation), e.g. (40), (41), (45), or only optional, e.g. (43). Secondly, the third
person singular masculine 'negative pronoun' rn 'hux is used, though typically
with a shortened vowel, as an unmarked negative in constituent negation (e.g.
rn 'hux hawn 'not here'), and even shows a tendency in the current language to
replace the negative pronouns with distinctions of person, number, and gender
(e.g. rn 'in torn-x in (40), rn 'hi-x in (45».
7Note that it is also possible to negate the second verb, with a different
meaning:
28 B. COMRIE
Werner Abraham
The University of Groningen
The Netherlands Institute for Advanced Sciences at Wassenaar
In this paper three themes are taken up from recent literature on linguistic
problems of Dutch and English, and developed in some detail. Its purpose is to
show how, in certain questions of detailed linguistic analysis and description,
the morphology. of case inflection, verbal prefixation, adverbial marking,
morphological indication of directionality vs. local non-directionality as well
as the closeness of an argument indicated by lexical compounding can essential-
ly contribute to, and enhance, the quality of the identification and
interpretation of syntactic structure. This would seem to be of specific weight
when looking at related languages such as three Westgermanic ones (English,
Dutch, and German). Their typological characteristics are still much disputed
(English SVO; Dutch SVO in independent, SOY in dependent sentences; German
- V-in independent, --V in dependent sentences - this if we follow a non-
transformational surface typology), and it is even doubtful whether the three
languages are to be judged on a par with respect to their categorial analyzibility:
English is beyond doubt a language characterized by the prominence of
grammatical relations (configurational language). Is the same true for German
with its widely free word-order and a good degree of topic prominence (non-
configurational ?)?
It is against the background of such questions that this attempt at a
comparison, with focus on three linguistic details, has been made. No conclusive
answers will be given, however. Rather, what lies at the base is my feeling that
what occasionally appears to be a heuristic and methodological misunder-
29
30 W.ABRAHAM
Booij (1981) draws the conclusion from (1) and (2) that (2), apparently the
passive conversion of (1) with some restrictions on the constituent een haag
loon "a high salary" that he regards as minor, presents proof of the fact that
zijn medewerkers "his staff' is DO (transitive or direct object) in (1).A careful
distributional analysis (one in which he would have to give a strict account of
een haag loon in (I» however shows that the constituent in question is 10
(intransitive or indirect object) in (1), and that (2) is not directly related
syntactically. Note that there can be no doubt whatsoever in German
(presenting, as it were, the consonant-shifted version of the Dutch examples).
It is hard to see why the author would not have tried one of the well-known
tests for the IO-status (aan-conversion, nominalization). It is intriguing to believe
that a practice common in generative semantics (to pay a good salary = to pay
wei£) may have induced the above analysis. But this is pure speculation. The
phenomenon acquires a specific dimension by the fact that the status of 10
in colloquial Dutch is increasingly giving way to the re-analysis as ditransitive
object (see e.g. van Langendonck 1968; Nieuwborg 1968; more recently v.d.
Toom 1981; Zaalberg 1981).
(6) (7)
==
NP NP
,/ v"'" -V
(10) (DO) / ' ""
NP V
(10)
In opting for (6), Daalder and Blom 1976 (henceforth D & B) make use of
Chomsky's asymmetrical superiority relation (Chomsky 1973: 246).
(10) " ... the category A is 'superior' to the category B in the phrase marker
if every major category dominating A dominates B as well but not
conversely".
32 W. ABRAHAM
D & B (1976: 404f.) claim that (6), but not (7), is an adequate deep structure
analysis of (8), which contains a reciprocal pronoun (elkaar). They assume that
an interpretive rule is operative on the level of the underlying deep structure by
way of which antecedents are attributed to the pronouns. This antecedent-
pronoun relation is warranted, among other general constraints such as the
Specified Subject Condition and the Tensed Sentence Condition, by (10): the
10, after all, is the antecedent of the DO as in (11) or of a reciprocal pronoun
incorporated in DO, as in (8). Consequently, 10 has to be superior.
From among (6), (7), and (12), (13), (14), all except (6) are rejected for the
analysis of (8), (9), and (11): (7) because the required IO-relation does not
satisfy the superiority relation; (12) and (13) since the first major category
dominating the antecedent, PP, does not in turn dominate DO; and (14) because
Chomsky's constraint is violated that the dominating hierarchy must not hold
conversely likewise (I am following Hoekstra's argument: Hoekstra (1978:
48; cf. Daalder and Blom 1976: 404f.)
/,"
(12) (13) (14)
/,p",,-
NP PP V NP NP V
(DO)/ "'- (10) (DO)
P NP
(10)
The underlying structures (12) and (13) have been devised by D & B (1976)
to account for the fact that Dutch lOs generally can have PP-(aan- )equivalents.
I will not touch on this claim in the rest of the discussion (see Hoekstra 1976:
48 for comments on D & B's conclusions). It is perhaps to be noted that (12)
and (13) are not extensions of Chomsky's notion of the superiority relation
applied to sentences like (8) since Chomsky explicitly excludes PP from the list
of major categories.
1 would like to show that D & B's fundamental assumption about the
identification and distribution of grammatical relations in (8) and (9) is wrong.
Quite obviously this has its repercussions on the authors' final conclusions that
THE IDENTIFICATION OF GRAMMATICAL RELATIONS 33
they arrive at on the basis of such assumptions. Whatever their fault, however,
they seem to be on safe ground as far as the normal distribution of lOs and DOs
but not as far as their respective surface order in Dutch is concerned. My
argument, then, will be that distributional arguments per ana/ogiam (more
specifically here: between NP-IOs and NP-DOs, on the one hand, and
reciprocal/reflexive pronouns of such relational status) need not be conclusive
and should, in the absence of better, non-distributional arguments, be
abandoned.
Let me first sketch what is normally held to be a good distributional
argument per ana/ogiam that whould seem to corroborate D & B's initial
assumptions (D & B: 404f.).
D&B WA
(15) Ik gee! die mensen 1 het adres van die mensen 1 IO-DO IO-DO
The analyses sketched in the boxes on the right-hand side are, if for no better
reason, analogical inferences from (15) and (17) to (16) and (18), respectively.
However, as the analyses under "WA" suggest, D & B's relational distribution
for (18) is doubtful, whereas it seems correct for (16). Now, D & B could have
signalled the correct analysis had they made use of another distributional and
transformational test. See the following sentences which are aan-(PP-)
equivalents to (9) and (17).
Note that in order to obtain DO-status for the reciprocal in (22) below some-
thing like the following analogy would be necessary.
(22) DO PP 0 10 DO
.,
+-'
;:3
Ik gee! een been 0 een been
aan de hand till Ik gee! de hand
0
~
Ik gun die mensen aan elkaar t:: Ikgun die mensen elkaar
'"
/f~
(13)
NP PP V
(DO) / "-
P NP
(10)
Since the IO in (13) has the status of a PP and not, as has been argued before,
that of an NP, (23) and (24) turn out to be ungrammatical. Hoekstra (1978:
48ff.) has presented evidence from Dutch that the assumption of (6) plus
Chomsky'S superiority principle cannot be held to explain that (23) and (24)
are non-sentences in Dutch. First, by way of inducing a false structural analogy,
he shows that (26) should be correct if D & B's arguments were extended by
analogy from the structurally equivalent (25). However, (26) is out on all
fOunts.
THE IDENTIFICATION OF GRAMMATICAL RELATIONS 35
Furthermore, Hoekstra shows that (13) maps certain sentential structures which
are by no means ungrammatical thereby disproving the validity of D & B's
disclaimer. See (27).
It is safe to say that D & B's analysis, for one, rests on insufficient distributional
evidence and, furthermore, disregards one stringent surface condition explaining
the phenomenon of reflexives and reciprocals, namely that they are excluded
from the superficial antecedent position. Note that this is a necessary, but by
no means a sufficient, requirement for a correct analysis of reflexives and
reciprocals. Furthermore it leaves intact the option that any such structural
principle as the superiority condition is operative.
It is to be noted that a language such as German, while very close to Dutch
historically and synchronically, quite effortlessly avoids these pitfalls. See the
following equivalents to the Dutch examples above.
(8') Ich gab DEN (dative) Leu ten {gegenseitig IHRE (accusative) Adressen
DIE (accusative) Adressen voneinander
(11 ') Ich gonne DIE (accusative) jungen Leute { e~nhander .,
SIC gegenseltlg
In (8'), dative object and accusative object are morphologically distinct. (11 ')
is just as unambiguous since the preceding object NP is in the accusative leaving
but the dative, and consequently the 10, for the reciprocal. Misinterpretations
such as D & B's for the functional distribution in (11) [(18)] are impossible, be
it merely for reasons of morphological transparency of the functional
distribution. Appeals to (transformational) control characteristics as those
employed by Hoekstra for Dutch, are superfluous. The morphological
distinctions in German accompany a number of further functional and
distributional properties 3 .
36 W.ABRAHAM
sentences (i.e. with two objects) is the 10 (Hoekstra makes a passing remark to
this effect without giving it any weight in his critique of D & B; cf. Hoekstra
1978: 49 below). What this amounts to, in brief, is that D & B's basic
assumption that (6), and not (7), is an adequate analysis of three-valued
sentences in Dutch is correct. However, it is obtained by virtue of incorrect
structural arguments. Specifically, their hidden ordering premise is mistaken,
although in line with much of what is being held in generative grammar
nowadays; and they have failed to see one important condition, namely that the
reciprocal in 3-place sentences needs to be in the 8-role (grammatical relation)
of 10; see (31) to this effect 4 . It is worth noting that, of course, no such
condition holds in two-place sentences. Possible counter-examples to the
assumed relational condition, namely RECIPROCAL = 10 in 3-place sentences,
are (32) and (33).
(32) and (33) seem to have readings which are not equivalent with readings
with reflexives. Other tests including verbs like geben, hingeben, garantieren,
verpflichten, das Leben kosten, verbieten, [ehren, failed to render readings
with REC = DO, NP = 10 (dative). However, I am somewhat at a loss at having
to decide whether (32) and (33) are fully correct and whether they would con-
sequently invalidate the rule REC = 10 (dative) with 10 and DO.
One final point addressed to the condition that there must be a (pro-)nominal
antecedent when the reciprocal elkaar/einander is to be realized (Hoekstra 1978:
50; 0 & B 1976: 408ff.
(34) Einander j hatten sie j gewiss lieb, aber filr Aussenstehende hatten
Each-other- had- they-certainly- dear, but- for-outsiders- had-
Such clause-initial reciprocals are acceptable also for any of the 3-place examples
where any of the two objects (but not the subject) stands in an antecedent
relation to the REC-relation. However, the examples below show that the
sentence-initial position is a special case that has to be exempted from the other-
wise valid restriction that the REC has to have an antecedent to relate to.
Compare (8') and (11') on page 7. The pragmatic function of the reciprocal
in (34) - (36) is that of contrasting focus. I will not dwell on the question
whether this movement can be handled within the rule that excludes (37) and
(38). It remains to point out that the position of the REC in the last two
examples cannot be saved with the function of focus; i.e. no reading can be
obtained by contrastive accent on these elements.
In Janssen 1977, it is argued that the (free) object of the possessive dative
and the (obligatory) internal object are on a par with respect to their syntactic
status, whereas the (likewise obligatory) external, i.e. prepositional, object is
structurally and semantically different. In order to account for both the
difference and the syntactic and semantic relationship between internal (pre-
positionless) and external (prepositional) objects in 3-place sentences, Janssen
chooses a format of lexical description developed by Jackendoff 1975. Jacken-
doff establishes a symmetric relationship "is lexically related to" by way of a
redundancy rule which, in case the prepositionless (dative) object is realized,
assigns a lexically empty P-node to the sentence. This amounts to a format of
the lexical entry where the P-node is assigned to both underlying structures,
but which, provided the dative N-node is filled, assigns a lexical null-element
to the P-node 5 .
THE IDENTIFICATION OF GRAMMATICAL RELATIONS 39
( 42)
==~S,,= ..
Nl /V",
I V
...P
_I \
Peter
(Prep Obj)
N2
/"=N3
V Prep
(D10) I I I
de krant geeft aan Fred
similarity seems to be due to the specific relation between aan en geven. A wide
number of simple/complex pairs such as aan + geven do not readily lend them-
selves to the relation of similarity. What is more important, aan in (41) has no
separate categorial status, but is a separable prefix to geven. The lexical entry,
in its canonical infinitive, is aangeven. It is part of the distributional properties
that sentences with aan + V-predicates can double up onaan (albeit rated some-
what clumsy stylistically) :
(45) Der Bauer hob den Heuballen flYf den Wagen hinfWf
The-farmer-lifted- the- truss-of-hay- on- the- wagon-to-up
(infinitive: hinaufheben)
It is in line with this observation that Janssen's conclusion from (51) and (52)
is incorrect (I 977: 15).
There are, as far as I can see, two ways to a structural assignment of aehter/
tegen : one is an adverbial modifying wind (wind van aehteren/uit de tegenrieh-
ting "wind from behind/from the opposite direction"). The second solution
assigns a modifying adverb to the verb hebben : note the quotative form de wind
tegen/mee hebben, German den Wind mithaben/dagegen/aehterlings haben,
where Wind is in the transitive accusative and mit a verbal prefix, dagegen an
adverb to the predicate.
e. Janssen (1977: 16) claims that naar zieh toe in (58) is the full variant of the
elliptical toe in (59), just as mee in (60) is the ellipsis of met + PRO + mee.
But met zieh mee is no variant of mee in any interesting, non-trivial sense,
nor are naar zieh toe and toe. Rather, mee and toe are separated verbal prefixes,
while the prepositional alternant with additional naar/met zieh mee is doubling
up on the semantic interpretation that the respective sentences have just as well
without the prepositional phrases. Note that English would render Dutch mee
in (55) with the adverbial along (without the possibility of doubling up),
whereas German makes a verbal prefix of it (m it nehmen. See von Polenz
1977 for German).
THE IDENTIFICATION OF GRAMMATICAL RELATIONS 43
Compare the slightly but crucially contrasting (60). What my argument amounts
to is that the dialectal variant of (61) without the reflexive pronoun cannot,
in the absence of other confrrming examples and given the counter-examples
(62) and (63), be regarded as an instance of a general PRO-ellipsis with the
preposition left stranded, but rather as an idiosyncratic instance. All other
analogous examples present themselves as structures with something else but the
prepositional category.
My arguments that Janssen's semantic identification of stranded prepositions,
bared of all morphologically and structurally/categorically supporting evidence,
is inconclusive, if correct, have been of a negative nature so far. If I set myself
the aim to make evident that the categorial status of prepositions depends
crucially on their cooccurring with nominals, then any identification as an
adverbial or verbal prefix hinges on a plausible structural distinction of pre-
positional phrases, adverbials, and verbal prefixes. In the remainder of this
chapter, I will dwell briefly on this question.
"Adverb", to begin with, is a functional class with a large and diversified
number of members (see, for example, Jackendoff 1972 and Bartsch 1972 for
a discussion of its semantic and syntactic varieties). From among this variety,
44 W.ABRAHAM
(64)
Adv v
Adv, of course, then is a grammatical category, and no longer a grammatical
relation. It will include a number of other categories, among them prepositional
phrases (but not prepositions alone), adjectives, and particle-like elements with
no derivational characteristic. Instances of this relational category are along
in the English versions of (60), most probably also all instances cooccurring
with hebben/have/haben and zijn/be/sein; cf. (53), (57) and (61). Note that
our hesitance, particularly in the German cases, whether to identify one element
as a verbal prefix or an adverbial argument, is best represented by this structural
closeness. It seems the most plausible solution. See also (44), (45). (46) and
(51). Naar zich in Janssen's variant of (59) (Janssen 1977: 16) is a different
case, however. Its underlying structure is probably something like (65).
(65)
THE IDENTIFICATION OF GRAMMATICAL RELATIONS 45
(NP', PP', V' etc. stand for the lexical interpretations of the respective
elements).
When we try, on the basis of this line of argument, to account for geven-passives
in Dutch we arrive at an impasse, since the structure corresponding to (71)
is unacceptable (if we accept Standard Dutch, ABN, only).
In sub stratal Dutch, however, the 10 with geven together with a vast number of
other indirect verbs can be passive-converted (v. Langendonck 1968, v.d. Toorn
1981, Zaalberg 1981) and would therefore seem to be accounted for by Gazdar's
proposal. In German, not only is the 10 not passive-convertible in terms of
DO ~ SUBJ (it is, however, in terms of the bekommen-passive comparable to
the Dutch, though greatly restricted, krijgen-passive) but there is no general
correspondence between 10 and PPd (alternation between dative and pre-
positional object). That is, German will only have one passive rule (as far as the
genuine passive with DO ~ SUBJ is concerned), namely (74).
equivalence of 10 and Prep + NP. Note that PPd in Gazdar's (68) can only
mean that the 10 is the closest argument to the verb. If this is compared with
(66) and (67), however, we note an unexplained contradiction: in (66b) the DO
(or, more precisely: the non-IO 7) forms a complex function with the verb,
with the 10 as an argument to this complex function. This difference will
obviously have to be resolved in either two ways: either the underlying
definition of TYP in (66) is correct and has to supplant (68); or (68) is the base
of TYP, in which case (66) will have to be replaced. If we can believe Faltz
(1978) or Ziv and Sheintuch (1979) there are undisputable syntactic differences
between "dative" and prepositional object. Such observations alone would not
make it impossible, however, to collapse the dative object and Prep-Obj into one
specific rule. All that (66) - (68) account for is the syntactic property of give
with respect to its syntax of passivization. What appears to be a stronger
argument in Keenan's type of reasoning for the degree of closeness of an
argument to the verb (Keenan 1977). It is the DO, not the 10 or Prep + NP,
together with the verb that yields a new, and often idiomatic, meaning as
compared to the simple verb: give the kicks, give a try, give a treatment. In
idioms without prepositional objects, the object is invariably a DO, never an
10 : kick the bucket, hit the sack, give way, playa prank. And, fmally, cognate
objects are always DOs: sleep a sound sleep, dream an exciting dream, run a
fast race/run, etc. All this justifies the conclusion that the DO is a closer
argument to the verb than the 10 (in three-place constructions). And this in turn
means that we have to opt for (68) as the most general passive rule.
An ever better case for such a general solution can be made for Dutch and
German. Both Dutch and German have a class of two-place verbs that take dative
(intransitive) objects only. While Dutch cannot identify those dative objects
morphologically they are clearly non-transitive since they cannot be passivized.
This observation (see Hoekstra 1981 : 87 for Dutch) justifies the following
hierarchical relation.
(75)
~IVP _ _ _ _ _ _ _
NP DTVP~
(10) / ~
NP TVP
(DO)
48 W.ABRAHAM
TVP stands for the class of verbs (or verbal complexes) that take a direct object
to form DTVPs. DTVP is the class of verbs (or verbal complexes) that take
indirect objects to form IVPs. DTVPs comprise: two-place verbs with one non-
accusative object only (Dutch bevallen "please", ge/ukken "succeed", German
gefallen, geUngen, passieren, folgen, walten, sich erinnern, sich schiimen, etc.)
as well as three-place verbs with a surface DO. IVP is the category of intransitive
verbs (or verbal complexes) that take a subject to form a sentence. This category
comprises one-place verbs like laugh, run, and complexes made up from two-
and three-place verbs such as discussed above. This hierarchical, structural,
tripartition is well motivated for with respect to the grammar of passivization:
English can passivize both TVPs and DTVPs, but not IVPs. Dutch can passivize
TVPs and IVPs (er wordt gezongen "there-is-sung" = "there is singing"), but not
DTVPs (except for a restricted class of verbs that allow the krijgen-(3 ~ 1-)
passive). German resembles Dutch closely in all respects except for the 3 ~ 1-
passive: The class of verbs that can undergo the bekommen/kriegen-passive is
much less restricted than that in Dutch (Eroms 1980). Hoekstra (1981: 87)
correctly observes that the strategy to account for the differences in the
grammars of passivization by way of a categorial distinction thus receives an
independent justification_ Moreover, such a categorial distinction allows for
a unified treatment of lOs that is amiss in Gazdar's proposal.
Let us turn back to (67). We have argued that Gazdar's bipartite treatment
of promise is ill-conceived because (a) it is in contradiction to (68), and (b)
it fails to account for the closeness-evidence of DOs. Now, let us have a look
at German. The syntactic analogy between the accusative NP-complement and
the sentential complement (without case marking) is so forceful that the
transitive character of the clausal complement is felt to be self-evident. This is
so notwithstanding the fact that the clausal DO cannot be promoted to subject
by way of a regular passivization. See the following examples:
die Gefolgschaft }
(78) Dem Konig wurde vom Vasallen [der Austritt aus der versprochen
Gefolgschaft
(78) is the version corresponding to (67). Note that the infinitives of (76)
have been conflated in the verb-derived NPs in (78).
I would like to suggest that there are a number of fully acceptable and less
acceptable structures that relate between the German equivalents of (66) and
(67), (76) and (77).
?
(79) . Dem Konig wurde vom Vasallen die/dessen ( /*seine) Gefolgschafts-
leistung versprochen /zugesagt.
(79) would be a passive version of (76) in the strict sense, although it is felt to
be clumsy (bureaucratic style). die Gefolgschaftsleistung is the passive subject,
nominalizing the infinitival complement in (76) word by word, and as such is
structurally reminiscent of the DO-status in the corresponding active sentence.
This faces us with a dilemmatic setting: should we account for the fact that
versprechen + infinitival does not allow a passive conversion, or are we to give
credit to the syntactic closeness of the infinitival complement to a DO-nominal-
ization ? The consequences to either alternative are exclusive of one another:
phenomenon one forces us to register two separate lexical entries; the second
fact suggests that we describe versprechen in a unified lexical format.
The solution seems straightforward enough, since it is desirable to account
for both properties, i.e. formulate one single lexical entry but distinguish the
two different syntactic properties depending on the categorial analysis of the
complement. Compare (80).
(80)
(81) and (82) show that there is more involved than simply the question whether
or not the complements/verbs such as versprechen are structurally complex or
not.
zugesagt }
(aller Arbeitsnehmer) { zuges tan den
versprochen
all- workers)- { conce.ded}
promIsed
(82) {zugestanden }
Dem Direktor wurde vom G. *zugesagt (aUe A.) auszusperren
*versprochen
Note that the TVPs zugestehen, erlauben other than zusagen, versprechen as
matrix verbs do not render acceptable passive versions. See also (83) and (84).
(84) *versproChenJ
Dem G. wurde vom K. { zugestanden den Laden zu verlassen
. erlaubt
ENGLISH
(i) Arguments to the effect that the preposition less (dative) object and to
+ NP with 3-place verbs like give are identical with respect to their
relational status (i.e. 10), bear on semantic intuitions, but not on syntactic
control properties (no common passive conversion) and distributional
characteristics (dative phrase only to the left, never to the right of the
DO; to + NP appears to have a less restricted distribution).
(ii) The relational identification of 10 and local adjunct seems more or less
based on accidental, rather than generally valid, syntactic evidence
(Kruisinga and Erades 1947). Faltz (1978) presents semantic arguments
in favour of the identification of 10 and dative, while distributional
characteristics are a disclaimer to /0 = to + NP (see Faltz 1978; Ziv and
Sheintuch 1979).
DUTCH
GERMAN
Conclusion
FOOTNOTES
1(25) will probably have a bisentential deep structure in which Jan (subject in
the deeper sentence) satisfies the superiority relation with respect to the DO,
foto. Exactly this holds also for (26) : kinderen is superior to elkaars fotos.
~s~
NP /VP"
NP V
I I
~s------:..aat
NP ~VP~
I /'" ~
Jan NP V
de kinderen I I
een foto zien
elkaars foto
THE IDENTIFICATION OF GRAMMATICAL RELATIONS 53
D & B's claim that the required antecedent relation is spelled out in the deep
structure, quite obviously, makes the incorrect prediction that (26) is an
acceptable sen tence.
2(27) satisfies the superiority condition: the DO, de kinderen, is superior to the
PP, aan elkaars auders. Furthermore, (27) fits the structural description of
(13) and should consequently be unacceptable. There is no doubt, however,
that (27) is a good sentence.
3The following relations between case markers and grammatical relations seem
to hold in German :
- 0 (not correct): dative NP ~ 10 (since there are "free" datives in German)
- 0 (not correct): 10 ~. dative NP (since there .are lOs in the accusative)
- 0 (not correct): accusative NP ~ DO (because of accusative lOs)
o (correct) : DO ~ accusative NP (with the restruction that, by force of
idiosyncratic lexical distinctions, DO subsumes a variety of syntactically distinct
properties; see Saltveit 1968). .
4The structural relation between 10 and DO as sketched in (6) will not be
defended here. See section 3 of this paper for one independent line of reasoning
that supports (6). In order to clarify my own point of view, let me emphasize
here that I do not claim that the (logical) conjunction of (6) and the superiority
relations account for sentences such as (11) : Ik gun die mensen (DO) elkaar
(10). What I do claim, however, is that D & B, under the force of their
assumption that this conjunction does in fact explain the grammaticality of (11),
give the two objects different functional interpretations (Le. die mensen = 10,
elkaar = DO) so as to make them fall under these restrictions.
5My presentation lacks in accuracy. What Iackendoff aimed at was a format to
relate to one another verbal pairs such as the 2-place and the I-place break.
What I have presented here is Ianssen's transfer of Iackendoffs arguments to
the Dutch option between the dative object and the prepositional object.
6Dieter Kastovsky (Vienna) pointed out to me that (72) is a tolerated form of
the passive restricted, as it seems, to give and write. It belongs to the terse,
abbreviating register of speech employed by speakers who as recognized
authorities can afford to set a new norm. What this would amount to is a
modified(68),namely: {PP'i}
give' ( ) (NP'j)
riP \
7I assume that the distinction between DO and non-DO is the crucial distinction
that Gazdar has in mind. This, of course, considerably weakens my suggestion
that this treatment of 10 is inconsistent. However, since Gazdar does not
indicate a separate categorial status of non-DO in contrast to 10, I see no reason
not to identify his non-DO as 10.
8Van Riemsdijk (1981), in the vein of other recent generative linguistic work,
suggests for German a hierarchy of NP- (or argument-) closeness to the verb
ranging from accusative/direct object via dative/indirect object to genitive/
oblique. The arguments for this hierarchy is based on observations of errors in
the case morphology of parenthetical object doublettes. Clearly, this suggestion
wants further empirical substantiation.
VERB COMPLEMENTS AND SENTENCE COMPLEMENTS:
TWO DIFFERENT TYPES OF RELATION
Claire Blanche-Benveniste
Universite de Provence
it is said to relate the "idea of happiness" with the verb. Used as a "sentence
complement", as in :
55
56 CL. BLANCHE-BENVENISTE
1 .1. In both cases, authors propose one identical relation and distinguish
only between different secondary properties. To use the same word "comple-
ment" in both cases implies that there exists a relation in some way identical,
be it with a verb or with a sentence. The differences would then come from the
basis of the relation. It is nothing more, then, than a question of parenthetizing 4 .
1.2. It is described as the same relation, but with different degrees. The
relation between the verb and its complement is viewed as "more integrated",
"tighter". It is considered to be a more "peripherical" or "looser" one between
the sentence and its complement. This concept of degrees in the relation mayor
may not be related to other syntactic relations 5 .
1.3. It is described as the same relation, but the properties are different.
The differences are explained by the fact that the verb is the predicative center,
or because there are various "areas" in the sentence, or because there is a
distinction between what is in the sentence and the mode of insertion of the
sentence in the utterance 6. Sometimes it is difficult to see why the same
relation is maintained in both cases 7, if the different levels are supposed to be
distinctive. Such is the case with generative grammars which relate the "sentence
complement" to a higher sentence, "outside the proper sentence" 8.
1.4. It is not described as the same relation: one is grammatical and the other
is not. For the "verb complement", it is a syntactic relation established on
grammatical schemas. For the "sentence complement", it is a discursive relation,
without syntactic rules.
This position was often assumed in French description of spoken French or
dialects; "sentence complements" were identified with "non-built comple-
ments", escaping the usual rules of syntax, "expressive phenomena belonging to
the style more than to the syntax" (Camproux, 481). This kind of analysis was
frequently used for the description of facts exterior to standard French
grammar, such as :
1.5. Both relations are considered to be within. the grammar of the language,
but they belong to distinct orders; this is the position I will assume.
I think it useful to consider examples of spoken French and dialects ,10,
but I think the sentence complement has a grammatical status.
By relating a verb and its complement, the relation links two morpho-
syntactic categories. Obviously, the verb is a morpho- syntactic category in
French; but we have to admit that its complement, by the fact it is built as a
verbal complement, is also a defmite morpho-syntactic category. Such is not the
case for the so-called "sentence complement" in its relation to the sentence.
They are held together by the virtue of what Bolinger calls "togetherness", but
the "sentence complement" is not organized in systematic paradigms as is the
verbal complement. When we seek its properties in terms of categories, we only
find negative ones.
As I intend to avoid the fallacious word of "complement", I shall from
now on use the terminology of "government" for the verbal syntax: the verb
"governs" an element which is inside its "government". The "governed" element
is governed by its verb 11. For the relation between the sentence and the
adjuncted element, I shall use the term of "association". In "quant amoi, ya va"
(As for me, it is all right), "quant a moi" (as for me) is associated to the verb,
it is not "governed" 12. Government rules and association rules both belong
to the grammar, as I think we can prove from the fact that some languages -
as for instance XIIth century French - have morphological clues to distinguish
between the two types of rules.
2.0. A given lexical unit is cften involved in both relations, so that it is not
possible to give regular lists of governed vs. associated elements. Some are
specialized in one kind of relation, such as the clitic elements "la, lui", always
governed by a verb. Some others, such as "quant amoi" (as for me), are always
associated and never governed. These clear examples will be used as typical
tokens in the beginning, until ambiguous examples can be classified by
distin~tive PI"?perties.
S8 CL. BLANCHE-BENVENISTE
In the actual use of spoken conversational French, governed elements have the
same mobility :
17ansila(AG4719)
(17 he is)
But in this case, there is an interesting into native fact: the governed element,
located in front of the verb, has a melodic schema analogous to the final schema
of the utterance 13. There is a descending schema on :
heureusemenl,
It would be the same for a governed element such as "a lui" (of him) in an
affirmative utterance :
a lui~ tu pens:s ?
The melody of the preposed element reproduces that o!the final element, with
less amplitude, and never contrasts with it. Such a melody ·shows a kind of
integration into the verbal government. On the contrary, there is a contrast for
the associated element. When preposed to the verb, an element such as "quant
a moi" (as for me), must be pronounced according to a non-fmal intonative
schema, contrasting with that of the final:
1 ------~
_ _ _ _ _ _- J
and never:
2.2. Modalities
quant a moi cela ne va pas bien (as for me it is not going very well)
cela va-toil bien, quant Ii toi ? (is it going well, as for you ?)
According to this property, it has often been proposed to separate "parce que-
phrases", and "puisque-phrases" 15 :
but it excludes modalities for the typically associated interpretation (it is lucky
that) :
_ _ _ _ _ _ _ _ _ _- J / _ _~----~
They are always linked with the special "place and intonation" schema of
associated elements :
There can exist no contrast between different modalities of "peur", and this
impossibility seems to be related to the impossibility of acting as a governed
element.
The governed type has a regular paradigm, allowing several contrasts; this regular
minimal paradigm reveals a larger one, which is lacking for the associated type.
64 CL. BLANCHE-BENVENISTE
This constant relation with pronominal forms has been used as a basis for a
systematic analysis of French syntax, in the studies proposed by K. Van den
Eynde and myself 17. Pronouns such as "a lui", "ainsi", "comment", can be
used as useful clues as long as they act as generic and classificatory terms. All
the elements of the paradigm have the same constant relation with this kind of
term. An element such as "Ie petit doigt" (the little finger), in "i! ne levera pas
Ie petit doigt", not having the same relation with the pronoun "Ie", cannot be
considered as a governed element of the verb "lever".
The paradigmatic affJliation established between an adverbial form like
"heureusement" and a generic pronoun "ainsi" or "comment" permits us to
neutralize the lexicon and consider only the syntactic relation of the governed
element.
The effect of grouping into one and the same series is lost; moreover, we observe
that interrogative "comment ?", which occurs in the governed elements, does
not occur in the associated ones:
----~/----------~
*comment, IVa s'est termine~
Many typically associated elements such as "quant a moi" (as for me) have no
pronominal correspondence at all. No pronominal equivalence would fit here.
There is no constant relation between:
a
Quant moi / ainsi / a!ors / c'est termine
(as for me / in this way / then / it is all over).
3. Associated elements
upon more or less conclusive differences in apparent meaning 18. For instance,
we have to control whether, when a so-called "movable" element is placed in
front of the verb, with a non-final intonation, it thereby loses the possibility of
receiving the set of modalities and the proportional relation to the pronouns.
There exists no proportional relation with the pronouns, for instance with
the interrogative one :
alors, il etait 11 h
(then it was 11),
but this relation is not linked with a whole paradigm. The same demonstration
would fit for examples such as :
Such examples give good results because verbs such as "it was 11" or "it was
too late" tend to reject any temporal governed element. To ask "when was it
11 ?" sounds inconsistent, as would be the case for most stative verbs. Semantic
features of this kind entail an interpretation of the "complement" as being
outside the verbal government. Such stative verbs can be associated to temporal
indications as long as they stay outside the verbal government. What comes
under discussion then is not the mere possibility of applying a temporal
indication to the verb but the type of relation existing between the temporal
indication and the verb. With a non-stative verb as in "ses fils sont partis" (his
sons went away), both interpretations can hold, and we grasp the slight
difference in meaning by opposing the two into native patterns:
- governed pattern,
which would be the case if the temporal part stood for a governed element:
Within this interpretation, the temporal part cannot commute with the inter-
rogative pronoun :
--/
*quand, ses fils sont-ils partis ?
les gens quand ils rient ils ont pas honte de rire (AG)
(People when they laugh, they are not ashamed of laughing.)
(?) quand est-ce qu'ils n'ont pas honte de rire ?
(? when is it that they are not ashamed of laughing ?)
parce que quand on a fini ses devoirs heu qu'on veut jouer on aime-
rait plutot jouer dehors (Nelly E3 4)
(because when you've finished your homework and you want to play
you'd rather play outside)
hi-bas quand il y a des mariages ils durent au moins trois jours (Tujk
1020)
(over there when they have weddings, they last at least three days)
(?) quand est-ce qu'ils durent trois jours ?
(? when is it that they last three days ?)
The associated "quand-phrase" often goes with a cleft sentence such as "quand ...
c'est que ... " (when ... it means that ...) :
quand tout marche bien c'est que les chefs sont bons (apud Chetrit
134)
(when everything is all right, it means that the bosses are good).
VERB COMPLEMENTS AND SENTENCE COMPLEMENTS 69
je connais des femmes moi quand leur marl il rentre ala fin de la quin-
zaine prrt les so us a moi (AG 4 40 9)
(I know women I do, when their husband he comes home at the end of
the fortnight, hep ! the dough for me).
b) si-phrases
- Governed by the verb, and placed in front of the verb, it has a final-like
intonation :
si Mara a la scene est un miracle, ala ville elle est une star (Newspapers)
(while M. is a miracle on stage, in town she is a star)
Commutations with "dans ces conditions, en ce cas, alors ..." do not maintain
70 CL. BLANCHE-BENVENISTE
c) avec-phrases
and it often goes with a contrastive effect: "avec vous et pas avec quelqu'un
d'autre" (it is with you and nobody else). Modalities do not change this type of
relation:
The governed element forms a paradigm with many pronominal items: "avec
qui, avec quelqu'un, avec lui" (with whom, with somebody, with him).
je crains Ie car moi deja je crains Ia voiture mais enfin avec man mari
avec Ie vent ~a va (Lic 2 7)
(I get sick in coaches I do, even in cars I get sick, but with my husband
and some fresh air, I can manage).
by implying a contrast, "with my husband but nobody else ... " Modalities would
be odd; interrogative is excluded and the equivalence with pronominal type like
"ainsi" (like that) does not parallel with "avec hti" (with him).
d) com me-phrase s
c'est plutot comme elle peut que com me elle veut que ...
(it is rather the way she can that the way she wants ... )
The "comme-phrase" falls into the paradigm of "ainsi, comme cela, comment,
autrement" (so, how, like that).
Relations with pronominal forms of the previous paradigm do not give a good
serial effect:
- when governed by the verb, these phrases have the form "pas de N" or "pas
du N" if there is a contrast :
"Pas du N" occurs in front of the verb, whether there be a formal contrast
or not, with final-like intonation:
"Pas du N" is the only form compatible with modalities such as restriction
or interrogation:
pas }
rien que + du cafe
seulement
pas de }
:;;:n + cafe
nen comme
It does not contain any modality. The two different analyses can be presented
as follows:
pas du pas de
rien que beaucoup de cafe du
plutot quatre beaucoup de cafe
aucun
rien comme
In an associated use, locative or temporal elements can occur without any pre-
position and there does not seem to be any taxinomic organization to rule its
occurrence :
A mere relation of "togetherness" seems to hold here, without any hint of the
previous organization characteristic of the governed elements.
Many authors insisted on the fact that what we call associated elements
characterize the way the utterance is viewed more than the verb itself 21
VERB COMPLEMENTS AND SENTENCE COMPLEMENTS 75
cannot be reversed :
------~/------------~,
rue d'Alesia, il va au cinema aParis,
Examples such as "cette villa on ne peut pas dorrnir" (this cottage nobody can
sleep), as far as they have been described, have often been analysed as having
a theme, "cette villa", linked to the predicative part of the utterance 25.
Throughout these various meanings, there appear some constant character-
istics. Associated elements often give the effect of a preliminary framework for
the content of the verbal utterance. The relation is explicitly given as such
when the lexicon renders explicit the relation, as in "legalement parlant" (legal-
ly speaking). The relation looks hazy when such is not the case. In "cette villa
on ne peut pas dormir" (this cottage nobody can sleep), it can be inside the
cottage, with this cottage nearby, because of this cottagr, since there is this
cottage, when dreaming of this cottage, etc. Various interpretations are open to
us. The relation is not shaped by a paradigmatic organisation and thus appears
hazy and of a rudimentary type. No special meaning can resume all the meaning
possibilities implied by associated elements. Many are essentially characterized
by opposition with governed elements.
In front of the clitics -and-verb block, there occur elements such as "ne", pre-
fix reo, or adverbial onques (34). After the block, adverbial or, negatives pas
or mie. The whole forms the "verbal zone" (hereafter noted within bars).
A. Examples:
B. Examples:
Word order A occurs when no governed element (not even the subject of the
verb) precedes the verbal group (cf. 368-418). This is the case when nothing
precedes the verb, or only an exclamative "ha", or a coordinator "et" :
- a prepositional complement :
- an adverbial phrase:
Skarup represents these facts in different "zones". The verbal group and
the governed elements constitute the "proposition"; associated elements are
"outside the proposition". The proposition itself is divided into "preverbal
zone", "verbal zone" and "post-verbal zone". If the preverbal zone is empty,
we fmd word order A. If the preverbal zone contains a governed element, we
find word order B.
outside outside
the prop. proposition the prop.
- V + pro
+ pro + V
or as a governed element (in the preverbal zone), and then complement clitics
come before the verb :
ceus qui m'ont escute, / lor pri iou qu'il n'oblient/ (184)
the complement clitic is in front of the verb; this means that the nominal group
which comes before is governed by the verb and cannot be redundant with the
clitic, it has to be translated by : "those who have listened to me, I pray them
(other ones) not to forget".
Such terms as "in the proposition" and "outside the proposition" embarrass
Skarup, and he explains his choice (416,429). It leads him to distinguish between
subordinate clauses which are "in the proposition" and others which are "out-
side". But terminology is of little importance. His classification selects very
interesting semantic facts, among which we often find parallels with modern
French. Here are some clear oppositions given for the same phrases, in governed
or associated use.
Used as a governed element, "si com" has the meaning of comparison, and
could be translated by "the same way ... "
si com je voel / m 'en sui jou / par mon gre tomes (283)
(the way I wanted did I leave as I willed)
- "quant" phrases
- "se" phrases
- "apres" phrases
comma:
puis que vendra au grand destroit / Ii rendrez vos / trestout son droit
(283)
(when then he reaches the great strait will you give him all his due)
5. Conclusion
NOTES
Alexandra Colen
aspirant NFWO - R.U.G. Engelse taalkunde
I Introduction
85
86 A.COLEN
II. Analysis
1. That-clauses.
a. IllEJ. she was still alive consoled him.
b. Mary consoled him.
c. She was still alive.
2. Wh-interrogative clauses
a. I cannot imagine what made him do it.
b. I cannot imagine the house.
c. What made him do it ?
The nominals in the above examples all fulfill the functions of nouns and
NPs in their matrix sentences. All of them may function as subjects, objects and
complements of the copula. Apart from the fact that they fulfill the same
functions as simple nouns, they also have in common the fact that all of them
can be paraphrased by sentences (as in c.). These paraphrase sentences make
explicit the structural relations which hold between the constituents within the
nominals. These structural relations are interpreted as grammatical relations
(such as those between subject and predicate, or between verb and object)
typical of sentences, rather than as modifier-head relations typical of NPs.
This indicates that the structures underlying these nominals are not NP
structures. Derived nominals like 6-7, which have the surface structures of NPs
containing simple heads and underlying structures which are not NPs, must be
regarded as products of processes which radically change their internal
structure. They are referred to as "structure-changing" nominais, to distinguish
them from the "structure-preserving" nominals of examples 1-5, whose under-
lying and surface structures do not conflict. The nominals in examples 1 to 7
represent sentences in their entirety, i.e. sentences which are inserted into empty
noun slots of the matrix sentence (Vendler 1968, 31). They are called "sentence
nominals". Their structure can be represented as follows: (US = underlying
structure; SS = surface structure).
e.g.
--- consoled him} -(US) [she was still alive] consoled him
she was still alive _(SS) that she was still alive consoled him
complement to the head (Vendler 1968, 31). The structure of such RCNs
("restrictive complement nominals") is represented as :
e.g.
The man --- met hert _(US)[The man [the man had quarelled]]met her
The man had quarreled)
. ~(SS) the man who had quarreled met her
term "VP" is used in this sense rather than in the restricted sense in which it is
used in generative grammar (viz. to denote only the predicate of the sentence).
The number and type of complements within a VP depend on the valency of
the head. The dependents within a VP are NPs. The valency structure within a
VP differs in kind from the dt-dm structure within a NP. Valency accounts for
the obligatory constituents of a sentence (those which represent the participant
cases of case structure : see Fillmore 1970).
The optional constituents of a sentence, viz. adverbial adjuncts of various
kinds, represent another type of component structure. Similar optional
constituents are found in NPs. In sentences these constituents belong to two
types: sentence-modifiers (which can be regarded as modifiers of higher per-
formative verbs) and modifiers of the main verb, such as time, place and manner
adverbials. The former are not considered here. The latter modify the heads of
VPs in sentences but are not inherent parts of these VPs. The corresponding
optional modifiers in NPs are similarly not regarded as inherent parts of the NPs.
Such modifiers constitute the "circumstancial contextualization" of the VPs
and NPs in which they occur. The grammatical relations between verbs and their
adverbial modifiers (i.e. their circumstancial contextualizers) are different from
those between verbs and their valency dependents.
A third type of structure which is reflected in sentences determines the
relation between main verbs and the morpho syntactic markers of tense, aspect
and modality which are associated with them. The relations of temporal,
aspectual and modal contextualization also differ in kind from the valency
relations within sentence VPs. A fourth type of structure which is reflected in
sentences but is not inherent in the structure of VPs is that which distinguishes
between subject and predicate functions. This structure accounts for the
syntactic relation between subjects and predicates which is not explained by the
valency structure of the VP. This relation is marked by the morphosyntactic
phenomenon of concord. It may be argued that the subject-predicate relation
is a grammatical reflection of one aspect of information structure. This
grammaticalized level of information structure may be referred to as "thematic
structure" in order to distinguish it from those aspects of information structure
which are not grammaticalized (e.g. emphatic components of information, and
the distinction between presupposition and focus are signalled by the
nonsegmental phonological component of language rather than the syntactic
component).
The distinction of different types of structure which are incorporated into
the sentence is set up on the basis of differences in the functions fulfilled by the
THE SYNTAX OF ENGLISH NOMINALS 91
11.2.1. The structure which underlies all SNs is the dependency structure of
VPs. In all types the non-adverbial and non-adjectival complements within the
nominal represent NPs or Ns which function as obligatory complements of the
main verb in the corresponding sentence paraphrase (non-predicating adjectives
as in presidential address, have a special status in this regard, see: Levi 1978).
That this is true also of NPs headed by derived nominals is apparent in the
following comparison with NPs containing simple heads:
13. whether-clauses
She could not decide whether she should go or not.
She could not decide whether to go or not. deleted subject = she
of subject deletion and time neutralization in these cases indicates that thematic
structure and temporal contextualization are in some way closely related or
interacting components of clause structure. Modal contextualization is also ex-
cluded in non-finite clauses. The non-finite forms in wh-clauses cannot be
marked for aspect (constructions expressing progressive or perfect aspect are
only marginally acceptable in standard English:
The selection restrictions of fail and refuse exclude objects denoting ongoing
actions or events. The meanings of the verbs imply that the action denoted by
the object has not occurred, and consequently cannot have duration. On the
other hand, process denoting verbs, such as finish and practise can only have
nominal -ing clauses as complements, as their meaning requires that the action
or event denoted by the object be marked as ongoing, i.e. as progressive.
Similarly, the semantic difference between "putative" and "factual" meaning
that is often associated with the choice between to-infinitive and ·ing clauses is
restricted to certain classes of matrixes, such as those containing the verbs
96 A.COLEN
begin, cease, start, or those containing emotive verbs such as like, love, hate,
prefer, the verbs regret, remember and forget, and verbs of permission such as
allow, authorize, advise, permit, recommend (Jespersen 1914-1949; Dekeyser
et al 1979). In all cases this semantic distinction can be explained as a function
of the meaning of the matrix verb and the progressive, respectively
non-progressive aspect inherent in the non-finite clauses. Consequently, the
semantic restrictions on the use of the to-infinitive and -ing clause nominals do
not provide sufficient evidence for regarding them as two distinct and unrelated
types of nominal clause.
The -ing clauses, then, may be analyzed as clauses containing a progressive
form of a verb in the to-infinitive form (i.e. the neutral tense) from which the
element to be has been deleted (see example 18). The semantic redundancy of
these forms increases the likelihood of their being deleted in these cases.
Similarly, the deletion of to from the non-progressive verb form accounts for
the occurrence of clauses containing the bare infmitive as non-progressive
counterparts of -ing clauses, as in 19.
Though to is obligatorily deleted in the subject clause of see in a. (cfr. the un-
grammaticallity of c.), it is not deleted in the corresponding passive matrix
with subject-raising from the nominal clause to the matrix (example e.). This
indicates that the deletion of to is a surface structure phenomenon. The
corresponding progressives occur with or without deletion of to be. The non-
deleted progressive form of d. is more acceptable than the corresponding non-
progressive without deletion in c.
The SNs with deverbal noun heads cannot be marked for the progressive and
perfect aspects in the same way as the corresponding sentences :
THE SYNTAX OF ENGLISH NOMINALS 97
As far as tense is concerned, these SNs do not contain any forms apart from the
-ing form which occur in sentences as tense-markers. As sentence constituents
they share the temporal contextualization of the matrix main verb. This is the
case for all nominals, both non-finite clauses and NPs. The nominal constituents
of a sentence share its temporal contextualization by the implication that their
referents exist or occur simultaneously with the action, event, etc. denoted
by the main verb. The fact that derived nominal SNs cannot be marked for
aspect may indicate that they are not marked for time, i.e. rather than being
located at an unspecific point on the time dimension, they have no temporal
contextualization at all. In this respect derived nominal SNs are closer to NPs
with simple noun heads than to nominal clauses.
The above observations are summarized in table 1., which shows how the
various components of sentence structure are reflected in sentence nominals.
'"
~
..c::
....
1:
~
=00-
c::: u
Q)
"0 =
VP str. + + + + + -
thematic
structure + + + + - -
aspectual cont. + + + + - -
temporal cont. + + + - - -
modal cont. + + + - - -
structure is accounted for outside the base component (i.e. the generative com-
ponent) of the grammar. In our proposal the sentence is regarded as a basically
complex structure. If any of the obligatory component structures is missing the
sentence is defective, i.e. the construction in question is not a sentence and
cannot be used as an independent sentence.
The component structures of temporal, aspectual and modal contextualiz-
ation are relevant to the gradient distinction between sentences and various
types of SNs. This is obscured in the models of standard generative grammar.
There, temporal, aspectual and modal distinctions are formalized as morphemes,
specifically as optional constituents of a category "aux". In this way the
corresponding distinctions between SNs can be captured (Lees 1960). Never-
theless, in reducing functions operating on sentences to constituents of a non-
immediate sentence constituent there is a certain loss of explicitness. In the pro-
posed model these functions are formalized directly as functions operating on
sentences, and consequently as indispensable elements of sentence structure.
If the definition of a sentence as a complex of distinct co-occurring
types of structures is accepted, it is possible to account for the syntactic
characteristics of different types of SNs in a simple systematic way. Some NPs
have an underlying structure which resembles that of sentences, i.e. of
thematicized and contextualized VPs. Other are similar in underlying structure
to VPs which are thematicized but only partially contextualized (see table I.).
Yet others correspond to unthematic and uncontextualized VPs, in this way one
can account for the intermediate position which SNs occupy on the scale which
has at one extreme all the syntactic characteristics of sentences and at the other
extreme the syntactic characteristics of NPs. In other words, this proposal
explains the existence of the phenomenon of "nouniness" (Ross 1972; 1973).
Potential occurrence as a sentence is the defining characteristic of VPs as
opposed to NPs. This explains why they can be interpreted as having predicative
force (Jespersen formalized the notion of latent versus explicit predicative force
by means of the concept of "nexus": Jespersen 1924; 191449). In the
proposed model of syntactic structure the primitive linguistic categories are
NPs and VPs. The syntactic structure of SNs indicates that the initial category
of a generative PS component should be the VP rather than the sentence. The
latter can be defined in terms of the former. (The way one accounts for SNs
in terms of underlying sentential structures depends on one's implicit definition
of the sentence. Depending on whether greater importance is attached to its
dependency structure, its thematic structure or its contextualization, different
100 A. COLEN
between VPs and NPs also has its semantic correlate in the distinction between
role-structure (reflected in valency structure) and specification (reflected in
noun modification). Role structure is inherent in the relational concepts ex-
pressed by verbs, whereas the semantic relation of specification obtains between
nonrelational concepts (such as the individuals and entities denoted by simple
nouns: Lyons 1977). Apart from having a role-structure, relational concepts
are also those which can be contextualized on the time dimension. This provides
the semantic (and ultimately cognitive) explanation of why the same elements
are pivotal with respect to valency structure and contextualization.
The functional distinction between VPs and NPs is made in terms of thematic
structure. VPs function as predicates, and their heads as predicate verbs. NPs
may function as subjects or as non-pivotal elements of the predicate. Outside
of thematic structure the predicating function of the VP is latent. In nominals
which are derived from VPs there is a conflict between their latent function and
their actual function. They are VPs, i.e. potential sentences, but fulfill the
functions of NPs as the internal structure of non-thematic VPs does not differ
from that of NPs (both types of phrases consist of a head and a number of
optional dependents).Non-thematic VPs in nominal functions are realized as
NPs in surface structure: their actual nominal function overrides their latent
predicating function (and consequently their latent valency), and determines
the internal structure of the construction. When the underlying VP is actualized,
by thematicization, in its predicative function, its valency-structure is also
actualized. The valency of a VP overrides its nominal function in the matrix
and determines its realization in surface structure. Thematic VPs in nominal
functions are realized as clauses. Non-thematic VPs (i.e. VPs whose latent
predicative function and valency are not actualized) in nominal functions are
realized as NPs.
It seems highly plausible that the level at which the underlying sentence
components are defined is not a purely syntactic level, but a logical one, i.e.
one where lingUistically organized semantic and syntactic structure are simul-
taneously represented. Both sentences and sentence nominals can be related to
logical structures containing many-place predicates and arguments
(corresponding to the verb and its dependents in the VP). On this level the se-
mantic structure of messages is reflected in the distinction between the logical
102 A.COLEN
Renaat Declerck
Katholieke Universiteit Leuven-Kortrijk
1. Introduction
105
106 R.DECLERCK
2.1. The reason for assuming (1 ,b), with two occurrences of the NP Mary, is
that in (I ,a) Mary will generally be interpreted as functioning both as direct
object of saw and as subject of come in. However, the fact that Mary can be
interpreted as object of saw follows pragmatically from the fact that IPVC's
always express direct perception and from the nature of the perception verb:
'I saw Mary' is a necessary inference from 'I saw Mary come in'. However, with
other perception verbs this pragmatic inference may fail. Thus, (2,a) does not
imply that 1 heard Mary herself and as noted by Gee (1977 : 474), it does not
follow from (2 ,b) that 1 felt Tom:
Since there is no thematic object-of relation between the perception verb and
the NP following it (though such a relation may often be pragmatically inferred),
the NP in question cannot be generated in (deep) object position. Further
evidence of this is that the head of the NP may be an abstract noun, whose
referent cannot be directly perceived (see De Geest (1973), Kirsner & Thompson
(1976)) :
2.2. The 'subject' of the IPVC may be one of various elements that cannot be
generated as objects in underlying structure, e.g. existential there (which is not
base-generated but transformationally inserted), an idiom chunk that must be
a deep subject, or the 'expletive'it "
(4) (a) We saw there arise over the meadow a blue haze. (example from
Kirsner & Thompson (1976»
(b) We heard all hell break loose. (example from Gee (1977).
(c) I heard it chime one o'clock as I was turning out of the gate.
2.3. The superficial 'object' of the perception verb cannot normally undergo
Object Shift (Tough-movement), although base-generated objects normally can
(cf. Lightfoot (I 976 : 27 6)) :
(5) *The choir was easy (for us) to hear sing a song.
INFINITIVAL PERCEPTION VERB COMPLEMENTS 107
2.5. In sentences like (8) (in which be is normally deleted from the IPVC) it
results from extraposing a subject clause and cannot, therefore, be generated
in (deep) object position :
(9) John forced/persuaded Bill to take the examination, but he did not
force/persuade Tom [to take the examination].
(lO)(a) I've seen/heard Ann come in, but I haven't seen/heard Tom yet.
(b) I've seen/heard Ann come in, but I haven't seen/heard Tom come
in yet.
2.7. It may be pointed out, finally, that an Equi analysis of IPVC's is definitely
ruled out if one assumes the Chomskyan version of transformational grammar.
On the assumption that IPVC's are S's (cf. below), there are a number of
restrictions which, in the Chomskyan framework, must be accounted for in
terms of the S.S.C. (or Opacity Condition). For example:
(12)
~s~
NP VP~
I v·~Jp VP
I I ~
saw Mary come in
However, sentences like these make clear no more than that Mary and take a
bath do not form a constituent on the level of surface structure. Assuming that
IPVC's are no S's in underlying structure is not the only way of accounting for
this. Any analysis in terms of a transformation which breaks up the S node
from underlying structure (e.g. Subject Raising, or Verb Raising) also offers an
explanation. Moreover, the claim that IPVC's never behave as constituents is
disproved by the fact that entire IPVC's can be coreferential with pronouns,
as in
The argumentation in Gee (1975) is equally weak, since all the syntactic
arguments adduced concern not perception verbs but other verbs followed by
'naked' ('bare') infinitives, like help, let and make. However, the mere fact that
all these verbs are followed by an infinitive without to can hardly be a sufficient
reason for assuming that there cannot be a structural difference between them.
There is only one argument that concerns perception verbs directly, and it is of
a semantic nature. Gee points out that when an intensional verb is followed by
INFINITIV AL PERCEPTION VERB COMPLEMENTS 109
(15) *Mary believes your brother stole her car, though she doesn't realize
it was stealing.
Gee also points out that perception verbs too set up an intensional context.
For this reason a VP dominated by an S functioning as a PVC cannot (according
to Gee) be transparent with respect to the perception verb :
(16) *Mary saw that your brother was stealing her car, though she did
not realize it was stealing.
(17) Mary saw your brother steal a car, but she didn't realize it was stealing
which make clear that what is being expressed is Bill's interpretation of the
event. It follows that the IPVC can be interpreted 'de re' in (IS,a), but not in
(lS,b) :
(19)(a) Bill saw John steal a car, but he did not realise it was stealing.
(b) *As far as he was concerned, Bill saw John steal a car, but he
did not realize it was stealing.
So, neither the claim that perception verbs always create an intensional context
nor the claim that IPVC's can always be interpreted 'de re' appears to be correct.
Gee's argument consequently has no force at all.
While there seems to be no real evidence for Gee's and Akmajian's non-
sentential analysis, there is certainly plenty of evidence against it :
3.1. The analysis assumes that the NP following the perception verb is base-
generated in object position, thus raising again several of the problems that
have been pointed out in connection with the Equi analysis (e.g. the fact that
the NP in question need not be interpretable as object, or that it may not be
generable in object position, for example when it is the existential there, or the
expletive it, etc.).
3.2. In sentences like (20), the reflexive pronoun refers back to the NP follow-
ing the perception verb, not to the subject of the head clause:
each other refers to the children, not to the subject of the matrix. On a clause
mate analysis this means that there must be a complement S containing each
other and the children, but not Peter and John. In the Chomskyan framework
the IPVC must also be a complement S since the children acts as a specified
subject blocking coreferentiality between each other and Peter and John.
3.5. As appears from (23), emphatic compound personal pronouns like myself
can 'float away' from the 'subject' NP's of IPVC's. This proves that the NP's
in question must be true subjects at some stage in the derivation (rather than
being directly generated in object position), since it is well-known that emphatic
pronouns can float away from subject NP's only (cf. (24,a-c)) :
(23)I've never seen John steal peaches himself, but I've seen his wife
do so.
3.6. As noted by Postal (1974: 196), adverbials like by himself can occur in
connection with subjects only3 :
However, adverbials like by himself can easily be related to the 'subject' NP's
of IPVC's :
112 R.DECLERCK
(26)(a) I've· seen lohn paint the whole house by himself before.
(b) I've never seen Mary go out by herself.
This proves again that the relevant NP's must be subject of an S at some stage
in the derivation.
3.7. As noted by Gee (1975) and Kirsner & Thompson (1976), it is possible to
apply there-insertion to IPVC's4 :
(27)(a) I've never seen there be anyone executed here without being
given a chance to confess first.
(b) I saw there arise over the meadow a blue haze.
This possibility of inserting there proves that the IPVC must be an S when
there-insertion takes place, since (a) there-insertion is a sentence-cyclic trans-
formation (cf. Bresnan (1971 b)), and (b) it is well-known that there can only be
inserted in subject position.
3.8. In sentences like (28,a-b) the nonfinite adverbial clause is taken to have
the same subject as the matrix. The object NP of the matrix cannot be the
controller NP.
(28)(a) Peter met lohn before leaving the country.
(b) Peter watched lohn without saying a word.
Sentences such as (29) are ambiguous, because the adverbial clause can be
embedded in the highest S as well as in the complement S, as shown in (30,a-b).
It follows that the NP controlling the intesretation of the nonfinite clause in
(29) can be I or John, but it cannot be Mary .
Consider now :
Sentences (31 ,a-·b) are ambiguous in exactly the same way as (29). The fact
INFINITIVAL PERCEPTION VERB COMPLEMENTS 113
that the controller NP can be John (the 'object' NP of the perception verb)
can only be explained on the assumption that John (or its trace) is in subject
position at the time when Equi-NP deletion (or whatever rule governs the
possible interpretations) applies. An underlying structure of the form of (I 2),
in which John is generated directly in object position is therefore inadequate:
it wrongly predicts that only Peter can be the controller NP in (31 ,a-b),
3.9. A similar argument can be derived from sentences involving infinitival result
clauses:
(32)(a) Bill had to read the poem to us so often as to get tired of it.
(b) Bill had to read the poem to us often e,nough to get tired of it.
Again, only the subject NP (Bill) can be taken to be the subject of the non finite
clause. However, in (33) not only John but also Bill can be the controller NP.
In (34) the latter interpretation is even the only plausible one:
(33)John heard Bill read that poem to us so often as to (or: often enough
to) get tired of it.
(34)John saw Bill hit the girl so hard as to (or: hard enough to) hurt
her badly.
It follows that in (33) and (34) Bill (the superficial object NP of the perception
verb) must be a subject NP at some stage in the derivation.
3.1 O. The nine above arguments concern English directly, as they are based on
sentences from this language. However, it may be interesting also to point out
that in some languages the infinitive in IPVC's is inflected to agree with the
'object' NP of the perception verb. Thus, Perlmutter (1972 : 88) notes that in
Portuguese "we find sentence pairs like the following:
(35,a) has the ordinary infinitive correr, but in (35,b) the infinitive has the
plural ending -em, in agreement with its subject (as cavalos)".6 Since I know of
no attested cases where Number Agreement is not a clause-internal rule involving
a verb and its subject, I think this can be taken as evidence that os cavalos is
114 R.DECLERCK
subject of co"erem on some level, hence that the IPVC is (derived from) a
sentential structure.
The general conclusion from this discussion of the non sentential analysis
must be that there is ample evidence - eleven arguments have been advanced -
that the analysis is untenable, both within and without the Chomskyan frame-
work.
Since IPVC's are S's in underlying structure, the most plaUSible analysis
within the framework of Chomsky would seem to be a 'bare S' analysis similar
to the analysis which Chomsky assigns to sentences like I believe John to be
an idiot. (The term 'bare S' was initially used in the sense of "an S without a
complementizer'. However, since the S's in question can be argued to have a
null COMP, I will use the term here in the sense of 'a null complementizer S'.)
On a bare S analysis the sentence I saw Mary come in is assumed to have a
structure something like (37) on both the underlying and the superficial level.
No transformation is involved in its derivation.
4.1. Chomsky (1973) proposes a bare S analysis for sentences like (38,a-b),
which Postal (1974) wants to derive by means of Subject Raising:
Chomsky concludes from this that the complex NP's stories about NP and
pictures of NP are true subjects in (41, a-b), i.e. cannot have been turned into
objects by Subject-to-Object Raising. When applying this kind of argument to
sentences involving IPVC's we notice that NP's can easily be extracted from
complex NP's that are subjects of IPVC's :
4.2. As pointed out by Chomsky (1973), some of the conditions proposed hold
not only for transformations but also for rules of interpretation. For example,
for the interpretation of a sentence like (43,a) it is necessary that not be
associated with enough rather than with understand. This association is blocked
by the S.S.C. in (43,b), so that this sentence "receives no direct interpretation".
Consider now :
(44)(a) You haven't heard me utter enough English sentences (to be able
to conclude that I speak English perfectly).
(b) I haven't seen him visit sufficient pubs (to conclude that he is
a hard drinker).
Sentences like (44,a-b) are impeccable. Yet they would" have to be blocked
by the S.S.C. if the 'subject' of the IPVC were really in subject position in the
surface structure. (Note that the S.S.C. does operate as expected in (45), which
therefore receives no interpretation:
4.3. In English (like in many other languages) it is possible for a negator whose
scope is an entire sentence to be incorporated in one of the lexical items of that
sentence. Thus, structures that have the logical form of something like (46,
a-c) may lexicalize not only as (47,a-c) hut also as (48,a-c) :
I will not go here into the question of whether this incorporation should be
accounted for in transformational terms or in terms of a rule of interpretation.
What is important is that in both cases the rule is subject to the general
conditions pointed out by Chomsky. For example, although it is possible for a
structure with the logical form of (49,a) to reach the surface as (49,b), (SO,a)
cannot be realized as (SO,b) because of the S.S.C. (the specified subject being
John's).
Consider now :
In all these sentences the negator which is incorporated in one of the lexical
items of the IPVC really bears on the matrix. (For this reason the sentences
can be supplemented with and neither did John, but not with and John did
so too.) If the IPVC's were S's on the surface (as would follow from a bare S
analysis), the 'subject' NP's of the IPVC's would be true specified subjects and
the sentences would be blocked by the S.S.C.
4.4. Sentences like (51 ,a-c) disprove a bare S analysis in still another way. The
process of negator incorporation is clause-internal in the strictest sense: the
negator cannot be incorporated in any lexical item belonging to a higher or
lower S. This is clear from examples like (S2,a-b) and (S3,a-b), where the (a)
and (b) sentences are not synonymous: in the (a) sentences the negator bears
on the matrix (so that we can add and neither did John), whereas in the (b)
sentences the incorporated negator bears only on the embedded S, leaving the
matrix positive (so that we can add and so did John) :
Since the negator which is incorporated inside the IPVC in (51 ,a-c) really
bears on the matrix, it follows that (51 ,a-c) must be superficially simplex,
hence that a bare S analysis (as well as Rosenbaum's Equi analysis) is
disconfirmed.
4.5. In spoken English the pronouns them and him are often reduced to 'em
and 'm. This reduction happens independently of whether the pronoun in
question is a normal object (as in I saw 'em, I heard 'm) or is the 'subject' of
an IPVC (as in I saw 'em come in, I heard 'm say it). Now Bresnan (1971 a : 6)
argues that an item that undergoes vowel reduction must necessarily be a surface
'syntactic dependent' (sister constituent) of the item to which it adheres.
Because of this, the fact that want to go can be reduced to wanna go is taken by
Bresnan to mean that there must be a transformation turning to (which is
initially a syntactic dependent of go) into a syntactic dependent of want.
If Bresnan's theory is correct, it follows that reduced pronouns like· 'em
and 'm must be syntactic dependents of the perception verb. In other words,
in surface structure we must have the configuration (54), which is incompatible
with a bare S analysis.
(54)
/Vp~
V NP
I
saw
,mI
4.6. It is inherent in a bare S analysis that the S in question must not be
dominated by an NP node. If there were an NP over S, both the A-over-A
principle and the Subjacency Condition would prevent moving any NP out of
the S and into the matrix. :
In actual fact, however, NP's can be extracted from the complement clause :
Because of sentences like these a bare S analysis has to assume that the S which
dominates the IPVC is not itself dominated by NP. However, there are several
indications that, as far as IPVC's are concerned, this assumption is dubious.
INFINITIV AL PERCEPTION VERB COMPLEMENTS 119
To mention only the most obvious one: some of the perception verbs after
which IPVC's are found are prepositional verbs, as in the following sentences 7 :
Postal bases his derivation (among other things) on the claim that reflexivization
is a clause-internal operation. Chomsky (1973) rejects this claim, saying that
reflexivization can apply across an S boundary as long as none of the general
conditions are violated. Thus, reflexivization applies directly to (S8,b).
According to Chomsky there is therefore no evidence for a Raising analysis
and (S8,a) simply has a bisentential surface structure. It should be noted that
this bare S analysis implicitly rejects the claim that only object NP's can be
reflexivized (or marked as coreferential with the subject NP by a rule of
construal). Postal (1974) by contrast, maintains this claim, on the basis of
examples like
But Chomsky holds that the ungrammaticality of (59) follows from the violation
of the Tensed S Condition (or PIC or NIC), not from the fact that a subject
NP is reflexivized.
However, there is one problem for Chomsky's analysis that does not arise
on the assumption that subject NP's cannot reflexivize. Bisentential underlying
structures similar to (58,b) seem to be required for several classes of verbs in
English, but if the two clauses have the same subject, the second subject NP
cannot be reflexivized. Rather it has to undergo Equi-NP deletion. For example:
120 R.DECLERCK
Sentences like these appear to bear out the claim that subject NP's cannot
reflexivize (since Chomsky's conditions will not block reflexivization in the (b)
sentences)8. Sentences like (58,a) are then evidence against a bare S analysis,
since on this analysis the reflexive pronoun is in subject position.
When we try applying this argument to structures with IPVC's we find that
these can undergo reflexivization, but not Equi-NP deletion:
Sentences like these tend to disconfirm a bare S analysis of IPVC's and suggest
that Raising applies in their derivation.
4.8. In English it is possible for a time adverbial to occur between a verb and its
direct object, provided the latter is a superficial clause or has undergone
Complex NP Shift:
4.9. As noted by Postal (1974 : 83), Complex NP Shift cannot operate on NP's
that are subjects at the point of application 9 :
(65)(a) *1 regret the fact that were destroyed so many of our relics.
(b) *Are happy all of the men who recovered from mononucleosis.
INFINITIVAL PERCEPTION VERB COMPLEMENTS 121
However, sentences like (66 ,a -b) appear to be grammatical. This suggests that
the 'subjects' of the IPYC's are no longer subjects when Complex NP Shift
applies.
(66)(a) Into the garden I saw run the fattest cat in Urbana. (example
from Bolinger (1977 : 516))
(b) I saw come into my house a respectable English family with
four children
4.10. A further weak point of a bare S analysis is that, as has already been
noted, IPYe's never behave as single constituents in the surface structure. In
spite of the fact that a bare S analysis treats them as superficial S's, they cannot
appear as S constituents in any of the constructions that are commonly used
as tests for constituency (even if the constituents in question need not be NP's).
Compare:
(67)(a) I heard, but I did not see, that someone fired a shot. (Right
Node Raising) 10
(b) *?I heard, but I did not see, someone fire a shot.
(68)(a) What I heard was that someone opened the door. (Pseudo-
cleft)
(b) *What I heard was someone open the door.
(69)(a) That someone had opened the door, I saw at once. (Topicalization)
(b) *Someone open the door, you will see at once.
(70)(a) What did you see? - That someone went into the house.
(b) What did you see? - *Someone go into the house.
Sentences like these suggest that IPVC's are no longer constituents in the surface
structure. (Notice that this fact is consistent with a raising analysis).
(72)(a) We saw the boys drink so much whisky that they got drunk.
(b) We saw the boys drink so much whisky that we could not believe
our eyes.
In (72,a) the contents of the that-clause are a function of the quantity of whisky
that the boys drank; in (72,b), on the other hand, the contents of the that-
clause are a function of the quantity of whisky that we saw the boys drink.
In the Chomskyan framework this means that the rules of interpretation must
relate the phrase containing so much to drink in (72,a) and to saw in (72,b).
However, on a bare S analysis the rule establishing a relation between so much
and saw should be blocked by the S.S.C., since the boys is a specified subject.
The fact that the S.S.C. does not operate in (72,b) therefore disconfirms the
bare S analysis. (Notice that the S.S.C. does operate as expected in senten~es
like (73,b) an:d (74,b), which are therefore semantically deviant:
(73)(a) We resented their drinking so much whisky that they got drunk.
(b) ·We ·resented their drinking so much whisky that we punished
them.
(74)(a) We saw that they drank so much whisky that tht!y got drunk.
(b) ·We saw that they drank so much whisky that we could not be-
lieve our eyes.)
Concluding this discussion of the bare S analysis of IPVC's we can say that
evidence of various kinds appears to disprove this analysis. Some arguments
concern the fact that IPVC's do not always satisfy the conditions (more
specifically, the Subject Condition and the S.S.C.) which they should satisfy on
a bare S analysis. Other arguments make clear that the 'subject' NP of an IPVC
must be a syntactic dependent of the perception verb in the surface structure,
and that the S underlying the IPVC must be dominated by NP. Still other pieces
of evidence show that the 'subject' NP of the IPVC can no longer be subject in
the surface structure, that IPVC's cannot be clausal constituents on the surface,
and that sentences involving IPVC's must be superficially simplex.
5. RaiSing
It is clear from the above summary that a raising analysis along the lines of
Postal (I974) will encounter none of those problems. On this analysis the IPVC
INFINITIV AL PERCEPTION VERB COMPLEMENTS 123
5.1. For many linguists the most serious objection against this raising analysis
will no doubt be that it is quite incompatible with the Chomskyan framework
(see also below). Indeed, Postal (1974) explicitly rejects Chomsky'S conditions
in favour of a theory that crucially involves 'clause mate rules' like
reflexivization, reciprocal marking, passivization and the inclusion constraint
(disjoint reference). As pointed out by Lightfoot (1976 : 275--6), the existence
of a rule of raising would therefore ruin the otherwise good results that come
from the Chomskyan framework. On the other hand, it must be admitted that,
in the particular case of IPVC's, raising yields far better results than the
Chomsky an analysis.
5.2. An argument that has often been adduced against Subject-to-Object Raising
is that the putatively raised NP cannot undergo Object Shift (Tough-movement)
(see e.g. Gee (1975 : 352), Ughtfoot (1976)) :
(75)(a) *The plane was difficult for us to see fly at that distance.
(b) *John was easy for me to hear come into the house.
However, I do not think that this is a particularly strong piece of evidence. The
logic of the argument is that the possibility of applying Object Shift is not only
a sufficient but also a necessary test for establishing whether an NP is an object
in the surface structure. However, it is not logically necessary that this should
be the case. There is no a priori reason why the possibility of Object Shift should
not be restricted to base-generated objects only. As a matter of fact, precisely
this claim is made in Berman (1973 : 39), where it is argued "that Tough-move-
ment may not move any NP that has been previously moved by any other trans-
formation, or, otherwise stated, that Tough-movement may move a NP only
124 R.DECLERCK
from its position in underlying structure". Berman's examples involve not only
sentences with putatively raised NP's but also sentences in which the superficial
object NP has already been moved by a prior application of Dative movement,
About-movement (cf. Postal (1971 )), and passivization. If Berman is correct, the
ungrammaticality of (75,a-b) does not disprove the raising analysis but actual-
ly confirms it.
5.3. A truly difficult point about the raising analysis is the question of whether
or not the S should be pruned after its subject has been raised. According to
the pruning convention argued in Ross (1967 : 26), the S node (and the NP
above it) should be pruned because the node does not branch any more (i.e. it
dominates only VP) once raising has applied. However, Postal (1974: 231-
232) explicitly rejects that pruning takes place. The reason is that, if the S is
pruned, the ungrammaticality of sentences like the following becomes inex-
plicable :
On the assumption that raising and S pruning have operated here, (76) is super-
ficially simplex and clause mate rules like reflexivization and Reciprocal Marking
should yield grammatical results. The ungrammaticality of (76) therefore obliges
Postal to reject that raising is followed by S pruning.
However, if S is not pruned after raising (that is, if the output structure of
raising is not superficially simplex), a raising analysis of IPVC's is inadequate,
since we have seen that sentences involving IPVC's are superficially simplex (cf.
especially 4.4.). A further piece of evidence confirming this point is provided
by sentences like (66,a-b) (partly repeated here) :
(66)(a) Into the garden I saw run the fattest cat in Urbana.
(67)(a) *Into the garden I saw that ran the fattest cat in Urbana.
(b) *She asked whether it looked like rain a man who was near her.
(c) *1 persuaded to help me the senator who came from Texas.
6.1. As we have seen, the fact that IPVC's can follow prepositional verbs like
look at and listen to seems to be an indication that the S dominating the IPVC
is itself dominated by NP. This is in accordance with the raising analysis but
presents a problem for the bare S analysis since, in the Chomskyan framework,
the A-over-A principle and subjacency must block any transformation (even
WH-movement) extracting an NP from the S, whereas in actual fact extraction
does not yield ungrammatical results at all (cf. above). If raising is incorporated
within the Chomskyan framework this problem crops up again. Even raising itself
cannot apply then without violating subjacency if it operates on an S dominated
byNP.
6.2. In the discussion of the bare S analysis I have pointed out several types of
126 R.DECLERCK
sentences involving IPVC's which are quite grammatical but should be blocked
by the S.S.C. (or Opacity Condition) on the bare S analysis. On the Raising+t
analysis these cases are equally inexplicable, since the t that is left behind when
the subject NP of the IPVC is raised is the specified subject no less than the
original NP so that the S.S.C. should again block the sentences in question.
7. Conclusion
7.1. When comparing the advantages and disadvantages of the bare S analysis,
raising and raising+t we come to the following conclusions:
a. The bare S analysis raises numerous problems, which have to do with (a)
the impossibility of having NP above S; (b) the fact that in several cases the
S.S.C. does not operate as expected;(c)the fact that in at least one case the
Subject Condition does not operate as expected; (d) the fact that the 'subject'
NP of the IPVC cannot be treated as a syntactic dependent of the perception
verb; (e) evidence that the 'subject' NP of the IPVC is no longer subject in the
surface structure; (f) evidence that IPVC's are no longer clausal constituents on
the surface; (g) evidence that sentences involving IPVC's are superficially
simplex; and (h) the impossibility of applying Object Shift to the object of an
IPVC.
b. A raising analysis (without t) creates none of those problems but raises
some difficulties in connection with S pruning and (possibly) the impossibility
of applying Object Shift to the raised NP. It also ruins the (otherwise fairly
good) results of the Chomskyan framework.
c. A raising+t analysis raises both problems (a), (b), (g) and (h) of the bare S
analysis and the difficulties connected with raising (without t).
INFINITIV AL PERCEPTION VERB COMPLEMENTS 127
FOOTNOTES
IThis conclusion does not hold for Dutch (and probably German and French)
IPVC's. which can be satisfactorily analysed in terms of Verb Raising. (see
Declerck 1981a).
2This phrase-marker has been simplified so as not to contain such nodes (e.g.
S, COMP, AUX, Tense) as are irrelevant to the argumentation.
3The restriction in question holds only for the adverbial use of the by-phrase.
It does not prevent it being used as subject complement or object complement
(predicative adjunct) ;
(i) I have never seen John (as he was) by himself.
4Examples of this construction are not very numerous because there-insertion
is not very usual with nonstative verbs, whereas IPVC's only exceptionally take
a stative verb.
5Actually, interpretation (30,a) is less acceptable than interpretation (30,b)
because (30,a) violates Ross' Internal S Constraint (which is a surface constraint
saying that a sentence containing a clause-internal S in surface structure has a
reduced acceptability).
6The reference numbers in this quotation have been adapted.
7For some speakers sentences like these may only be marginally acceptable.
The reason is that 'active' perception verbs like look at and listen to imply a
durative perception and are therefore normally followed by PPVC's (Participial
Perception Verb Complements) (e.g. I looked at the servant making up the fire).
However, as argued in Declerck (1981c), IPVC's differ from PPVC's only in
that the former are non progressive whereas the latter are progressive. (That is,
I saw John coming is in fact a reduction of 'I saw John be coming'). Apart
from this difference of aspect there is no structural difference between IPVC's
and PPVC's (at least if we disregard special uses of PPVC's - see Declerck
(1981 c, 1981 d». It follows that the argument that there must be an NP above
the S dominating the IPVC is unaffected even if it should be necessary to
substitute PPVC's for the IPVC's in (57,a-b).
8Chomsky (1976) attempts to obviate this problem by assuming that reflexiviz-
ation always takes place but that the reflexive pronoun "is obligatorily deleted in
the context [or - VP. Thus, he claims that I want to leave at once is derived
from 'I want for myself to leave at once' by obligatory deletion of [or myself
However, this is an ad hoc solution which necessitates assuming a [or-comple-
mentizer after some verbs (e.g. remember) which can never be overtly followed
by [or. Moreover, the solution does not cover cases like (62,a-b) where the
128 R.DECLERCK
complement clause isa gerund, since gerund clauses by definition do not involve
the complementizer for.
9Bresnan (1976 : 486-487) challenges this claim on the basis of sentences
like the following:
(i) Near the fountain sat a large purple gorilla.
However, Gazdar & Pullum (1980) show (on the basis of facts concerning
Number Agreement) that there has been deleted from Bresnan's sentences,
hence that the complex NP's were not subjects when they were shifted.
Bresnan's objection is therefore vacuous.
10For a discussion of the usefulness of Right Node Raising as a test for
constituenthood see Postal (1974: 125), Bresnan (1974), Bresnan (1976:
493) and Abbott (1976 : 639).
PASSIVE IN A SEMANTIC-SYNTACTIC DEPENDENCY
NETWORK
W. Van Langendonck
Katholieke Universiteit Leuven
129
130 W. VAN LANGENDONCK
As Kirsner (1976 : 389) states, "in both its uses, passive morphology (worden +
past participle) is one linguistic sign signaling the single, relatively abstract
meaning high participant not focussed, which may be roughly paraphrased as
'the logical subject is not the grammatical subject"'. The latter principle recurs
in Hoekstra & Moortgat (1979 : 148). It is even claimed to be a universal. How-
ever, these two authors do not consider passive subjects, passive morphology,
passive agent-phrases to be essential for the characterization of passive. If this
is correct, as I think it is, it entails passive status for patterns as exemplified in
Hoekstra & Moortgat (1979) :
where there is coreference between the affected person and the agent. Moreover,
a typical passive auxiliary is absent. Nevertheless, we can speak of a background-
ing of the agent in this case as well, since an active counterpart with a fore-
grounded agent can be constructed: She dressed (herself). The possibility of an
active counterpart is another passive test. According to this criterion, (3) looks
more passive than (4) - (9), because only (3) clearly corresponds to an active
sentence with a foregrounded agent: Ik hoar Piet een fuga spelen 'I hear Pete
playa fugue'.
In light of these observations, it would probably be wise to adhere to a fuzzy
passive concept. As fully passive we would view cases like (1), where all passive
features show up. The socalled pseudopassive ill (2) is less passive because it
lacks an affected NP, and so on. On the other hand, we would retain the hard
core of passive: the backgrounding of the logical subject. Hence, we see no
passive status at all in sentences like:
As Pollmann (1975 : 151) points out, geboren is not a passive participle but an
adjective since no infinitive corresponds to it. This explains that an agent-phrase
is out in (11). The verb worden does not function here as a passive auxiliary,
but as a copula combinable with an adjective and meaning 'become'. No agent
or logical subject seems to be implied at all, so we cannot speak of a passive.
In (12) we encounter a by-phrase, even a true agent-phrase, but we cannot say
that the agent is backgrounded, for semantically, the preposition by functions
as the main predicate unlike in passives.
2.1. The background position of passive by-phrases correlates with the fact that
they are mostly optional, sometimes even impossible (cf. 10). The majority of
English and Dutch passives simply lack an agent-phrase (80 to 87 %) according
to Stein 1979 : 35 and Kirsner 1976 : 392). Coetzee (1980 : 208) mentions for
English that "the long passive is of later date, and perhaps a literary invention".
This squares well with the observation that in the dialects of Dutch, passive
agent-phrases seem to be extremely rare. In a few languages such phrases do not
even exist.
By-phrases share their backgroundedness with other adverbial prepositional
phrases. The latter are also mostly optional. In just a few instances, agent- and
other adverbial phrases are obligatory, compare:
(13)a. His first insult was followed by an even worse one. (Stein 1979 :
103)
b. She put it on the table.
PASSIVE IN A DEPENDENCY NETWORK 133
A similar case is
While English would use a form with to do to replace the process verb, Dutch
can insert the verb gebeuren 'happen'. But this is an intransitive active verb.
Nevertheless, it obtains the same door-phrase as the antecedent passive clause.
We could leave out the expletive form dit gebeurt 'this is done' in (16) without
harming the meaning of the sentence at all.
That the agent phrase preposition preserves its meaning is also suggested by
the circumstance that sometimes a different preposition is utilized. Coetzee
(1980 : 205) adduces a couple of English examples:
(19) In ons land wordt er opvallend veel getrouwd tussen hier wonende
buitenlanders en Belgen.
In our country there becomes strikingly often married (= there is an
awful lot of marrying) between foreigners living here and Belgians
2.2. With respect to passive auxiliaries, the same line of argument can be
followed as in the case of by-phrases. As a rule, in transformationalist
approaches the passive auxiliary is not a basic lexical item, but a derived one.
This is even the case in the 'nontransformationalist' phrasal analysis of passive
in Keenan (1979). Passives here are VPs derived from transitive VPs (TVPs),
e.g.
In this approach we still need an operation converting the operator PASS into
[be ... + -ed]. This entails that there are at least two verbs be, one being the
normal auxiliary, the other being transformationally derived in view of passive.
More embarrassing is the fact that be is not the only passive aUxiliary in English.
Coetzee (1980 : 205) mentions others: get, become, stand, e.g. in :
(25) Wat er gebeurde was dat plotseling de hele zaak door iedereen werd
(*was) verstaan.
What happened was that suddenly the whole matter was understood
by everybody [event]
Finally, it can be remarked that worden and zijn require different dummy
subjects when the real subjects are extraposed clauses: worden entails the use of
er 'there', zijn the use of het 'it'6:
3. The formal analysis of passives will have to meet the above considerations
and requirements. Especially, we will have to account for the autonomy of
passive with respect to active, the backgrounding of the logical subject, the
lexical status of the passive auxiliary and the preposition of the agent-phrase
(if these elements figure in the sentence), and the irrelevance of word order or
transformational operations.
A suitable framework to handle passive (and other structures or operations !)
seems to me to be dependency grammar. This format enables us to separate
constituent structure rules from wordorder rules. If one allows networks in
addition, it is possible to deal with a lot of 'operations' for which otherwise,
transformations or even global rules would be needed (see Hudson 1976;
Schachter 1978; Hoard 1979). As for the notion of dependency itself, one can
distinguish between syntactic and semantic dependencies (Heringer e.a. 1980;
Nichols 1978). Certain authors only work with semantic dependencies (Hoard
1979; Langacker 1978), others only with syntactic dependencies (Hudson
1976). It may well be that we need both of them (Schachter 1978). A good
starting-point is Langacker's theory of space grammar. This grammar yields
semantic structures of sentences where form reflects meaning in a direct way.
Langacker claims that "even simple sentences like He may see her and He was
sad involve several propositional layers or STRATA, each having a distinct
predictional function in the composition of the complex type of predicational
unit we call a sentence" (Langacker 1978 : 854-5). The above two sentences
are assigned a semantic or PROPOSITIONAL structure as exemplified in figures
(30a-b). Important to the present approach is the fact that Langacker uses
dependency trees, "though these are to be regarded as convenient abbreviations
for a more basic SHELL or LAYERING structure representing part-whole
relationships" (ibid.), cf. fig. (30c) :
138 W. VAN LANGENDONCK
GROUND G (G) G
I I
EPISTEMIC MAY (E) DIST
I I
EXISTENTIAL DO (3) BE
I (OC)
I
OBJECTIVE SEE SAD
CONTENT /"""HER
HE HE
I
Fig. (30a) He may see her Fig. (30b) He was sad. Fig. (30c)
In these figures, an epistemic path leads from the speaker to the objective
situation. The latter is designated by the lowest stratum: OBJECTIVE
CONTENT (OC). The EXISTENTIAL stratum (3) predicates existence of this
situation, whereby do predicates the existence of a process, while be indicates
the existence of a state. The EPISTEMIC layer (E) contains an optional modal
as well as tense. (In 30a there is simply no tense; in 30b the DISTAL label re-
places the traditional past tense). The term GROUND (G) specifies the speech
act involved and refers to the speaker. In the present paper, the main concern
is of course the level of OBJECTIVE CONTENT. As Langacker (1978) essential-
ly deals with the epistemic stratum, the analysis of the objective content layer
is not very elaborate. It is e.g. not clear whether his dependency structures
are networks or just trees. The difference is highly relevant to the analysis pro-
posed here. It amounts to the following 7 : if the arguments in the sem~ntic
representations are to have only a single predicate, we speak of a dependency
TREE; if the arguments are allowed to have more than one predicate, a
dependency NETWORK is created. The predicates are displayed above their
arguments and are said to 'dominate' them. Each predicate plus the arguments
it dominates form a proposition. The propositions themselves may in turn
function as arguments of complex propositions. The resulting semantic repre-
sentations thus contain only propositions and their elements.
(31)'
S BY
I.
terronsts
BE
I
execute
~Homer
L'>.
( AtJ~s~rr~~sts
As can be read from the diagram, there is some overlap between semantic and
syntactic dependencies: for both kinds of dependencies, EXECUTE depends on
BE, Athens on IN and territorists on BY. Apparently because of their secondary
character, the secondary predications take the form of (backgrounded)
adverbials in syntactic structure and therefore depend syntactically on the (non-
fmite) verb, though semantically they dominate it (plus its arguments). The
DlST AL predicate depends morphologically on the auxiliary. The greatest
discrepancy regards the passive subject. Whereas semantically it is the object of
EXECUTE, syntactically it depends on BE, not also on EXECUTE, as Hudson
(I976) would hold. Deciding on syntactic dependencies is facilitated by two
criteria given in Nichols (1978) :
i) There is syntactic dependency between X and Y if wordorder rules have
to refer to X in establishing the position of Y or vice-versa. Applying this to
(32), we can state that subject and fmite verb are in a syntactic dependency
relation because they follow each other directly in this and most other
sentences. This does not obtain for the subject and the past participle
execu ted 10.
ii) The second test says that elements with a syntactic dependency relation
can form a minimal utterance. As I can say Homer was, but not Homer
executed, this criterion confirms the first one.
3.2. To prove the usefulness of syntactic dependency relations for the analysis
of certain typically passive phenomena, let us turn to more sophisticated
patterns.
142 w. VAN LANGENDONCK
3.2.1. For English we have to deviate sometimes from the above principle
that the first simplex specified argument under BE obtains the status of
syntactic subject. In certain cases where the active sentence displays an indirect
as well as a direct object, the corresponding passive allows either object to
become the subject, e.g.
The difference between (33b) and (33c) can only be read from the syntactic
part of the stemmas :
(33)'b. (33)'c.
!J. happiness me
•
this
In Standard Dutch, as in a lot of other languages, structures like (33c) are not
allowed. However, in informal, colloquial and dialectal Dutch, they are fairly
frequent (cf. Van Langendonck 1968). As in many other languages where such
patterns do not occur freely, they are mostly found with possessive NPs and/or
in adversative contexts 11 (Davison 1980), e.g.
(38)' DISTAL
(I~
hfIJG~~k ~lEDEN
t
een
3.2.3. Hoekstra & Moortgat (1979: 155ss.) treat of Dutch passive patterns
exemplified by :
In a lexicalist framework, they set up the following syntactic phrase structure for
(39), in an SOV order:
(39)'
PP v
I I
door Piet spelen
Surprising is their claim that een fuga would be the direct object of hoar and
at the same time the syntactic subject of spelen. The reason for this move is
apparently that in thi.s nontransformational phrase structure grammar, there is
hardly another way to analyse the like patterns. But from a methodological
point of view we should be careful in assigning to an NP the functions of
syntactic object and subject simultaneously, even if different verbs are involved.
In such an approach, the definition of subject and object is certain to become
extremely complicated and ad hoc. It is then a pity that any motivation for the
move is lacking. Indeed, a pronominalization test proves that een fuga in (39)
can only be a direct object by normal standards: Ik hoar haar spe/en (door
Piet). The pronoun haar is an accusative form, hence cannot be a syntactic
subject. And as long as no special evidence is given, we have to assume the most
obvious interpretation, viz. that een fuga is simply the direct object of spelen,
not of hoar.
Via a particular dependency criterion, a dependency approach reaches this
conclusion. In Van Langendonck (forthcoming) the following test to identify
a syntactic dependency relation between words is set up (for Dutch) : words
which in a main clause can occur together before the finite verb, are related
syntactically. Application of this criterion to the crucial elements of (39) yields:
PASSIVE IN A DEPENDENCY NETWORK 145
This means that there is a dependency relation between een fuga and spe/en.
In other words, een fuga functions as the direct object of spe/en 17 .
In agreement with the above considerations, sentence (39) receives the
following dependency diagram :
(39)"
ik
pRE~ P7
SPELENpl
t::.
/~fuga
+
een
Hoekstra & Moortgat (1979 : 157, fn. 18) remark that in the complement of
verbs like horen 'hear' no pseudo-passives, i.e. passives with intransitive verbs (as
in patterns like 2) can occur. Sentence (41) is deemed ungrammatical:
The awkwardness of (41) is, however, rather due to a pragmatic factor. Such
marked passives as the pseudo-passive do not take agent-phrases as easily.
Especially, if an agent-phrase is chosen, it should have some length and second,
the NP in it should not refer to a concrete individual like John in (41) but rather
to a non-particular referent, which is likely to occupy a background position 18.
Observing this pragmatic condition allows us to construct better examples than
(41) :
( 42)'a
,sI
NIET
'(In
TAL
.r-~TEN
(/ ~ ~OOR
hotelier ONTBIJTEN I)
iedereen
dt t
I tegelijk
(43) DISTAL
(I
WORDEN~ DOOR
er "'---;:'~SEN ~ ied~feen
I
f,
The markedness of this type is also illustrated by the constraint that the logical
subject should refer to a human agent. This obtains for the logical subjects in
(42) as well.
3.2.5. The next type requires an agent as logical subject; it concerns a peculiar
Dutch pseudo-passive with the verb gebeuren 'happen' as head of the sentence.
The proper content of the utterance is given in the subject of gebeuren, which
is a nominalized structure. Moreover, a door-phrase can be attached to gebeuren,
though it is an active intransitive verb 19 ! Consider the example (and compare
sentence 16) :
r./ ""-
het'
VERVEN~
~
-van
}
•
de
[:, hUIS
•
het
3.2.6. Last, but not least, we want to say a few words about so called 'raising'
in passive structures, as occur in Latin and English (but not in Dutch), e.g.
The 'raising' of /apides from (logical) object to subject can be accounted for by
the joint semantic and syntactic structures of the following tree:
(45)' ESSE
I)
COEPISSE
I)
CONJICERE
/~lapides
t:,
(46)' DI"AL
BE
~I)
( _ BELIEVE"
th". /~)
BE
,
!::.
IU~
mice
~~barn
the
The deeply embeddded logical subject mice of IN becomes the syntactic subject
of BE by the above principle: take the first simplex specified argument as the
syntactic subject. The same principle is applicable to the Latin sentence (45).
The element there can be assigned the role of a presentative functor as was
proposed in Hetzron (1975).
Notice that for these raised structures we do not even need the network
device : a single dependency tree is sufficient.
FOOTNOTES
I For detailed recent evidence see Langacker & Monro (1975), Kirsner (1976),
Hoard (1979), Stein (1979), Coetzee (1980), Keenan (1979).
2For these notions see Mourelatos (1978).
3The reason why occurrences are the unmarked class in passives lies in the
circumstance that passives can only be formed with verbs (not with adjectives),
and verbs normally refer to occurrences.
4Pollmann (1975 : 23ss., 146ss.) overlooks cases with experiencers and mentions
only agents. In general, for the use of worden in passives, not agency is relevant,
but the notion of occurrence, as posited above; see examples like (25). Even so,
problems remain (cf. Pollmann 1975 : ISO).
50nly with locational predications do adverbial phrases, possibly by-phrases,
seem obligatory: Het huis is omgeven door het bos 'The house is surrounded
by the forest' (*Het huis is omgeven); Het terrein is gelegen op het platteland
'The plot is situated in the countryside' (*Het terre in is gelegen).
6For other phenomena in connection with er and passives, see Pollmann (1975 :
112ss.).
7See especially Hoard (1979); also Hudson (1976; 1980a,b), Schachter (1978).
8As Keenan (1979: 13) himself admits, his other criticisms of the sentential
account exclusively apply to transformational analyses, not to bisentential
accounts like Langacker & Munro (1975) or, consequently, Hoard (1979).
9The father of dependency grammar did not even mention semantic dependen-
cies (Tesniere 1969). Hudson (1976) mentions them, but works exclusively
with syntactic dependencies.
I OJ ackendoff (1977: 32, fn.2) makes a similar claim with respect to the
sentence John is tall: he admits a syntactic relation between John and is, not
between John and tall, though there are se1ectional restrictions between the
latter two . .Interestingly, he uses quite different arguments to corroborate his
claim.
11In an adversative context the event described possesses an adverse character,
e.g. in (34).
12Notice that deze stoe/en 'these chairs' can function as topic in colloquial
Dutch, the preposition being stranded: Deze stoelen is nog niet op gezeten.
Agreement is not (yet) possible. In English it is, cf. Sinha (1978 : 450), Davison
(1980 : 44).
13 For the treatment of co-predicates in phrase structure and in dependency
grammar, see Nichols (1978) and Van Langendonck (forthcoming).
14 In this way we avoid Johnson's (1977: 159, fn. 14) dilemma whether the
passive subject (in equivalent German structures) is a case of indirect object
150 W. VAN LANGENDONCK
Annie Zaenen
Harvard University
In their important 1978 paper, Kayne & Pollock argue that Stylistic Inversion
in French provides independent evidence for the successive cyclic movement
hypothesis argued for in Chomsky (I977) and references cited there, on the
basis of island phenomena. Bresnan (I975) and Grimshaw (I975), on the other
hand, show that both in the comparative construction in Modern English and in
relative clauses in Chaucerian English, the successive cyclic wh-movement hypo-
thesis runs into trouble and that in these constructions, unbounded deletion
should be allowed. Nevertheless, both comparative subdeletion and presumably
relativization in Chaucerian English obey the "island" constraints that the
successive cyclic wh-movement hypothesis was meant to explain. The same
state of affairs obtains in Irish, according to the analysis given in McCloskey
(I979), in Kikuyu as argued in Clements (I979), in Old English (Allen (I 977)),
and Old Icelandic (Maling (I 976)).
The findings of Kayne & Pollock (1978) and those of Bresnan (I 975) and
Grimshaw (1975) are not incompatible: it could be that the French
constructions that they are analyzing are to be derived by successive cyclic
movement, whereas in other cases "extraction" phenomena have to be
accounted for by unbounded deletion 1. Chung (forthcoming) for instance, de-
fends explicitly an analysis under which relative clauses in Chamorro are derived
via unbounded deletion, whereas questions are formed by successive cyclic wh-
movement. The drawback of such an approach is that in the cases referred to
above, the unitary explanation for island constraints would be lost, except if
successive cyclic wh-movement and unbounded deletion are augmented with
some other mechanism like the indexing procedure outlined in Bresnan &
151
152 A.ZAENEN
1.1. Reiativization
Underlying structure :
Wh-movement:
Wh-deletion :
The postulated wh-deletion rule in Icelandic has to be different from the one
proposed for English by the proponents of a similar analysis of English that-
relatives. In English we have both that and wh-relatives, hence, one can say that
there is no obligatory deletion in COMP, but just an output constraint
prohibiting doubly filled COMP nodes in the surface. In Icelandic the wh-consti-
tuent must be obligatorily deleted, leaving no overt sign that wh-movement has
applied in the derivation of any relative clause.
The problem arises in connection with the pied piping cases. From a source
(3), we can get (3a) :
But now a simple-minded deletion rule will produce either (3c) or (3d) :
Both these sentences are ungrammatical; (3c) is obtained by assuming that all
the material in COMP preceding sem has been deleted ,whereas in (3d) only the
wh-word has been deleted.
Instead of having a rule that deletes wh-elements in COMP, we could also
have a filter that excludes the ungrammatical sentences once they are generated.
A possible filter would look as follows:
It would rule out all prepositional phrases in COMPo But such a filter would· not
be strong enough : prepositions are not the only kind of pied piping that is dis-
allowed in relative clauses. Possessives have also to be excluded. Whereas we have
the questions in (5), the corresponding relatives are not well-formed:
but:
So it is not enough to rule out PP's; one has also to rule out some types of
NP's, but of course not all of them, otherwise there would be no relative clauses
left under the proposed analysis.
One can achieve this result e.g. in the following way: first one deletes the
wh-word (and only the wh-word) in COMP and then if there is still some
material left in COMP except for sem, one filters out the result. So we have the
following two rules:
1.2. Topicalization
Hence the topic is in its surface position in the base; what is moved is a wh-
constituent that is moved up from its base position to the COMP position of the
156 A. ZAENEN
S in a successive cyclic fashion and that is deleted there. This proposal is meant
to account for both topicalization and Left-Dislocation by the same base rule.
A first problem that arises in Icelandic is that topicalization but not left-
dislocation is allowed in embedded clauses. A relevant contrast is given in (9) :
(9) a. *Hann segir ab 6lafur. hann hafi ekki heyrt fni honum.
He said that Olaf (N), he had not heard from him.
b. Hann segir a~ 6lafi hafi hann ekki heyrt fni
He said that Olaf had he not heard from.
If we allow for the introduction of topics and left-dislocated elements with the
same PS rule, we have to invoke a separate mechanism to avoid the generation
of sentences like (9a).
Another type of problem arises in cases where topicalization is possible but
question movement is not. These are admittedly rare in Icelandic but we find the
following contrast :
Munu is one of the rare verbs that subcategorizes for VP but not for NP and in
Icelandic, as in English, there are no pro forms for VP's. Hence the wh-element
that should be moved under Chomsky's (1977) analysis is not available. One can
invoke an abstract pronoun here, but note that one has also to give it a
distribution that is quite different from that of overt pronouns, in that it has to
have a VP-form. A prediction that would have been made by the wh-movement
analysis ohne weiteres, namely the parallel behavior of pre posed topics and
preposed wh-phrases, is not borne out and a special assumption is needed to ex-
plain the lack of parallelism.
A third problem is that it is not clear how reflexivization facts should be
handled under Chomsky'S (1977) analysis. Reflexivization is a sentence internal
process in Icelandic. The conditions on it are too complicated to state here (see
Thrainsson (1976) for data and Zaenen (1980) for a summary). However
complicated, the conditions are perfectly well stateable in terms of the structure
prior to topicalization under the standard assumption that in topicalization, the
topicalized constituent itself is moved (and not base generated in situ as under
Chomsky's proposal). But if we assume the PS rule under (8) as the underlying
structure for topicalization, we can not account for the reflexives in topic
VERB-FIRST CLAUSES IN ICELANDIC 157
position at all, especially not given the fact that they are impossible in Left-
dislocated structures as shown by the following contrast:
(Icelandic like all Scandinavian languages has a special reflexive form for the
possessive ).
This problem cannot be solved by assuming that the moved abstract wh-word
is marked for +- / - reflexive because as the example in (II) shows, the topicalized
constituent itself doesn't have to be a reflexive; it can contain one. Even in these
cases reflexivization is obligatory as the ungrammaticality of the following non-
reflexive version of (11) (a) shows:
2.i. The Kayne & Pollock (J 9 78) analysis of French Stylistic in-
version
Kayne & Pollock (1978) discuss French sentences of the type illustrated in
(12) :
In (12) the verb precedes the subject in the embedded clause contrary to the
normal French word order where it follows. Kayne & Pollock (1978) argue that
this rule of Stylistic Inversion (henceforth SI) should be stated as follows:
Stylistic Inversion:
Verb-pronoun inversion:
The effect of the rule in (13) and the assumptions made is that SI will be
possible in all clauses that intervene between the position of the wh-word in
underlying structure and the position where it ends up in the surface. I will
call these clauses, including the one the wh-word originates in, a binding domain.
It can be defined as follows (adapted from Clements (1979)4 :
(14) A binding domain consists of all the clauses dominating a bindee and
not dominating its binder. (Where the bindee designates the indexed
trace and the binder, the wh-word in its surface position).
binding relation
...-/\" S
(movement: or ~
deletion) XPjS - bin dee
outside of _ _ _ ~
binding ~
domain
Now as the definition just given shows, this domain can be easily defined
without appealing to successive cyclic wh-movement. So the condition on SI
can be restated: SI applies in binding domains.
But under the assumptions made in the introduction to this paper, namely
that both successive cyclic wh-movement and unbounded deletion are allowed,
there is an empirical difference between the formulation in (12) and one that
would use the notion 'binding domain' as defined in (14) : following (12) we
would predict that no SI phenomena will be found in cases where it can be
argued that unbounded deletion has taken place, whereas under an account
160 A. ZAENEN
incorporating (14) we make the prediction that both wh-movement and un-
bounded deletion can give rise to SI-type phenomena (if we assume that both
rules leave indexed traces)5. I will now show that the latter prediction is borne
out in Icelandic.
In Mating & Zaenen (1981), it is argued at length that there exists a word
order constraint in Icelandic that requires the tensed part of the verb to be in
second position in all embedded clauses 6 . This constraint was proposed to
account for the following contrasts:
,
(15)a. Eg held a~ trimin muni taka smalann a morgun.
I think that the trolls will take the shepherd tomorrow.
a
b. *Eg held a~ tri)Jlin smalann muni taka morgun.
I think that the trolls the shepherd will take tomorrow.
a
c. *Eg held a~ muni trollin smalann taka morgun.
I thin,k that will the trolls the shepherd take tomorrow.
,
(16) a. Eg held a1"l smalann muni trimin taka a morgun,
I t)1ink that the shepherd will the trolls take tomorrow.
a
b, *Eg held a~ smalann trollin muni taka morgun.
I think that the shepherd the trolls will take tomorrow.
These examples illustrate that the tensed verb is in second position under the
highest S-node and assume that the structure is (18) (a) for indirect questions,
and (J 8) (b) for topics:
(18)a. s (b)
'"
/~
COMP s
I ./
WH NP .~ ..
v .....
VERB-FIRST CLAUSES IN ICELANDIC 161
Here a intransitive verb has been passivized, hence no new subject is created
but the old one has disappeared. What is interesting for our purposes here is the
conditions under which the dummy ba~ shows up7. It appears only when other-
wise the verb would not have been in second position. (20) illustrates this
pattern :
,
(20)a. Eg held al'J pal'J hafi veril'J dansa~ f gaer.
I think that there has been danced yesterday.
b. Eg held a~ f gaer IJab hafi veri~ dansal'J.
c. ~g held al'J (gaer hafi veril'J dansa<'l.
d. ~!} held al'J (gaer hafl t>a~ veri<'l dansab.
e. *Eg held a~ hafi verii'J dansab (gaer.
f. *Eg held al'J hafi ~al'J veri() dansa~ (gaer.
In (20) (b) and (d) a constituent has been fronted and pa?J has been inserted
at the same time; we see this is ungrammatical whether the insertion is before
or after the verb. In (c) a constituent has been fronted and no dummy inserted
and the sentence is grammatical. (£) is ungrammatical with a dummy inserted
after the verb and (e) is again ungrammatical because no fronting has occurred
and no dummy has been inserted. (a) is grammatical because the dummy has
been inserted and nothing else has been fronted. These examples show that
pa?J does not behave like a subject in the sense that there and it in English
do : pa(J only occurs in preverbal position and only when the verb would other-
wise not be in second position.
A similar situation is found with respect to indefinite NP post-position8 .
This construction is illustrated in (21) :
,
(21)pal'J drekka margir Yin aIslandi.
There drink many (people) wine in Iceland.
latter rule only postposes the subject to the position immediately after the
tensed verb, whereas the rule of indefinite subject postposing can put the subject
after an indefinite number of (adjacent) verbal elements. The contrast is
illustrated in (22) : (a) -- (c) are attempts to do subject-verb inversion (the only
rule that could apply with a definite subject), whereas (a') - (c') are cases of in-
definite subject extraposition.
The pattern of pafJ insertion found in this construction is exactly the same
as that found with impersonal passives, as shown in (23) :
a
(23)a. Eg held ai'> pal'> drekki margir vfu islandi.
I t)1ink that th~re drink many wine in Iceland.
a
b. *Eg held ai'> Islandi pa~ drekki margir vm.
c. Eg held a~ ~ ISlandi drekki margir vi"n.
d. * ~g held ai'> l Islandi drekki ~ai'> ,margir vfu.
e. *~ held ai'> drekki margir vi"n a Islanpi.
f. *Eg held ai'> drekki pai' margir vm ~ Islandi.
Again the dummy shows up when, otherwise, the verb would not be in second
position.
There is, however, one exception to the pattern sketched above that is not
discussed in Maling & Zaenen (1981). Contrary to what our exposition up to
now would lead us to suppose, there are some cases of verb-first embedded
clauses. Some examples are given in (24) to (26) :
It is fairly easy to see what all these cases have in common: first they all
have a verb-first clause that is subjectless because either impersonal passive or
indefinite subject post-position have applied, and second, the sentences involve
the application of an extraction rule, question movement in (24), relativization
in (25), and topkalization in (26)9.
It remains to be seen what exactly the relation is between this verb-first word
order and extraction rules. In the examples given up to now, the extraction site
is in the clause that has the verb-first word order; this is, however, not
necessarily so. We find also examples like the following:
So, the verb first pattern can also be found in clauses that are "between" the
extraction site and the controlling position. It cannot be found, however, in
clauses that are embedded further down than the extraction site. An example
is given in (28) :
~ /
(28) *1 Russlandi sagCli hann aCl drekki margir vodka Ii Islandi.
In Russia said he that drink many vodka in Iceland.
The presence of the locative phrase d /standi insures that the locative i Russlandi
cannot be construed as belonging to the lower clause. The contrast in
grammaticality between this sentence and the previous bnes, shows that what is
at issue is the binding domain, as defined in section 2.2. In (28) the lower clause
is outside of the binding domain and the verb-first word order is ungrammatical.
To show that the verb-first pattern is limited to subjectless sentences, I will
now give a few ungrammatical examples where definite subjects (i.e. the only
164 A. ZAENEN
ones that are clearly subjects) are in post-verbal position in the same environ-
ment.
(29)a. *Hvac) heldur ~u al'> telur J6n a~ drekki margt f61k i Russlandi ?
What think you that believes John that drink many people in
Russia?
b. *Vodka er drykkur, sem eg veit a~ telur J6n a(') 61afur drekki.
Vodka is a drink that I know that believes John that Olaf drinks.
c. *Vodka veit eg al'> telur J6n al'J drekki margt f61k i Russlandi.
Vodka know I that believes John that drink many people in
Russia.
As can be seen from these examples, verb first order is impossible in clauses
that have subjects.
Finally the following examples show that no pao-insertion can occur under
the conditions under which the verb first word order is possible:
(30)a. *Hann spurl'li hvar pa<"> drekki margt f61k vrn. (cf. (25)a)
H~ asked where there drink many people wine.
b. * A slsipinu held eg a~ ~a~ var dansac). (cf. (26)b)
On the ship think I that there was densed.
c. *Vodka er drykkur, sem l>a~ er talit'J ab 61afur drekki (cf. 27(b».
Vodka is a drink that there is believed that Olaf drinks.
This does not mean that the verb necessarily has to be first. There is a minor
fronting rule in Icelandic that allows for the pre posing of some elements like
participles in the clause type that I am discussing here. I will not discuss its
effects, for more information see Mating (1980).
It is clear by now that the domain in which dummy insertion is not allowed
in subjectless clauses, even if nothing precedes the verb, is exactly the binding
domain as defined in section 2.2., hence the same domain in which SI applies
in French. In section I., however, I argued that neither relatives nor topicalized
sentences are an instance of successive cyclic wh-movement in Icelandic but
that the former is a case of unbounded deletion and the latter a case of another
type of unbounded movement rule. Under these assumptions the analysis given
by Kayne & Pollock (1978) cannot be extended to these new cases without
further ado, whereas the account in terms of binding domains as proposed in
Clements (1979) can.
VERB--FIRST CLAUSES IN ICELANDIC 165
The previous section showed that contrary to the prediction made by the
account given in Kayne & Pollock (1978) together with the arguments that both
movement and unbounded deletion have to be admitted as possible rules in the
grammars of natural languages, SI-type phenomena are not limited to construc-
tions in which it can be argued that successive cyclic wh-movement has applied.
It shows that the account for SI-type phenomena has to be generalized so as
to allow their occurrence also in contexts in which unbounded deletion or other
"extraction" rules have applied. We have now reached a rather paradoxical
situation because both types of arguments for successive cyclic movement,
namely "island constraints" and "SI-type-phenomena" are now shown to be
independent from successive cyclic wh-movement itself and also to apply in
contexts where for independent reasons a successive cyclic wh-movement
analysis is undesirable. Hence it seems that the attempt to capture the
generalizations involved by way of the successive cyclic wh-movement
hypothesis was wrong and that we need a mechanism that would generalize
over both movement and deletion cases. As said before, such a mechanism was
proposed in Bresnan & Grimshaw (1978). Their account, however, seems to
invoke a redundant number of mechanisms: not only have we a movement and
a deletion rule, in top of those two processes we have also an index mechanism
that basically retraces that path of the transformational rules. An easy way to
avoid this type of redundancy is to assume that, in fact, no transformations
have applied in the derivation of either questions or relatives or topic
constructions, but that all of them are base-generated and subject to a kind of
binding that I will call syntactic binding. This type of binding can capture island
constraints and SI phenomena in a way similar to successive cyclic wh-movement
but without suffering from the inadequacies presented by that account in the
cases of so-called unbounded deletion. In this section I will sketch one such
account based on the mechanisms available in Kaplan & Bresnan (I 982)'s
Lexical Functional Grammar.
(31) S~NP VP
(tSUBJ) =~ t=~
The equations here tell us that the NP to the right of the arrow is the subject
of its mother node, i.e., of the S; and that all the information about the VP is
also information about the S in functional structure (f-structure). This latter
equation is actually redundant under the assumption that both Sand VP are
projections of V because LFG states explicitly what is often assumed in X bar
theory, namely that the features and the functions of the head of a category
are transmitted automatically to the dominating expansions of that category
and vice versa. This convention that I will call the head convention (adapting
the terminology used in Gazdar, Pullum, & Sag (I 981), who propose a similar
principle in Generalized Phrase Structure Grammar (GPSG)), will play an
important role in the statement of the proposal that I will make in the next
subsection. The equation added to the NP illustrates the second way in which
information can be transmitted locally in LPG, namely by stating that a
particular node has a particular function with respect to its mother node.
In LPG, syntactic binding or constituent control is effected through the
instantiation of another set of equations, namely linking equations, which are
represented as tt and H. An example is given in (32)10 :
(32) S ~ XP S
~=H t=~
XP~ e
t=H
This PSR together with the other PSR's required for the expansion of Scan
create a tree configuration like the one illustrated in (33) :
VERB-FIRST CLAUSES IN ICELANDIC 167
(33) S
/~ S
XP x'"
.1 ~
~=H XP
I X45
e
t = tt
All the nodes in (33) are equipped with variables and what the instantiation
procedure of Hand tt's does is to insure that x3 = x45 in the example given.
As discussed in Kaplan & Bresnan (1981) and in Zaenen (1980), the domain
in which this instantiation can occur has to be constrained. A first constraint is
that the binder has to c-command the bindee. (I would like to state a stronger
constraint: only sentential categories can be the right sister of a binder. There
seem, however, to be exceptions to this stronger version: tough-constructions
in English and topicalization in Makua as described in Stucky (1980). A weaker,
more adequate, constraint might be that only projections of V can be the right
sisters of binders.)
Further, I will assume that all S-nodes are bounding nodes (i.e., "boxed"
in the Kaplan & Bresnan notation) except if stipulated otherwise. To illustrate
this with a diagram, in the configuration under (34), only the highest S (i.e.,
what is above the dotted line) constitutes the binding domain.
(34)
/s~ \
XP S \
~=~~ ~,
s~
that S
' . . . _// A.
More deeply embedded S's can be made accessible to binders by PSR's that
introduce tt = H equations. For instance in English, we have the following
rule:
(35) S -+ that S
tt=H t=~
168 A. ZAENEN
(36) S~ XP [+wh] S
~=H t=~
(37)
Another notion that I will appeal to is that of root node. For a definition I
refer to Kaplan & Bresnan (1982). Here it is sufficient to point out that the
notion of root node as defined there treats the circled S in both cases in (38)
as a root node.
(38)
(a) /~ (b) /~
X S y fs\
~=H tt=H \V
~ ~
Intuitively: a root S is an S that is the right sister of a node with a H. (Other
cases are aJlowed : see Kaplan & Bresnan (1982). For the purpose of this paper
the generalization just stated is sufficient.)
VERB-FIRST CLAUSES IN ICELANDIC 169
(39)
A first way to do this that comes to mind is by way of a filter; e.g., the following
would do :
(40) *
It is, however, not clear that it is desirable to allow filters like (40) in LFG.
Note that (40) is not strictly a "surface" filter: it refers to the functional in-
formation attached to the first node (because in general of course, the
configuration in whichpa~ follows a S - initial element is allowed.) In that sense,
it is not a filter on strings or even on strings in a certain tree configuration, but
one that combines information of two levels, the constituent structure level and
the functional level. Instead of adding this new type of mechanism, and hence
the assumption that this mixed information is needed in filters, it would be more
interesting to see if the facts can be described with mechanisms already available
170 A. ZAENEN
structure depends on other rules. The crucial fact however is that in a binding
domain there is no evidence for the downstep at all : it is deleted. A contrast
is given in (42) :
The change in the tone pattern associated in (b) with ate and mote are the
consequence of this downstep deletion rule (A slightly more detailed description
is given in Zaenen (1981); for full details see Clements (1979)).
Again, Clements (1979) argues that unbounded dependencies in Kikuyu
should not be captured by successive cyclic wh-movement, but are an instance
of unbounded deletion. So again the formalization proposed in Kayne & Pollock
(1978) could not be extended to this case.
What strikes one when looking over the array of facts of these different
languages, is that only two elements in the sentence seem to be clearly affected
by their occurrence in a syntactic binding domain: the verb (either "morpho-
logically" as in Kikuyu or optionally in some Irish dialects, or "positionally"
as in French and in Icelandic 14), or the complementizer as in all versions of
Irish. We never see, for example, the morphology of the subject noun, or the
direct object or an adjective or an article being changed. Data from four
languages are of course insufficient to claim that this is an unshakeable universal,
but then data of 400 languages would be similarly insufficient. What I will do
in the next subsection is assume that the generalization just stated is correct,
and investigate what type of representation of the facts it suggests, and hence
what insight it gives us into the interdependencies that can be found in natural
languages.
of dependency that we find, for example, in English between that and a + FIN
(finite) verb form, or in French between avant que and the sUbjunctive.
Interestingly enough, we find in most of the languages discussed here, and
in other ones, evidence that the mode system of the language and the possibility
for binding within certain domains are related. As Kayne & Pollock (1978)
point out, SI is in fact not limited to binding domains but can also occur in
subjunctive clauses. Hence, we should actually allow for two cases: verbs that
are "bound" and subjunctive verbs. In Kikuyu, downstep "deletion" is not
limited to binding domains but can also be found in a few other, e.g., the
negative, tenses regardless of syntactic context.
Moreover, there are languages where the mode of the verb determines the
binding possibilities. It has long been assumed that in Russian, binding into
embedded clauses was impossible. Pesetsky (personal communication), however,
has found that this only holds for indicative clauses, and that extraction out of
subjunctives is possible. Hence the following contrast:
(45)a. Indien hij hier was/ware, zou dat probleem al opgelost zijn
If he here was/were, would that problem already solved be.
b. Was hij hier, het probleem zou al opgelost zijn.
Was he here, the problem would already be solved.
It seems thus reasonable to assume that the BND feature belongs with the
mode/tense indicators in natural languages and also that word order can be used
as an indicator in that system as well as morphology (or phonology for that
174 A.ZAENEN
The only language specific problem that Icelandic presents is that the PSR
that generate impersonal passives and indefinite subject postposition sentences
have to combine on the one hand with ]Jab and on the other hand they have to
be able to occur in binding domains, without being able to occur in both these
environments at the same time. Within the LFG framework as I have developed
it here and in Zaenen (1980) there is only one way to do this, namely, by
assuming that pa~ is introduced by a PSR that at the same moment stipulates
that the S that is the right sister ofpalJ is a +BND sentence. This PSR rule is
given in (46).
(46) s ~ pab s
(tBND)c = +
This rule does not associate a linking equation with paiJ, insuring that this S
cannot be the root of a binding domain. It nevertheless characterizes it as a
+ BND sentence to allow for the following rule introducing impersonal passives
and indefinite subject sentences. (I represent what follows the verb here by an
X in order to leave open the question of what the exact c-structure is that should
be given to impersonal passives and indefinite subject sentences in Icelandic.)
(47) s~ V X
(tBND) = +
This is intuitively not totally unsatisfactory; what the dummy actually does
is to allow for the sentence type in which nothing is topicalized, not even the
subject that in general acts as a discourse topic by default; hence the dummy
acts as a contentless topic 15 . Unfortunately, we cannot equipfta~ with a binding
equation because it doesn't really bind anything, so our formalism does not
treat pa'J as a topic. The sentence type that combines with it is, however, the
sentence type that normally combines with topics and this is captured by cha-
racterizing that type in both cases as being + BND.
VERB-FIRST CLAUSES IN ICELANDIC 175
(48)
In the system developed in section 4.3., the equivalent of a clause with a wh-
element in COMP is a root-sentence and, as we have seen, root-sentences are
universally constrained to be +BND; hence to achieve the effect of (13) as
diagrammed in (48), the only thing we have to do is to insure that SI is intro-
duced by a PSR that contains a verb that is + BND, as for example, the following
will do:
(49) S~ (NP) VP
(tSUBJ)=.j, t=.j,
VP~ V x NP y
(BND)= + (tSUBJ)=.j,
t=.j,
way that we get sentences with double subjects. French doesn't have such
sentences nor does any other language to my knowledge. In LFG they are ruled
out by the principle of functional consistence that requires every grammatical
feature of each grammatical unit to have a unique value. If we choose an
expansion of the PSR in (49) that would give us two values for SUBJ, the re-
sulting f-structure would be filtered out by the consistence requirement. (For
further discussion see Grimshaw (1981) and Kaplan & Bresnan (I 981 ».
The variables in the expansion of VP are an expression of ignorance: the
exact surface structure of sentences with SI has not been worked out. (See
Kayne (1972) for the most detailed description)16.
Other PSR's introduce V's that are not specified for the function BND,
which can occur both in bound and in non-bound sentences. This immediately
explains the optionality of inversion.
Kikuyu and Irish can be handled in an equally straightforward way. The
interested reader is referred to Zaenen (1981) for the relevant rules.
5. Conclusion
In this paper I have assumed that the evidence gathered in several languages
warrants the conclusion that not all extraction phenomena can be handled by
successive cyclic wh-movement. Given that assumption, the formalism proposed
in Kayne & Pollock (1978) to handle French Stylistic Inversion predicts that
phenomena like French Stylistic Inversion will only be found in cases where
there is no evidence against a wh-movement analysis as compared to a deletion
analysis or an analysis involving another type of movement rule. I have first
argued that this is not the case. This negative conclusion then warrants a treat-
ment of binding domain phenomena that abstracts away from the difference
between movement and deletion. Such an approach can be found in the non-
transformational analysis of unbounded dependencies proposed in Kaplan and
Bresnan (1981). I have taken this account as a basis and worked it out to
account not only for French SI, but also for Icelandic V/1 clauses. The
mechanism proposed also applies trivially to Irish and Kikuyu "binding domain"
phenomena. Hence it has the advantage of being universally applicable, whereas
the proposal in Kayne and Pollock (1978) is limited to French and only
generalizable to cases of successive cyclic wh-movement.
VERB-FIRST CLAUSES IN ICELANDIC 177
NOTES
*The following people are thanked for help and comments: J. Bresnan, N.
Clements, E. Engdahl, J. Grimshaw, J. Higginbotham, R. Kaplan, J. Maling, D.
Pesetsky, J. Schindler, and S. Weisler.
IThroughout this paper I will use transformational metaphors when describing
syntactic phenomena. The scientific terminology used in the theory I will adopt
in section 3 is explained there.
2Examples that show that the rule is indeed unbounded will be given in
section 3.
3This is not to say of course, that topicalization is not a movement rule in
Icelandic (under a transformational analysis), just that the movement rule does
not fit into the theory proposed in Chomsky (1977).
One might consider this a minor problem and propose that any element (but no
that-element) in COMP could have the effect that Kayne & Pollock ascribed to
a WH-element. But notice that even under that hypothesis, we do not
immediately get the right results in Icelandic: wh-elements in COMP do not
trigger inversion, whereas topics would. If we assume that both are in COMP,
we have to complicate the word order rules to make them sensitive to the
feature +/- who This treatment no longer allows for a straightforward statement
of the word order constraints discussed in Maling & Zaenen (1981). Notice also
that the distinction between -+ wh or -wh will only be relevant on the cycle
where the moved element ends up, and not on the intermediate cycles, as would
be predicted by the successive cyclic movement analysis.
4In Clements' own terminology, an "open sentence" is defined as "a clause
dominating an anaphoric element bound by an obligatory rule of grammar to
an antecedent which the clause does not dominate." Notice that the notion of
open sentence allows for the statement of the French SI-facts in an EST frame
work without successive cyclic wh-movement and without running into the
problems Kayne and Pollock's straw men run into.
SOne theoretical drawback of this formulation is, however, that it does not
allow us to state the condition on SI without reference to essential variables.
One might consider the most important achievement of subjacency theory to
be that it allows the restatement of rules with essential variables as rules not
involving such variables. The account that I will give in this paper is an account
with a program that attempts to exclude the use of such unbounded variables
and hence, limits the grammar to local rules.
6Some version of this generalization has been noted long ago. E.g. Wackernagel
(1892) discusses it. Wackernagel seems to want to reduce to the same "law"
the position of clitics in several ancient Indo-european languages and the
position of the verb in modem German main clauses. There are problems with
this reduction: first, there is no evidence for Wackernagel's conjecture that the
verb in older stages of Indo-european, being unstressed, would be put in second
position; and secondly, the second position relevant for clitics is the position
after the first word in the sentence (or after the first word and other clitics),
whereas the German verb is in second position after a major constituent. This
178 A. ZAENEN
Ludo Melis*
Katholieke Universiteit Leuven
III'yenvoie.
181
182 L. MELIS
-1 +
2 + +
3 + +
4 +
S + + +
6 + +
7 + +
8 + +
9 + +
10 + + ?
II + + +/-
12 + +
13 + + +
14 +
IS +
16 ')
+
17 +
Table 1
The table presents overwhelming evidence for the claim that the V Vinf
NP 2 construction is radically different from the other two constructions; it also
permits the conclusion that the two first constructions are much alike, differring
chiefly in their behavior with respect to the first two properties 4 . These proper-
ties are linked : the figurative use of the verb commands the modification of
the selectional restrictions on NP I . So we can conclude that there is only one
basic construction, NP I V NP 2 (d) Vinf, and that the two observable construc-
tions are different realisations of that basic construction; the opposition
concerning (1) and (2) needs to be explained in semantic terms and the comple-
mentary distribution between the bare infinitive and the prepositional one is,
perhaps, commanded by the same facts S .
The results for (6), (7), (8), (9), (15), (16) and (I7) indicate that the
184 L. MELIS
infinitive functions as a constituent distinct from both the main verb and NP 2
(see also (3)); properties (10) and (11) permit an interpretation of the infinitive
as a locative complement, a PP with the preposition d or ~; (3) leads to an
analysis of NP 2 as the direct object of the sentence. So we propose the following
analysis for
/s~
~P /IP~
V NP PP
I /~
I P VP
:
I I
V/ ~NP
I I
I
I I I I I I
Jean envoyer Marie ~ chercher du pain
Cela mener Pierre a fuir Marie
That binodal structure doesn't appear in the syntactic structure; it is only ne-
cessary to explain the selection of NP2 and Vinf as a whole and the interpre-
tation of NP 2 as the 'subject' of Vinfo
Such a situation does also exist with the perception verbs (voir, entendre,
sentir, iaisser, ... ) in the construction V NP Vinf (Morin 1978). The other verbs
we are alluding to -- dissuader, forcer, ... -, the so-called EQUI-verbs, possess
also an underlying binodal structure but the two elements, the object NP and the
infinitive, are selected separately and, in current analyses,the infinitive is seen as
the remnant of a sentence-like structure and it can, in certain circumstances,
be replaced by a subordinate clause:
For the V Vinf NP2 construction, properties (4), (15), (16) and (17) indicate
that an analysis considering the movement verb and the infinitive as a single
constituent of category V composed of two lexical V's, is correct; such an
analysis, and in particular the assignment of the category label V instead of VP
to Vinf' is also supported by (5), (6), (7), (8) and (9). The results for (10) and
(11) indicate also that the infinitive is not a complement of the main verb and
that it has no function but being part of the verbal predicate. On the other hand
(3) and (4) provide an analysis of NP2 as a direct object of that complex verb.
So we propose to analyse
as NP [V V V] NP
~S~
NP /P~
/V~ NP
V V
I I
I I I I
I I
Jean envoyer che~cher Marie
du pain
Perception and causation verbs (jaire, iaisser, voir, entendre) can occur in a
construction with absolutely the same properties on all levels 11 . It is reasonable
to consider both movement verbs and perception and causation verbs who
enter the IPVC with complex verb as belonging to the same class, but only with
respect to this construction.
FOOTNOTES
for the opposition between literal use and figurative use of a lexical item. As
noted in the appendix, properties (6) and (8) are more frequently observed
with the prepositional construction, but those differences are minor. A
weakening of the contraints also is typical for figurative use.
5The alternation can be described by a deletion rule, though the constraints
governing such a rule are difficult to describe in syntactic terms. Notice that an
underlying d and a deletion rule are posited on other grounds by Emonds (1978)
and by Seelbach( 1978) for the intransitive movement verbs.
6An analysis of the first two groups is given by Blanche-Benveniste and Van Den
Eynde (1977); an analysis ofIPVC as 'verb -j- object + predicative complement'
is implied by Damourette and Pichon (par. 1060 and passim), Blanche-Benvenis-
te and Van Den Eynde (1977) and Morin (1978); the commutation with other
predicative complements, present participle or relative clause, points to the
same analysis (Rothemberg 1979).
7 Pending on the theoretical framework adopted, that binodal structure can be
interpreted either as a syntactic element or structure linked to the 'surface'
structure by a transformation, probably 'raising', or as a component of the
lexical or of the interpretative structure, linked to the syntactic structure by
rules of interpretation. We avoid discussion of this problem; for us, it is
sufficient to indicate that such a structure exists and to describe for which pur-
poses it must be established: selection of Vinf and NP2, interpretation of the
links between these constituents.
8This position is taken by Damourette and Pichon (par. 1060), stating that the
two classes are alike in that the verbs are constructed with an "about dicephale
binodal" but that they differ on other points: movement verbs are followed by
a "progredience" and perception verbs by a "conspicience". Gross (1975: 168)
says that laisser belongs to the same class as the causative movement verbs;
this is certainly correct for the locative interpretation ('laisser' = 'faire ne pas
mouvoir') but it is unclear whether that analysis can be extended to cover the
other uses of the verb.
9In the case of the intransitive movement verbs (aller, entrer, monter, ... ) the
infinitive is linked to NP I; Lamiroy (1981) says that the infinitive also has
locative properties, but she doesn't accept an underlying bi-nodal structure; this
question seems however pending.
10 As indicated in the appendix such a two-step selection mechanism is necessary
to give a uniform treatment of idiomatic and non-idiomatic uses. If the idio-
matic expressions are to be considered as lexical items of a particular kind, they
must be inserted as such; in that case the bi-nodal structure cannot be posited
at a syntactical level, but it must belong to the interpretative component.
II See Radford (1975), Kayne (1975), Morin (1978), Blanche-Benveniste and
Van Den Eynde (1977), Emonds (1978) for analyses of that construction;
they all consider that there is a complex verb in the (superficial) syntactic struc-
ture. In our discussion, differences between these authors on other points may
be disregarded.
THE INFINITIVE WITH CAUSATIVE MOVEMENT VERBS 189
APPENDIX
0), (2) :
Jean envoie Marie chercher du pain, et, pour cela, elle doit traverser la ville
Jean envoie chercher du pain et, pour cela, Marie doit traverser toute la
ville
* Cela envoie Marie chercher du pain
* Cela envoie chercher (Marie/ du pain)
a
L'attitude de Pierre conduit Marie rejeter cette solution;
In this case, an animate subject is possible, but with 'a non agentive interpre-
tation :
Pierre conduit Marie a rejeter cette solution par son attitude negative.
(3), (4) :
Jean envoie Marie chercher du pain
Ill'envoie en chercher
* Ill'en envoie chercher
* II envoie I'en chercher
The set of restrictions is based on pragmatic facts: the infinitive reports the
action to be performed at the end of the movement; such an action cannot be
accomplished or be in course of accomplishment (5,6,7); it cannot be in the
focus of a modality operator, or of the negation (7,8).
(9) :
Jean envoie Marie chercher du pain
* Jean envoie chercher Marie du pain
Cela mene Jean a fuir Marie
(10) :
Habitually, mener requires the presence of a PP :
Pierre mene Marie au bal
* Pierre mene Marie;
sentences with an infinitive instead of a PP are also acceptable:
Pierre mene Marie danser
Pierre mene danser Marie
a
Cela mene Pierre fuir Marie.
Other infinitival phrases, such as final pour + infinitive, cannot replace the PP :
* Pierre mene Marie pour danser
* Cela mene Pierre pour fuir Marie.
So the infinitive plays the same role as the PP. In the case of the V Vinf NP2
construction, things however, are not so clear; sentences without PP are
acceptable, but a locative PP can be adjoined and it must be attached to the
main verb and to the infinitive:·
THE INFINITIVE WITH CAUSATIVE MOVEMENT VERBS 191
(11) :
In the V NP2 Vinf and V NP2 d Vinf constructions, the infinitive can be coor-
dinated to a locative PP (SandfeId 1965 : par. 102).
"elIes n 'avaient pas manque de demander a ce qu 'on les menat dans les
petits theatres et sou per au cabaret" (Sandfeld 1965 : par. 102).
Cela l'a mene a abandonner Paris et donc a la solitude morale la plus
complete;
it corresponds also to the locative clitic pronoun y (Sandfeld 1965 : par. 102):
Quand me conduiras-tu ramasser des fraises des bois? Je t'y conduirai ce
midi meme
Est-ce que cela pourrait Ie mener a abandonner les negociations? Cela
l'y menera certainement.
it is a good response to the locative question ou ? (Sandfeld 1965: par. 102;
Gross, 1975: 168).
"Ou Paul envoie-t-il Pierre? -- Voir Marie" (Gross 1975 : 168).
OU cela Ie menera-t-il ? A rejeter toute certitude;
it forms finally with a locative PP a complex but unique constituent:
On l'a envoye a Rome voir Ie Pape
On l'a envoye a Rome en ambassade.
These properties don't exist in the case of the V Vinf NP2 construction:
* elles n'avaient pas manque de demander a ce qu'on menat dans les petits
theatres et puis souper les invites.
* Quand enverras-tu chercher Marie? J'y enverrai ce midi meme.
* Ou enverras-tu Marie? - J'enverrai chercher Marie.
* On a envoye a Rome voir Ie Pape.
(12) :
In all three constructions, the infinitive reports an action, a 'faire' according to
Lamiroy (1981 : 49, 116ss). This depends not only on the infinitive but also on
the links between the infinitive and its subject, as can be seen in the following
cases:
Pierre envoie Gerard amuser Marie
192 L. MELIS
(13) :
Jean envoie Marie chercher du pain et elle cherche ce pain
Cela mene Pierre a rejeter la solution et i1 la rejettera effectivement dans
une lettre fameuse
Pierre envoie chercher Marie et elle cherchera pendant plusieurs heures.
(14) :
* Jean envoie Marie chercher et on la cherchera pendant des heures
* a
Cela mene Pierre chercher et on Ie cherchera pendant des heures
Pierre envoie chercher du pain et on en cherchera longtemps.
(15) :
The arguments given by Radford (1975) for the analysis of je ferai pendre Jean
as a construction with a complex verb do all apply in this case.
(16) :
There exist two series of idiomatic expressions involving the construction V Vinf
NP2; in the first series we can list expressions as
envoyer paitre, promener, bouler, se faire pendre ... ;
in the second
ne pas Ie lui en voyer dire.
For the first series, the dictionaries propose also the fomula envoyer quelqu 'un
promener (with the construction V NP2 Vinf) but they don't give any example
of this construction. Informants interpret habitually
II a envoye Pierre promener
as
'il a envoye Pierre en promenade'
and
II a envoye promener Pierre
as
'il s 'est de barrass6 de Pierre'
(17) :
As seen under (16) envoyer promener and similar expressions are equivalent to
'se d6barrasser de'; Damourette and Pichon (par. 1112) consider envoyer
chercher as a lexical unit.
GENERALIZED UNION
Gilles Fauconnier
Ecole des Hautes Etudes en Sciences Sociales - Paris
Universite de Paris VIII
I. Prolegomena
Since Kayne (1975)'s study, the following facts about taire are well known:
(1) J'ai fait envoyer Ie paquet a Paul. (I had the parcel sent to Paul)
(2) Je lui ai fait envoyer Ie paquet. (I had the parcel sent to him)
Paul in (I) may be either the sender or the receiver ofthe parcel, i.e. "logical-
Iy", the subject, 1, or the indirect object, 3, of the clause which is itself the
object of taire. Similarly, the clitic lui in (2), just like its counterpart Paul in
(1) may be interpreted semantically as the subject or the indirect object of
envoyer (sender or receiver).
195
196 G. F AUCONNIER
On the other hand in (3), where we find both the cIitic lui and the full form
a Paul, the possibility of having two interpretations vanishes : lui is necessarily
the sender of the parcel, Paul is necessarily the receiver. The second interpreta-
tion (lui receives, Paul sends), which might be expected given (I) and (2), is
excluded.
There have been attempts to relate these observations to the interaction of
clitic movement with the derived structure of the faire faire construction. But
there exists a distribution quite parallel to (1), (2), (3), which involves neither
faire nor the "movement" of the cIitic into a higher clause :
(6) Luc lui semble fidele d Jeanne. (Luc seems to her to be faithful to
Jeanne)
Like (1), example.(4) allows two interpretations: "it seems to Jeanne that Luc
is faithful" or "it seems that Luc is faithful to Jeanne". Similarly in (5), the
corresponding clitic, lUi, may be interpreted as a complement of sembler, or as
a complement of fidele : "it seems to her that Luc is faithful" or "it seems that
Luc is faithful to her".
But (6), just like (3), allows only one interpretation: "it seems to her that
Luc is faithful to Jeanne" and not "it seems to Jeanne that Luc is faithful to
her", i.e . lui complement of sembler and d Jeanne complement of fidele.
Floating quantifiers also reveal a parallelism between the two constructions.
As we know, a quantifier may quantify a superficially distant noun phrase:
In (8) and (9) we find the same ambiguities as in (I) and (4) :
(9) Ce sont des gens auxquels cette fille semble d tous tres attachee.
(8) may be understood with the people (Ies gens) either sending or receiving
a parcel. In (9), either it seems to them that the girl is very attached (to some-
GENERALIZED UNION 197
thing or someone), or it seems that the girl is very attached to them. As in (3)
and (6), the presence of a clitic lui excludes one interpretation:
Ol)Ce sont des gens auxquels cette fiUe lui semble tres attachee.
In (10), the people (les gens) can only be the receiver, while in (11) auxquels
must be the complement of attachee, and lui the complement of seem
(semble).
Now, in these configurations, floating quantifiers, although possible with
both interpretations in (8) and (9), become entirely excluded:
(12) *Voila les gens aqui je lui ferai d chacun envoyer un paquet.
(3) *Ce sont des gens auxquels cette fille lui semble d tous tn!s attachee.
(14) clause a
1st stratum
r ,A
clause b
,
Equi
, ir
....... ~
je faire a Ie lapin [ ~ avaler la sa lade 1
1 P 2
(15)
2nd stratum clause a
~
rje faire avaler fa safade
~
Ii Ie lapin
PI P2 2 3
(16)
1st stratum clause a
~
---
~
/ claus~ b
r[ -..
je faire fondre 1a glace]
predicate object
P 2
object
../
P 2
(17)
-
2nd stratum clause a
je faire
./""-
fondre
--....
la glace
PI P2 2
After agreement, this yields the sentence Je [ais [andre la glace (I make the ice
melt.
Assuming the perspective just outlined, we must answer two general types
of questions regarding Union:
(2) When two clauses a and b unite to form a single one, a, what
relations do the elements of the "initial" clauses a and b bear
with respect to the resulting clause a ?
c) 3 to 2 advancement in french
(22)
vole
on
For some verbs like fournir 6 , this advancement is possible in the presence of a
direct object, 2, which then becomes a "chomeur" :
(23)a. On fournit des armes aux soldats. (The soldiers are provided with
I 2 3 weapons)
What is of interest in the present context is that among the verbs which
advance their indirect object in the absence of a direct object we find the
faire+ V combinations produced by Clause Union :8
(30)Alice fait voir Ie tableau d Georges. (Alice shows (lit. makes see)
1 2 3 the painting to Georges)
So the superficial direct object in sentences like (32) is actually the indirect
object of faire, after 3 - 2 advancement in the absence of another object, as in
19, 20, 21. This is confirmed by the possibility for many speakers in some sen-
tences of not applying advancement, leaving the object with relation 3, as in
(27); Kayne and Postal cite the following examples, whose form is indeed widely
attested in everyday spoken French:
(33)Elle lui fera telephoner a ses parents 9. (She will make him phone his
parents.
202 G. FAUCONNIER
(34)Cela lui fera penser a sa mere 10. (That will make him think of his
mother).
Other arguments based on unaccusatives and double datives will be given below.
-----
(35) clause a
r- ----....
clause b
-
2 3
........... /
2 oblique
(36) clause a
~-----------~---------~-
je les lui fais voir
2 3
(35') a
les
(36')
les
The following formulation of the "final I" (or final subject) law 11 will
be defended :
(39) * pleut
In (40), p/euvoir has no final subject, but the corresponding clause is not final.
The corresponding networks are :
(38')
a
~
pleut il
(40')
Dieu
As we shall see, under the present analysis, there is no rule of Passive; rather,
constructions traditionaIly called Passive arise through a regular combination of
Union and Unaccusative advancement. In harmony with this overall analysis,
by-phrases will no longer be Passive chomeurs, but rather, more simply, Obli-
ques on the first stratum. Like" other obliques in some constructions, they may
GENERALIZED UNION 205
( 41)
We
telephone
Motivation for such a view will become clear as the overall analysis unfolds.
Notice at this point that Agent ~ 1 will often have to apply in order for the net-
work to meet the final 1 Law (sec. e) :
(43)*
II. Union
(46) Union
1. If two clauses a and b unite (b subordinate to a)1 then
- the elements of a keep their relation, except for b, which
"disappears" .
- any element e bearing the relation {3 in b, now bears the same
relation {3 in a, except if e is a term and some element in a already
bears relation (3; in this case, element e becomes a chomeur
in a (p).
2. Union is blocked if b has a full subject (1) on its last stratum.
(47)Order in French:
Examples:
la
couleuvre
(49) Le petit prince fait avaler la couleuvre au chameau. (The little prince
makes the camel swallow the snake)
In accord with sec. I.d final positions 2 and 3 in (48) can be cliticized, to give
sentences like:
par Ie
chameau
Network (51) is well formed since b has no subject; fa couleuvre and par Ie
chameau are respectively 2 and Agent in a by simple inheritance, since a has
no 2 after the eradication of b 12. The linear output, according to (47) is (52) :
208 G. FAUCONNIER
fait
Now consider the case where the downstairs clause b has an indirect object
3. Then Union will cause a stratal uniqueness conflict if faire already has a 3
(netwcrk like (48)) and this conflict, by 46.1 will lead to the chomage of the
downstairs 3; there will be no conflict and therefore no chomage if faire has
two arguments (network like (52)). Networks corresponding to these two situa-
tions are (54) and (60) :
three arguments:
(54)
- - - - - - : . u destinataire
fait
GENERALIZED UNION 209
Similarly (3) Je lui ai fait en voyer Ie paquet a Paul, has only the reading in
which Paul is the receiver, and (12) Voila les gens aqui je luiferai achacun en-
voyer un paquet, shows the impossibility of a floating quantifier associated with
a
a chOmeur (3), qui.
Of course, other relational positions, obliques in particular, may cIiticize
without conflict :
Finally, the conflict in (54) is the same if b has no direct object, 2, but
it is "superficially" masked by 3 to 2 advancement:
(58)*
lui
210 G. FAUCONNIER
(60)
Ie petit prince
par
Ie facteur
correspond to networks exactly like (60) except that b happens not to have
an element bearing the oblique relation Agent.
GENERALIZED UNION 211
A different case from the first two is that in which clause b is unaccusative.
Here again, Union with two-argument [aire will be possible, since b, having no
subject on its first stratum will directly meet condition 46.2. Verbs like taire
(to be quiet) are interesting in this connection because for them, Unaccusative
advancement (2 ~ 1) is superficially manifested by the presence of a reflexive
pronoun. In general, Unaccusative advancement may be reflexive or not,
depending on the verbs involved :
(62) (63)
la neige
tombe
(64)
les
enfants
The same analysis is valid of course for other unaccusatives that require a
reflexive in simple clauses but not in causative constructions:
Note how the final 1 Law operates in such cases: clause b has no final 1, but it
is not a final clause since Union absorbs it. So the absence of se in (65) corre-
lates with the absence of il (or Ie) in "Dieu fait pleuvoir" and perhaps in "j'ai
fait ressortir que X" (compare : "iJ ressort que X").
It may be useful to emphasize that nothing blocks Unaccusative advance-
ment in networks like (65), but then b acquires aI, and Union is possible only
if this 1 is empty, i.e. controlled by a 3 complement of faire :
GENERALIZED UNION 213
P7~~~----- SE taire
des idees
viennent lui
lui
214 G. F AUCONNIER
(72)S:a lui fait venir des idees. (It makes ideas come to him)
(72) is good: the clitic lui, corresponding to an indirect object of venir is a real
term, 3, on the last stratum of a. The important contrast, then, is between (72)
and (59), which is ungrammatical, although superficially similar:
This contrast reflects the difference between networks (71) and (58); in the
latter as opposed to the former faire has a third argument whose 3 relation
induces the chOmage of the 3 element in b, by 46.1 (inheritance and stratal
uniqueness conflict).
This paper started out in sec. I by pointing out the analogy between sembler
and faire in certain constructions. These constructions are superficially
characterized by the presence of complement clitics, not in front of the predi-
cate of which they are "logically" arguments, but in front of the predicate
sembler or faire :
(73) Luc y fait entrer Pauline.
t t
(74)Luc y semble attache.
f· t
I will now proceed, in the spirit of Aissen and Perlmutter (1976), to consider
in general that the configuration
cl P P'
t t
GENERALIZED UNION 215
(77)
"'1..------ - ___
Y
~aller
attache
216 G. FAUCONNIER
(6) Luc lui semble fidele d Jeanne. (only one reading: "faithful to
Jeanne")
(5) may correspond either to network (79) or (79'), but (6) can only correspond
to (80) - network (80') is not well formed.
(79)
lui
(79')
we
(80)
(80') *
lui
The situation is therefore in all respects identical to the one for laire in
networks like (54), and as suggested already by the superficial parallelism, the
properties of sentences (6) and (3) are the same effect of the same cause:
Union 46.1 and the Stratal Uniqueness conflict which puts "extra" elements en
chomage.
The unacteptability of (13) has the same origin.
(13) * ... des gens auxquels cette fille lui semble Ii tous tres attachee.
(13) has two indirect complements auxquels and lui that correspond to positions
...
3 and 3 in (80). Being a clitic, lui cannot
,., be a chOmeur and so necessarily corres-
ponds to 3; auxquels must then be 3, and being a chOmeur, cannot be linked to
a floating quantifier. In (9) on the contrary, position 3 in a is not occupied,
auxquels is therefore an employed 3, hence the acceptability of the quantifier:
(9) Ce sont des gens auxquels cette fille semble a tous tres attachee.
In sum the analogy between sembler and laire comes from two shared proper-
ties:
- they both undergo Union
- they both have an optional 3 complement.
For convenience, I will use the following classification of verbal and adjectival
predicates by means of the features +V (verbal) and +A (adjectival) :
GENERALIZED UNION 219
adjectives: +A -v
past participles: +A +V
other verbal forms :-A +V
Predicates of the faire. laisser type allow Union only with a downstairs
predicate [ +V, -A] (a "pure verb") :
(81) On a fait poursuivre Cesar par les Gaulois. (They had Caesar chased
by the Gauls)
Very close to faire semantically, the predicate rendre differs by allowing Union
only with "pure" adjectives [+ A, -V] :
(87)Cette potion lui rendra Cerbere fidele. (This potion will make Cerbere
faithful to her/ him).
(88)
lui
cette potion
220 G. FAUCONNIER
Cerbere inherits its 2 relation, and lui inherits the 3 relation since rendre has no
initial indirect object. Therefore the corresponding positions are cliticizable :
(91 )On lui croyait Brutus fidele. (Brutus was believed to be faithful to
[-V,+A] him).
lui
(93)On lui trouvait Cesar fidele. (They found Caesar to be faithful to him)
Turning now to the unaccusative verbs which are Union triggers, we find
that devenir requires [-f A, -V] while etre, like sembler, requires only [+ A] :
lui
lui
222 G. FAUCONNIER
In (97) and (98), the Union stratum (second stratum of a) has no 1 and so
Unaccusative Advancement 2 -+ I operates (3rd stratum); if it didn't the
network would violate the final 1 Law.
c) Passive
(103)
~ _ _ _ _ par les
Gaulois
d) Impersonal passives
If Union occurs with etre, but there is no downstairs 2, there will subsequent-
ly be no upstairs 2 available for Unaccusative advancement; in order to meet
the final 1 Law a dummy il comes in, giving the so-called impersonal passive
224 G. FAUCONNIER
constructions :
(l04) II est obei au capitaine par les soldats. (The captain is obeyed
by the soldiers)
au
capitaine
(105) II est (souvent) danse dans les rues. (There is often dancing in
the streets)
dans
les rues
However, if Union occurs with a verb like [aire in such cases, the upstairs clause
will already have aI, and there will of course be no dummy-insertion:
(106) On fera telephoner Ii l'hOtel par Max. (We will have Max call the
hotel)
GENERALIZED UNION 225
a
l'hatel
(107) On fera danser dans les rues.(We will have dancing in the streets)
dans
les rues
(110) Le capitaine a ete respecte par les soldats. (The captain was
respected by the soldiers)
( Ill) La polka a ete dansee dans les rues. (They danced the polka in
the streets)
226 G. FAUCONNIER
And this is of course condition 46.2 in the definition of Union in sec. II. except
that it now applies regardless of the verbs or adjectives involved. It follows
that Union will also apply to tense auxiliaries like avoir, under the assump-
tion that they also trigger Raising and select [+ A + V]. Example (I 12) will
correspond to network (113) :
(113)
la
nadle
GENERALIZED UNION 227
Note also that under the present approach, the I AEX law will no longer
cover the same range of facts as in Postal and Perlmutter (1978). Impersonal
passives of unaccusatives will be out simply because the unaccusatives have no
first stratum agents and because a 2 is always available for 2 ~ I advancement:
(114) Les feuilles sont tombees. (resultative meaning: the leaves are
fallen).
In fact, if the type of analysis outlined here for French could be extended to
other languages, the 1 AEX Law would follow from a more general condition:
(116) is mere speculation at this point. It fits in with the "teleological" view
mentioned above that advancements and dummy insertion conspire to meet the
final I Law. Something like (116) might also hold, at least as a tendency, for
other advancements: it was noted in sec. I.c that 3 ~ 2 Advancement is highly
favored if there is no 2 on stratum i.
The Generalized Union hypothesis provides a straightforward analysis of
passive constructions with no Passive rule. It should be underscored that the
resulting view of passive is in itself by no means new conceptually. In fact,
rather remarkably, it is essentially "isomorphic" to the account suggested in
Langacker and Munro (1975). On the basis of Mojave and Uto- Aztecan data,
Langacker and Munro propose the following "underlying" structure for
passives:
( 117)
v
I
be
228 G. FAUCONNIER
and emphasize that "BE is considered as a predicate with real semantic content",
"no instrumental or agentive phrase is posited as an integral part of the passive
construction per se", "a passive sentence does not derive from the same abstract
representation that underlies the corresponding active sentence", "there is no
single rule that can be felicitously referred to as a "passive transformation" ".
The approach is also in harmony with the hypothesis in Hertz (to appear)
that "les phrases passives impersonnelles ne sont de fa~on generale pas derivees
via NP Pre posing (in the present framework 2 ~ 1 Advancement)". I would
also agree in spirit with Hertz's functional interpretation of passive morphology,
adding however that the very possibility of such morphology stems from
Generalized Union.
Finally, Clause Union properties of passive have been pointed out for
Japanese by Shigeru Sakahara. Note that French uses so.called causative laire
with a purely "passive" meaning, as in (118) :
(118)· Je me suis fait happer par une voiture (I was hit by a car)
FOOTNOTES
For suggestions, criticisms, discussions ... , thanks to J. Gibson, S. Moreira da
Silva, M. do Nascimiento, P. Postal, E. Raposo, S. Sakahara and especially D.
Perlmutter, who, among other things, first suggested that adjectives might
behave like other predicates in Union phenomena, a key idea for the analysis
attempted here.
1Cf. Perlmutter (1980).
2Cf. Postal and Perlmutter (1978).
3But a number of speakers, myself included, allow laire with Equi and no
Union: Je te ferai la regarder.
4Cf. also Leclere (1978).
521 b) is ambiguous, since people themselves may be "reqUisitioned".
GENERALIZED UNION 229
Simon C Dik
Institute for General Linguistics
University of Amsterdam
o. Introduction
(2) a. m-c8ul-r
you-wash-imperative 'Wash him !'
b. m-c8ul-v-{
you-wash-R-imperative 'Wash yourself!'
This paper discusses some properties of the verbal reflexive strategy. It argues
for the following points:
(i) Typically, verbal constructions of type (2b) are not specifically reflexive
constructions, but more general 'reduced' or detransitivized constructions which
can be used for a variety of purposes, only one of which is to indicate reflexive
relationships.
231
232 s. DIK
(ii) Within the framework of Functional Grammar (FG, cf. Dik 1978, 1980)
such constructions can be described by means of a rule of predicate formation,
which reduces the number of argument positions of a given predicate-frame
by one.
(iii) Such a rule of reduced predicate formation can be assumed to have a
unified semantic effect on the output predicate, the different uses of reduced
predicates can be understood as different interpretations of the general meaning
of the reduced predicate-frame.
(iv) Although reduced predicate formation can historically arise through a
reinterpretation of an earlier nominal reflexive construction, there is no reason
to assume that this is its only possible historical source.
The structure of this paper is as follows: section 1 summarizes some points
made in Faltz's (1977) typological study of reflexive constructions; section 2
presents a general theory of reduced predicate formation within the framework
of FG, and shows how different interpretations of reduced predicates ·can
be understood in terms of one unified underlying meaning; section 3 describes
the appearance of reduced predicates in four languages of different types;
section 4 gives some conclusions, discusses some implications of the analysis,
and sketches some wider perspectives. It also contains a brief comparison with
Langacker and Munro (I 975). who treated similar phenomena within a different
theoretical framework.
J. Faltz's typology
Thus, the rule takes in a two-place verbal predicate-frame, and delivers a derived
predicate-frame with only one argument position, marked by some element R
(= Reduction) which symbolizes its derived nature. The derived meaning is that
the original input relation is applied only to a single entity.
A more concrete example of this schema could be something like:
Now, the derived meaning of predicates formed in this way would seem to
present something of a paradox: how can a two-place relation be relevant to
only a single entity? This paradox, however, is precisely what I would wish
to regard as the whole point of the detransitivization operation. Indeed, the
THE STATUS OF VERBAL REFLEXIVES 235
. (5) TASK:
Given a two-place relation R2 and a single entity a, construct an
interpretation in which R2 is applied to a.
Let us first assume that the single entity a indicates an individual of a kind which
would normally appear as the first argument (in the case of (4), the Agent) of
the input predicate. In other words, let us assume a construction of the form:
In that case, there would seem to be two possible solutions to task (5) :
Another way in which task (5) can be solved is by assuming that a does not
only provide the first, but also the second argument to R2. This would lead to
an interpretation along the lines of :
(II) 0~ R2
Since in this interpretation both the first and the second argument are equally
specific (both are provided by a), there is no reason to expect any 'de-
actualizing' effect in this interpretation.
Note that we are assuming that, although construction (6) can in this way
get a reflexive interpretation, it is not, in itself, a reflexive construction: the
reflexive interpretation comes out as one of the possible solutions to the task
posed by the general meaning of the construction.
Let us now assume that the single entity rather indicates an individual which
would normally appear as the second argument of R2. For instance, let us
assume a construction of the form:
(14) 0 R2_0
Note that we do not say that (12) is a passive construction, but that it can get
an interpretation which is quite similar to that of a passive construction. Since
a reconstructed argument is involved, just as in interpretation (A), we might
again expect a possible de-actualization, along the lines of :
(16) 'The ball has the property of typically being hit by some x'
Note that, if a would represent some entity that could just as well be first or
second argument of R2, we might expect the construction involved to be three-
ways ambiguous as between interpretations (A), (8), and (C). Thus, we might
expect an expression such as (6) to also be open to a third interpretation, of
type (C) :
Let us now assume that the single argument of the detransitivized predicate
consists of some expression A, which indicates a set of entities ai' a2' ... , an"
Thus, suppose we have a construction of the form:
In this case, however, there is a further possible solution to task (5), namely:
or graphically:
(22)
(24)Ivan se je ubio
Ivan se aux killed
'Ivan killed himself'
(25)Djeca su se vidjela
children aux se seen
== (i) 'The children saw themselves'
(ii) 'The children saw each other'
(c) impersonal, as in :
(26)Spava se na podu
sleeps se on ground
'One sleeps on the ground'
(d) 'modal'
240 S. DIK
(30)lvan se je ubio
Ivan se aux killed
(0 'Ivan killed himself'
(ij) '(Something or someone) killed Ivan'
The constraint against animate terms in the pseudo-reflexive can now be inter-
preted as a measure to avoid this ambiguity of (30) : it excludes (ii).
Fact (iii), finally, was that pseudo-reflective constructions tend to assume
a generic or habitual rather than an 'actual' interpretation. For examples, see
Dik and Gvozdanovic (1981). This fact we have seen to be predictable when the
interpretation runs along the lines of (C), in which an unspecified first argument
is reconstructed. Note that a construction such as :
would hardly make much sense as the deSCription of some actual event. It is
quite understandable, then, that the interpretation takes the generic route:
'Mathematics is such that, if anyone wishes to learn it, the learning is a slow
process'.
The different uses of the se-construction in Serbo-Croatian thus seem to be
interpetable in terms of the theory sketched in section 2. A special property of
Serbo-Croatian is, that this construction can also be used in 'impersonal' appli-
cations. This fact could be incorporated within the same theory by the
assumption that se-predicate formation can also be appiied to intransitive input
predicates, which are then deprived of their only argument position and thus
end up as expressions which claim a certain property to apply, without
specifying an entity to which the property applies 3 .
242 s. DlK
Although there is no clear difference in meaning between (32) and (33), the
opposition is quite important in Dyirbal, because there are a number of
syntactic and discourse rules which require a nominal term which is coreferential
to another nominal term in the abso!utive, to also be in the absolutive. Thus,
if a transitive Agent is the target of such a rule, the antipassive can be used to
get it into absolutive position.
Notice that the derived antipassive verb has all the characteristics of an
intransitive verb. This also means that the dative/instrumental term in (33) is
related to the verb in another, more indirect way than the absolutive term
in (32).
There is another antipassive construction in Dyirbal, which takes another
suffix on the verb (-rry and variants), but otherwise has the same syntactic
characteristics as the -'lay antipassive. Compare the following pair:
THE STATUS OF VERBAL REFLEXIVES 243
The construction with.,ry can be used under the same circumstances as the
construction with "'jay, but there is a semantic difference: whereas the -t;ay
construction describes the actual occurrence of an event, the.,(y construction
is used to describe rather the potentiality of some event occurring. Thus, in (33)
the man in question must be actually spearing a wallaby, but (35) can be used
to express that the man is a fruit-eater, even though at the time he may not be
eating at all (Dixon ibid. 91).
The oblique (dative·instrumental) constituents in such constructions as (33)
and (35) are not obligatory: they can be left out, which again shows that the
nuclear construction is an intransitive predication, which can be extended by an
optional constituent in dative or instrumental. Thus, we can have:
(36)bayi dYangaymarinYu
det-abs eat
'He eats'
This construction, then, can either be interpreted along the lines of 'the man
is potentially involved in hiding-events', or as 'the man is hiding himself. There
is a way of forcing out the reflexive interpretation, by optionally adding the
suffix -dYilu to the argument of the -rfy construction. Compare:
244 S. DIK
(39)bayindYilu dYaIJgaymarinYu
det-self-abs eat
'He is eating himself'
Why is this impossible? The answer is simple: there is a much more direct way
of expressing the intended meaning. A peculiarity of Dyirbal is, that in the
transitive ergative construction the absolutive nominal is obligatory, and the
ergative nominal optional.
Thus, given a construction such as (34) above, the ergative Agent can be
left out, resulting in :
use the same suffix for both the reflexive and the reciprocal condition. In such
languages, then, we expect the relevant derived predicate, when applied to a
non-singular term, to be interpretable in the following ways:
It seems that even in languages. which use the same suffix for these different
conditions, reduplication of the verb may (optionally?) be used in order to
force out the reciprocal interpretation.
( c) reciprocal, as in :
(47)ki9ar-c vhe:-v-km
dog-Subj tail-?-incompl
'The dog is tailed', 'The dog has a tail'
First, it seems that -v- is here attached to a nominal rather than to a verbal stem,
and second, no reduction of argument positions seems. to be involved.
Presumably, this -v- expresses another type of predicate formation, in which
a nominal stem is verbalized into an intransitive verb.
We saw in section 3.2. that Dyirbal has a way of forcing out the reflexive
interpretation of the detransitivized verb. The same is true of Yavapai. Consider:
248 S. DIK
and compare this construction with (46), which was given a reciprocal inter-
pretation. (48) can only get a reflexive interpretation, and this is due to the
optional element yem, which can also be used in non-reflexive constructions
with an adverbial reading such as 'all alone', 'all by oneself. This comes out
quite clearly in the following opposition:
Thus, it would be correct to say that (48) and (49b) are reflexive constructions:
they necessarily get a reflexive interpretation. However, they are quite different
from the more familiar nominal reflexives : the reflexive interpretation comes
out as the resultant of a detransitivized verb which merely says: 'the underlying
two-place relation applies to the single entity x', plus an adverbial element which
says that x is involved 'all alone'. Another difference is, of course, that the
latter element is optional, and only used when otherwise the verbal construction
in itself would be too vague to achieve its communicative purposes.
Note that these interpretations again correspond to (8) reflexive, (0) reciprocal,
(C) unspecified first argument. Interpretation (iv), however, cannot be simply
equated with interpretation (A) unspecified second argument, since in that case
the interpretation should be The children taughi (someone or other), rather
than The children learnt'. Rather, (iv) would seem to be a specialization of
(i) or (iii).
Note that Hixkaryana has no regular passive construction, and that when
the detransitivized verb is used in its pseudo-passive application, it is impossible
for an overt Agent to be expressed (ibid. 90).
It would seem correct to say, then, that the construction is neither reflexive
nor passive as such, but another example of a detransitivized predicate with
'reflexive' and 'unspecified Agent' interpretations.
A detransitivizing predicate formation rule would thus seem to be the
correct means for describing this construction.
4. Discussion
We saw in section 1 above that Faltz (1977) assumes that verbal reflexives
can historically only develop from earlier nominal reflexives. That nominal
reflexives can develop into verbal reflexives (or middles) is beyond doubt:
such a development can be positively attested from the history of different
language families. That verbal reflexives only arise in this way, however, is a
guess based on (a) lack of evidence to the contrary, (b) the assumption that
a single historical source is to be preferred to a multiple one.
From our perspective, we have several reasons for assuming this guess to
be incorrect.
First, note that in our analysis a detransitivization operation is c:ttegorical-
ly different from processes underlying the formation of nominal reflexives.
We must assume that nominal reflexives can get historically reinterpreted in
terms of detransitivization. But there is no logical reason why a language could
not have a detransitivization rule quite independent of any antecedent nominal
reflexive construction.
Second, consider the situation in Australian languages, as described in
Dixon (1980). Most of these indicate reflexive relationships as in Dyirbal, by
means of a detransitivized verb involving a suffix which can even be retraced
to a common proto-stage *-DHirri-y (ibid. 447--8). In different Australian
languages, the reflexes of this suffix have different ranges of application,
but the common denominator always seems to be 'detransitivization' rather
than 'reflexive'. There is no indication that the reconstructed proto-Australian
suffix has some earlier, nominal reflexive source. Some Australian languages
do have nominal reflexive constructions (ibid. 434), but (a) in such languages
the verb appears to behave as transitive rather than intransitive, (b) the reflexive
elements found in such languages seem to bear no resemblance to the
detransitivizing suffix used in the other Australian languages. Thus, it does not
only seem unnecessary, but even improbable that the detransitivizing suffix
arose from an earlier nominal reflexive element.
Third, a positive counter-example to Faltz's guess is provided in Langacker
and Munro's (1975) discussion of the history of Uto- Aztecan 'passive'
constructions. For these languages, an original suffix *-t~wa can be
252 S. DIK
(ibid. 803). This would seem to constitute clear proof that a verbal reflexive
may not only arise through a reinterpretation of an earlier nominal reflexive,
but also through an extension (which, in our approach, is a 'natural' one) of
a detransitivization operation with a reflexive interpretation.
One might of course try to salvage Faltz's guess by claiming that both in the
Australian and in the Uta-Aztecan case, the original detransitivization suffix
itself goes back to an even earlier nominal reflexive. But, as far as I can see, this
would be merespeculation about developments which might have preceded even
the earliest reconstructed proto-stage of these languages. And, from our point
of view, there is no intrinsic reason for wanting such speCUlation to be true.
Langacker and Munro (I975) discuss data similar to those treated in the
present paper, in the context of an alternative transformational hypothesis
concerning the fundamental nature of 'passive' constructions. Their analysis
is based on (mainly historical) phenomena observed in the Uto-Aztecan
languages, and on (mainly synchronic) properties of Mojave,a Yuman language
related to Yavapai, which we di3cussed in section 3.3. above. But the analysis is
assumed to be extendible to passive constructions in general.
The main features of the analysis concerning the underlying structure of
passives are presented in the following diagram:
THE STATUS OF VERBAL REFLEXIVES 253
(52)
(54)a. x x V > x V
{~}
b. x x V > x V
t~}
(55)a. t. x V > V
{~}
x
t. >
{~}
b. x V x V
254 S. DlK
NOTES
Susumu Kuno
Harvard University*
1. Introduction
257
258 S. KUNO
Pronominalization will apply to (2) if Adverb Preposing does not take place,
and to (3) if it does:
(4) a. Johni found a snake near the girl hei was talking with.
b. *Hei found a snake near the girl Johni was talking with.
c. Near the girl hei was talking with, Johni found a snake.
d. Near the girl Johni was talking with, hei found a snake.
(5) [Johni found a snake near the girl [Johni was talking with]]
(6) [Near the girl [Johni was talking with], Johni found a snake]
Pronominalization now applies to this structure. (4c) and (4d) are both possible
because in neither derivation does he both precede and command John. Thus,
REFLEXIVIZA nON IN ENGLISH 259
the data given in (4) firmly establish that Adverb Preposing must be applied
before Pronominalization. However, this rule ordering is in direct conflict with
the ordering established for the data given in (I). Hence arises Lakoffs rule-
ordering paradox.
In the above, I have shown that (1) and (4), in the transformational theory
for pronouns, yield a rule-ordering paradox. The same paradox arises in the
interpretative approach to pronominal reference also. J ackendoff (1972)
assumes that a rule for establishing coreference linkage between a pronoun and
a full-fledged NP applies cyclically at the end of each cycle. This rule basically
follows the same precede-command condition that Ross and Langacker proposed
for their transformational approach. On the basis of the sentences of (1), J acken-
doff assumed that this interpretive rule must apply before Adverb Preposing,
but noted the difficulty such a rule ordering would face accounting for the data
of (4). He left the problem unresolved.
An interpretive approach that assumes that pronominal reference is
established on the basis of surface structure also fails to account for the data
under discussion. Such an approach will try to determine if the pronoun and
John may be interpreted as coreferent in (1) and (4). If the conditon for
pronominal coreference in this framework is to be stated on the basis of some
kind of command relationship, which must be the case in order to establish the
fact that there is no co reference interpretation between he and John in (1 b)
and (4b), it will have a great deal of difficulty in establishing an interpretation
of coreference for (I c) and blocking an interpretation of coreference for (ld).
Thus, this approach has basically the same problem that the transformational
approach with Adverb Preposing before Pronominalization has been sho\Vn to
suffer from.
Reinhart (1976, 1981) has proposed that what controls pronominal co-
reference is not the concept of precede-command, but that of c-command :
while A commands B if A's immediate parent S(entence) node also dominates
B, A c-commands B only if A's immediate parent node, whatever category it
might be, also dominates B. For example. observe the following structure:
In the above structure,John commands Mary and Mary commands John because
John's immediate parent S-node dominates Mary, and vice versa. In contrast,
although John c-commands Mary (because John's immediate parent node also
dominates Mary) Mary does not c-command John because Mary's immediate
260 S. KUNO
parent node PP does not dominate John. Reinhart hypothesized that the
conditions for pronominal coreference can be stated on the basis of surface
structure, without recourse to the concept of "precede", simply by stating that
a pronoun cannot c-command a coreferential nonpronominal NP. According to
this analysis, (1 c) is acceptable because him does not c-command John (because
the immediate parent node PP of him does not dominate John), but (ld) is
unacceptable because he c-commands John (because the immediate parent node
S of he dominates John). However, as Carden (1981) has shown convincingly,
there does not seem to be any way to develop this c-command account of pro-
nominal co reference so as to make the correct predictions on the acceptability
status of the sentences in (4)1.
(5) does not receive [+ reflexive 1marking because it is not a clausemate of the
lefthand John. Thus, after Adverb Preposing, the lefthand John of (6) is still
unmarked, and therefore, it can remain unpronominalized. Hence the accepta-
bility of (4d). According to this analysis, while Reflexivization applies cyclical-
ly, Pronominalization applies post cyclically , after all word order changes have
taken place.
The interpretive counterpart of the above analysis requires that nonreflexive
pronouns be treated as potentially [+reflexive]. To start with, it is clear that
the interpretive approach needs a rule for establishing coreference between an
NP and a clausemate reflexive pronoun and disjoint reference between an
NP and a clausemate nonreflexive pronoun:
In (11 a), himself must be coreferential with John, while in (11 b), him cannot be
coreferential with John. Let us call the rule which determines this the Reflexive
Rule. The Reflexive Rule applies only forward: namely, only when the reflexive
pronoun follows its coreferent NP. In everybody's analysis, the Reflexive Rule
would obligatorily assign disjoint reference to John and him of both (12a) and
(12b) :
In order to allow for the interpretation of (12b) and (13b) with coreference,
it is necessary to assume that all non nominative pronouns are optionally marked
as [+reflexive 1as well. Thus, we would have the following structures as inputs
262 S. KUNO
The Reflexive Rule applies, and obligatorily marks John and himself/him of
(14a, b) and (1S,a, b) as [+coref], and John and him of (14c) and (lSc) as
[-coref). A later low-level spelling rule would block the derivation involving
(l4b) on the basis of non syntactic conditions (to be described in 3.) which
would dictate that the reflexive form be used in this pattern, but it would accept
the derivation involving (1Sb). The same rule would mark the derivation
involving (lSa) as unacceptable for many speakers, unless himself is used
contrastively.
The Reflexive Rule applies obligatorily, before Adverb Preposing. Thus, given
the Reflexive Rule would mark he and John as [ -coref] because the rule cannot
apply backwards. This [-coref] marking cannot be cancelled by any later
operations. Hence the impossibility of obtaining a co reference interpretation
for (17) :
On the other hand, given (I8), the Reflexive Rule does not apply between
John and he because they are not clausemates :
(l8)a. [John found a snake near the girl [he was talking with))
b. [He found a snake near the girl [John was talking with))
Therefore, the Reflexive Rule does not establish a [-coref] linkage between
these two NPs. After Adverb Preposing, the following structures are obtained:
(19) a. [Near the girl [he was talking with], John found a snake]
b. [Near the girl (John was talking with], he found a snake]
REFLEXIVIZA TION IN ENGLISH 263
The Pronominal Rule now applies, and optionally establishes coreference linkage
between John and he both in (l9a) and (19b) because neither violates Ross-
Langacker's precede-command constraint.
I· have given above an outline of my Reflexive Pronoun Analysis of the
phenomenon under discussion, in the framework of both a transformational
and an interpretive analysis of pronominal reference. What was new about my
reflexive analysis was the blind application of the Reflexive Rule to all clause-
mate NPs, with the distinction between surface reflexive and nonreflexive pro-
nouns to be taken care of by a later low-level spelling rule. In the rest of this
paper, I will use the transformational framework for the presentation of further
data and their explication, but the interpretive analogue of the analysis
presented is readily available for those who are more comfortable with an
interpretive approach.
In Kuno (I975), I used the reflexive pronoun analysis to explain the contrast
between (20a) and (20b) :
lackendoff (1975) attributed the contrast to the fact that she denotes "Real-
Mary" in (a), but "Image-Mary" in (b). It is clear that such an explanation would
not work because the following sentences are acceptable even though the
italicized she denotes "Real-Mary", just like in (20a), and not "Image-Mary" :
(21)a. Judging from Maryi's portrait, shei must have been sick.
b. From Maryi's portrait, everyone thinks shei must have been sick.
My explanation for (20) is the following. (20a) is derived from the underlying
structure
where the second Mary receives a [+reflexive] marking before Adverb Preposing.
Therefore, there is no way to keep it unpronominalized, and hence, the un-
acceptability of (20a). On the other hand, (20b) is derived from the structure
which has in Mary's portrait in sentence-initial position as a thematic adverb,
and therefore, Mary's does not get marked as [+reflexive]. Hence, if other
conditions are met, it is possible to keep Mary's unpronominalized 3 .
264 s. KUNO
The above conditioning for spelling out NPs that are marked as [+ reflexive]
is not exactly right. We can be much more specific about the spelling condition
now. Observe the following sentences, in which him and himself are meant to
be coreferential with John. In order to avoid the interpretation in which himself
is used emphatically for contrast, both himself and him should be pronounced
with no stress on them.
The contrast like the above has led many to assume that pronouns appear in
locatives, and reflexives in nonlocatives. However, this kind of generalization
does not work for sentences such as the following:
It is difficult to justify the claim that the prepositional phrases in the above
are locative when him shows up and nonlocative when himself shows up. It is
also difficult to claim that the prepositional phrases are directional when
reflexives show up, and nondirectional when pronouns show up. The crucial
difference between the reflexive and pronominal versions seems to lie elsewhere.
Let us examine the subtle. semantic differences that exist between the
REFLEXIVIZATION IN ENGLISH 265
sentences in (26) with him and himself (26a) with himself implies that John
held the book with his hand and put it behind his back. The book was directly
touching him. (26a) with him can be used for, but is not limited to, the same
situation. It can also be used for a situation in which the book was on a chair,
and John was standing in front of the chair so that the book could not be seen.
In other words, the pronominal version is noncommital with respect to the
direct contact of the book with John. Similarly, (26b) with himself implies
that John put the blanket over his head and covered himself with it, perhaps to
hide under it. On the other hand, the pronominal version of (26b) does not
necessarily imply such direct action. Perhaps, John was in bed, under a
comforter, and he pulled the blanket and put it over the comforter. Likewise,
(26c) with himself implies that the blanket was touching John, while there is
no such implication necessary in the pronominal version. (26d) with himself
implies that John was sitting or lying on the blanket, while the pronominal
version can be used to describe a situation, for example, in which John was
sitting on a chair, and he put the blanket under the chair. (26e) with the
reflexive pronoun is acceptable, for example, in a tug-of-war situation, in which
John could keep pulling the rope toward his direction. In this tug-of-war
situation, the sentence implies that it was a big victory for John. On the other
hand, in the same kind of tug-of-war situation, the pronominal version most
likely implies that the rope moved only a little bit in John's direction.
(26f) with himself means that John tied himself with the rope. On the other
hand, the pronominal version of the same sentence can be used to describe a
situation, for example, in which there were four poles standing around John,
and he strung the rope around the poles.
The above observations lead us to the following generalization:
The direct object of verbs is semantically the direct target of the actions or
states that the verbs represent, and hence, reflexivization of direct objects is
always obligatory with clausemate triggers.
It would not be amiss here to compare (27) with the generalization that
Zribi-Hertz (1980) has proposed for the French nonclitic reflexive lui-meme
'himself (elle-meme 'herself, etc.) versus lui 'him' (elle 'her', etc.) in a similar
sentence pattern. Let us first observe some relevant sentences:
266 S. KUNO
While English allows only himself in both (28) and (29), French allows the
nonreflexive lui 'him' as well as the reflexive lui-meme 'himself in (28), but
not in (29). Zribi-Hertz hypothesizes that the following generalization holds,
among others, for the distribution of reflexive and nonreflexive pronouns in
the above sentence pattern :
The semantic relationship between the referents of NPI and NP2 under
discussion concerns such considerations as whether or not they occupy the
same position in time and space. For example, Zribi-Hertz accounts for the
contrast between (28a) and (29a) by stating that the object of Victor's mockery
in (28a) can be a Victor in the past, but that the object of Victor's jealousy
in (29a) has to be the contemporaneous Victor - Victor in the flesh. Similarly,
in (28c), the Victor who is ashamed and the Victor who is the source of this
feeling of shame do not have to be contemporaneous because one can be
ashamed of oneself with respect to the past action one has taken. Hence, lui
is used when the Victor some time in the past is meant, and lui-meme is used
when the Victor in the flesh is meant. In contrast, in (29c), the two Victors
that are involved must be the same person in time and space. Hence, only
lui-meme can be used.
I do not have space here to go further into the details of the French
facts about reflexive and nonreflexive pronouns that Zribi-Hertz has analyzed 4 .
It suffices here to note only that the French phenomenon briefly described
REFLEXIVIZATION IN ENGLISH 267
above is interesting both because of its similarity and dissimilarity with the
corresponding English phenomenon. The similarity lies in the fact that the
choice between reflexive and nonreflexive pronouns is made not on the basis
of syntax, but on the basis of nonsyntactic factors involving the relationship
between the two NPs involved. The dissimilarity lies in the fact that while
English depends upon whether NP2 is the direct target of the action/state
represented by the predicate and would treat (28) and (29) in the same way,
French depends on whether the two coreferential NPs refer to the same object
in time and space or to two separate stages of the same object. It would be
interesting to investigate the sources of this difference between English and
French, but I must leave it for future research.
(33)a. I hate the story about himself/*him that John always tells.
b. I told the story about *himself/??him that John likes to hear.
Jackendoff (1972) suggested that the contrast between (a) and (b) have some-
thing to do with the question of whether the subject is marked with. the
thematic relation Agent by the verb or not. Namely, tell has an Agent subject,
but hate does not. Although Jackendoff did not develop this idea any further,
it seems that picture-noun reflexivization is influenced by this factor.
Furthermore, observe the following contrast:
possessive NPs. There are speakers who consider the following sentences nearly
acceptable:
presume that many speakers, even though they might regard (a) and (b)
marginal, stilI consider them to be considerably better than (c). What we have
here is not an all-or-nothing phenomenon, but a continuum, and this continuum
cannot be accounted for by resorting to a syntactic constraint such as the
Specified Subject Condition.
There is a strong motivation for assuming that reflexive marking applies
even into picture nouns with possessive NPs. For example, observe the following
contrast:
Again, Jackendoff (1975) attributed the above contrast to the fact that she
denotes "Real-Mary" in (a), but "Image-Mary" in (b). But it is clear that this
explanation does not work :
I assume that (37a) is unacceptable for the same reason that (20a) is unaccept-
able, namely, Reflexive Marking applies before Adverb Pre po sing to mark
Mary in the picture noun as [ + reflexive] :
(39)[MarYi found a scratch in John's picture of MarYi [+reflexive]]
Since Mary in the picture noun has been marked as [+ reflexive], it cannot be
realized as an unpronominalized NP in the surface sentence.
The above analysis necessitates the abandonment of the Specified
Subject Condition for explaining the unacceptability of (31 b), and the adoption
of a more semantically-based condition for picture noun reflexives. Such a
reformulation is needed anyway for accounting for the facts of (37) and (38),
as well as for all the other facts discussed in this section. It is also in accord with
270 S. KUNO
the fact that I discussed in Kuno (1980) concerning extraction from picture
nouns with specified subjects. For example, observe the following two
sentences:
(40)a. This is the term that I don't like Chomsky's definition of.
b. This is the story that I haven't been able to get Mary's version of.
Both sentences violate the Specified Subject Condition, but they are acceptable
to all speakers of English as far as I know.
Those who oppose my reflexive pronoun explanation of (37a) on the basis
of the Specified Subject Condition would have to first solve the problem of how
to account for the acceptability of (40) despite the fact that it should be subject
to the same condition 6 .
FOOTNOTES
Reflexive Marking (or its interpretive rule counterpart) from applying to (39),
and would thus leave the contrast between (37a) and (37b) unaccounted for.
THE TEMPORAL REFERENCE OF THE ENGLISH FUTURATE
Carlota S. Smith
The University of Texas
There are two closely-related ways of talking about the future in English,
the will-future and the futurate. The constructions are not interchangeable,
because the futurate involves some kind of plan, schedule, control, or pattern
of events, while the will-future is not so restricted. This restriction on the
futurate is familiar and has been discussed by grammarians at least since
Jespersen 1931. Less familiarly, the futurate does not fit neatly into a general
syntactic-semantic account of temporal reference in English: neither adverbial
nor aspectual forms play their characteristic roles in this construction. Because
it is problematic, the futurate is of some interest for the study of temporal
reference. Particularly interesting are the truth-conditional and presentational
variations of aspect. The futurate in a general account of temporal reference
is the specific topic of this paper*.
The futurate serves also as an area in which to discuss and demonstrate an
approach to the general question of how syntactic and semantic structures
should be related in a grammar. I will argue for intermediate semantic structures
that represent abstract scope relations yet are relatively close to the syntax.
The rationale for this type of structure is that it would serve as input for
different types of semantic constructs, such as model-theoretic and text
structures.
As an introduction, consider the following futurate sentences and two
questions about their temporal interpretation. Present tense and a future time
adverbial are the hallmarks of the futurate construction.
273
274 C. SMITH
The ftrst question concerns the time that a futurate talks about, and the role
of tense and adverhial in specifying it. Intuitively it is clear that the futurate
spans two times, the present moment and the future time indicated by the
adverbial. It also seems clear that the futurate is "about" the present. In
semantic interpretation, to evaluate a sentence for truth or falsity we need to
know what time a sentence is about; to adopt the useful term of Reichenbach
1947, we need to know the reference time of the sentence 1. Generally the time
adverbial contributes to the specification of reference time, but this does not
seem to be the case with the futurate. The matter is discussed in section I below.
Now consider the difference in interpretation between the aspectual pairs
of the sentences in 1. The differences are, oddly, rather elusive. For example,
a and b do not differ - as one might expect - in that the progressive talks about
an ongoing activity of squash-playing whereas the simple form does not. Nor do
the other pairs differ in this way. Certain features can be found that differentiate
between particular aspectual pairs, but it is difficult to generalize them. For
instance in some futurates the progressive form seems to make a slightly weaker
prediction than the simple form; but the pair c and d show that this is not
always the case. Indeed, it seems most unlikely that one interpretive notion
such as uncertainty would be appropriate for the entire range of futurate
sentences. Rather, what is wanted is a semantic account of the futurate that
allows the pragmatic interpretations that occur. I present such an account in
the following sections.
tense and time adverbial and ask whether the event in question occurred at that
time. Thus for 2a one asks whether, at the time yesterday, the sentence Mary
work is true; for 2b one asks whether at the time 6 0 'clock past, the sentence
We swimming is true.
This is not the pattern on which futurate sentences are understood and
evaluated, however. Futurates are not evaluated at the time specified by the
future adverbial but at the moment of speech, or Speech Time (ST). To see this,
consider the examples of 3 :
To evaluate these sentences we do not find tomorrow and ask whether John
rehearse is true. Nor do we ask the question at ST; whether John is actually
rehearsing when these sentences are uttered is irrelevant. Rather, to evaluate the
sentences of 3 one looks at the situation at ST, to see whether it licenses the
prediction. Thus one might ask whether John's name is on the rehearsal schedule
for the following day, whether he will be in town then, etc. If the situation at
ST licenses the prediction John rehearse tomorrow, the sentences of 4 are true.
Thus the role of the future adverbial is not to specify the time of evaluation, but
rather another time - here, the time of the predicted situation. As noted above,
the time of evaluation of a sentence is Reference Time (RT) in Reichenbach's
terminology; the future time specified in a futurate is Event Time (ET).
I have suggested that the Reference Time of a futurate sentence is the
moment of speech, ST. There is some evidence for this from time adverbials.
Futurates can always have a second time adverbial, that indicates present time
adverbial:
Now
(4) At this moment John rehearses tomorrow
Today
The will-future is evaluated at the time specified by the future adverbial: one
looks at a future 3 0 'clock and asks whether Mary rehearse is true. But for the
futurate, one looks at ST and asks whether Mary rehearse at 3 o'clock is true.
And, tellingly, futurates are compatible with a question about what will actually
happen in the future whereas will-futures are not. Thus 6a is good, 6b distinctly
odd.
This is just what one would expect if the first conjunct of 6a is evaluated at ST
whereas that of 6b is evaluated at tomorrow. (The second conjunct of both is
evaluated at tomorrow.).
These points show that the reference time of the futurate is indeed ST,and
that the future adverb specifies a time that has another function in the sentence.
If a present adverbial is taken as underlying all futurate construction, their
temporal interpretation can be handled with general rules needed anyway for
the interpretation of English. I suggest in Smith 1978 a set of such rules, that
map combinations of tense and time adverbials into times in a framework based
on Reichenbach's account of temporal reference. In the system a combination
of present tense and a present time adverbial indicates a present RT.
I now suggest that the interpretive rules be written so as to map surface
structures, or sentences, onto abstract semantic structures. These structures
would represent the semantic scope of temporal expressions as well as their
interpretation in times4 . Such structures mediate between the syntactic and
semantic structures of a language, by giving information relevant for semantic
structures of different types; they would not be limited to temporal expressions,
of course. For the constructions under discussion here, I propose that abstract
structures roughly along the lines of 7 below, be set up to model the semantic
THE ENGLISH FUTURA TE 277
scope of temporal (and other) expression. Structures like 9 would be the input
to interpretive rules for e.g. model-theoretic or situational semantics, for
constructing text structures, etc. The higher nodes represent general notions
(such as reference time, event time, aspect) and the lower nodes represent
notions and expressions particular to a given language.
----- -------
(7) S"
/empo~ /S'~
I ,/ "- VP
Time Frequency, Aspect /'" S ""-
/" ~ Duration
Tense Adverb NP
I .
{ presentt !Progressive~
Past J tSimple )
In this structure temporal expressions appear under the Temporal and Aspectual
nodes; temporal expressions may also be located in the inner sentence, as shown
in 9. (For the purposes of this paper, I will consider only time adverbials of
location, ignoring frequency and duration; the analysis of modals, induding
will, is also ignored.)
Tense and time adverbial, giving the reference time of a sentence, are
dominated by the Temporal constituent. So, for example, 8 represents a
temporal structure associated with sentence 2a.
(8) S"
~ "----"
Temtoral /S'~
(9)
---:-- S" ~,
1
.
Tem 0ral /S ~
/Ti~ /S~
Int ::[ T /\
As)ect
ProW,gi.. Ad..,b
It should be noted that structures like 9 are not incompatible with the insights
of Reichenbach about temporal reference. The structure allows two places for
a time adverbial (the Temporal constituent and the inner sentence) and thus for
the specification of two times. Reichenbach's system has three times: Event
Time, Reference Time, and Speech Time. Speech Time is the keystone of. the
system, since past, present and future, the general reference. times, are
determined by their relation to ST. Past time precedes ST, future time follows
ST, present time is simultaneous with ST. Thus although ST and its relation to
RT is not directly specified in structures such as 8 and 9, it is an essential part of
their interpretation.
John was drawing a circle may be true, but John drew a circle may be false;
perhaps John changed his mind in media res and drew an eggplant instead. As
Vendler, Dowty, and others have emphasized, differences like this can arise
because certain events involve completion in their temporal schemata. Such
events have at least two essentially different stages (before completion and
completion itself) and are said to be heterogenous. Some events do not involve
completion - Vendler's example is pushing a cart - and they are homogenous8.
Unlike the aforementioned cases, sentences about homogenous events do not
differ in truth value if they differ aspectually. So, They were running is true
when They ran is true because there is no outcome or result with which the
process of running is associated.
The standard interpretation of the progressive does not hold for futurates,
as Dowty 1977 notes. Consider the interpretation of the following, for
instance:
am not actually walking to work yet. Consider in this regard the interpretation
of 11 :
12 says that at the time yesterday the process Mary rehearse was going on, and
this is a standard progressive. But for an event such as rehearsing no goal or
outcome is associated with the process. Yet 12b cannot be interpreted as saying
that the process of Mary rehearsing is actually going on; rather, a process
associated with Mary rehearse tomorrow is going on.
The problem with the notion of extended process and events such as
rehearsing is this: in the futurate perspective all events are heterogenous. An ex-
tended process involves an event with stages that are essentially different.
Applying this notion to a homogenous event would require a radical change
in the structure of the event. The appealing simplicity of the extended process
approach cannot be maintained, therefore. Since the notion of extended process
enables a general account of the progressive, there might be some interest in
revising it appropriately. Before attempting a revision, however, it will be worth
while to look more closely at the approach.
that is expected to continue into the future. In this interpretation, both pre-
liminary processes and the actual processes they are associated with are
presented in futurate sentences. Consider in this light a sentence like 13, a
standard example.
maintaining the idea that futurates involve two situations. The analyses make
different predictions about the aspect of the inner sentence, or future situation.
One analysis predicts that the aspect of the inner sentence is always perfective,
because there is no syntactic place for the form that marks imperfective. The
other analysis predicts that the aspect of the inner sentence varies with that of
the main sentence - that the aspect of the main verb has the inner sentence
(that presents the future situation) in its scope.
These predictions can be tested with diagnostics of the aspect of the inner
sentence. I suggest two such diagnostics, one using adverbials and the other
conjunction.
Completive adverbials (in an hour, etc.) are compatible with perfective but not
imperfective aspect, as noted in Vendler 1967. Compare for example the
sentences of 14 :
(I6)Tonight I'm cooking in an hour a dinner that took Julia Child all day
(17) Next semester we're grading all the papers in three days
These examples suggest that the inner sentence of a futurate has perfective
aspect: all three are perfectly grammatical.
One might object that these examples are not convincing, because adverbials
with in are not necessarily completives. They can be taken as part of an event-
description, as in the dialogue of 18 :
(19)a. Sarah was making a quilt but she didn't actually make one
b. Sarah made a quilt but she didn't actually make one
(20)Sarah is making a quilt tomorrow, but she may not actually make one
(21)Martin is building a model airplane next week but he may not actually
build it
Both these examples are contradictory, indicating that the inner sentences of
the first conjuncts have perfective aspect. (The judgment is difficult to make
because there is a different, more natural reading, in which the future event
will not occur at all; this is the reading alluded to in the discussion of example
6b above.) Note that the results of both tests go in the same direction.
We now have a fairly reasonable account of the analysis in which a futurate
involves two underlying situations. We have seen that the aspect of the present
situation may vary, but that of the future situation is semantically perfective -
for surface syntactic reasons. In this account, it is interesting to note, syntactic
possibilities constrain semantic ones. The analysis explains quite nicely one's
intuition, noted in the introduction to this paper, about aspect in the futurate
284 r.SMITH
Section 5. Two aspectual puzzles about the futurate were noted at the
beginning of this paper. On the one hand, futurate progressives do not indicate
events that are "in process" in the usual sense; on the other, the difference
between simple and progressive futurate is difficult to pin down. The first puzzle
is explained by the notion of preliminary process. We are now able to clear up
the second.
The aspectual difference between futurates is elusive for two reasons. The
first reason is that, as we have seen, both perfective and imperfective present
the future situation in the same way - as the final endpoint of a predictive
situation.
The second reason is that the aspectual difference in futurates is not based
on a truth-conditional difference. The same truth conditions hold for both
perfective and imperfective sentences. In other words, the situations in which
a perfective futurate are true, are just those in which an imperfective futurate
are true. Consider for example the sort of situation that whould license futurate
predictions such as 1a and b, repeated here :
FOOTNOTES
7These matters are discussed in Dowty 1981, Vlach 1981, Bennett and Partee
1972.
8Heterogenous events are, in the terminology of Vendler, achievements and
accomplishments; homogenous events are activities.
9This approach is discussed in Dowty 1977.
lOIt is not clear that situations are always presented with an aspect;
nominalizations fOl example may not have aspect underlyingly. The question
deserves investigation.
II This use of adverbials is not limited to in, e.g.
i A : What are you doing?
B: I'm running as fast as I can; it's good for the cardiovascular system
ii : I went to Paris for three days, and ended up staying for three months.
12This important point was suggested to me in discussions with Frank Vlach.
13 A stative situation is, very roughly, one that is stable and homogenous. For
discussion see Smith, forthcoming.
PROLEGOMENA TO A SEMANTIC ANALYSIS OF PHASE-
INDICATING VERBAL PERIPHRASES IN PORTUGUESE
A1arcelo l)ascal
Universidado Estadual de Campinas
Tel-Aviv University
where brackets indicate an optional element, and curled brackets, the obligatory
choice of one of the elements within them. The 'etc.' refers to verbs which are
less frequent than the three listed, but whose sense is sufficiently close to theirs,
so that the constructions containing them may be considered as belonging to
the same "family", from a semantic point of view. Among such verbs, we in-
clude : entrar a, por-se a, deitar a, deixar de, estar par, terminar, etc. A set of
examples of the constructions schematically represented by the above formula
is given in the appendix. Most of these examples are from relatively recent
Brazilian and Portuguese authors. Some constructions whiCh are no longer
used (e.g. (3)), deserve also to be included, as well as some examples from
spoken Brazilian Portuguese. Concerning the latter, I have relied on my
intuitions as a native speaker in order to provide the customary indications
about acceptability.
Although this study is primarily concerned with verbal periphrases, my list
289
290 M.DASCAL
of examples includes sentences where the relevant constructions are ~ot, strictly
speaking, verbs, but rather adjectives, participles, and even nouns. Their
relevance to this study can be however summarily justified if we recall that, at
least for some kinds of verbs, "the role of the verb melts into that of predicate,
and actions fade into qualities and relations" (Vendler 1967: 109).
As far as I know, a systematic semantic study of this family of constructions
in Portuguese has not been so far undertaken. There are syntactic analyses of
some of them (e.g. Pontes 1973), semantic analyses of other ones (e.g. Kloppel
1960, Castilho 1966), as well as accounts of analogous constructions in other
languages (e.g. Rohrer 1977 for Spanish, Bausch 1963 for French, Perlmutter
1970 for English). Nevertheless, in spite of the obvious relations among the
members of the family of expressions under examination, we do not have so far
an integrated and coherent semantic analysis of them as a whole. Furthermore,
although some of the authors mentioned make use of the methods and concepts
of formal semantics (e.g. Rohrer) - especially of the notion of truth conditions
and of the operators defined within the framework of tense logic - a semantic
application of such tools to this family of constructions, which might go a long
way towards making their semantic description precise and empirically testable,
is still a desideratum. My aim here is, therefore, twofold: to provide an
integrated semantic description of the set of constructions in question by means
of an application, albeit somewhat informal, of the basic notions of formal
semantics and tense logic. If such a purpose is achieved, this paper might well
be "the beginning of the end of a beginning", i.e. a true "prolegomenon" to
a semantic account of such periphrases.
II. I said that the set of constructions illustrated by examples (I) - (26)
will be treated as a semantic "family". I am well aware of the difficulties in
justifying claims of this kind, characteristic of most studies of "semantic
fields". As a matter of fact, the semantic "kinship" of these constructions -
which seems to me to be intuitively well expressed by the general label 'phase
indicators' (Lyons 1977 : 710) - will only be spelled out through the proposed
analysis. It is possible, however, to present from the outset some facts that grant
some initial plausibility to my claim and provide an empirical basis in the light
of which the adequacy of the analysis might be evaluated.
The synonymy of the constructions comerar+por+inf 'begin '-+by '+inf.'
and comerar+ger. 'begin+ger.' is quite uncontroversial l . A sentence such as (9),
for instance, can be paraphrased in a meaning preserving way by
THE PHASE-INDICATING VERBAL PERIPHRASES 291
(27)Teve ainda a arte maior de nao dizer nada aos dous, para que Candido
Neves, no desespero da crise, comer;:asse enjeitando 0 filho e acabasse
por a1canr;:ar algum meio seguro e regular de obter dinheiro.
'He also displayed .... , so that Candid~ Neves ... would begin rejecting
his son and end up by find a regular .. .'
and
are no doubt quite close semantically. Their difference has to do with the
more permanent, more state-like (even though only "initial" in the person's
career) nature of being-an-office-boy, as opposed to the more provisional (even
though recurrent) nature of the taking-and-bringing-messages. In some sense, the
state described by the first sentence of (29) is the result of the repetition and, so
to speak, institutionalization, of an activity described by the first sentence of
(28). When the verb in the gerund refers to a process that tends towards a
specific end, the relationship between a sentence with the gerund and the
sentence with the corresponding adjective is one of process to result-of-process :
Notice that (28) and (29), as well as (30) and (31), seem to differ temporally.
Let us assume, for the sake of argument, that process and product (result) or
activity and state are related as suggested above. Let us assume also that each
pair of sentences refer to the same series of events, so that they may be
compared temporally. One can, then, observe that (31) and (29) locate the be-
292 M.DASCAL
ginnings to which they refer at time points which are after those of the
beginnings referred to by (30) and (28), respectively. Graphically, we would
have, for the pair (30) - (31) :
other events of the series
tl t2 ST
C. begins C. begins (=time of utterance)
softening soft
Notice also that the beginnings under consideration cannot be conceived merely
as the initial phase of the process or activity described by the verb in the gerund
nor of the state referred to by the adjective or noun-phrase. They are rather
presented as the beginning of a series of events, processes or states, of which. the
softening or being-soft are the initial ones. The choice of (30) instead of (31)
pushes the beginning of the series back in time. But neither (30) nor (31) refer
directly to the stages of the softening itself. Both take this process (or its result)
as given en bloc and, as such, as being able to become members of a wider
succession of processes, events, or states. This is confirmed by the fact that, in
many cases, the action that marks the beginning of a series must be considered
completed, in order to be able to function adequately as the beginning of
a series:
All the examples considered so far in the present section have in common
THE PHASE-INDICATING VERBAL PERIPHRASES 293
the fact that they assume there is a sequence - not necessarily specified - of
events, processes or states, and present a certain event, process or state - usually
specified - as the one which begins the sequence. This is not the case, on the
other hand, with constructions of the type comerar+a+inf (examples (1),
(5), (6), (7)). In these cases, the beginning referred to is always the beginning
of the action, process, event or state described by the verb in the infinitive,
rather than the beginning of a series of other events 4 . Such a difference, which
is, no doubt, significant, might suggest that there is a deep semantic gap between
the two types of construction, leading even to the postulation of the existence
of two semantically distinct verbs, comerar 1 and comerar2' Such a position,
besides being anti-economical and anti-intuitive, is hardly justifiable in the light
of a possible integrated account of the semantics of both constructions.
Intuitively, there is no doubt that in both cases the same notion of "beginning"
is at stake: a shift from not-performing-an-action (not-happening-a-process,
not-being-in-a-state) to performing (happening, being) that action (process,
state) in its initial phase. In both cases, it is necessary to be able to conceive
of an action/process/state as segmentable in identifiable and temporally ordered
'phases". The difference between the two types of construction lies mainly in
the different scopes of the action/process/state to whose initial phase reference
is made. All such facts will have to be accounted for in the proposed analysis.
What has been said about the construction comer;ar+por+inf, comerar+
ger. and comerar+adj. can be applied, mutatis mutandis, to the analogous
constructions with acabar, as in (11), (12), (17), (18), (19), (22)-(26). To such
examples, one may add cases of acabar+ past part. : (20) and (21). As for
acabar+de+inf, there are some difficulties. Whereas some of the examples
of this construction correspond straightforwardly to the cases of comerar+
a+inf, in so far as they indicate the final phase of a process/action (cf. (13),
(16)), others seem to refer not to a final phase of the process/action, but to the
fact that the process/action has been completed at a time immediately prior to
the time of utterance (e.g. (I 5)). Other examples seem to admit both
interpretations (e.g. (14)). The difference between the two possible readings of .
acabar+de+inf is so neat that it seems to support the thesis that indeed they
correspond to two different, though homophonous, verbs (cf. Pontes 1973 :
122). To this, one may add the fact that, in other languages, the two meanings
are lexicalized in entirely different ways (e.g. Fr. venir de vs. finir). Further-
more, in Portuguese, although expressed by the same form, the two meanings
are in a relation of semantic opposition to two entirely different expressions,
294 M.DASCAL
namely estar por and comerar a, respectively. Finally, if one takes into account
Lyons' claim (1977 : 705) that the system of verbal tenses is essentially deictic
(i.e., every "tense" refers essentially to the time of utterance), whereas the
system of aspects (to which the notion of "phase" should belong) is not, one
would have to admit that the two readings of acabar de belong to different
semantico-syntactic systems, so that it is reasonable to assume that they
correspond indeed to two different homophonous verbs.
In spite of the weight of such arguments, I will still try to include the
apparently independent meaning of acabar+de+inf into the semantic
description of the "family" here considered. The arguments above deserve care-
ful discussion. Here, only a few preliminary remarks are offered. Firstly, notice
that the integrated semantic description of the "family" does not preclude
differences in meaning - even quite radical ones - between its members. What
the analysis purports to do is to disclose the semantic "kinship" between them,
i.e. to explain how their different meanings are, nevertheless, related to each
other. By postulating two homophonous verbs, on the other hand, one merely
acknowledges a difference without taking into account the similarities, and,
therefore, without accounting for them. From a methodological point of view,
this procedure amounts to remaining at the lowest possible level of empirical
generalization, namely, that of a mere compilation of distributional data.
Semantic analysis, however, should go beyond this level. A comparison with
other languages, with its implicit suggestion that there is some universal- (or
merely general) phenomenon underlying the diversity of all (or some) languages,
is already a step ahead. In many cases such a step seems to be fully justified.
For example, the fact that many languages have two different words (Sp. sino
vs. pero, Ger. sondern vs. aber. Hebr. ela vs. aval) for the two different uses of
Eng!. but, Port. mas. Fr. mais, may serve as an argument supporting the thesis
that there are in fact two but, two mas and two mais (cf. Ducrot 1976; Dascal
and Katriel 1977). But, unless there are other ways of supporting such a
hypothesis, the argument in question might be turned upside down: given the
existence of only one but, one mas and one mais, in English, Portuguese and
French, it is equally plausible to suppose, prima facie, that sino and pero,
sondern and aber, ela and aval are semantically equivalent or, at least, very close,
deserving therefore a single, integrated, semantic analysis.
Finally, regarding the difference between the aspect and tense systems, even
if one admits the existence of a radical distinction (in terms of the opposition
non-deictic vs. deictic) between these two systems, it does not follow that they
THE PHASE--INDICA TING VERBAL PERIPHRASES 29S
III. The frequent use in the previous section of the expression 'intuitively'
and similar ones already indicates the difficulties that any semantic analysis of
the constructions considered will have to face. It will be necessary to go beyond
a mere intuitive terminology in order to set up the foundations for a satisfactory
formal account. In fact, most of the traditional analyses, although providing
extremely useful material, are satisfied with the appeal to an intuitive
terminology.
A case in point is the notion of verbal aspect. As we have seen its relevance
for the description and differentiation of our constructions is in general
acknowledged. In some languages, special verbal morphemes are used in order
to express precisely what is expressed in Portuguese by means of periphrases.
But the traditional semantic characterization of the aspects is far from explicit,
so that the use of their names cannot be taken as explanatory. According to
296 M. DASCAL
a commonly held view, verbal aspect refers to the 'point of view' (usually that
of the speaker) out of which an action/process/state is considered. Thus, the
temporal reference of two sentences can be the same, whereas they differ only
in their 'aspect'. Consider for example:
(35)0 povo acabou por aceitar essas visitas como uma rotina inevitavel.
'The people ended up by accept(ing) these visits as an inevitable
routine'
and
(36)0 povo comer;:ou por aceitar essas visit as como uma rotina inevitavel.
'The people began by accept(ing) these visits as an inevitable routine'.
Assuming that both sentences refer to the same event (the acceptance of the
routine), one can say that both locate it temporally in the same way as far as
the times of utterance (ST), of reference (RT), and of occurrence (ET) of·the
event are concerned. It is only a difference in "points of view" (in this case of
the speaker) that is expressed in the presentation of the event either as a "be-
ginning" or as an '.'end" of a sequence of events. The first point of view reflects
the "expectation" that other events of the same sequence will follow; the
second, that other events preceded it and in some sense led to it. Furthermore
there is in the former the expectation of an impending change in the situation
generated by the event, whereas in the latter there is an assumption that the
event came about against the background of adverse circumstances or as the
result of an "effort preceding the conclusion of the process" (Castilho 1966 :
65).
All these components will be traditionally considered as a part of the
difference in "aspect" between (35) and (36). The question, however, is to
determine more precisely the nature of such differences. Are they representable
in terms of differences in the truth conditions of the two sentences? Or else
only as differences in their semantic or pragmatic presuppositions? The
traditional notion of "aspect" does not provide an answer for these questions.
The extremely suggestive descriptions (for they are not really definitions)
offered by Tesniere (1969 : 76) of the two aspects relevant for the analysis of
(35) and (36) are clearly insufficient for capturing the semantic peculiarities
of these examples:
Besides being insufficient for the description of (35) and (36), such definitions
do not even attempt to clarify certain fundamental questions. Thus, does the
'a la fois' in them indicate a logical conjunction of two components that belong
to the same logical level (duration+beginning, duration+end), i.e., of two
components such that the falsehood of any of them would imply the falsehood
of the whole sentence? Or else should 'a la fois' be understood as referring
to components belonging to different levels, one of them explicitly asserted
whereas the other is only presupposed? Maybe these examples fit rather what
Tesniere considers to be another aspect of verbal aspect, namely the "point de
vue affectif' (1969 : 77). Such "point of view", introduced in order to account
for the use of the adverbs encore 'still' and deja 'already', is responsible for the
expectations of prior termination and posterior beginning of the action:
"De ce point de vue, encore correspond a l'aspect terminatif, et exprime
que Ie proces n 'est pas encore termine au point du temps ou l'on s'at-
tendait a ce qu'il Ie filt, ou bien OU on aurait souhaite qu'il Ie filL."
(38)Felicio correu.
'Felicio run'
Let us assume that these sentences are uttered at the same time. Whereas (38)
follows from (37), the same cannot be said of (40) and (39)6. Since the set of
logical implications of a sentence is very intimately connected with its semantic
content, the semantic analysis of (39) and (37) will have to account for their
difference in this respect, in terms of the differences between the two verb
phrases (run and write a novel) they contain.
One might distinguish the two verb phrases in terms of the aspectual
dichotomy imperfective vs. perfective. Some languages, like Russian and Greek,
employ morphological means to mark the difference between an action when
conceived as complete or as tending towards a definite end and an activity not
necessarily completed or successful. The same difference can be expressed, in
THE PHASE-INDICATING VERBAL PERIPHRASES 299
(Where -r stands for an infinitive and -ou '-ed', for a past morpheme). The same
restriction must be imposed on the inferential schema:
The features of aspect so far considered, however, are not the only properties
of the verb phrase which affect the semantic interpretation of the constructions
here examined. The type of "situation" (a generic term for state, process, event,
and action) normally referred to by a verb is another important factor to be
taken into account. It is what German linguists call 'Aktionsart', and Lyons
(1977 : 706) proposes to call the "character" of the verb. Conhecer 'to know'
and reconhecer 'to recognize', for example, differ in "character" : the former
denotes a "state"; the latter, an "event". Events, in Lyons' terminology, are
"non-extended dynamic situations that occur momentarily in time". Being
momentary, they cannot have, strictu sensu, neither duration nor phases. It is
such a "character" of a verb that disqualifies such sentences as :
300 M. DASCAL
Such a sentence may be acceptable, but it cannot mean that Varela began the
process or act of recognizing the murderer at 9 p.m. Its interpretation might
perhaps be: "Varela began to produce signs of recognizing the murderer at
9 p.m.". Similarly, "Varela esta vencendo", 'Varela is winning', cannot mean
"Varela is in the process of winning", but only something like 'Varela's
performance renders his victory probable' (Cf. Lyons 1977 : 712). It is also the
"character" of reconhecer that disambiguates (45) yielding an interpretation
referring to what we have called the post-final phase:
The "character" seems also to be responsible for the total exclusion of (4:6) :
and for the possibility of interpreting (47) at most in the sense of (48) but not
of (46).
"It has now become clear that such notions as duration, completion,
momentariness, inception and termination (which were listed earlier as
being among the more common notions to which reference is made in
general discussions of aspect) are not all applicable to every kind of
situation ".
IV. Let us review now some of the semantic accounts offered for the family
of constructions under consideration here.
Kloppel (1960) studies constructions of the form comerar+(prep) +inf ,
to which he attributes seven different "functions" -- i.e. meanings. These
functions vary with the kind of verb in the periphrase (it can be either
"perfective" or "imperfective"), with the preposition used (a, de, por, or ~),
and with the number of the subject of comerar. They are described in terms of
"Aktionsarten", whose names are suggestive by themselves:
302 M.DASCAL
J
century
singular a, de, ~ {perfeCtiVe INCEPTIV-FREQUENTATIVUM
imperfectiv or ITERA TIVUM (53), (54)
eingUlar}
plural par
{perfective ,J
1st act/phase of an action/
imperfectiv process of which it is a part (8), (9)
In order to account for cases such as (4) and, more typically (52), Kl6ppel
introduces a distinct function, the INCEPTIV-SUCCESSIVUM :
(S2)A 'tega sanha que em taaes feitos nehuuna cousa esguarda, come'tou
tamto darder nos emtemdimentos do poboo, que aa porta prim 'tip all
da egreja estava, que come'tarom de braadar altas vozes aos de ~ima,
que estavom fazendo que nom deitavom 0 Bispo afumdo ?
The blind hatred that, in such circumstances, does not spare anything,
began to bum so much in the minds of the people assembled at the
main door of the church, that (they) began to shout out at those
above them, what are they doing that prevents them from throwing
the bishop down ?'
(56) ... apelos para fazer qualquer coisa ... cometyaram a ser recebidos pelo
Pe. Prudente ...
' ... demands to do something ... began to be received by Father Pru-
dente .. .'
Since their subjects are in the plural, would these sentences now be said to have
the successive, rather than the frequentative (iterative) aspect, that their
corresponding active sentences are supposed to display? In view of the lack of
a more convincing semantic distinction, the frequentative function should not
be considered as substantially different from the successive.
As for the LENTO-PERFEKTIV, it is characterized by Kloppel as implying
that the action/process, to whose beginnings the peri phrase refers, reaches its
end, though not instantly. Apparently, in terms of the inferential schemata
discussed above, such a characterization amounts to saying that schema (41)
applies to these cases, as it does to atelic verbs. From (5), for example, it follows
that the night fell completely; from (6), it follows that the moon reached the
horizon. Furthermore, since this periphrase keeps its perfective character,
schema (57) too seems to be valid in these cases JO .
"Erstreckt sich eine Handlung oder ein Vorgang auf einen langeren Zeit-
raum und werden dabei die einzelnen Akte oder Phasen, die zwischen
Anfang und Ende der Handlung, bezw. des Vorgangs liegen (die ihrerseits
306 M.DASCAL
first moments
of an action 12
implies
completion - - - - - LENTO-PERFEKTIVUM
the peri-
phrase the actions in the FREQUENT ATIVUM
refers to sequence resemble ITERA TIVUM
each other SUCCESSIVUM
first action of (paradigmatic INCEPTIV -SUCCESSIVUM
a series of relation)
actions
The actions in the
sequence are not
necessarily similar
(syntagmatic relation) .. "First Phase"
+
- - - - - -..roI-t ST "egressive phase" acabo de fazer
'I have just done'
(ST = utterance time)
This condition, however, can be criticized on two counts. Firstly, notice that
even punctual verbs do admit periphrases with camerar. Recall example (6),
as well as :
(60) Foi quando resolveu trocar os pneus, na vigesima quinta volta, que
Piquet come~ou a veneer 0 Grand-Prix de M~naco.
'It was when he decided to change the tires, in the twenty-fifth lap,
that Piquet began to win the Grand-Prix of Monaco'
308 M.DASCAL
(63)Vit6ria esquiou.
'Vit6ria skied'
That is to say, the definition validates the inferential schema (41) for all verbs.
But, as we know, such inferences should be blocked for telic verbs. Maybe this
is one of the reasons that leads Rohrer to propose the alternative definition:
Since (58~ is suitable for some cases, and (64) for others, Rohrer ends up with
the disjun~ion (58) v (64) as the general definition for the periphrases under
analysis. But, in fact, this does not help to solve the "paradox", for,
independently of the occurrence of A or -A at t, the inference of schema (41)
remains valid, for all A, because of the third conjunct in the definition. That
conjunct ensures that A will be true, if not at t, then at some time interval
immediately after t. This is enough to guarantee that, at an utterance time after
t (not contiguous to it), one may correctly assert that A occurred. In order to
solve the problem it would be necessary to modify not only the second, but
also the third conjunct of the definition, which is shared by (58) and (64).
Aquist et al. (I977 : 74) propose a much more complex disjunctive definition
than Rohrer's, containing as many disjuncts as needed in order to account for
the diversity of behavior of the periphrase with different kinds of verb:
v ( ~ A&-A&~I-"I A»
310 M.DASCAL
Nevertheless, not even this disjunction is able to block inference (41) for verb
phrases like 'write a novel', since for each one of its disjunctions it is a valid
inference. Notice also that, unlike (58) and (64), (65) allows for the use of
camerar with punctual verbs, thanks to its second disjunct. It does not seem to
me, however, that the analysis such a disjunct gives of constructions of the
type camerar+punctual verb explains adequately their meaning, as discussed
above. Furthermore, one might stress that a disjunctive definitions, although
combining in a single formula the truth·conditions of a set of expressions, dis·
play only a superficial generality. For one thing, they do not correlate the
various disjuncts with the lexical properties of each verb, thus leaving unexplain·
ed the selection of one or another disjunct in each particular case.
In his analysis of the "imminential phase", Rohrer is mainly correct in
distinguishing it from the "near future", constructed with the auxiliary ir
'to go' in the present tense. He characterizes the difference as follows: in the
"imminential phase", ST = RT (utterance time = reference time) and both
precede ET (event time); in the "near future", RT = ET and both are preceded
by ST. The identification of RT with either ET or ST is made via the widely
held hypothesis (cf. Smith 1975) that temporal adverbs always indicate RT.
This hypothesis, together with the equation RT = ST in the imminential phase,
would exclude sentences like
(67)Amanha estou porsair.
'Tomorrow I am about to leave'
I have shown elsewhere (Dascal 1979) that there are a few difficulties with the
temporal adverbial hypothesis. But, no matter how one solves them, it seems
clear that there are some important semantic similarities and differences between
the imminential and the inceptive phases, that should not be overlooked. In
both, there is the idea of a passage from the non-occurrence of an action to its
occurrence, at some time. In the imminential periphrase, the passage does not
have to materialize in order for the sentence to be true. There is no contra·
diction in
(68)Fhivia est.i por dar II luz, mas 0 parto nao tern chances de se realizar.
'Flavia is about to give birth, but the delivery has no chances to
come about'
indeed takes place. Another difference lies in the fact that, whereas the latter
contains also the idea of a continuation of the action beyond its initial phase,
the former is neutral in this respect. It is not "concerned" with what happens
after the beginning of the action, focussing only on its pre-initial and initial
phases. Hence, its combination with punctual verbs (which are "monophasic")
presents no troubles at all (compare with (59)) :
On the other hand, the equivalence we have observed between the construc-
tions comerar+por+inf and comerar+ger., is analogous to a similar equivalence
for estar por, which holds, however, only for punctual verbs: estd chegando
'is coming' = estd por chegar 'is about to come', but estd correndo 'is running'"*
estd por correr 'is about to run'. In other words, only when estar+ ger. does not
admit the progressive reading it is equivalent to estar+por+ in! This is a further
indication of the fact that this peri phrase excludes any mention to a possible
or real continuation of the action (the progressive does make such a mention
- see (72)). It seems to me that the condition that requires the being or function
denoted by the topic of the sentence to "exist" (Rohrer 1977 : 117), derives,
ultimately, from the factors just mentioned. Rohrer calls attention to the fact
that sentences such as
(70)Jaime construini uma nova teoria que esta par destruir definitiva-
mente 0 morfema zero.
'Jaime will build a new theory that is about to destroy entirely the
zero morpheme'
are inacceptable due to the non-existence, at RT, of the object denoted by the
topic of the sentence (Jaime's new theory). But this only shows that the
periphrase focusses on the pre-initial phase: there, not only the topic must
exist, but also all must be practically ready for the action to occur, once time
reaches the ET; on the other hand, nothing is asserted by the sentence about
what will happen after the ET; even the topic may suddenly cease to exist then,
without falsifying the assertion. The periphrase in question expresses, so to
speak, the completion of the "preparations" for the action, according to the
speaker.
Rohrer's treatment of the other two periphrases (acabar de and parar de)
is analogous to his account of the two we have examined, and does not need
312 M.DASCAL
to be reviewed here.
V. At this point we have on the table many of the pieces of the puzzle I
called "an integrated treatment" of this family of expressions. A full solution,
taking into account all the observations presented above, would require a far
more detailed work. Here, in accordance with the prolegomenic nature of this
paper, I will only assemble some of the suggestions already made and propose
a tentative analysis which accounts for the semantic as well as for some
pragmatic characteristics of the constructions analyzed.
Let us recall, first of all, that the "paradox" involving the implications of
the periphrases with acabar and camerar, is analogous to the "imperfective
paradox" pointed out by Dowty, which has to do with the progressive (see
note 6). In order to solve this problem - and not only for that purpose - Dowty
(1977) proposes the following definitions for the operators BECOME (0) and
PROGRESSIVE (~).
(71 )(BECOME (~» is true at interval I if, and only if, (a) there is an
initial boundary interval J for I such that -~ is true at J, (b) there is
a final boundary interval K such that. is true at K, and (c) there is
no non-empty interval I' such that I' C I and such that (a) and (b)
hold for I' as well as for I.
This operator is, according to Dowty, a component of every telic verb. Leaving
aside his doubts concerning the status of condition ( c) (would it be a part of the
truth conditions, or a "felicity condition", in Searle's sense, for the appropriate
use of such verbs ?), which stipulates that the interval I must be the minimal
interval which is able to satisfy conditions (a) and (b), we can represent in the
following diagram the essentials of the definition:
t ) (
ST
...
Linda is not a painter Linda is a painter
ofthe picture of the picture
-~ ~
(72)(PROG (tn is true at I and at a possible world w if, and only if, there
is an interval I' such that I C I' and there is a world w' for which ~
is true at I' and w is exactly like w' at all time preceding and in-
cluding I.
r- 1-,
..
I I
f w
I
:-,
I I
r,I Pedro eats
..
(~)
I w'
I I
4-1'
I
The fact that I' exceeds I both in the direction of the past and the future is
designed to capture the idea, expressed by the progressive, that the described
action goes beyond the interval (or time point) considered. But the fact that I'
belongs to the possible world w', but not to the actual world w, expresses the
idea that the progressive represents such an extension of the action as conceived
(Le. true in a possible world), but not necessarily as actualized.
314 M.DASCAL
Whenever the progressive occurs with a telic verb (e.g. "Linda is painting a
picture"), this corresponds semantically to a combination of the two operators,
namely: PROG (BECOME (~)). A conflation of the two diagrams (cf. Dowty
1977 : 57) will then clearly show that the sub-interval K of I must be conceived
(in w') as a sub-interval of 1', i.e. the action must be completed in w', where
Linda indeed becomes a painter. But, since such a completion needs not occur
in w, the inference "Linda is painting a picture; therefore Linda painted a
picture" is invalid. On the other hand, the hearer has the right to believe that
the conclusion does follow, if the speaker does not provide additional
information that cancels it. That is to say, such an inference has the formal
status of a Gricean implicature 15 .
Periphrases with comerar share some features with the progressive, as we have
seen. They suggest that the action is extended towards the future, beyond an
initial interval or time point, although such an extension is represented as merely
possible (perhaps even probable), but not necessarily actual (cf. example (61)).
On the other hand, comerar itself is, semantically, a telic verb. That is to say,
it must contain, as the other telic verbs do, the BECOME operator. Two of
Dowty's ideas can", therefore, be directly applied to such periphrases. To them,
one must add another notion, discussed above, namely the notion of a temporal
partition of an action in successive phases or stages. The BECOME component
of camerar must apply, in the actual world, at least to the first stage of the
action; otherwise, there would not be, properly, a "beginning" of that action.
What can be missing, however, in the actual world, are the other stages of the
action, without affecting the truth of the assertion that the action began. On the
other hand, if one refers properly to a "beginning" of the action at a time
interval I, then it cannot be the case that the action, at I, is already completed
or even advanced beyond a certain "middle" stage. For instance, it would be
highly misleading to claim that the building of a house is in its beginning if all
that remains to be done is to paint the ceilings. A "beginning", then, must end
somewhere before the "end" of an action, or before "too much" of it has been
performed. The notion "end of the beginning" is notoriously vague, and will
depend heavily on the kind of verb included in the periphrase. Nevertheless,
the analysis will have to take it into account, somehow.
Putting together all the preceding remarks, we propose the following
analysis:
) (
~l w
..
I
I
I (-~1) •
I
I . r: BECOME 02 -,
I I
..
I
'"
I w'
I -~1 ) ( ~
fl L BECOME ~3
...-...- BECOME ~l---l
~----------------I'-- ________ ~
Some comments will clarify the clauses of this definition. The nature of the
partition mentioned in (a) is not specified in the definition. It is precisely the
different possible kinds of partition that characterize the different types of
verbs. The only general constraint on such partitions is that ~ 1 be a necessary
component of ~. Such a constraint ensures, together with (b), the truth of -~
at least in an interval preceding its beginning, whatever the partition one has
316 M. DASCAL
in mind.
In an atelic verb, such as run, the action is partitioned in such a way that it
repeats itself (whatever else is also done) at each successive interval, if at all.
This can be expressed by the disjunction:
By means of a double set of indices, the upper ones specifying the time interval
and the lower ones, the kind of sub-action, we can abbreviate the above formula
by:
A punctual verb does not allow for a periphrase with comerar because it is
"monophasic", i.e. it does not allow for any partition. It is, therefore, unable to
fulfill condition (a) of the analysis. As noted, the members of a partition need
not be distinct. Verbs that denote states or activities allow for periphrases
with comerar and can, therefore, be associated with partitions, although the
members of such partitions should probably be all identical, in the sense of being
activities or states of the same kind as the one denoted by the verb. Some
variation on (75) should account for them. It is also in terms of varieties of
partitions that the "frequentative", certain peculiarities of the "successive",
and the "1ento-perfektiv", can be accounted for. Notice that the different types
THE PHASE-INDICATING VERBAL PERIPHRASES 317
of partition validate or block inferences under (42), thus explaining away the
"paradox" : a verb which is characterized by a partition like (75) renders the
inference valid, whereas verbs whose characteristic partition is either (76) or
(77) block it.
Condition (c) of the definition takes care of both the continuation of the
action (in a possible world) and the vague notion of "end of the beginning".
It stipulates that at least one of the components of the partition be actualized
in w', in an interval of time posterior to 11' Notice that atelic and telic verbs
receive different treatments, under conditions (cl) and (c2), respectively. The
operator BECOME cannot be used in (cl), since ~1 - the only component of
The central idea here is that the action/process/event ~ is conceived as the initial
phase of a "broader" action/process/event, 1/1. Such an idea has been defended
in the preceding sections. Condition (c) has been abbreviated (if compared to
(73» because, whenever one uses properly the come~ar+por+inf periphrase,
318 M.DASCAL
Notice that, except for ~ I' the other stages of the partition of ~ are not even
mentioned in condition (c), since they are irrelevant for periphrases with estar
por. A weaker version of the analysis proposed would replace (b) by the req uire-
ment that ~o simply be true in 10 and w, rather than becoming true precisely
in that interval. Something might indeed be about to happen merely because of
the current state of affairs, which happens to be without any noticeable change
for a long time, now. Still, it would be odd to say, for instance, "it is about to
rain" at t if, at some time interval prior and relatively close to t, absolutely
nothing connected with the expectation of rain became true. Hence, I prefer
THE PHASE-INDICATING VERBAL PERIPHRASES 319
APPENDIX
(I) Come~o a con tar do meio da minha vida e ja estou com 39 anos.
'I begin to tell from the middle of my life and I am already 39 years old'.
(7) Quando nao havia pesca, nem biscate, nem uma galinha ingenua, quando
enfim, a vida come~ava a significar uma azeda melancolia, 0 Barba~as
abria entao 0 peito aos grandes sacrificios, aceitando uma empreitada
qualquer, ceifa ou colheita de azeitonas, cortes de lenha, etc. - a que ele
se entregava com uma gana de quem tinha pressa em voltar ao repouso.
'When there was no fishing, no small jobs, not even an inocuous chicken,
when, in short, life was beginning to mean a sour melancholy, Barba~as
would then open his chest to the big sacrifice, by accepting any contract
work, harvest or picking of olives, cutting wood, etc. - to which he would
lend himself as someone that hurried to return to rest'
320 M. DASCAL
(8) Come~ou a querer aprender tipografia, mas viu cedo que era preciso algum
tempo para compor bern, e ainda assim talvez nao ganhasse 0 bastante ...
'He began by want(ing) to learn typography, but he soon realized that
some time was needed in order to set up (type) well, and even so maybe
he would not earn enough .. .'
(9) Teve ainda a arte maior de n~o dizer nada aos dous, para que Candido
Neves, no desespero da crise, come~asse por enjeitar 0 filho e acabasse
alcan~ando algum meio seguro e regular de obter dinheiro.
'He also displayed an even higher skill by not tell(ing) anything to both
of them, so that Candido Neves, despaired by the crisis, would begin by
reject his son and end up (by) finding some regular and sure means of
getting money'
(10) A tinta ja anda meio desbotada por aqui e algumas tra~as se locupletaram
em alguns adjetivos, mas a hist6ria come~a falando sobre urn triangulo
de terras que pertencia aos indios amoa~a, arara, canamari e ipurina.
'The ink is already half washed off here, and a few moth succeeded in
becoming adjectives, but the story begins telling about a triangle of land
that used to belong to the amoa~a, arara, canamari, and ipurina indians'.
(II) Ele vai ser como os outros. Come~a duro, acaba amolecendo. Nao e,
Amancio? -
'He will be like the others. Begins hard, ends up softening. Isn't it so,
Amancio ?'
(12) A carreira de Pereira foi a jato: em menos de seis anos comer;ou estudante
e acabou catedrttico.
'Pereira's career was jetlike : in less than six years he began (as a) student
and ended up (as) full professor'
(13) Mandovi esperou 0 homem acabar de achar gra~a, mas vendo que isto ia
demorar urn pouco aproveitou 0 tempo para apanhar as rodilhas inteiras,
deixando os cigarros soltos, que no seu en tender nao lhe pertenciam mais.
'Mandovi waited for the man to finish to find fun (= to laugh ?), but
seeing that this was going to take some time, he took advantage of the
available time in order to collect the whole packages, leaving out the single
cigarettes, which, in his opinion, no longer belonged to him'
(I5) Urn dia 0 amante chego!J tarde e falou-Ihe de urn encontro que acabava
de ter com urn amigo na Africa.
'One day the lover arrived late and talked to her about a meeting he just
had with a friend in Africa'
THE PHASE-INDICATING VERBAL PERIPHRASES 321
(16) Diziamos que esse raio de comboio nao acabava de passar, home ...
'We used to say that that hell of a train didn't finish to pass, man .. .'
(17) Pouco a pouco, acabei por passar muito rente it porta e demorar-me na
vitrina aver coisas que nao me interessavam.
'Step by step, I ended up (by) pass(ing) very close to the door and to stand
by the show-case watching things that did not interest me'
(18) 0 pulso acelerava-se-lhe, 0 cora9ao pulsava mais forte e ele acabava por
mergulhar numa sonolencia lucida, em que a realidade e 0 sonho se mistu-
ravam.
'His hart beat would accelerate, his heart would beat more strongly, and he
would end up by plunging into a lucid somnolence where reality and
dream mix up'
(19) Longe, a familia, a qual nao escrevera em dois anos, continuaria por mais
algum tempo it espera dele, ou de noticias : mas acabaria por esquece-lo.
'Far away, the family, to whom he had not written for two years, would
go on waiting for him for some time: but it would end up forget(ing)
him'
(22) Quando ele bate, lento e 0 som de sino fica dan9ando no ar como um
choro, como a voz duma pessoa que esta se queixando, Tina pensa na
vida, na morte, no passado e acaba chorando, chorando desatadamente.
'When it strikes, slowly, and the sound of the bell remains dancing in
the air like a cry, like the voice of a person who is complaining, Tina
thinks about life, death, the past, and ends up (by) crying, crying in
bursts'
(26) Depois de tantas entrevistas e revelayao dos segredos mais intimos diante
do analista, 0 paciente acabou indefeso.
'After so many interviews and the revelation of intimate secrets to the
analyst, the patient ended up defenseless'
NOTES
I There are, however, some rare examples, spotted by Castilho (1966), in modern
Brazilian authors. where comerar+ger. is equivalent to comerar+a+inf rather
than to comerar+ pod inf :
A partir de outubro a cheia comeya baixando lentamente.
'Since october, the flood begins lowering slowly'
Contempiando essa gente do segundo andar, me ponho imaginando a
classe a que pertence.
'Looking at these people of the second floor, I begin imagining the class
to which they belong'
21 give more or less word by word translations. In English the infinitive cannot
be used in some positions, as in these cases.
3In this case, the temporal difference observed between (30) and (31) does
not seem to occur between (32) and comerou autor de um conto e acabou
escritor prollfico 'He began (as the) author of a short story and ended up (as a)
prolific writer', perhaps due to the different nature of the verb phrases.
4when the aspect is the so-called "successive" (e.g. (2), (3», there is indeed a
sequence of actions, events, etc., but all must be of the same type, and this
type is described by the verb in the infinitive.
5Por further discussion of "aspect", see section III, below.
6This paradox is analogous to the "imperfective paradox" for the progressive,
noticed by Vendler (1967) and dealt with, among others, by Dowty (1977)
and Konig (1979). Whereas (ii) follows from (i), (iv) does not follow from (iii):
(i) John was pushing a card (ij) John pushed a card.
. (iii) John was drawing a circle (iv) John drew a circle.
1Starting with the attempts at systematization by the philosophers Vendler
(1967) and Kenny (1963), the work on verb typology seems to present a certain
convergence, leading to quite similar classifications, such as those of Lyons
(1977) and of Mourelatos (1978). We assume the reader is familiar with at least
one of these works. As for the formaliZation of the semantics of aspects, see
Hoepeiman (1976, 1978).
8The syntactic aspects of these and other periphrases have been discussed,
among others, by Mattoso Camara (1972), Perini (1976), Pontes (1973), Barbara
(1975), in ~onnection with the work of Chomsky (1957, 1965), McCawley
THE PHASE-INDICATING VERBAL PERIPHRASES 323
Marc Dominicy
Universite Libre de Bruxelles
Vrije Universiteit Brussel
325
326 M. DOMIN ICY
I. Let us have a look at the adverbs which are normally used to translate
I'enir de + infinitive 2 :
Romantsh: (1) Eu sun be (apaina, gust) gnu
(Bezzola-Tonjachen: 364,460, 572; DRG : J, 313; AIS :
VIII, 1646)
"I have only (hardly, just) come"
Contemporary English has the group only just (Jespersen 1927, IV, 63) :
(12)(a) What's eating him is all that happened last night and he's only
just be rung in on it.
(R. Chandler, The Big Sleep, Harmondsworth, Penguin Books,
109).
(b) Much of this was described, in 1955, in a lengthy mimeographed
monograph ( ... ) which was made available to interested scholars
and university libraries but has only just been published.
(J. Lyons, Chomsky, Hassocks, Harvester Press, 1977, 47).
Similar facts are found in several Italian dialects (AIS : VIII, 1646), and in 19th-
century German, which uses nur erst "only only" (Behaghel : III, 172; Curme :
348; Grimm : III, 993) :
In Latin, the group modo vix occurs, in which the presence of modo "only"
reinforces the restrictive value of vix "hardly" :
If just, gUst, erst, modo, etc. have a restrictive meaning in such examples, then
the very existence of but (on(v) just, be gUst, nur erst, modo vix, etc. can easily
be accounted for. Indeed, languages often express restriction in a redundant
way: a clear illustration is provided by Italian soltanto "only", which stems
from Latin solum tantum "only only" (LOfstedt : 22-3).
1.2. Since we assume now that all the adverbs of (1-10) possess a (lexically
or contextually) restrictive meaning, we have to discover what they modify. Let
us come back, with this view in mind, to the English just (see OED: V, 640) :
(l9)Here made (... J represents C.. ) an event that had fallen out some
considerable Time before: for had it happen'd but just then. whilst
the speaker takes notice of it, or but a little before: hath would
have been the proper expression for it. (White (1761), in McCoard :
33).
In such examples, (but) just modifies a time adverb (now, then). Accordingly,
my conjecture is that an adequate grammar of English must assign to sentence
(4) an underlying structure of the same type (cf. Jacobson: 70). This
hypothesis, which entails that a transformational rule optionally deletes the
time adverb, applies to many other languages, e.g. :
TIME, TENSE AND RESTRICTION 329
The conditions which trigger the deletion rule would have to be specified for
every language. Some restrictive adverbs, like Rumanian numai and Italian
cognate words, do not allow this transformation to take place (ALR : V, 1490,
VI, l801;AIS: VIII, 1646):
Generally speaking, deletion seems to reduce the range of the possible pOSitions
of the restrictive adverb, which is closely tied with the perfect tense. In English,
just now may follow the main verb :
whereas just occurs either between the auxiliary verb and the past participle or
in pre-auxiliary position (when the past participle has been deleted)4.
2. To sum up, the sentences we have studied so far derive from an under-
lying structure which contains: (i) a "perfect" morpheme; (ii) a "restriction"
morpheme which modifies a time adverb like "now" or "then". In the following,
I will argue that such a "deep representation" can be interpreted through a
semantic theory dealing with restriction and the perfect tenses.
330 M. DOMINICY
It follows that a sentence like John eats bread must be false if (28b) is true 5 .
Although I accept the notion of focus, I do not believe that interpretation
obligatorily takes place at the surface level. On the contrary, I have just admitted
that the focus of some restrictive words is (or can be) deleted by a trans-
formational rule, Within an entirely explicit theory, condition (27b(a)) would
perhaps be rephrased. Furthermore, I assume that the standard theory of pre-
supposition is basicaliy correct. I do not think that the main argument of my
paper would be crucially affected if I considered (27a) as a "conventional
implicature" (Karttunen-Peters; Ladusaw: 103, 165-6). However, there is a
significant difference between Ladusaw's description and mine. According to
Ladusaw, (30b) entails (31b) :
but this entailment is "obscured" by the fact that (30b) implicates (30a) and
that (31b) implicates (31a). If (31b) presupposes (31a), which must be false if
(30b) is true, then (31 b) is neither true nor (classically) false if (30b) is true,
and (30b) is false if (31 b) is true.
2.1.1. The theory which has just been sketched also applies to sentences
like:
Indeed:
In sentence (32d), only focusses on the numeral two. If three> two> one/
(- )Num in the class is (are) clever, then (32d) presupposes (32a) and entails
not-(32c).
A lot of evidence supports principle (33) (see, e.g., Ducrot 1973; Fauconnier;
Horn). Nevertheless, some particular effects of (33) are blurred by well-known
pragmatic constraints. If, for example, a speaker S utters (32b), the hearer H
will normally infer that S believes that (32d) is true. In order to explain this
phenomenon, I have to define a set of conversational implicatures which are,
in fact, much too powerful (see Seuren : 110; Van Der Sandt: 10-1), but which
fit my present purpose :
332 M. DOMINICY
(36) Let p and q be two sentences such that p entails q and q does not
entail p. If S utters q, then H will infer that S believes that p is false.
Obviously, H's inference rests on his assumption that S complies with Grice's
first maxim of quantity (see also Fauconnier 1976a : 265; Gazdar: Horn 1972 :
37-42).
Though the "no more than" meaning of only amounts to a sub case of its
general meaning, any attempt to describe sentences like (32d) without making
use of some "scalar principle" is doomed to failure (see, e.g., Bartsch: 203-7
and Keenan). On the other hand, it seems impossible to maintain that only is
a "negative scalar operator", i.e. a downwards-oriented mapping between two
points on a scale (Shanon; see also Ducrot 1973). Neither Bartsch-Keenan nor
Shanon-Ducrot account for the ambiguity of sentence (28b) :
(see also Altmann 1976 : 208-15). Moreover, it comes out that adverbs meaning
"hardly" or "just" cannot get a restrictive value if their focus does not belong
to a scale. Consider, e.g., the French sentences in (38)6 :
(38b) and (38c) do not exhibit the ambiguity of (38a). When adverbs of the
same category become leXically restrictive words (Portuguese apenas, English
TIME, TENSE AND RESTRICTION 333
Latin: (44) lam modo nunc possum contentus vivere parvo (Tibullus)
(Dominicy 1974)
"Now, only now am [I] able to live contented with little'"
In such cases, the restrictive adverb focusses on what Vet (105-26) calls "un
adverbe de temps proprement dit", i.e. a word or a phrase which denotes, even
vaguely, a time point or a time interval and which is, in itself, a correct answer
to the question "When ?". The hypothesis I want to defend here makes the
following claim:
This apparently strange fact is easily accounted for if the expressions which
denote time points (time intervals) prior to now are higher than avo "now"
on the relevant scale.
Oi) In some Romance languages an adverb meaning "hardly" may substitute for
the restrictive adverbs of (39-42) (AIS : VIII, 1646;ALR : VI, 1801; Sandfeld-
Olsen: 11,235; Tiktin : 1,4) :
According to our analysis of 2.1.1. this proves that the focus (hier, ahora,
adesso. acum) belongs to a scale.
(iii) The French expressions au moins "at least" and en tout cas "in any case"
are used when the speaker wants to shift on a scale from a higher point to a
lower one, i.e. when he wants to blur the effects of the first maxim of quantity;
the facts are roughly similar in English and other languages (see Ducrot 1972 :
135; Horn 1972; Kempson: 154; Smith) :
The fact that (49a) is normal whereas (49b) is definitely odd directly follows
from hypothesis (45): jeudi "on Thursday" is higher than vendredi "on
Friday" on the relevant scale.
(iv) Hypothesis (45) also accounts for the striking parallelism between (50)
TIME, TENSE AND RESTRICTION 335
and (51) :
(52)If S utters p, H will infer that S believes that H believes that p is false.
Here H's inference rests on his assumption that S complies with Grice's second
maxim of quantity (see also Ducrot 1972). Principle (52) predicts that (SOb)
and (51 b) conversationally implicate:
(53)(a) "s believes that H believes that more-than-all the students voted
for Hubert"
(b) "s believes that H believes that Mary answered Peter's question
earlier-than-as soon as possible"
In other words, the speaker who utters (SOb) or (SIb) acts as if the hearer
believed in the truth of a sentence which is known to be logically or factually
false.
(v) Shanon (66-7) points out the contrast between (S4a) and (54b) :
Let us assume, for the sake of simplicity, that still focusses on the (deletable)
time adverb now, and that (56a) presupposes or implicates that, in a near future,
John will not be working. From this presupposition or implicature the oddity
of (56b) directly follows. Indeed, hypothesis (55b) claims that it cannot be the
case that John has been working and that he will not have been working 8 .
If (55b) turns out to be correct, we come to a crucial result. Let p and q
be two sentences such that: (i) p means that some event E is accomplished
at t; (ii) q means that E is accomplished at 1'; (iii) t precedes t'; (iv) p and q
conform to the same propositional schema. Let X and X' be the expressions
which denote t and t', respectively. According to (33), X is higher than X'
on the relevant scale, since p entails q whereas q does not entail p. Hence, hypo-
thesis (45) predicts that any restrictive adverb which focusses on X or X' will
get the meaning of "not until". In other words, the sentence:
ting diachronic corollary. Consider, e.g., the examples in (59) and (60) :
(63)
j
4
3 maintenant
2
I
If it is true that Peter has only read three books so far, then there was a time
prior to now at which Peter had read less than three books and there will
normally be a time at which he will have read more than three books. Thus,
erst and pas differ from nur and maar in a very simple way: their focus must
belong to a scale which is isomorphically related to some future-oriented scale
of time points (time intervals). This hypothesis accounts for the fact that, in
any example similar to (62), nur/maar may substitute for erst/pas, whereas the
converse substitution often proves impossible.
Karttunen and Konig fInd it puzzling that erst and pas also possess the
meaning of "not until" :
are syntactically and semantically ambiguous, since erst/pas "only just" may
focus either on a deleted nun/nu "now" or on the predicate ll . As a matter of
fact, an indirect argument in favour of such an interpretation can be found in
the history of some restrictive expressions. Rumanian uses numai ce "only that"
in order to translate venir de + infinitive (cf. ALR : V, 1490, VI, 1801; DLR :
I, 2, 243, VII, 553; Sandfeld-Olsen : III, 329; Tiktin : 1,308,319, II, 1064) :
The fact that the complementizer ce "that" introduces a main clause seems
rather strange; however, Latin, Polish, Russian and Bulgarian do have similar
constructions. The origin of this phenomenon becomes understandable if we
look at complex sentences such as :
Here, the focus of numai "only" is the clause fee) am pus capul pe pernt,
340 M. DOMINICY
(70) "When I fell asleep, I had only put my head on the pillow"
On the other hand, abia is also able to focus on a time adverb (see 2.1.2), so
that sentence (3) exhibit the same ambiguity as (67) :
If an analogical process applies to (69), (71) and (3), it will create a restrictive
adverb from a sentential connective:
4. In this last paragraph, I will try to show that the hypothesis of 2.2 can be
extended to venir de + infinitive. More specifically, I will argue that venir q.e +
infinitive is a surface construction which conflates an underlying restrictive
morpheme (RESTR) and an underlying "perfect" morpheme.
4.1. Evidence shows that the formation of the periphrasis venir de + infini-
tive dates back to the 15th century (see Damourette-Pichon; Flydal; Gougen-
heim; Werner; Wilmet 1970). This fact conforms to the predictions of 2.2 :
during the same period, perfect tenses in French begin to shift from an
"aspectual" to a "temporal" value. Moreover, the underlying presence of a
"perfect" morpheme accounts for some properties of venir de + infinitive.
In contemporary French, any clause introduced by the connective apres que
"after (that)" requires the use of a perfect tense except for a very few
exceptions (Wilmet 1976) :
Now, instead of a perfect tense, we can find venir de + infinitive (Wilmet 1976:
144) :
In such cases, subjects generalize a rule which, under normal circumstances, only
applies to perfect tenses:
This redundancy comes down to a sub case of the general tendency to redupli-
cate the underlying restrictive morpheme (cf. 1.2). In other words, venir de +
infinitive combines with seulement or (tout) juste in exactly the same way as
just combines with but or only in sentences (11) and (12).
However, my hypothesis is mostly supported by semantic arguments. In
examples (79) and (80), venir de + infinitive lies within the scope of a negative
or interrogative marker:
sitions, the denial of (79c) and the questioning of (80) only bear on the truth-
condition.
Sentence (82) conveys an information which precludes the use of the
question word Quand? "When ?". Consequently, venir de + infinitive may be
used to answer a "when-question" (cf. Soemarmo : 362) :
4.3. In 3, we saw that expressions meaning "not until" may, at a later stage,
focus on a term which belongs to a scale isomorphically related to a future-
oriented scale of time points; and conversely, French ne faire que "de (nothing)
but" is a restrictive expression whose focus necessarily contains the main verb
(Piot) :
Examples like (86) and (87) are already found in 15th-century French
(Werner: 281-4). Since the 16th century, the infinitive which follows ne faire
que may be introduced by the preposition de (= litt. "from", but more akin to
English to before infinitive).
The presence of de does not imply, nor preclude, the existence of a scalar
isomorphism (see Damourette-Pichon: VI, 222-3; Gougenheim: 129-32;
Moignet 1973: 129 and 1974: 39,43; Tobler: III, 92-3). Yet the 17th-
century grammarian Vaugelas formulated a normative rule according to which de
must (resp. may not) occur when the focus of ne faire que belongs (resp. does
not belong) to a scale isomorphically related to a future-oriented scale of time
points. Vaugelas' rule did not enforce itself upon the French speakers, but it
344 M.DOMINICY
(89)Elle ne venoit que recevoir ung petit poulet de papier de son amy.
(Brant6me, ibid., 341).
"She had just received a little letter (of paper) from her friend"
The formal mixing up of venir de and ne faire que (de) shows that a pragmatic
equivalence holds between the periphrasis venir de + infinitive - i.e. the under-
lying restrictive morpheme which focusses on a time adverb - and the
expression ne faire que (de), when the focus of ne .. , que belongs to a scale
isomorphically related to a future-oriented scale of time points.
FOOTNOTES
1See Dietrich: 147. On Portuguese: Castilho : 86, 93; Cella: 524-5; Rolanda:
1476; Sten : 306.
TIME, TENSE AND RESTRICTION 345
Jan Koster
Tilburg University
School of Language and Literature
1. Levels of representation
One of the basic problems of the theory of core grammar is the determination
of the levels of representation for the sentences of a language, and the character-
ization of the mappings that interconnect these levels of representation.
Traditionally, a distinction was made between deep structure and surface
structure, which were thought to be connected by transformational derivations.
,
At present, the most common model has the following form 1 :
(1) D-structure
"move a"
i
S-structure
PR
~~ LF
D-structure (formerly deep structure) is generated by the base rules, while the
transformational component that yields S-structure is reduced to the schema
"move a" . S-structure is more abstract than traditional surface structure in that
it contains empty elements, some base-generated, like PRO in (2a), others as a
residue of the original positions of moved elements, such as the trace t in (2b)2 :
347
348 J. KOSTER
Both PRO and TRACE are coindexed with their antecedent John in (2).
The level PR (= Phonetic Representation) is close to the traditional notion
of surface structure. It is derived from S-structure by applying certain deletions,
and perhaps stylistic rules that may be seen as instances of "move a".
LF stands for Logical Form and is a level that minimally differs from S-
structure, mainly by application of rules of LF-movement, that can again be
regarded as instances of "move a", although there is no consensus as to the
properties of rules of LF-movement.
In this article, I have nothing to say about PR, only a little about LF, and
somewhat more about D- and S-structure. One of my conclusions will be that
the distinction between the latter two levels is not justified by the available
evidence. Clearly, then, the determination of the number and nature of levels
is an empirical issue.
The major rule leading from S-structure to LF is Quantifier Raising (QR),
as developed by May (1977). This rule picks up a quantified NP and adjoins
it to the minimal S that contains it. Everyone in (3a), for instance, is treated in
this way, giving (3b) :
QR can be seen as an operation that makes the scope of a quantifier more trans-
parent, assuming that the scope of a quantifier is the c-command domain of
that quantifier. Clearly, the existence of a level of LF is dependent on the
existence of rules like QR and the crucial question, therefore, is whether there is
any justification for QR, i.e. evidence beyond the observation that the scope of
a quantifier is the domain defined by the minimal S containing the quantifier.
Evidence appears to be rather minimal and the most interesting claim, namely
that QR has exactly the properties of "move a" (particularly Subjacency), also
appears to be the most problematic. As an example, consider the fact that wh-
phrases cannot be extracted from PPs in Dutch 4 :
(5) *Wie praatte je met t
Who talked you with
'Who did you talk with'
THE CONFIGURATIONAL MATRIX 349
If QR had the same properties as "overt" movement rules, one would expect
the same pattern for QR. Quantifiers are always possible in PPs in languages
like Dutch and, therefore, we see an unexplained difference between QR and
wh-movement 5. There are other differences, and I therefore agree with Chomsky
(1981, ch. 4) that LF-movement does not have the properties of overt
"move a". Since as it stands, scope properties. cannot be derived from the
standard properties of "move a", the most important piece of potential
evidence for QR remains problematic. The other evidence is slight as well, and
for a sceptic, LF-extensions of S-structure might seem rather stipulative at the
moment. This is an active and interesting area of research, however, and perhaps
the issue can be settled in the near future 6.
On the whole, it seems that apart from PR, S-structtire is the best-established
level of syntactic representation. A separate level of LF remains interesting
but problematic while, in my opinion at least, the traditional level of D-structure
has become near vacuous in a theory that incorporates traces.
2. From D- to S-structure
The -s of likes depends on the local subject John, which is singular. If we do not
want to complicate agreement facts without limit, we have to assume a level of
D-structure, because subjects can drift indefinitely far away from the relevant
subject position:
generated in the D-structure position indicated by ___ , where the local process
of subject-verb agreement applies. This is a rather representative argument for
D-structure and it is compelling, I believe, in a theory without traces. Under the
assumptions of trace theory, however, such arguments loose all force. The reason
is that the essential information that was lost in a theory without traces, is
preserved in a theory with traces. Thus, the trace-theory equivalent of (7) still
has a local subject at the place where subject-verb agreement applies, name-
ly ti :
This trace can serve as a singular subject, because it is linked to a singular lexical
NP, namely which man.
Traces are always locally bound, in their governing category (cf. Chomsky
1981). If such local links are repeated, we have what is called a chain:
Let us call O'n the functional position (the D-structure position in terms of older
theories). Functional positions are governed, subcategorized, 8-marked, or case-
marked. The first position of the chain, 0'1' is called the head of the chain, which
contains the lexical material of the chain, if present. From a general uniqueness
condition, to be discussed later, it follows that only one member of a chain can
have lexical content. From the binding theory of Chomsky (1981), it follows
that only 0' I of a chain can have lexical content.
Together, these assumptions entail that each chain has one functional
position, O'n' and at most one position with lexical content, 0'1' In general, the
lexical content of a chain has to match its functional position. For instance,
if O'n is 8-marked as an agent, a lexical 0'1 has to contain material that is
compatible with this agent function.
That trace theory undermines the classical arguments for the distinction
between D- and S-structure, follows from the un controversial fact that chains
THE CONFIGURATIONAL MATRIX 351
Therefore, we have to assume that the deep structure of (14) is like (13), so that
the idiomatic sense, based on adjacency, can be determined.
If chains are represented at S-structure, such arguments have no force.:
352 1. KOSTER
One of the explanations given is that the structure underlying (I8) contains a
trace between want and to :
(19) Whoi do you want lei to see Bill]
blocks contraction8 .
Van Riemsdijk and Williams rightly point out that (19) is analogous to (17),
and that in both cases there is lexical material (Bill and Who, respectively) that
is responsible for the fact that contraction does not apply. From such facts,
they conclude that there is a level of NP-structure, in which (19) is exactly
analogous to (I7) in that there is actual lexical material between want and to.
NP-structure precedes Wh-movement so that Who i is still in the position of ei
in (19).
I agree with the observed parallelism between (17) and (19), but I see no
argument for NP-structure on the basis of these facts. Also without NP-structure,
(19) is exactly analogous to (17a). This follows from a proper definition of a
lexical NP (i.e. the kind of NP that blocks contraction). If a lexical NP is simply
an NP with lexical content, there is no difference between (17a) and (19).
In both cases, this fact can immediately be read from S-structure by checking
the content of the chain (= al) to which the embedded subject b~longs. In
(17a) this content is Bill, in (19) Who. In both cases the content (under al)
is lexical, so there is no relevant difference 9 .
It should perhaps be stressed that this conclusion can be drawn on the basis
of S-structure without "reconstruction" in the sense discussed by Van Riemsdijk
and Williams. So, it is simply false that we need either reconstruction or NP-
structure. Neither is necessary, given the uncontroversial fact that chains are
represented at S-structure.
Summarizing, then, we see that there is no reason to distinguish D-structure
from S-structure. Nor is there a compelling argument for NP-structure. Pre-
sumably, theories with only S-structure are notational variants of theories with
additional levels like D-structure or NP-structure 10. If this is true, the latter
theories can only be saved from Occam's razor by conceptual arguments of some
sort. Some such arguments will be discussed in the next section.
Since the beginning of trace theory, Chomsky has stressed that it is possible
to generate S-structure directly, without a transformational derivation from D-
to S-structure 11.
Chomsky has not considered this step very meaningful, however, for a
number of reasons. First of all, "move a" is not really eliminated if the
properties of it show up elsewhere, namely in the interpretive rules that connect
354 J. KOSTER
This format mentions two bounding nodes, ~ and W, while the domain state-
ments for bound anaphora (cf. Clause Mate Conditions) are usually formulated
with one bounding node:
Even then, the differences are not overwhelming. But we can make a stronger
THE CONFIGURATIONAL MATRIX 355
claim: (21) is also the right format for movement rules. This is a non-trivial
unification, with empirical consequences that clearly favor (21) for movement
rules, in my opinion.
All in all, two different research strategies originated from trace theory.
Either one could seek to maximize the differences between the theory of "move-
ment" and the theory of anaphora, or one could seek to unify the two. The
Lectures on Government and Bindmg adopt the former strategy, while, personal-
ly, I believe that the latter approach is correct. At any rate, the issue has
appeared to be productive and led to new insights and facts.
First, I believe that the issue has led to a considerable body of facts that
shows that (21) is the right format for movement rules. I have little to add to
what I have said elsewhere about this topic, and I believe that no new evidence
in favor of (20) has appeared 13.
Second, there have been attempts to dissociate the conditions for anaphora
from the conditions for movement in other respects. Rizzi (I 978), for instance,
argues that the SSC and the NIC do not apply to movement rules, but only to
rules of bound anaphora l4 . Both the NIC and the SSC have not really survived
in the Government-Binding framework. The NIC has disappeared altogether, and
can therefore no longer be considered a condition that distinguishes two kinds
of rules. The SSC has also disappeared, but some elements of it have been pre-
served in the notion "accessible subject" : domains for bound anaphora must
have an accessible subject, in contrast with domains for traces 15. The notion
"accessible subject" is too intricate to discuss here, but things can be illustrated
with an example:
(22)[ S 1 They think [ S2 iti is a pity [Si that pictures of each other are
for sale ]]]
Both TRACE and PRO are empty NPs ([Npe ]), but TRACE is supposed to be
a residue of movement, while PRO is considered a base-generated element.
Originally TRACE and PRO were seen as two distinct primitives, while recently,
in Lectures on Government and Binding, the two elements have been defIned
as two contextual variants of the same empty element 17. This recent
reformulation goes in the direction of a closer identifIcation of the two
elements. The differences are now believed to be contextual: TRACE is always
governed, while PRO cannot be governed. This difference in governance has
consequences for the locality conditions. While TRACES are always locally
bound (in their (minimal) governing category), PROs can also have long-distance
antecedents:
This observation seems correct, but it overlooks the very important fact that
there are two kinds of PROs (cf. Williams 1980). Some PROs are just like
traces: they are obligatorily and locally bound by a c-commanding antecedent.
Only optional PROS, i.e. PROs in for-complements· and gerunds in English,
can have deviant properties such as long-distance antecedents. These differences
can be explained by assuming that the Ss containing obligatory PROs undergo
S-deletion, just like complements to raising verbs. If this explanation can be
maintained, TRACE and PRO in (23) have the exact same properties l8 . I have
tried to show elsewhere that this explanation cannot only be maintained, but
that it also predicts new facts (cf. Koster 1981).
All in all, it seems to me that at present the evidence for the distinction
THE CONFIGURATIONAL MATRIX 357
(25)a. obligatoriness
b. uniqueness of 0:
c. prominence of 0:
d. locality (0: and 'Y in same domain)
In this structure, there is more than one VP that assigns the subject role to the
NPJohn.
Another mechanism constrained by uniqueness is the assignment of lexical
content (case) to functional positions for NPs. Under the assumptions of free
indexing, an ungrammatical expression like *John hits has two possible
structures :
In the first example, (27a) ej is not bound at all. This violates the obligatoriness
condition on such elements: it must be bound. In (27b), ei is bound by the
subject John. But this structure is also ill-formed. Assignment of lexical content
(expressed by case) is obligatory. So (27b) can also be seen as a violation of
obligatoriness, i.e. if ei has no lexical content at all. If we interpret John as
the lexical content of the object ei (a chain interpretation), we would have a
violation of uniqueness. John would not only be the lexical content assigned to
the object position (objective case assigned by hit), but the lexical content
assigned to the subject position (nominative case assigned by INFL).
Prominence principles are the familiar notions of c-command and subject-
hood. Usually 0: c-comrnands 'Y in all processes of core grammar. This is true for
rules of anaphora, "movement", subject-verb agreement, predication (cf.
Williams 1980), complementation (sub categorization , case assignment, and
a-marking), and several other processes. In some cases, 0: must also have the
subject role. This is a somewhat stricter version. There is some variation in this
respect, from language to language, but also from rule to rule.
Locality principles are of the form (21). The (usually governed) element 'Y
must be bound by Q within ~ in :
where the value of ~ varies slightly, again from language to language, and from
rule to rule 21 . .
The configurational function characterizes processes as different as
interpretation of bound anaphors, interpretation of traces, and subject-verb
agreement. It forms a separate module of grammar that is independent from the
rule content of the processes that it constrains. Although many details remain
to be clarified, it is the configurational function that I consider the most
important result of the fruitful linguistic theorizing of the last twenty years22.
The configurational function is intimately connected with the best established
level of syntactic representation, namely S-structure.
NOTES
4For extraction from PP in Dutch, see Van Riemsdijk (1978), and also Koster
(1978) and (1980), where it is assumed that PP is a bounding node for
subjacency (or a proper reformulation of it).
5May (1977) notices some of these discrepancies, but does not offer a solution
in all cases. New suggestions concerning this crucial problem will be given in a
revised version of May (1977) (personal communication).
6There is a large literature on LF. Relevant references can be found in Chomsky
(1981). See also Higginbotham (1979), Kayne (1980), Aoun et al. (1980).
For an interesting discussion of LF and conditions on movement, see Huang
(1981).
7The concept of a chain assumed here is somewhat more general than the
function chains discussed by Chomsky (1981,ch.6), where al is always an NP
(cf. Chomsky 1981, p. 333). The generalized chain conception assumed here,
also covers chains in which a 1 is a Wh-phrase of types other than NP.
8See, for instance, Chomsky (1977).
9Similar considerations apply to the binding of anaphors. Thus, there is no
relevant difference among:
(i) They saw pictures of each other
(ii) Which pictures of each other did they see '!
(iii) They saw each other
In (i) and (iii), each other is (part of) the content of an NP with the minimal
chain length (n == I). In (ii), each other is contained by the head of a chain of
two elements (n =:: 2). In all three cases, each other is (part of) the content of the
chains in question, which is under ai, by definition. Therefore, I assume that a
proper formulation of the binding theory refers to the notion "content"
(== al): anaphors must be (part of) the content of categories that are
c-commanded by the antecedent. Thus, the content itself need not be
c-commanded by the antecedent but only the functional "anchor" of the
chain (= an). That head (al) and functional "anchor" (an) sometimes fall
*"
together (n =:: 1) and sometimes not (n I) is irrelevant in my view. In all (ases,
the crucial insight is that even with a chain of one element (n =:: I), content (a I)
and functional position (an) can be distinguished. This is in fact, I believe, the
idea underlying trace theory: the content of a category is moved, while its
functional "anchor" is left behind.
10Thus, I do not deny, the possibility of empirical distinctions, nor do I have a
strong a priori preference for theories with S-structure only. My point is simply
that the existing empirical evidence for D- or NP-structure is not very
convincing. The available evidence for D-structure is summarized in Chomsky
(1981, p. 346). See also op. cit. p. 89 ff.
II See Chomsky (1981, p. 91) and the references mentioned there.
12See Chomsky (1981,42 ff.)
l3 See Koster (1978) and (1980).
14For the SSC and the NIC, see Chomsky (1980).
15See Chomsky (1981), p. 209 ff.).
360 J. KOSTER
James D. McCawley
University of Chicago
A number of recent articles (e.g. Braine 1971, Baker 1979) have taken up
the question of how a child can learn a language without reliance on large
amounts of negative grammaticality data, i.e. data of the form 'X is not
grammatical as a sentence of language L'. The authors note that the most
obvious source of negative grammaticality data, namely corrections of the child's
speech by adults and older children, in fact plays only a small role in language
acquisition. Baker also notes that in the case of the most commonly attested
corrections, namely those relating to morphological irregularities ("Don't
say dood, say did"), the positive aspect of the correction provides all the
data that the child needs: since specific rules (as Panini, Kiparsky, and
Koutsoudas et al. tell us) take precedence over general rules, when the child
learns that the past tense of do is did, he will automatically stop saying dood
regardless of whether he ever learned that dood is unacceptable. The child is thus
in an important respect not a little linguist: negative data provide much of the
factual basis for argumentation by big lingUists, but little of the data that the
child has available for use in learning a grammar.
In this paper, I will present some highly speculative suggestions relating to
the more general question raised by Braine and Baker, that of how an account
of language acquisition can be given that makes realistic assumptions about
the nature of the data that the child uses and about what the child can do with
those data at the time at which they are available to him.
The dilemma that Braine and Baker wrestle with emerges from Gold's (1967)
theorems about the learnability of languages on the basis of positive together
with negative data or on the basis of positive data alone. Gold's theorems involve
361
362 J. MCCAWLEY
.punsters possess that ability, and a plausible case can be made that it represents
a pathological condition. In any event, an ability to make reliable
'grammaticality judgements', as they are usually conceived, is of such rarity that
its relationship to general questions of linguistic competence is about as
tangential as that of, say, Broca's aphasia.
The bulk of this paper will be concerned with the acquisition of languages in
the ordinary sense, not with the acquisition of 'languages' in the technical sense.
If it should turn out, through what I would regard as a bizarre fluke, that
acquisition of a language in the ordinary sense has as a side effect the acquisition
of a 'language' in the technical sense, that result would certainly be amusing,
though it is not clear that it would be of any importance. Inter alia, I regard
the question of whether a child acquires a 'language' in the technical sense as
completely independent of the question of whether the linguistic competence
that he acquires has, loosely speaking, the form of a transformational grammar.
I will in fact take as a starting point the assumption that linguistic competence
has, loosely speaking, that form, in the sense that it involves surface structures
that have the form of ordered labeled trees, underlying structures that are also
of that general form (with the qualification that I do not wish to assume that
left-to-right order plays any role in underlying structure), and rules relating
structural configurations in underlying and surface structures. This leaves open a
great number of questions about the details of linguistic knowledge and
linguistic structure, and I will argue below that some of those questions can be
answered on the basis of considerations of language acquisition, that is, certain
answers to these questions but not others can be made on the basis of a scenario
for language acquisition that is plausible in relation to what is known about
developmental stages, individual variation, and data available t(') the learner.
Transformational grammars are generally taken as involving both a system of
transformations, which specify how the different stages of a syntactic derivation
are related, and a two-part system of base rules that specify what deep structures
are possible. The two parts that comprise the base rules are the phrase-structure
rules, which specify what gross combinations are possible in the deep structures
of the given language (e.g. the rule S ..... NP VP specifies that an item of category
S may consist of an item of category NP followed by an item of category VP),
and the lexicon, which specifies what lexical items the language has and what
restrictions there are on the occurrence of each item in deep structure (e.g.
put must be followed by a NP and a directional PP or adverb). A tacit (and
gratuitous) assumption has always been made that the GROSS
364 J. MCCAWLEY
a member (be) of the lexical class V followed by something of the form V, and
would do so even though he has not yet learned rules that tell him what passive
clauses are possible 4 :
For the child to execute this scenario, he will have to identify the lexical
categories of the relevant items. While I have no clear picture of how a child
would do that, I am reasonably sure of the following things: (i) lexical
categories are 'fuzzy', and peripheral members are assigned to them 011 the basis
of the same sorts of principles that determine hO'.v cognitive categories in general
are extended to take in peripheral members (Rosch et al. 1976), (ii) morphology
plays a major role in the identification of lexical categories, i.e. the participation
of an item in the inflectional paradigm corresponding to N, V, or A is a
SUF FICIENT condition for assigning it to that lexical category, and (iii) the
cores of lexical categories can be characterized in semantic as well as in morpho-
logical terms.
On the basis of this picture of how the acquisition of syntax might work,
I tentatively conclude that the acquisition of syntactic categories can be to a
large extent a by-product of the learning of word meanings and of morphology
and that even incomplete learning of syntactic category distinctions is sufficient
to give a child the means to learn rules of surface well-formedness that can then
be put to work in the analysis of more complicated syntactic constructions.
This conception of categories and rules of surface well-formedness allows for
incomplete knowledge about the well-formedness of a particular possible surface
structure, e.g. it makes available to the child a surface structure and a semantic
interpretation for combinations of a given verb with a following V even if the
child does not know what form the verb requires the V to be in (whether
infinitive with to, infinitive without to, present participle, or past participle)
or whether a given V can be used in that form. I lay particular emphasis on this
point because I think that in a number of cases acquisition via a stage of in-
complete learning provides the most reasonable account of a linguistic
368 J. MCCAWLEY
phenomenon. For example, I conjecture that Finnish children first learn that
there are environments in which consonant alternations take place and only
later learn what exactly those alternations areS.
For a further example, consider the acquisition of auxiliary verbs in English.
Suppose that the child has developed to the stage of having a surface
configuration [V Vl Y' As soon as the child identifies the various auxiliary verbs
as verbs, he will, under my hypothesis of maximal exploitation of surface con-
figurations, interpret them as forming V's with the immediately following V,
thus yielding the right-branching surface constituent structure argued for in Ross
1969 6 . Note that this surface structure will be availahle to the child
even if he has not yet acquired knowledge allowing him to judge what sequences
of auxiliary verbs are possible. Furthermore, if he has acquired the
transf<?rmation of Raising (or an equivalent semantic interpretation rule) that
associates certain [V Vl y configurations with semantic structures in which the
lower V is in a sentential subject of the V, and if he can assign to the auxiliary
verbs meanings involving sentential subjects, he is then in a position to under-
stand sentences involving auxiliary verbs piled to any depth. Limber (1973 :
177) observes that 'most auxiliary verbs appear several months after instances
of the complement verbs'. This is exactly what I would expect, given that I
hold (cf. fn. 6) that auxiliary verbs are peripheral members of a fuzzy category
'verb' : a given surface configuration should first be learned for core instances
of the relevant categories and then extended to peripheral members. According
to Limber, the principal exception to his generalization is the early appearance
of modal auxiliaries in yes-no questions such as Can I knock it over? The
v
exceptions thus do not involve the surface configuration [V \fl and hence
should be acqUirable independently of combinations involving that
configuration 7 .
So far I have taken up only surface well-formedness conditions that have
the general form of context-free phrase structure rules: rules that specify what
sequences of syntactic categories can be the daughters of a node of a given
category. There clearly are surface well-formedness conditions of other types,
such as the constraint (Ross 1972) against 'double -ing' constructions such as
*They disapprove of your stopping working, and the constraint in Dutch and
German that a verb is in second position in a 'root' clause but otherwise is final
in its V. 8 I have not yet carried out the obvious project of making a catalog of
well-supported surface constraints and sketching possible scenarios for their
acquisition. I will, though, take up briefly one remaining detail of the 'phrase
PLAUSIBILITY IN THEORIES OF LANGUAGE ACQUISITION 369
e.g. failures in the child's attempts to assign the sentence a satisfactory inter-
pretation in which the it is meaningful. In any case, here semantic data that do
not fit comfortably into the positive/negative dichotomy can do work that
might otherwise be thought to require strictly negative data.
While the scenario of language acquisition that I have sketched above assigns
a tremendous role to positive data (e.g. a child learns a surface phrase-structure
rule V ~ V V simply by identifying instances of the structure [V V]y), I
conjecture that the learning of meanings, of the rules for the correspondence
between semantic structure and surface structure, and of contextual restrictions
on linguistic structures involve data that have both positive and negative aspects.
Moreover, I think that lingUists have been overly hasty in giving up on purely
negative data as a significant factor in language acquisition, their undue haste
being the result of the policy of identifying data with grammaticality data and
of regarding each grammaticality datum as serving only to cause one sentence
to be included in or excluded from the set of sentences to be generated. In an
alternative conception of the function of data in language acquisition, data can
affect the status of a (tentatively learned) rule and thereby affect the role to be
played by subsequent data relevant to that rule; for example, learning that one
particular predicate does not allow Raising or Dative-movement could cause the
learner to classify it as a 'governed rule' and thereafter apply it only with
predicates for which he had learned from positive data that it was applicable.
Note that in that case a little negative data can go a long way: one negative
datum would render unnecessary the scores of negative data that Baker (1979)
sees as being required if a child is to learn to apply a Dative-movement trans-
formation only to the right verbs.
But let us turn to the status of positive data. In an earlier paper (McCawley
1976), I raised doubts about the standard conception of positive data. I noted
that the child's language acquisition is based directly not on what people around
him are actually saying but on his perceptions of those utterances, that his
perception works in terms of his current linguistic system, and thus that he may
systematically misidentify or even fail to perceive elements that are present in
adults' speech II. This suggests to me that to the mind of an 18-month old child,
adults are speaking baby talk but with a severe speech defect, manifested in the
interruption of normal speech by non-speech sounds such as I~~], [:>v], and
liz], in which case the child may go through a stage in which such sentences
as Cookie on plate are not only what his grammar generates but are indeed
primary linguistic data for him - they are adult speech as he perceives it. But
PLAUSIBILITY IN THEORIES OF LANGUAGE ACQUISITION 371
somewhere along the line his linguistic competence changes in such a way that
many of these 'positive data' become 'ungrammatical' for him.
In my 1976 paper I took this as showing that the set of data that the child
uses in language acquisition do not form a 'corpus' - the child can't be
constructing a grammar to fit the whole set of data that he has experienced,
since those data do not all relate to the same language (even if all the ad ult
speech around him has been confined to a single dialect). While I still accept that
conclusion, I now realize that one could treat the earlier 'grammaticality' of
Cookie on plate as not in conflict with the later ungrammaticality of Cookie
on plate by distinguishing properly between negative information and mere
lack of information. Let us take the various 'data' as involving not strings of
words or even whole syntactic structures but rather fragments of linguistic
structures. Suppose that the child learns that a noun may be the head of a NP
at an early stage of his development but does not learn until later what other
constituents a NP mayor must have (or whether the head noun is oJligatorily
present). At the earlier stage, it is not really correct to say that Cookie on plate
is grammatical or ungrammatical as a whole. There are only respects in which it
is grammatical (like its consisting of a NP followed by a predicate expression) 12
and respects in which it is indeterminate as to grammaticality (the child's
grammar might at this point be noncommital as to whether a NP requires any
constituents beyond a noun or whether the predicate phrase of a main clause
requires a verb as its head). While a major part of his language acquisition will
be the acquisition of rules of surface well-formedness, it will not be until a
relatively advanced stage of his linguistic development, if even then, that those
rules will exhaustively partition the class of possible surface structures into a set
of (superficially) well-formed ones and a set of ill-formed ones. I in fact reject
an assumption that I hitherto have appeared to acquiesce in, namely that incom-
pleteness in the coverage of the surface well-formedness conditions is ultimately
weeded out in the course of linguistic development; as far as I can see, it is of no
consequence whether such completeness is ever achieved.
I wish to draw two principal morals from this highly speCUlative and in-
conclusive ,discussion of language acquisition. First, the learning of rules of
surface well-formed ness (in contrast to what would presumably be required to
learn 'base rules' as normally understood in transformational grammar) can be
accomplished on the basis of information that the child acquires in the process
of learning other parts of the grammar, in particular, learning the lexical
categories and meanings of the various lexical items and the correspondance
372 J. MCCAWLEY
FOOTNOTES
SIt may be appropriate to formulate the Finnish consonant gradation rule using
curly brackets in which the different alternations are simply listed. The curly
brackets, however, are then not an 'abbreviatory device', since the child learns
the curly brackets before he learns what goes inside them and indeed never
learns the 'rules' that they supposedly abbreviate.
6A similar structure is proposed in Akmajian, Steele, and Wasow 1979 except
that there the different YP's are assigned to different categories (y3, y2, Y 1),
modal auxiliaries are held to be daughters not of a yi node but of the S-node,
and passive be is treated as a sister rather than an aunt of the main verb. I should
qualify my glib reference to a lexical category 'verb' that includes modals. The
defective morphology of modal auxiliary verbs reduces considerably the
morphological basis for identifying them as verbs. Not only do modals (other
than is to) normally exhibit no agreement with the subject, but most of them do
not even have a past tense: must, may, might, should, and for many speakers
also ought and modal need, have only a present tense form. While the child is
thus under far less pressure to identify must as a verb (and not as, say, an adverb
or something unclassifiable) than he is to identify kiss or throw as a verb, if he
hits on the idea of treating it as a verb it will fit into that lexical category quite
well, since it occurs in a verb position (namely as left sister of a V), it fits In to
the same surface configurations (including 'inverted' constructions) as do have
and be, whose morphological status as verbs is clear, it is the only item
identifiable as the .tense bearer of clauses in which it occurs, and it conforms to
the same rule for the relation of semantic structure to surface structure as do
such verbs as seem. However, it would not surprise me if there were consider-
able individual variation as to what modal auxiliary verbs different children
identify as verbs.
7Moreover, the child may well not yet have identified the verbs in question
as belonging to the lexical category Y when he begins using them in apparent
yes-no questions. Cf. Braine's (1976 : 46-47) discussion of a 22 month old child
for whom initial can-I was simply a force indicator, marking the utterance as a
request, e.g. Can-l fiX it 'Please fix it'.
8See Koster 1975 for reasons why the right generalization is 'final in its
Y' rather than 'final in its S '.
9 Since reading Bolinger 1977, I feel uneasy about calling anything semantical-
ly empty. Bolinger makes a persuasive case that most of the items alleged to be
semantically empty carry subtle but non-null semantic content. Even if Bolinger
is right, the notion 'semantically empty' may remain of relevance to language
acquisition. I conjecture that the items in question are initially learned as
semantically empty and that the learner subsequently finds semantic functions
for them to fulfill.
IOCL Chomsky'S (1981 : 9) suggestion that 'a not unreasonable acquisition
system can be devised with the operative principle that if certain structures or
rules fail to be exemplified in relatively simply expressions, where they would
be expected to be found, then a (possibly marked) option is selected excluding
them in the grammar, so that a kind of "negative evidence" can be available
even without corrections, adverse reactions, etc.'.
PLAUSIBILITY IN THEORIES OF LANGUAGE ACQUISITION 375
II Some weak support for this suggestion is found in Brown and Fraser's ( 1963)
report that children at age 2; 0 uniformly omitted articles from imitations of
adult sentences and children at age 2; 1.5 only sporadically produced articles
in the mimicry task. See, however, the study by Katz, Baker, and Macnamara
(1974), in which presence or absence of an indefinite article in phrases in which
IS-month old children were taught made-up words influenced whether they
interpreted the word as a common noun or as a proper noun.
l2Here 'predicate expression' means something of a category X. The child at this
stage need not have learned that main clauses require a tensed V and not just
any old X.
l3The existence of some sort of general-purpose learning mechanism is clear,
in view of the ability of human beings to learn esoteric perceptual tasks such
as chicken sexing and aircraft spotting (see Gibson 1969 for discussion of several
such tasks). It is many millenia to early for us to have genes specific to the
identification of aircraft.
ACKNOWLEDGEMENT
REFERENCES*
*Each reference is followed by the name of the citating author(s). (first letters).
377
378 REFERENCES
GUPTA A. 1980. The logic of common nouns. New Haven, Yale University
Press. MA
HARRIS Z. 1962. String Analysis of Sentence Structure. The Hague, Mouton,
Papers on Formal Linguistics 1. BL
HATCHER A. G. 1949. (Don't) buy but one. American Speech 24: 49-53. DO
HAYES B. 1976. The Semantic Nature of the Intervention Constraint. Lin-
guistic Inquiry 7.' 2, 371-376. KU
HAYS D. G. 1964. Dependency theory: a formalism and some other obser-
vations.Language40: 511-525. COL
HERINGER H. J. & STRECKER B. & WIMMER R. 1980. Syntax. Fragen,
Losungen, Alternativen. Miinchen, Fink. LA
HETZRON R. 1975. The presentative movement or why the ideal word order
is V.S.O.P. In C. Li (Ed.) Word order and word order change: 346-
388. Univ. of Texas Press. LA
HIGG INBOTHAM J. 1979. Pronouns and bound variables. Linguistic Inquiry
11 : 679-708. KO
HOARD J. 1979. On the semantic representation of oblique complements.
Language 55 : 319-332. LA
HOEKSTRA T. 1978. De status en plaats van het Indirekte Objekt. In J_ G.
Kooij(Ed.) : 40-69. AB
HOEKSTRA T. 1981. De theorie van funktie-argumentstruktuur. In Studies
voor Damsteegt : 78~-91. AB
HOEKSTRA T. & MOORTGAT M. 1979. Passief en het lexicon. Forum der
Letteren 20: 137-161. LA
HOEPELMANN J. 1976. Tense logic and the semantics of the Russian aspects.
Theoretical Linguistics 1 : 158-182. DA
HOEPELMANN J. 1978. A note on the treatment of the Russian aspects in a
Montague-grammar. In C. Rohrer (Ed.) 1978 : 49-98. DA
HOEPELMANN J. & ROHRER C. 1980. "Deja" et "encore" et les temps du
passe en fran9ais. In J. David & R. Martin (Eds.) : 119-143. DO
HOLANDA FERREIRA A. B. de. Novo dicionario da lingua portuguesa. Rio de
Janeiro, Nova Fronteira. DO
HORN L. 1969. A Presuppositional Analysis of "only" and "even". CLS 5 :
97-107. DO
HORN L. 1972. On the Semantic Properties of Logical Operators in English.
Doctoral dissertation, Univ. of California. DO
HUANG J. 1980. Move WH in a language without Wh-movement. Mimeo-
graphed, MIT. KO
HUDSON R. A. 1976. Arguments for a non-transformational grammar. Uni-
versity of Chicago Press. LA
HUDSON R. A. 1980a. Constituency and dependency. Linguistics 18: 179-
198. LA
HUDSON R. A. 1980b. A second attack on constituency: a rt::pJy to Dahl.
Linguistics 18 : 489--504. LA
JACKENDOFF R. S. 1972. Semantic Interpretation in Generative Grammar.
Cambridge, MIT Press. AB-DO-KU-MA
JACKENDOFF, R. S. 1975. On Belief-Contexts. Linguistic Inquiry 6 .' I, 53-
93. KU
REFERENCES 385