You are on page 1of 41

SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and

Linguistic Theory, 5:4 (1987:Dec.) p.561

KEN SAFIR

THE SYNTACTIC PROJECTION OF LEXICAL


THEMATIC STRUCTURE'"

1. INTRODUCTION
It has always been a fundamental task for syntactic theory to determine
what role our knowledge of the lexicon plays in our knowledge of syntax.
In recent years this task has taken on new urgency, especially in light of
many recent studies that suggest that syntactic principles are maximally
general, and that most of syntactic structure arises merely as a result of
the interaction between lexical specifications and general principles.
This paper is concerned with the slightly more specific, but no less
fundamental question of how thematic structure in the lexicon is mapped
onto syntactic structure. My main conclusion, based on evidence from
grammatical mapping in nominals, will be that the notion EXTERNAL
ARGUMENT cannot be defined independently of its structural context:
rather the external argument must be defined relative to an internal
argument or maximal projection. This phenomenon, which I will call
GRAMMATICAL FUNCTION (GF) RELATIVITY, and some of the theoretical
conclusions I will draw from it, will have important consequences for the
distribution of implicit arguments in nominals, and, more generally, for
Chomsky's (1981) Projection Principle. l
Towards the end (Section 7), I will emphasize the consequences of my
analysis for the representation of grammatical relations in general. It will
be argued that the results of this research favor the Government-
Binding approach to grammatical relations over theories such as Rela-
tional Grammar (see Perlmutter, 1983; Perlmutter and Rosen, 1984) and

• I would like to thank Frank Heny, Richard Kayne, Howard Lasnik, David Lebeaux, Joan
Maling, Luigi Rizzi, Tom Roeper, Beatrice Santorini, Ur Schlonsky, Tarald Taraldsen,
Esther Torrego and three NLLT reviewers for useful discussion and commentary. I have
also benefited from the audiences at the Conference on Comparative Grammar at Prin-
ceton in March, 1986, at the Conference on Language and Communication at Syracuse
UniverSity in April, 1986, at the Lexicon Seminar at MIT in May, 1986, and at the Third
Conference on Comparative Germanic Syntax in Turku, Finland, in June, 1986, where
various stages and portions of this work have been presented. An early version of some of
these ideas, now much revised, appeared in Safir (1986a).
1 The nature of grammatical mapping in nominals is currently an area of vigorous ongoing
research, and so many of the sources cited in this article are currently unpublished. In so far
as is possible, I will try to provide the re ler with an adequate context for the proposals
cited from these sources.
Natural Language and Linguistic Theory 5 (1987) 561-601.
© 1987 by D. Reidel Publishing Company

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

562 KEN SAFIR

Lexical Functional Grammar (see Bresnan, 1982). In the latter two


theories, grammatical functions that map thematic relations onto struc-
tural positions are stated in the lexicon where grammatical functions such
as 'subject' and 'object' are assumed to be primitive and universal,
although syntactic instantiations of the primitive grammatical functions
may differ cross-linguistically. In Government-Binding Theory, however,
it is assumed that grammatical functions are not primitive, but rather are
constructed from the interactions of a variety of subtheories and lexical
specifications. In the final section I will argue that only a theory that
constructs grammatical relations, as in the Government-Binding ap-
proach, can account for the dependence of external arguments on
internal ones in an explanatory fashion.
Discussions of grammatical mapping tend to be fraught with covert
assumptions that leave researchers working in slightly different frame-
works talking past one another, so it is worthwhile to clarify some key
terms right at the outset. Two related notions that will figure prominently
in what follows are the relations PROJECTED and LINKED as stated in (1).
(1) An argument is
a. linked if it is mapped onto a structural position at D-struc-
ture,
b. projected if it is syntactically non-inert at. D-structure.
It follows that all linked arguments are also projected, but it will be
maintained that some arguments, namely, implicit arguments, are pro-
jected without being linked (this issue is not addressed until Section 4).
Linking of arguments corresponds to what most linguists mean by
grammatical mapping. For example, in a sentence like John examined the
fish, the theme of examine is linked to a syntactic position (sister of V) ty
the direct object relation (or internal argument, following Williams
(1981) and so the fish is understood to be the theme of examine.
Likewise the agent of examine is linked to the NP daughter of S by
Williams' external argument relation, which assigns a thematic role to an
NP which is outside of the maximal projection (in X terms) of the
predicate in question. Both arguments are linked at D-structure and so
both arguments are projected. This means that the same grammatical
relations must be expressed at every syntactic level, as required by the
Projection Principle. When a predicate projects both its internal and
external arguments, it has projected its full set of thematic roles, or its
full thematic array. Both linking and projection will be slightly
elaborated in later sections.
Although most of the other notions that serve to present the main

Copyright (c) 2004 Pro Quest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 563

issues are stated in the text, I shall also be assuming the general
framework of Government-Binding Theory (GB), especially with respect
to issues I do not address directly.
Given the terminology just introduced, the force of GF Relativity can
now be stated a little more precisely: the grammatical relation 'external
argument' must be reformulated so as to be defined relative to the
presence of a linked internal argument in a given structure. As evidence
for the existence of GF Relativity, it will be argued that only nominal
predicates that link internal arguments will be able to have true external
ones.
In order to establish this set of claims I must first show that there is a
syntactic effect that serves as a diagnostic for the presence or absence of
an external argument (and hence the full thematic array). Then I must
show that this effect is susceptible to manipulation based on whether or
not an internal argument is linked. To do this I need a syntactic context
where internal arguments are only optionally required, and it is for this
reason that the bulk of my discussion will focus on deverbal nominal
constructions.

2. ARGUMENT LINKING IN NOMINALS

The distribution of thematic roles in nominals raises an urgent question


for Chomsky's Projection Principle: does the Projection Principle ever
apply to the structure of nominals in the way it applies to the structure of
sentences? The obvious fact about nominals is that it is simply not
necessary for a deverbal nominal to express all of the thematic roles that
the corresponding verb requires. For example, the noun discussion
requires a theme argument only optionally, although the verb discuss
clearly requires a direct theme.
(2)a. John discussed * (the issue) for a long time.
b. John's discussion (of the issue) lasted a long time.
The question then arises as to whether the Projection Principle has any
relevance at all for nominals. If the Projection Principle is in force, it
should account for the distribution of traces - i.e., thematic roles must be
assigned to argument positions in the same way at every level, and so if
ari argument moves, it can only maintain its thematic role by association
with a trace in a grammatically linked position. Thus any account of the
grammatical mapping that takes place in nominals must examine the role
of NP-movement in nominals, sinr ~ NP-trace is only required where the
Projection Principle is in force.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

564 KEN SAFIR

In what follows, it will emerge that the distribution of internal


arguments in nominals will crucially determine the force of the Pro-
jection Principle with respect to all arguments of nominals.

2.1. Movement in Nominals and the Adjunct Restriction


There are at least two different accounts of the relationship between the
nominals in (3a, b).
(3)a. a photograph of John
b. John's photograph
One account is that of Chomsky (1970) in which the surface structure
position of John is derived by movement of that NP from postnominal
position. By this account, the S-structure of (3b) as in (4a) is analogous to
that of the gerundive passive in (4b).
(4)a. [John's; photograph e;]
b. [John's; being photographed e;]
The gerundive passive is generally assumed to have sentential structure,
and those who argue that (4a) is the correct representation of (3b)
generally assume that the genitives in (4a) and (4b) are both 'subjects'.
A different analysis has been proposed by Williams (1982), however,
who suggests that examples like (3b) are base generated as they appear
on the surface. He observes that the prenominal genitive can bear
various thematic relations to the nominal head. For example, John could
be the photographer or the owner of the photograph in (3b), but in (4b)
John can only be the theme of photographed. The fact that the
pre nominal genitive may bear various relations to the head does not,
however, preclude the possibility that the theme reading for John in (3b)
could be generated by movement, as in (4), even though the possessor
reading (among others) might arise in the manner Williams suggests.
Indeed Anderson (1983) defends precisely this analysis: the thematic
object of deverbal nouns is derived by movement, even though other
assignments of thematic roles -to prenominal genitive position may be
assigned directly.2 If Anderson's analysis is correct, we would expect,

2 Andenon (1983) demonstrates the independent necessity of having an s morpheme


count as a theta-role assigner for the 'possession' relation, which may be interpreted
variously as 'authorship', 'inalienable connection' or 'ownership', etc. She proposes that the
•s morpheme which can be lexically inserted at D-structure, assigns the possessive relation
in nominals like John's sugar, and presumably optionally in nominals like john's photograph
and John's lTeatment.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 565

given the Projection Principle, that a trace would be found in post-


nominal position when the prenominal genitive NP (PGNP) corresponds
to the object of the deverbal noun (i.e., the object of the corresponding
verb).
Some evidence bearing on whether or not nominals contain post-
nominal traces may be derived from a curious restriction on the dis-
tribution of adjectival adjunct predicates in English which is illustrated in
(5a-c).
(5)a. the photograph of John sick
b. Bill's photograph of John sick
c. *John's photograph sick
While the of-object John in both (5a) and (5b) can be modified by the
adjectival adjunct, neither the agent prenominal genitive in (5b) nor the
theme prenominal genitive in (5c) can be adjunct-modified. This pro-
perty of prenominal genitives does not hold of the subjects of gerunds
taking direct objects, which are known to have more verbal/sentential
properties.
(6)a. Bill's photographing John sick irritated us.
b. John's being photographed sick irritated us.
In (6a), either John or Bill can be understood to be the one sick, while in
(6b) the surface subject of a gerundive passive serves as the antecedent
for the adjunct predicate. The contrast between (5b, c) and (6) seems to
argue against the analysis of non-gerundive nominals in terms of move-
ment, because in (5c) the movement analysis ought to permit a post-
nominal trace to license the adjunct predicate, just as the of-NP does in
(5a, b).
Although I shall argue against movement in nominals on the basis of
the adjunct restriction evidenced in (5c), the restriction on adjunct
modification of PGNPs is not so simple, and we must understand its
subtleties before we can use it as a diagnostic for the presence of
postnominal traces or anything else. Notice that some nominals do allow
PGNPs to support postnominal adjectival adjuncts. Compare (7a-b) with
(5b).
(7)a. John's treatment of Bill naked started a riot.
b. Joe's discussion of this issue stoned created confusion.
In (7a), the adjunct naked can modify John as well as Bill, especially if
John is a doctor who likes to work in the buff. In (7b), Joe is clearly
modified by the adjunct, as there is no ambiguity possible. One possible

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

566 KEN SAFIR

reason for the difference between nominals like photograph, on the one
hand, and those like treatment and discussion, on the other may be that
the latter two nominals describe an event or process, whereas a pho-
tograph is not an event. If we consider only (5a, b) and (7a, b) we can
state at least a part of the restriction on PGNP modification as follows.
(8) An adjunct can modify a PGNP only if the nominal describes
an event or process.
I shall not have much to say about (8), except to note that adjuncts of the
sort occurring in (5) and (7) apparently have the force of conditionals,
and thus require the context of the event interpretation (cf. Pesetsky,
1982; Sproat, 1985; Roberts, 1985). So far, (8) says nothing about
postnominal traces.
Now let us focus for a moment on the interpretation of (7a) where Bill,
the theme, is adjunct modified. That such an interpretation is possible is
no surprise, since a postnominal NP also supports an adjunct in (5a, b).
Moreover, if (8) is generally a sufficient condition on adjunct
modification, then it ought to be possible for the PGNP to support an
adjunct, whether there is a postnominal trace or not, since examples like
(9a) clearly refer to a process or event: but (9a) is not a well-formed
reference to the treatment Bill received while he was naked. 3
(9)a. *Bill's treatment naked started a riot.
Under an analysis of (9a) that assumes movement from postnominal
position (Analysis A), we would expect the postnominal trace to facilitate
adjunct modification as in (9c), just as it should (but does not) in (9b)
(=5C).4
(9)b. *[[John's]j photograph [e;] [sick]]
c. *[[Bill's]j treatment [ej] [naked]]
But even if we assume that there is no movement in (9a), then we still
have to explain why that example is ungrammatical under either the

3 I shall not make use of result adjectival adjuncts, such as those recently discussed by
Carrier and Randall (1986), as these may be subject to different structural conditions and
lexical selection - all the examples used here are tempoOiry state adjectivii adjuncts. I have
chosen this test because it reveals the pattern of data that raises the relevant issues. It is
possible to choose adjuncts that do not produce such clear judgements, but this is, of
course, pointless. For example, it is possible to improve some of the unacceptable adjunct
interpretations by adding while as in JaM's lTealment while drunk. I do not know why this
improvement should come about.
4 I am not assuming here that John sick forms a small clause constituent in a photograph of
JaM sick, rather I assume the adjectival adjunct is a daughter of N' or N". See Safir (1984).

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXicAL STRUCTURE 567

interpretation where Bill is agent (Analysis B) or one where Bill is theme


without movement (Analysis C), given that (8) is satisfied.
Although at this point the adjunct restriction may seem quite complex,
it can nonetheless be reduced to a fairly simple generalization if we begin
by abandoning the view that the 'theme' interpretation of nominals like
those in (5c) and (9a) is derived by movement (i.e., if analysis A is
rejected). Then under either analysis B or C of (9a), the property that
distinguishes it from the readings available in (7a, b) is that in (7a, b) but
not in (5c)/(9), there is a postnominal argument NP, i.e., an. internal
argument of the deverbal nominal corresponding to the internal
argument of the corresponding verb. Thus the restriction on adjuncts
must be made sensitive to whether or not the nominal in question has an
internal argument NP in postnominal position, i.e., an of-NP. (Williams
(1981) among others also treats the of-object as the internal argument of
the nominal.)
The peculiar restriction on adjunct modification of PGNPs may now be
stated provisionally as in (10), which replaces (8).
(l0) The Adjunct Restriction
An adjunct can modify a PGNP only if:
a. the nominal describes an event or process (= 8) and
b. the nominal in question links an internal argument.
If the condition in (1 Ob) is correct, we now can use the success or failure
of aojunct modification of a PGNP as a diagnostic for the presence or
absence of an internal argument in a nominal. It appears that the
condition in {lOb) is related to a fairly wide set of superficially in-
dependent properties (see Finer and Roeper, 1983; Roeper, 1984,
1986a; Lebeaux, 1984; Pustejovsky, 1985 among others).
For example, the internal argument condition (lOb) also appears to be
the crucial factor determining thematic interpretation of the' PGNP in
examples like (11): if the internal argument is present, then the PGNP
cannot have anything but the external argument interpretation.
(11)a. Attila's destruction of the city
b. John's treatment of Bill
c. Our discussion of the issue
d. Ron's reorganization of the papers
Thus (lIb), for instance, cannot refer to the treatment of Bill that John
owns, and it is marginal, though not impossible, as a reference to the one
he prescribes. Rather the most transparent interpretation of (lIb) is as a
reference to the event of John treating BilL The nominal discussion

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

568 KEN SAFIR

allows for an interpretation in which we own a commentary by someone


who has discussed the issues, and we are talking about this thing that we
own, but under the event interpretation it is clear that the PGNP could
only refer to the discussants (cf. also Anderson, 1983, p. 5). This forced
thematic interpretation is symptomatic of the presence of a complete
thematic array.
One way of accounting for the external argument interpretations in
(11), is to simply state the generalization in (12).
(12) The PGNP is interpreted as the external argument of a
nominal N if N links its internal argument.
What (12) requires with respect to the examples in (11) is that if the
internal thematic role of a nominal is linked, i.e., mapped onto a
postnominal position (typically the of-argument), then the PGNP must be
interpreted as the external argument of the nominal predicate, and the
full thematic structure (both internal and external arguments) of the
predicate are linked. S
Now consider nominals like Bill's treatment under our assumption that
the PGNP has not been derived by movement, where there is thus no
internal argument, and where both agent and theme interpretations are
available for the PGNP. It is clear that (12) says nothing about such
cases. How are thematic roles assigned to the arguments? When the
internal argument is not linked, I shall assume that the interpretation of
the PGNP is then relatively free. This claim is already plausible for
examples like Bill's treatment, where the PGNP may be interpreted not
only as one of the roles in the thematic array of treatment (e.g., agent or
theme) but also as a possessor or author or any other pragmatically
accessible thematic role. (To satisfy the general requirements of the
theory, I will assume that any freely interpreted PGNP position - if one is
realized syntactically - counts as a theta-position with respect to the
Theta Criterion.)
This possibility of free thematic interpretation for the PGNP means
that the agent role can be interpreted by other means than the gram-
matical mapping imposed by (12); i.e., the agent role can be linked to the
thematic array of the nominal without being an external argument. The
differences between grammatical mapping via (12) and free thematic

S I have avoided discussion of nominals which have internal arguments other than
of-arguments, such as Don's dependence on Mary or Paul's participation in the group. The
empirical issues involved are complex. Theoretically speaking, the simplest result would be
if there were no difference between other PP complements of nominals and of-NPs.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 569

interpretation of the PGNP will be examined further in Sections 3, 4 and


6.
To summarize briefly, the Adjunct Restriction in (10) and the linking
condition in (12) have been introduced, both of which make crucial
reference to internal arguments of nominals. For examples where (12)
does not hold, an additional means of assigning thematic roles to PGNPs
has been introduced: free thematic interpretation. Notice that the
diagnostic value of (lOb) is enhanced by (12), since whenever (lOb)
holds, we may expect (12) to hold too. In fact (12), though I have
introduced it casually as an empirical generalization, is a provisional
version of the GF Relativity Hypothesis, and will playa crucial role in all
of my subsequent discussion.

2.2. Predicting the Distribution of Movement in Nominals


Up to this point, I have argued that (lOb) can capture the right general-
ization only if examples like (9a) have no post-nominal trace. It is now
possible to go a step further and exploit the linking condition in (12) to
provide an explanation of why nominals like Bill's treatment naked
cannot be derived by movement.
If the presence of a linked internal argument entails the presence of a
full thematic array, as required by (12), then we expect that the PGNP
position in such a nominal will have to be a theta-position at D-structure,
as it is at every level as a result of the Projection Principle. It then
follows that movement from the internal argument position to the PGNP
position should be excluded by the interaction of the Projection Principle
and the Theta Criterion, for there is no movement into a theta-position.
In effect, I am interpreting the derived nominal with an internal
argument as having the same thematic structure as an active verb, and as
such it projects an external theta-role onto the PGNP position if that
position is structurally realized in syntax.
But now consider what is predicted when the nominal is derived from
an unaccusative verb. If the nominal is ergative (i.e., has only an internal
argument) then there is no thematic role besides the internal one to
project; thus it ought to be possible to move the underlying object of
these intransitives to prenominal position, and if the resulting nominal
permits an event interpretation, then it should be possible for the PGNP
to be modified by an adjunct. This prediction is borne out, as illustrated
in (13).6

6 Pustejovsky (1985, p. 35) marks examples like John's arrival nude as ungrammatical, but
I believe this is the wrong interpretation of the data, especially when one compares it with

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

570 KEN SAFIR

(13}a. Andy's arrival drunk created controversy.


b. Ellen's entrance smashed embarassed her mother.
c. Rhett's return alive disappointed Scarlet.
d. Sam's (sudden) appearance naked shocked Sharon.
According to Burzio's (1986; p. 161) -er test, all of the intransitive verbs
related to the nominals in (13) are unaccusative ("ergative" in Burzio's
terms) - *a good arriver, *a good enterer, *a good returner, *a good
appearer - while non-ergative intransitives such as run and walk - a good
walker, a good runner - do not support the adjunct well, as illustrated in
(14).
(14}a. *John's walk drunk didn't clear his head.
b. *John's run naked shocked the neighbors.
Notice that there is nothing wrong with the nominals John's run or
John's walk indt:pendently - the appropriate interpretations could be
derived by free thematic interpretation, but the adjunct is only supported
by a true external argument licensed by an internal argument, as in
John's discussion of the issue drunk, or else by a postnominal trace, as in
(13).
Moreover, our conclusion (from (lOb}), that the Adjunct Restriction
on PGNPs is a diagnostic for the presence of a linked internal argument,
is now confirmed for postnominal trace as well as for of-NP. Successful
adjunct modification of a PGNP may now be reduced to either of two
cases: (i) when the PGNP binds a postnominal trace, it is really the trace
that is adjunct modified, or (ii) when the postnominal adjectival adjunct
modifies the PGNP and not the internal argument (i.e., an of-NP), in
which case the PGNP is always the external argument. The second clause
of the Adjunct Restriction, (lOb), as it interacts with (12), now provides a
diagnostic for whether or not the external argument function has been
mapped onto the PGNP. The ungrammaticality of examples like (14)
shows that in such cases the external argument function has not been
mapped onto the PGNP, even though agenthood has been mapped onto
this NP by free thematic interpretation. Similar remarks extend to agent
and theme interpretations of (9a).
If movement is possible in nominals, however, then we may ask

John's treatment nude or Bill's discussion nude. The interpretation of the data proposed
here is supported by data from German (Section 2.1) Dutch (Note 11) and Norwegian
(Note 14). See also Torrego (1986) and Giorgi (1986), who independently argue for the
existence of ergative nominals in Romance.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 571

whether the failure of raising in nominals, as illustrated in (15), is


consistent with what (12) predicts.
(15)a. *John's appearance to win
b. *Mary's likelihood to win
Notice that while predicates like appear and likely take internal senten-
tial arguments, there is nonetheless an important difference in inter-
pretation between the examples in (16) and the corresponding nominals
in (17).
(16)a. It appears (to me) that Irving is innocent.
b. It is likely that Willa will win.

(I7)a. The appearance (*to Bill) that Irving is innocent is his greatest
asset.
b. The likelihood that Willa will win.
While in (16a) the S that Irving was innocent is in relation to a dative
perceiver, the S in (I7a) describes the content of the appearance itself,
and the dative argument is impossible. A similar fact holds for the
contrast in interpretation between (I6b) and (I7b), where in (17b),
likelihood is not an assessment of a degree of probability, but rather a
label for the status of the proposition that follows. (See Stowell (1981) for
some closely related observations.) My claim here is that the Ss following
these deverbal nominals are not arguments of the nominals, but modifiers
of some sort. (Stowell calls them "appositives".) If so, then no thematic
role has been mapped onto them in spite of their structural position; they
cannot represent internal arguments. If there is no internal argument
linked in (15) and (17), then (12) does not hold. Finally, if (12) does not
hold, then the PGNPs in (15) will always be interpreted via free thematic
interpretation, which means John in (15a) and Mary in (15b) should be
arguments related to appearance and likelihood,' respectively, but no
plausible thematic interpretation is available. Moreover, since the PGNP
position is a theta-position, it cannot be a landing site for movement. 7
To summarize this section, the linking condition stated in (12) predicts
that movement to PGNP in nominals is possible only in ergative nominal
constructions8 (but d. Section 5). The correctness of this prediction has

7 The proposal in the text is not incompatible with Kayne's suggestion (1984, p. 142) that
N cannot be a proper governor across a clause boundary, and that therefore the trace of
raising in a nominal would be excluded by ECP.
8 This conclusion is at odds with Williams (1982), Higginbotham (1983), Rappaport
(1983), Zubizarreta (1986) and Grimshaw (1986) in that they argue that no movement is
allowed, but I also differ with Kayne (1984), Roeper (1984) and Anderson (1983) in that I

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

572 KEN SAFIR

been corroborated by the pattern of adjunct modification in accordance


with (lOb). The success of (12) in accounting for these facts suggests it
would be advantageous to derive it from general principles.

2.3. The Role of va


Before I attempt to derive some of the patterns of argument linking in
English nominals from UG, it is perhaps useful to anticipate a potential
objection. Is it not overambitious to assign a property to UG based on a
narrow set of phenomena found in a single language? There are two
sorts of answers to this objection, one theoretical, the other empirical.
The first answer is that a solution involving UG is implicated because
the syntactic knowledge involved could not be plausibly learned on the
basis of experience. For example, it is not at all obvious that the presence
of the of-argument should require any specific interpretation for the
PGNP. It is harder still to imagine what evidence would permit a child to
make the distinction between nominals that permit PGNPs to support
adjuncts and those that do not, since no direct evidence is available to
indicate what is not possible. So even if a child heard nominals like
John's arrival sick or Delia's discussion of the issue drunk, it would have
no positive evidence to the effect that nominals like *John's run drunk
and :* Delia's discussion drunk are impossible. Thus we would be obliged
to conclude that deep principles or parameters are involved, even if no
cross-linguistic evidence were available.
The second objection, the empirical one, is that the phenomena
discussed in Sections 2.1 and 2.2 are not confined to English. Indeed
nominals in German show the same sets of constrasts even though
German nominals have some slightly different structural properties.
The structure of nominals in German is quite similar to English in that
it has prenominal genitives and postnominal arguments corresponding to
the complement structure of the related verb. As in English, if the
internal argument is linked, then the PGNP is interpreted thematically in
accordance with (12). The interpretations of the examples in (18) cor-
respond almost exactly to their English counterparts in (11).9

do not assume that the city's destruction is derived by movement. See also Weinberg, Aoun,
Hornstein and Lightfoot (1985) for a movement analysis designed to account for the
missing patient readings for examples like that treatment of John's.
9 I would like to thank Beatrice Santorini for her help in formulating the relevant German
examples and eliciting the data in this section.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 573

(18)a. Attilas Zerstorung der Stadt


Attila-GEN destruction the-GEN city
b. Johns Behandlung des Patientenl? von Bill
John- GEN treatment the- GEN patients! of Bill
c. Unsere Erorterung des Problems
our discussion the- GEN problem- GEN
d. Rons Neuorganisierung der Akten
Ron-GEN reorganization the-GEN papers
One difference between English and German is that there is more than
one way for the internal argument of the nominal to be linked; it can
appear as a postnominal Genitive NP or as the object of the preposition
von. This enables us to set aside as contingent the details of how the
internal argument of a nominal is syntactically linked. For English, I have
assumed that the internal argument is the of-argument, while in German
more than one linking for the internal argument is available. 10 Neverthe-
less, as we shall see, the generalizations concerning the interpretation of
English and German nominals are very similar. Thus it ~ppears that as
long as there is a manner of expressing the internal argument in nominals
in both languages, the generalizations examined here are unaffected.
In German, the adjectival adjuncts that have served as an analytic
probe in English are never supported by even a postnominal NP as in
(19), but the same sort of tests can still be constructed using a pre-
positional phrase like in krankem Zustand meaning roughly 'in a sick
state'. Compare the German examples in (20) to their English counter-
parts in (5).
(19)a. *das Foto von John krank! krankem
the-NOM photo of John sick-NOM!sick-DAT
b. *Johnsj Foto krank;! krankes;! krankemj
John- GEN photo sick-NOM! sick- GEN! sick-DA T

10 The conditions under which the you-phrase is preferred to the postnominal genitive will
not concern us here although they are of some independent interest (see Curme (1922, pp.
476-485». The use of YOD+ DAT is common when the object is a proper name or ends in a
sibilant or often if the object is of indefinite (plural) number. Some speaken find the use of
both a prenominal and a postnominal Genitive to be awkward in some cases. I have not
looked into all of the conditions involved in the choice between the two possible in-
stantiations. See Curme pp. 476-485 for discussion of the Genitive in nominals and pp.
507-516 for discussion of Genitive with verbs and adjectives.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

574 KEN SAFIR

(20)a. das Foto von John; [in krankem Zustand];


b. Bills, Foto von Johnj [in krankem Zustand]j/.I
c. *Johns; Foto (in krankem Zustand];
Just as in English, a. thematic interpretation for the full nominal makes
support of the adjunct PP possible - compare (7a, b) with the German
counterparts in (21a, b) - but if the internal argument is not linked then
the adjunct modification fails, as in (21c) parallel to the English example
in (9a).
(21)a. Johns; Behandlung von Bill j [in nacktem Zustand]i/j
lohn- GEN treatment of Bill in naked state
fuhrte zu einem Aufstand
led to an-DAT uprising
b. Bills; Erorterung dieses Problems
Bill- GEN discussion these- GEN problems- GEN
[in betrunkenem Zustand]i
in drunken state
c. *Bills; Behandlung [in betrunkenem Zustand]i
Bill- GEN treatment in drunken state
fuhrte zu einem Aufstand
led to an-DAT uprising

Moreover, although the German ergative nominals in (22a, b) are ap-


parently somewhat less acceptable than their English counterparts in
(13a, b), the same contrast between ergative (22a, b) and non-ergative
intransitives (22c) (corresponding to (14a» is still found.

(22)a. Andy's Ankunft [in betrunkenem Zustand]


Andy- GEN arrival in drunken state
erregte ein Kontroverse
excited a controversy
b. Ellens Eintritt [in besoffenem
Ellen-GEN entrance in (roaring) drunken
Zustand] brachte ihre Mutter
state brought her mother
in Verlegenheit
in embarassment

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 575

c. *Johns Spaziergang [in betrunkenem Zustand]


lohn- GEN walk in drunken state
lies ihn nicht klarer denken
allowed him not clearer to- think
The cross-linguistic nature of the phenomenon confirms our conjec-
ture that deep properties of grammar are likely· to underlie the pattern of
data involved. I I Naturally further studies are likely to fine-tune our
analysis of the central generalizations.

3. LEXICAL STRUCTURE AND ARGUMENT LINKING

We are now faced with an intriguing question. Why should the structural
presence of the internal argument be crucial to the presence of a full
thematic array, and hence a completely thematic interpretation of the
PGNP? Conversely, why should a thematic interpretation be more
idiosyncratic when the internal argument is not represented? The initial
distinction between linking of an external argument under (12) (where
successful adjunct modification is a diagnostic) and linking by free
thematic interpretation suggests how I will go about it, but it is time for a
more concrete proposal.
The central observation, or at least the one that I think provides the
key to these problems, is that the projection of internal arguments
determines the way that the external argument is projected. This sug-
gests that the theory of lexical representation should not specify how
external arguments are linked in syntactic structure, as this should follow
from linking of internal arguments. A theory of lexical representation
with almost exactly this property has been proposed by Hale and Keyser
(1985) (see also Hale (1983».

\I A Dutch reviewer finds that Dutch examples corresponding to those in (21) also show
the effect, but that the effect is much weaker, though still present in (22a, b). In particular,
(s)he remarks that with a PP adjunct as in eeD wadeIiDg ill dro.... eD toestaad is
0llUllle_ 'a walk in drunk condition is unpleasant' is "quite OK". But the reviewer
adds: "Although bare adjectival adjuncts are as impossible in Dutch as they are in German,
they are quite good when the adjective has a complement. So, J _ .....aadeliDI YO Wim,
droDkeD Y" de Y. cockt", veroorzatte veei _ 'John's treatment of Bill, drunk
[from] the many cocktails, created a lot of turmoil' is OK, J .... 1Je1ludeliD&. dro....eD YO
de vele cock tills 'Johns treatment, drunk [from] the many cocktails' is definitely a star.
Similarly EUe_ UDkcn.t, droDkea YU de veIe cocUds, brac:bt IuIar moeder ill YeriegeD-
heid is wellformed, 18_ "lIDdeling, dronteD Y" de vele cocktails, Hacbt geeD klaarbeid
in zijn geest 'John's walk:, drunk [from] the many cocktails, did not bring clarity to his
mind' is clearly bad and merits a star ... Hence, for this type of sentence, the predictions
the proposal makes are born out."

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Klnwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

576 KEN SAFIR

Hale and Keyser (henceforth, H&K), propose that the lexical thematic
structure of a verb like cut should only specify that cut requires linking of
its theme argument to an internal position, and that it has an agent. Their
lexical representation of cut is illustrated in (23), where X is the agent
and Y is the theme as stated in the lexical conceptual structure (LCS)
below the tree diagram.
(23) cutv v
~
V NPI

I
I

x Y
LCS: X produce a linear separation in the material integrity
of Y by sharp edge coming into contact with Y.
The tree representation in the lexical entry, which H&K call the "lexical
structure", is mapped onto a syntactic structure at the level of lexical
insertion, i.e., D-structure (H&K assume the existence of NP-movement
in a number of constructions). In particular, the dashed line in (23)
implies that the relevant NP in the lexical structure is mapped onto an
NP in the D-structure. This mapping is more precisely what is meant by
LINKING of an internal argument. Some general procedure, the one that
defines how external arguments are assigned (see below), should then link
the lexically unlinked argument onto the subject position in syntactic
structure. The separate statement of the meaning of the word cut is not
necessarily itself linguistic, as it has to do with the concept that the word
cut stands for, not its syntactic instantiation.
To be slightly more precise, let us assume (24).

(24) If a predicate P has a lexical structure, then the lexical


structure of P consists of the first branching X projection
above P or the maximal projection of P elsewhere.

This means that the lexical structure of a verb like cut is V because cut
selects a sister to V, but for verbs like run, which assign no internal
argument, the lexical structure is the unbranching maximal VP. It is
assumed that the relevant X projections for a predicate of type X can be
expressed in the lexicon.
The three key properties of Hale and Keyser's representations that I
will exploit are (i) aspects of grammatical mapping are represented in

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 577

configurational terms in lexical entries, (ii) only the linking of the internal
argument is specified in the lexical entry and (iii) the LCS is not
necessarily affected by an adjustment in the way that arguments are
linked or thematic arrays are represented. Property (iii) makes it possible
to suggest that one may appeal pragmatically to the lexical conceptual
structure of a predicate and bypass its lexical argument structure - which
is exactly what I propose to be the content of free thematic interpretation
of the PGNP.

3.1. Linking in Nominals


Turning now to nominals, I presume that the lexical entry for a nominal
like investigation will be as in (25).
(25) investigationN [N NP
I
I
I
I
x Y
LCS: The activity of X investigating Y.
I shall treat this entry as if it were underived, although obviously
derivational morphology is added to the verbal stem of investigate and
the argument structure of the verb is inherited by the derived form.12
The key property is that investigation is a nominal capable of having a
full thematic array corresponding to the related verb, and it is the
conditions under which this is possible that concern us here.
Now let us reconsider the notion EXTERNAL ARGUMENT. Up to now
we have assumed that only the linking of the internal argument is
represented in lexical structure, while the presence of the remaining
argument is simply noted. The leftover argument can now be defined as
the external argument only if the definition in (26) is met.

12 I ignore the question of how a derived nominal inherits the argument structure of a
related verb. See Sproat (1985), Randall (1984), Roberts (1985), Selkirk (1982) and Roeper
(I986a) among many others, for some recent discussion and references. I shall concentrate
on those affixes that allow complete thematic inheritance, such as -lion, -ance, -ai, -ment,
etc. I assume (contra Sproat) that the pattern of internal argument linking is not a property
of nominal affixes but a property of nouns, as all affixes that permit full thematic inheritance
act in this way, provided they permit the requisite interpretation of eventhood with respect
to supporting adjuncts (cf. (8». Thus the constraints on affix types that add or delete
arguments are not relevant here (but see especially Roeper (1986a) for discussion). For the
same reasons I will not assume that implicit arguments are generally properties of affixes, or
affixes interpreted as A-positions as has been suggested in various forms by Jaeggli (1985),
Fabb (1984), and Roberts (1985).

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

578 KEN SAFIR

(26) The external argument of a lexical predicate P is the unique


argument of P that is assigned outside the lexical structure of
P.
The term 'outside' here is crucially relative. It follows from (26) that if P
does not have a branching lexical structure in its lexical entry, then no
external argument can be defined for P inside the maximal projection of
P. For a predicate that has a lexical structure, moreover, an external
argument is defined in syntax only if lexical structure is linked to syntactic
structure. The key result obtained by the application of (26) to the H&K
lexical structures may be stated as in (27).
(27) GF Relativity
The external argument can be defined in syntax only when the
lexical structure is linked.
It is now also possible to state very clearly when the Projection
Principle must be satisfied:
(28) If the lexical structure of a predicate P is projected at D-
structure, then the arguments of P must be projected at every
syntactic level.
If a lexical predicate does not project its full lexical structure, then it will
follow that the Projection Principle is not invoked. It thus follows that
the nominals in (29) do not violate the Projection Principle since the
lexical structure of examination has not been projected at all. 13
(29)a. The examination was terrible.
b. John's examination was terrible.
Since the external argument is not defined for the nominals in (29), it
follows that the nominal can either be without arguments altogether as in
(29a), or, if a PGNP has been generated as in (29b), then that PGNP can
receive a theta-role by free thematic interpretation. The theta-role

13 An interesting proposal made in Lebeaux (1984) also predicts the dependency of the
internal argument on the external one in nominals. Lebeaux suggests that nominalizing
affixes can raise in logical form (cf. Pesetsky (1985) on affix raising). Affix raising to N will
mean that the selectional properties of the verb will be uncovered, and the internal
argument is required to be present (i.e., as required by the verb without the affix). Affix
Raising to N permits the '~ubject' role to be missing. Affix Raising to the highest NP bar
level allows the full thematic structure to appear, including the PGNP (presumably as
subject). This does not derive (12), however, as Lebeaux's account would allow for a case
where affix movement is only to N and yet the PGNP is not interpreted as an argument of
the head with the internal argument that follows.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 579

assigned by free thematic interpretation to the PGNP may be selected


either from the LCS of examination or on the basis of any other salient
interpretation, e.g., John can be interpreted as agent or theme, owner or
even author of the nominal examination. 14 If, however, the lexical
structure is linked (where John is the internal argument of examination),
then the PGNP is interpreted as a thematic external argument position,
and, as discussed earlier, movement is excluded by the Theta Criterion.
Now that the definition of the external argument has been introduced
in relative terms, it is possible to be more precise about the consequences
of this definition in the simpler cases. In a sentence, for example, the
external argument will be the NP that is a sister to VP, as discussed in
Safir (1985, p. 74), because the sister of VP is outside of VP, hence
outside of the lexical structure for the V in question. Thus the unassigned
argument of cut will be mapped onto the subject position, and an
ergative intransitive like arrive will map no thematic argument onto an
external argument position. As will be pointed out in Section 7, this is not

14 For a large class of deverbal nominals, the theme (patient) interpretation is not only

available for the PGNP but preferred when the internal argument is not linked, as in John's
destnu:lion. My account of free thematic interpretation does not capture this fact. Various
accounts of thematic hierarchies such as those of Randall (1984) and Pustejovsky (1985)
have been proposed to predict the distribution of this interpretation, but see Sproat (1985)
and Levin and Rappaport (1985) for critiques of this approach. The preference for the
theme reading in examples like John's destnu:lion seems to be a property of interpretation
that varies across languages, however, and thus may indeed be determined by factors
independent of what determines the general pattern of data examined in this essay.
Consider the following data from Norwegian (provided by Tarald Taraldsen).
(i) Bill'sl behandling av John! syk l,.!
Bill's treatment of John sick
(ii) *John's behandling syk
John's treatment sick
(iii) John's ankomst syk
John's arrival sick
(iv) *John's lop syk
John's run sick
Although Norwegian appears to show the same class of Adjunct Restriction distinctions
that English and German do, it contrasts with English in that the Norwegian PGNP cannot
in general have the theme interpretation corresponding to the object of the related verb.
For example, (ii) would not be improved if syk were missing, at least not under the patient
reading for the PGNP, although the agent reading is possible.
Similar issues arise for examples such as the Chinese invasion versus China's invasion. A
reviewer points out that Kayne (1984) accounts for the lack of a theme reading for the
Chinese invasion by assuming that the adjective would not bind a trace, and that this
account is not available to me, since I do not assume a trace is available in China's
invasion.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

580 KEN SAFIR

the same thing as identifying the [NP, S] position as 'subject' directly, as it


is not part of the [NP, S] specification to require that the subject be out-
side of anything.
The latter distinction may be highlighted by considering an intransitive
like run, which has no branching lexical structure at all under this
account, as it has no internal argument. Run may still have an external
argument assigned outside of its maximal projection, however, as in the
case of the [NP,S] position. Thus run may have an external argument
without having a lexical internal argument, and this possibility contrasts
the [NP,S] position with the PGNP position in nominals. By contrast, the
PGNP position is not external to the maximal projection of the noun run
and the noun run has no lexical structure by virtue of which its single
argument could be defined as external. Nominals like John's run can
therefore only have free thematic interpretation by appeal to the LCS of
run. John gets a formal theta-role - to satisfy the Theta Criterion - from
the possessive marker in this case, but the content of that role is filled in
by free thematic interpretation. It is precisely this distinction between the
noun run and the verb run that the adjectival adjuncts are sensitive to
(d. (13) versus (14), (2Ia, b) versus (21c), (22a, b) versus (22c)).
Thus, due to the fact that the PGNP position is internal to the maximal
projection of the nominal, the PGNP can be defined as an external
argument position only by virtue of the linking of an internal argument.
Thus (12), repeated below, can now be derived from the interaction of
the Projection Principle with the definition of external argument.
(12) The PGNP is interpreted as the external argument of a
nominal N if N links its internal argument.
If the internal argument is dominated by a branching N induced by the
linked lexical structure, then the sister to N may be defined as the
external argument. The Projection Principle will then require that the full
thematic array be projected, and this will require the PGNP to be
interpreted as the external argument. But in what sense does the Pro-
jection Principle require the presence of a full thematic array when the
internal argument is linked and there is no PGNP which can be defined
as the external argument? This leads us to the next set of issues.

4. IMPLICIT ARGUMENTS

I turn now to what have been called "implicit arguments" in nominals,


which are essentially a form of 'missing' argument which the inter-
pretation of some nominals appears to imply (see Roeper, 1984, 1986a;

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 581

Williams, 1985; Chomsky, 1986a for recent discussion). As Chomsky


(1981) points out, not all nominals permit implicit arguments. NPs like
the book in (30b) do not have the interpretation 'arbitrary person's book',
parallel to the overt prenominal genitive in (30a), nor does (30b) have the
reading available for the gerundive nominal in (30c).
(30)a. John's book is interesting.
b. The book is interesting.
c. Owning the book is interesting.
While a possessor reading for the pre nominal genitive in (30a) is possible,
this reading is not available in (30b), though it is, in (30c). Considerations
like these led Aoun and Sportiche (1982) to propose a theory of
government that distinguished the two structures (31a, b) so that PRO
would be governed and hence excluded in (31 a) but not in (31 b).
(31)a. [NP [Det the] book]
b. [NP [s PRO [vp owing the book]]]
The same properties hold in nominals like (32a) and (32b), where both
nominals, unlike book, could potentially have thematic structure. IS
(32)a. *Treatment sick inspires others.
b. *A discussion stoned does not impress judges.
It is a consequence of the theory developed in the last section,
however, that the full thematic array of a deverbal nominal is projected if
the internal argument is linked. If the presence of adjunct modification is
diagnostic of the presence of linking with lexical structure, then when the
of-NP is present, it should suddenly become possible to construe an
adjunct with an implicit argument which, if it were overt, could appear in
prenominal genitive position - the external argument. The examples in
(33) bear out this prediction.

15 Tom Roeper (personal communication) points out examples like (i) and (ii) which appear
to be counterexamples to the claim in the text.
(i) a cure unencumbered by children.
(ii) E trip sick is no fun.
I do not think that (i) has the appropriate interpretation however, as unencumbered by
children appears to modify the cure rather than the one undergoing it. The second example
is more puzzling. I do not find it acceptable, but it does seem better than John's treatment
sick. Perhaps the underived status of trip contributes to this idiosyncratic behavior.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

582 KEN SAFIR

(33)a. Discussion of these issues stoned rarely produces satisfactory


results.
b. ?The treatment of diabetics stoned must stop.
c. *Photography of these trees stoned rarely produces satisfactory
results.
Example (33c) is particularly interesting in this regard, as the adjunct
interpretation is possible only if the nominal photography receives a
process reading (see Randall (1984)), as we would expect on the basis of
(8). The contrast between (32b) and (33a) is also evidenced in similar
German examples.

(34)a. Die Diskussion dieser Probleme in betrunkenem


the discussion these- GEN problems in drunken
Zustand fuhrt zu befreidigenden Ergebnissen
state lead to satisfactory results
b. *Eine Diskussion in betrunkenem Zustand
a discussion in drunken state
imponiert Richtern nicht
impresses judges not

Although up to now I have confined my discussion to adjectival


adjuncts, essentially the same set of contrasts emerges in nominals with
rationale clauses, as pointed out by Roeper and Finer (1983), Roeper
(1984, 1986), Pustejovsky (1985) and Chomsky (1986a) and exemplified
in (35).

(35)a. John's destruction of the city to prove a point


b. The destruction of the city to prove a point
c. The city's destruction to prove a point

The crucial factor here is whether or not there is any controller for the
PRO subject of the rationale clause. The overt PGNP serves as a
controller in (35a), where the internal argument is linked and the PGNP
is thus defined as the external argument; (35b) works the same way,
except that the external argument is implicit - yet it serves successfully as
a controller. No implicit controller is available in (35c), however. In our
terms this is expected because there can be no external argument where
the theme is the PGNP. If the theme is the PGNP then this can only have
arisen by free thematic interpretation, not argument projection, hence no

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 583

external argument is implicit, and control fails. 16 Moreover, the account


proposed here extends to the contrast in (36), where John is interpreted
as an agent.
(36)a. *John's destruction to prove a point will not influence Mary.
b. ?Destruction of the city to prove a point will not influence
Mary.
Once again, the key distinction is whether or not the lexical thematic
structure is linked - in (36a) it is not.
Two key points emerge from the data in (33-36). First, in each
example, the argument modified by the adjunct is unambiguously the one
that would be defined as the external one of verbs corresponding to these
nominals, thus indicating that the full thematic array of these nominals
has been projected. Second, the possible presence of an implicit
argument (hence the full thematic array) is predicted by the presence of
the linked internal argument. By contrast, free thematic interpretation
does not license implicit arguments because lexical structure is not linked
and so no grammatical function is implied.
Given that we now have a means of predicting the syntactic contexts in
which the implicit argument in nominals can appear, a question now
arises, as it has iri the literature, as to whether the implicit argument
should be linked - represented by a structurally present empty category
such as PRO - or merely projected, a possibility that has been suggested
but not been made quite precise.
If we assume that the representation of nominals with implicit
arguments is (37) for example, then the PGNP position is linked and the
Projection Principle applies.
(37) [NP [PRO] destruction of the city ...] ...
If this is the right approach, then it must be explained why the PRO is
only licensed when the internal argument is linked. Such a theory might
serve as an alternative to the Aoun & Sportiche (1982) theory of
government which is supposed to account for the difference between
(30b) and (30c). In such a theory, PRO is not licensed in (30b) by the
projection of a full thematic array, though it would be for (33a);

16 Notice that the argument for lexical structure goes through whether or not it is as-
sumed that the PGNP in (35a) (an implicit argument is (35c» is the controller of the
rationale clause in the grammatical cases, as Roeper (1984) assumes, or it is the whole
event that serves as controller for the rationale clause, as Lasnik (1985) and Williams
(1985) argue. The key factor is the presence of lexical structure as opposed to free thematic
interpretation.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

584 KEN SAFIR

government has nothing to do with it. Clearly, if the implicit argument of


nominals is PRO (as has been argued by Roeper (1986a, b), .among
others) then the claim that PRO is ungoverned is severely compromised,
or else government must be made sensitive to argument linking in a
rather complicated, and probably undesirable way. Although I do not
have any strong evidence that a theory of PRO with the right properties
could not be devised (an account based on government, or on the
projection of full thematic arrays, or whatever) I choose, for the sake of
argument, to develop an alternative account.
Suppose that the implicit arguments in (33), (35b) and (36b) are
projected but unlinked arguments. This means that they are defined
relative to linked internal arguments, but they are not mapped onto a
syntactic position - it is not necessary to invent an empty category to
which they correspond. One way to implement this proposal is to assume
that the projection of lexical.structure is sufficient to imply the presence
of the external argument, sinc!,! the external argument is defined in the
lexicon according to (26). Thus if the lexical structure is linked, then the
external argument is projected whether it is linked or not. The external
argument is quite literally 'implied'.
Under either approach (PRO or unlinked projected argument), the
fact that only external arguments can be implicit (a disputed inter-
pretation of these facts - see Section 6) can be accounted for within the
assumptions of Section 3. The crucial mechanism in both cases is the
triggering effect of internal argument linking, which provides for exter-
nal argument linking in the 'PRO account', and external argument
projection in the 'projected but unlinked argument' account. Whether
implicit arguments are PRO or not. GF Relativity will still hold, so the
question as to whether the implicit argument is PRO or not involves only
the legitimacy of projection without linking as proposed here.
Williams (1985) has argued that independent evidence indicates that
implicit arguments are not PRO on the basis of examples like those in
(38-40) (which are similar to examples pointed out by Anderson (1983».
(38) Yesterday's discussion of the issue stoned didn't clarify mat-
ters.
(39) ?Yesterday's arrival drunk compromised Mary's credibility.
(40) ??Yesterday's treatment of the patients drunk was a disgrace to
the profession.
The relative acceptability of (37a-c) creates a structural problem for the
PRO analysis that does not exist for the unlinked argument analysis

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 585

proposed in Williams (1985): there is nowhere to put the PROY


Moreover, the PGNP is still being interpreted within the full thematic
array (as required by the Projection Principle) here if, as Sproat (1985)
has suggested, the derived nominal has a bound event argument place
which the PGNP can modify.
Although I do not consider the matter settled (see Roeper, 1986b), let
us assume that the implicit argument is not PRO. In so far as the implicit
argument enters c-command relations, I will assume that it is a syntac-
tically active incorporated argument of the predicate that projects it, a
view similar to that of Williams (1985), who also treats the implicit
argument of a nominal as a property of the nominal head. Clearly
implicit arguments can act as controllers, but it may be that they cannot
be the antecedent of trace. In all of the above respects, the implicit
arguments discussed here closely parallel those discussed by Rizzi (1986).
Rizzi identifies the missing dative object of verbs as pro in Italian. It can
both be a controller of PRO and a binder of anaphors. But the missing
dative in English, which serves as a controller (in the right contexts), yet
fails as a binder, he identifies as an unsaturated argument: in effect a
projected but unlinked argument.
I differ crucially from Williams (1982, 1985), however, in that I have
established that the PGNP can be the external argument parallel to the
subject of a sentence. Moreover I have argued that the distribution of
implicit arguments in nominals thus derives from the definition of exter-
nal argument (27) and the treatment of the Projection Principle (28).
Finally, but less crucially, I have suggested that contrasts between the
properties of PRO and those of implicit arguments may be capturable by
the unlinked projected argument analysis, which I shall henceforth adopt.

5. 'PASSIVE' NOMINALS

This section is devoted to some loose ends in my account of movement


and argument linking in nominals. The question is whether external
argument linking in nominals can be extended to by-phrases, and, if so,
what the consequences are. The first relevant problem concerns the
derivation of PGNPs by movement.

17 In Safir (1986b), this argument is examined further on the basis of work by Clark (1985)

concerning the retroactive gerundive construction. Clark argues that the structure of these
gerunds is nominal, hence a PRO position is available in NP. By contrast, I argue that the
structure of these gerunds is sentential, and so the presence of PRO in these structures is
not an argument for PRO in nominal structure.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

586 KEN SAFIR

Up to this point, while I have argued that nominals such as (29) are not
derived by movement, no commitment has been made with regard to
'passive nominals' with by-phrases, such as those in (41).
(41)a. the examination of the patient by the doctor
b. the patient's examination by the doctor
The nominal examination is carefully chosen so as to comply with the
Affectedness Constraint pointed out by Anderson (1979) which limits the
ability of a theme to become a PGNP if it is not somehow changed or
'affected' by the action of the process nominal. Nominals like discussion,
for example, are more awkward, if possible at all (e.g., ??this issue's
discussion by the president). The key fact about (41a, b) is that the PGNP
must be interpreted as if it were the internal argument when the by-phrase
is present, and this is not independently the case for examples like (42):
(42) the patient's examination
Here, the patient could be agent. This suggests that the by-phrase is
linked to the argument structure of the nominal in some systematic way.
Moreover, as has been observed by many others (see Lasnik (1986) for
references), the by-object is incompatible with an agentive PGNP, as in
(43).
(43) Pete's destruction of the city (with Joe)/(*by Joe)

Unlike with-objects, the by-object cannot be added to the PGNP to


mean that both Pete and Joe destroyed the city. These incompatibilities
are expected if the by-object fills a thematic role with respect to the
nominal, and is thus subject to the Theta Criterion. It seems we must
conclude, that the by-object competes with the PGNP because they both
count as argument positions capable of satisfying the external argument
slot of a nominal with projected lexical structure (a linked internal
argument).
On the other hand, we have seen that nominals without by-phrases fail
the adjunct test, and so it was concluded that nominals like Bill's
treatment (with the theme interpretation), as in (9a), were not derived by
movement. The same test applied to appropriate nominals with a by-
phrase yields different results:
(44)a. Tom's treatment/examination by the doctor naked.
b. Tom's treatment/examination naked by the doctor.
(45)a. Tom was treated/examined by the doctor naked.
b. Tom use treated/examined naked by the doctor.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 587

It is true that the adjunct, where it is understood to modify the PGNP,


seems somewhat awkward both inside and outside the by-phrase;
however, this awkwardness also arises in sentential structure such as (45).
Perhaps the nominals are marginally worse than their sentential coun-
terparts, but the key point is that the examples in (44) are reported to be
noticeably better than examples like (9a). The facts appear to indicate
that the presence of the by-phrase licenses movement of the internal
argument to PGNP position. 18
In Section 3, movement was ruled out in examples like John's treat-
ment because the nominal was assumed to have linking properties similar
to those of an active transitive verb: movement to PGNP position from
postnominal position yields a Theta Criterion violation. In order for the
theory to permit such movement, the presence of the by-phrase must be
assumed to trigger some sort of dethematization of the external argument
position of the nominal, thereby freeing the PGNP position as a landing
site for movement. It is crucial to such a proposal, however, that when
the by-phrase is absent, there is a reason it cannot be implicit (projected
but not linked), or else we would expect movement in examples like the
city's destruction where the PGNP position is dethematized by leaving the
external argument unlinked to a syntactic NP position, as is routinely
possible in sentential passive (e.g., John was killed). Thus dethematiza-
tion in nominals, while similar to that induced by passive morphology,
must differ from it in crucial ways.
One distinction between nominal dethematization (NO) and passive
dethematization (PO) is that PO appears to be induced directly by
affixation while NO is not. After all, a variety of process/event nominal
types with affected objects permit the possibility of full thematic expres-
sion, including some nominals without overt affixes, and many of these
permit 'passive' nominals with by-phrases.
(46)a. Ron's apprehension by the police (?drunk) cost him his
license.
b. John's arrest by the authorities (?nude) took place at 3:00 a.m.
c. Tom's internment by the authorities (?nude) created a
scandal.
d. John's dismissal (from the hospital) by the doctor (??drunk)
led to litigation.

18 My own intuitions about the data in (44) and (45) are not quite clear and one reviewer
has the same reservations. I include these examples, however, because at three separate
presentations of this work, members of the audience proposed similar examples claiming a
distinction between (44a, b) and (9a).

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

588 KEN SAFIR

Rather than reducing this regularity of by-phrase linking to a coin-


cidence of affix properties, it seems necessary to treat by-phrase linking
in nominals as independent of specific affixes (ct. footnote 12). For
nominals, then, we may assume that the external argument, defined as the
theta-role assigned outside lexical structure, can be assigned to either a
by-phrase or to the PGNP position, but not to both, as a result of the
Theta Criterion. Thus when a by-phrase is present, and not interpreted
as instrument (ct. Section 7.1), the PGNP position is dethematized. If
there is no by-phrase, there is no dethematization, so the PGNP position,
if filled, is a theta-position.
Passive morphology, by contrast, does specifically dethematize in-
dependently of the presence of the by-phrase. 19 In this sense, one may
say that verbal passives involve lexical dethematization, an operation in
the lexicon, while dethematization in nominals is determined by syntactic
context.
There is some additional evidence to support the distinctions made
between sentential passives and 'passive' nominals. Consider the exam-
ples in (47).
(47)a. John's treatment lasted three hours.
b. John was treated for three hours.
John's treatment in (47) may mean the treatment that John received
without necessarily implying that anyone administered that treatment,
while the passive of the verb treat seems to have such an implied agency.
For example, while (47a) can describe the method by which John's chills
were overcome - perhaps by supine exposure to the sun for three hours,
(47b) cannot be used to describe the same situation - someone must have
administered treatment in (47b).
This point is reinforced by the systematic absence of adjectival ad-
juncts supported by an implicit internal argument when the PGNP is
theme. The passives in (48), where the adjunct can modify the implicit
agent, contrast sharply with the corresponding interpretation for the
nominals in (49), where no reading is available for the adjunct (in fact the
agent reading is even worse than the theme reading, if that is possible).20

19 Perhaps this distinction also accounts for the fact that nominal by-phrases can only be
agents or instruments in nominais, not experiencers. See Section 7.3.
20 I am assuming that the external argument of a passive can support an adjunct in certain
contexts, contrary to what is often assumed. The conditions under which such inter-
pretations are possible will not concern us here. The contrast between nominals and
passives may, however, be irrelevant if Lasnik (1986) and Williams (1985) are right in
claiming that rationale clauses are supported by the whole clause, not the implicit agent,

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 589

(48)a. ?Whenever a patient is treated drunk. the hospital should be


sued.
b. ?When an alien is examined stoned, strange perceptual effects
are observed.
(49)a .... A patient's treatment drunk almost always leads to his death.
b. ... An alien's examination stoned enables one to observe strange
perceptual effects.
The facts in (49) are unsurprising, given that the adjectival adjunct is
only possible where the lexical structure is projected - in which case the
theme reading for the PGNP (49) would violate the Theta Criterion (cf.
Section 2.2).
The upshot of this discussion is that a PGNP position in nominals with
by-phrases can be a landing site for NP-movement, but not a landing site
for movement in nominals without by-phrases. Once again, the adjunct
test provided evidence for the presence of the trace of movement. Thus
the city's destruction only can be derived by movement if the PGNP
position is dethematized. and in nominals only by-phrase linking is
available to achieve this. By contrast, the morphological affix of verbal
passives triggers dethematization of the external argument position as
well as permitting an implicit (unlinked but syntactically· non-inert)
argument.

6. FREE THEMATIC INTERPRETATION AND IMPLICIT


ARGUMENTS

A major prediction of the theory developed in the earlier sections is


that free thematic interpretation cannot license implicit arguments, but
that internal argument linking (and hence projection of a complete

and if adjectival adjuncts are like rationale clauses. Compare Roeper (1986a, 1986b). It is
not clear that the points raised by Williams (especially on pp. 310-315) and Lasnik extend
to rationale clauses in nominals, however. Compare (i), (ii) and (iii), where a playwright is
devising the plot.
(i) Andy arrives to provide an exciting climax.
(ii) Andy's arrival to provide an exciting climax left the audience unimpressed.
(iii) Andy's arrival provided an exciting climax.
While Andy's arrival can be a subject for provide in the appropriate sense in (iii), it fails to
support the interpretation of playwright agency in (ii). Rather it seems that in (ii), Andy
must be the understood controller of provide. Moreover, it is not clear that adjectival
adjuncts act at all like rationale clauses modifying sentences. For example, Andy's arrival
drunk does not mean that the arrival event was a drunken one, rather it means that Andy
was in a drunken state when he arrived.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

590 KEN SAFIR

thematic array) can. There are a number of contexts where it has been
suggested in the literature that implicit arguments are available which are
not consistent with the. latter prediction, and some of these contexts are
examined in this section.
For example, it is well known that picture NPs can support reflexives,
as in (50a), even though the same sort of NP does not allow its PGNP to
support an adjectival adjunct as in (SOb).
(50)a. Pictures of oneself nude can be upsetting to one's parents.
b. Paul'sj pictures of Addiej drunk j/. j.
The failure of the PGNP to support the adjectival adjunct, however, is
unsurprising, since a picture cannot be an event, and we have seen that
this is a necessary precondition for such adjuncts (as expressed in (8».
Thus it might be possible for us to assume that nouns like picture do have
thematic structure and an implicit lexical external argument, and that it is
present in (SOa) because the internal argument is linked. While I would
not wish to exclude the possibility that some non-event arguments might
have projected thematic structure, it is clear that such an analysis is not
at all consistent with the meaning of (50a). That sentence does not imply
that the pictures in question are owned or created by the person depic-
ted, and so there is no reason to suppose an implicit argument is licensing
these cases. Given that reflexives are normally bound and that the one in
(50a) is not, the grammaticality of the reflexive in (50a) remains a
mystery. The theory of implicit arguments does not solve this mystery,
but given the interpretation of examples like (50a), there is no reason to
assume that an implicit argument is present in such examples anyway.
One the other hand, there are contexts where assuming an implicit
argument provides a natural account, as in well-known cases like
(51a, b).
(51)a. Pretentious discussion of oneself may offend potential
employers.
b. Any attempt to leave causes trouble for the authorities.
In this instance, the reflexive is indeed related, perhaps bound, to the
implicit external argument of discussion. Similarly, the understood agent
of attempt can act as the implicit controller of the PRO subject of the
infinitive to leave, and this corresponds to the expected interpretation for
the external argument of attempt. 21

21It might be argued, as pointed out to me by Ken Hale (personal communication), that
onsel! appears more freely without a binder than himself does, and so freer appearance of

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 591

I will shortly consider some other purported cases of implicit


arguments, but first I must briefly describe another phenomenon that
serves as a useful diagnostic for the presence of a PGNP as external
argument. Pustejovsky (1985) discusses a class of extraction facts that
appear to be conditioned by whether (in my terms) a full thematic array
has been projected or not. He notes contrasts like those between (52) and
(53).
(52)a. ·What did John eat Bill's loaf of?
b. ·What did Mary drink John's bottles of?
c. ·What is John reading Bill's play about?
(53)a. Which theory did you read Kripke's proof of?
b. Whose book did you read Bill's comments on?
c. The opera that we saw the MET's performance of?
d. The city that I witnessed the enemy's destruction of?
Pustejovsky makes what I believe is the correct observation, however it
is instantiated theoretically: extraction across a PGNP is possible only if
the PGNP is interpreted as participating in the thematic structure of the
nominal. The success of extraction across a PGNP thus indicates that the
PGNP is the external argument.22 Note that if this is the right general-
ization, then it would appear that an event interpretation is not crucial in
order for the full thematic matrix to be present, as is clear with respect to
the nominals in (53a, b).
We can make use of the diagnostic just introduced to deepen the

oneself may not be a good diagnostic for the distribution of implicit argument antecedents.
On the other hand, insofar as. implicit arguments are often interpreted as arbitrary in
reference, they will not be appropriate antecedents for reflexives other than ontSelf.
22 See Pustejovsky for an extensive discussion and references. Torrego (1986) makes a
related set of observations about Spanish within the framework of Chomsky t1986b). She
attempts to relate the contrast between internal argument extraction across thematic
genitive NPs versus possessive genitive NPs to a difference between subjacency violations
versus ECP violations. Issues of how arguments are linked are not emphasized in her
account, since her primary interest is extraction asymmetries, but she does suggest that
possessive genitives are classed with the determiner/specifier system in a way that blocks
extractions from NP, while argument NPs do not act like specifier/determiners in the same
sense. This part of Torrego's analysis appears compatible with the proposals made here.
Some slightly different issues arise for the Romance languages that suggest areas for further
study. For example, there are no non-pronominal PGNPs, so external argument genitives
can appear postnominally. This will require more discussion of how internal arguments are
distinguished from external ones in lexical structure, but compatible structural assumptions
have been defended by Giorgi (1985) developing proposals by Cinque (1980). For an
analysis of Romance nominals along the lines of the countertheory in Section 7, see
Zubizarreta (1986).

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

592 KEN SAFIR

investigation of implicit arguments. Consider the question of whether the


thematic interpretation of nominals like discussion in (54) appeals to
lexically projected arguments or not. At least one interpetation of (54) is
that the proposal is what is to be discussed, hence proposal corresponds
to the internal argument of discussion.
(54) Connie contributed the proposal for discussion.
It does not seem that such examples require any syntactic relation,
however, but rather a pragmatic appeal to the lexical conceptual struc-
ture of the nominal discussion.
To see why an appeal to LCS is more plausible for (54), consider (55),
which might be part of a description of a pot luck dinner.
(55) Connie brought veal marsala for (our) dinner.
It seems reasonable to suppose that the implicit patient interpretation,
the one which leads us to expect that the veal marsala is to be dined
upon, is an aspect of similarity between nominals like dinner and dis-
cussion that is in direct contrast to their behavior in other domains.
However, the comparisons in (56) and (57) illustrate how different these
two types of nominals are.

(56)a. *John's discussion was the proposal.


b. Don's dinner was fish.
(57)a. *What sort of fish did you buy John's dinner of?
b. ?Which proposal did you respect John's discussion of?

Consider the semantic predication relations involved. The nominal Don' s


dinner of fish does not mean that Don dined on fish, but rather that Don's
dinner consisted of fish, as illustrated by the success of the same inter-
pretation in the copula in (56b). This (i.e., the interpretation of (56b» is
not a thematic interpretation, as can be seen by the comparison with
(56a), where the proposal is not predicated of discussion, but rather the
proposal is an argument of discussion. The relative success of extraction
across a PGNP in (57b) as opposed to (57a) suggests that the distinc-
tion between (56b) and (56a) is the same as that between (52) and (53),
which again implies that at some level the proposal is part of the
argument structure of discussion in (56a), while fish is not an
argument of dinner at any level.
Yet as different as these two nominals are, they do both allow an
implicit patient interpretation. It seems wrong to attribute this to a
similarity of syntactic argument projection, however, as it would be
necessary to treat dinner as having the same sort of theta-assigning

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 593

property as discussion, a position we already know to be false on the


basis of (56). Rather what is involved here is the notion that one does not
dine without eating, and one does not discuss without speaking of
something.
My argument against treating (54) as having the same internal
argument structure as (55) is that we can account for dissimilar syntactic
behavior between (54) and (55) by appealing to the differing pattern of
theta assignment to syntactic positions in the two cases. Though
difference in argument linking distinguishes these two types of nominals,
their similarity may lie in the entailment relations between an activity and
its consequence. Otherwise, to treat dinner and discussion as syntactically
parallel, it would be necessary to construct an argument structure for the
noun dinner that does not correspond to any thematic interpretation of a
linked internal argument.
Thus it appears that pragmatic appeal to an LCS in (54) and (55) is
what determines the apparently thematic interpretation in (54) for dis-
cussion and in (55) for dinner, and not any special syntactic relation
between them: no projection of lexical argument structure or of a
syntactic implicit argument is involved.
One might compare this account to that of Williams (1985), who, in
arguing against the existence of PRO in NP, argues that a 'control'
account of implicit argument antecedents would have to equate internal
and external arguments, whereas control only affects PRO subjects. I
differ from him here in treating "internal implicit argument~" as prag-
matically controlled. Williams notes that in John underwent an operation,
it is John that is understood to be the patient of the operation. Notice
however, that one can undergo much else in this syntactic context, e.g.,
John underwent Mary's egregious performance of the symphony, which
suggests that poor John is still patient, even though John cannot be
associated with any argument place of the nominal performance (see
Williams (1985) for further examples.)
The examples considered in this section probably do not exhaust the
contexts where understood arguments might be proposed as syntactically
active implicit arguments in nominals, but in so far as they take into
account the cases that have been raised, it appears that the theory of
Section 4 can be upheld: only syntactically projected external arguments
can be implicit arguments in nominals. 23

23 This appears to be directly at odds with the basic generalizations that Rizzi (1986) has
made with respect to Dative implicit arguments, if Dative arguments are to be considered
'internal' in the same sense.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

594 KEN SAFIR

7. THEORIES OF GRAMMATICAL RELATIONS

I return now to the issue mentioned at the outset: Should the theory of
grammatical linking include all grammatical functions as independent
primitives? If grammatical relations were each independent primitives,
we would never expect to find a context in which the expression of one
grammatical function of some predicate would be contingent on the
expression of any other unless we made a stipulation to that effect.
For example, suppose we define a grammatical function we call "sub-
ject" which maps an argument of any predicate P that has a "subject"
onto [NP,S] position in a sentence and PGNP position in nominal
constructions. Suppose we define another GF called "object" which
corresponds to the of-object in nominals and [NP,VP] position in a
sentence. Then we can state relativity as in (60) where the PGNP
position is assumed·to be optional, as was assumed earlier.

(60) GF Relativity' (GFR').


If a predicate P has an 'object', then it cannot have a 'subject'
unless the 'object' of P is linked.

Under these conditions, GFR' is descriptively adequate, but unlike GF


Relativity, GFR' must be explicitly stated in the grammar that assumes
grammatical relations are all independent primitives. GF Relativity, by
contrast, simply follows from the interpretation of H&K lexical structures
in terms of the definition of external argument - provided the GF of the
external argument is defined relatively, i.e., if grammatical relations are
not all independent primitives to which lexical entries refer. Only a
theory that requires the notion 'external argument' to be constructed by
reference to another independently definable entity, namely a lexical
structure, explains GFRelativity. Thus the internal argument within a
lexical structure is an independent primitive in this account, but the
external argument is not.
Since the foregoing argument for GF Relativity is an existence
argument, it can be refuted only by showing that dependence of external
arguments ('subjects' of nominals in some accounts) on the projection of
internal arguments (specifically of-objects) is not a factor in deriving the
linking patterns in the nominals investigated here. Otherwise it must be
concluded that theories that treat grammatical functions (particularly
'subjects') as independent primitives in lexical entries fail to be explana-
torily adequate in a fundamental way.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 595

7.2. The Uniformity Assumption

It may be countered that the two theories which this criticism is aimed at,
namely, Lexical-Functional Grammar and Relational Grammar do not
necessarily make the assumption that 'subject' and 'object' relations exist
within nominals, and in fact Rappaport (1983), working within LFG,
explicitly rejects such an assumption. Thus one could adopt GF Rela-
tivity for the GFs found in nominals yet claim it is irrelevant to 'true'
subject and object relations which are only defined for sentential struc-
tures. In this section I will consider some of the respects in which the
process of argument linking and projection in nominals is different from
linking in sentential structures, in order to determine whether GF rela-
tivity should still be generalized to both contexts of linking.
It is important to keep in mind that dissociation of argument linking in
nominals from the way that arguments are linked in sentences would
mean rejecting the strong assumption, essentially due to Chomsky
(1970), that there is uniformity of grammatical mapping across the
argument structures of related morphological stems. To reject what I will
call the UNIFORMITY ASSUMPTION totally would be to deny that the
grammatical mapping of a deverbal noun is anything more than ac-
cidentally related to the form of grammatical mapping of the cor-
responding verb - a very unpromising approach that none has treated
seriously. If matters were so simple, this would represent an a priori
advantage for generalizing GF Relativity to both syntactic contexts.
However, another view of the uniformity assumption, one that has
been adopted by Rappaport (1983), Grimshaw (1986) and Zubizarreta
(1986) among others, is to suggest that the mapping of thematic roles
onto positions in nominals is determined on the basis of the intrisnic
content of thematic structure, and that it is only this intrinsic content that
remains constant across lexical entries. Zubizarreta (1986, chapter 2)
describes the essence of Rappaport's proposal as foUows: "Noun phrases
lack semantically unrestricted GFs: i.e., SUBJ and OBJ. They only have
semantically restricted GFs: OBLIQUE OBJs (among which the OBJ
headed by the preposition of is included) and the POSS function (which
is proper to NPs)," (see also Rappaport, p. 134). Thus some level of
thematic structure, if not grammatical functions, is still uniform across
related morphological stems. It is not obvious whether this view of
thematic uniformity is compatible or not with the generalization of GF
Relativity across nominal and sentential structures. Hence it is worth-
while to look into one relatively explicit development of this view of
thematic uniformity in order to examine its consequences.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

596 KEN SAFIR

Zubizarreta (1986) (henceforth Z) argues further that what cor-


responds to the H&K lexical structure is not utilized in the mapping of
thematic roles onto structural positions in nominals: instead a different
level of structure, Lexico-semantic structure, where semantically restric-
ted grammatical functions are marked as such (e.g., locative arguments,
prepositional arguments), is utilized for nominal structures. Arguments
that are prepositional objects at Lexico-semantic structure may become
internal arguments or subjects in Lexico-syntactic structure, and verbs
access only the latter representation in their grammatical mapping. A
verb like enter, for instance, takes a direct object, but the corresponding
nominal must have a locative preposition. Similarly the grammatical
mapping of many 'psych' verbs is less "semantically transparent" (in Z's
terms) than the nominal, which requires a preposition.

(61)a. Ernie entered the room.


b. Ernie's entrance into/*of the room.
(62)a. The news terrified Bob.
b. *the news' terror of Bob
c. Bob's terror of/at the news

At the level of lexico-semantic structure, the verb enter would have a


locative argument, just as terrify has an at-object, but at lexico-syntactic
structure, enter takes a direct object and terrify takes an external
argument for the corresponding semantic roles. (The contrast between
verbs and nouns is supposed to follow from a theory of predicate types
which I will not review here.) Thus Z's approach argues for thematic
uniformity and only partial uniformity of grammatical mapping, but it
provides a means of accounting for the wide range of 'irregular' linkings
in nominals (those linkings in nominals that contrast with those of the
corresponding verbs).
Z's theory of irregular linkings in nominals provides an account of a
realm of facts that have not been addressed here, so it is reasonable to
ask if Z's proposals could be made consistent with GF Relativity. One
could imagine a version of Z's theory in which it would be possible to
incorporate an account of GF Relativity into Lexico-semantic structure
(which is accessed by nominal linking in her theory) for a subset of those
predicates that appear to have an internal argument in Lexico-syntactic
structure. For example, organize would have an internal argument in
both Lexico-semantic and Lexico-syntactic structure, while enter would
have a locative argument in the former and an internal argument in the
latter. Then if nominal linking draws on Lexico-semantic structure, it will

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 597

only link those internal arguments that are semantically transparent. 24 If


this revision were adopted, then GF Relativity would be extended to the
level of Lexico-semantic Structure.
If lexical structure is not expressed at Lexico-semantic structure
however (i.e., if Z's theory is not revised), then the evidence for dis-
tinguishing between internal and external arguments of nominals de-
veloped in this paper cannot be accounted for in Z's theory: the externall
internal argument distinction could not be accessed by nominal linking.

7.3. Against paNP as Adjunct


While extending lexical structure representations to Lexico-semantic
structure for some predicates may appear to be compatible with the
portion of Z's theory I have presented, Z would, in fact, reject it, since
she assumes that there is no movement in nominals. One reason for this
exclusion is presumably because the PGNP position would be empty
before movement, hence rendering it a semantically unrestricted GF,
which Z excludes in nominals (because of the assumed transparent
relation between Lexico-semantic structure and nominal linking). Indeed
the commitment to excluding movement in this fashion results in the
conclusion that both by-phrases and PGNPs count as adverbial adjuncts,
a view extended to PGNP positions as well by Grimshaw (1986). One
major advantage of this assumption is that the optionality of the by-
phrase and the PGNP is predicted, in that adjunct modifiers are always
optional. As this view is inconsistent with the conclusion that PGNPs and
by-phrases in nominals are external arguments, I shall attempt to refute
it.
An immediate disadvantage of this approach is that 'modifiers' can
normally modify overt arguments, but the PGNP and the by-object
("adjunct arguments", as Grimshaw calls them) can never modify overt
arguments or each other. The necessity of this assumption (stipulated by
Grimshaw) is demonstrated by the failure of (43), repeated here.
(43) Pete's destruction of the city (with Joe)/(*by Joe)

24 An apparent advantage of the semantic transparency claim pointed out by Grimshaw


(1986) is that deverbal nominals never pennit expletive PGNPs as in *itlthere's arrival of
John or *itlthere's destruction of the city by the army. This is expected in her approach since
there are no expletive modifiers, but does not foHow from the theory proposed here, though
it might follow from independent considerations. For example, if Genitive is an inherent
Case, in the sense of Chomsky (1986), then the 'Uniformity Condition' he proposes will
rule out Genitive Case assignment to an expletive element; the Uniformity Condition only
permits inherent Case to be assigned to an NP if that NP bears a theta-role of the Case
assigning head (Chomsky, 1986, pp. 194-196).

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

598 KEN SAFIR

Compare this notion of 'argument adjunct' with the functioning of the


commitative with-object which can modify a subject in the sense that the
by-phrase fails in (43), or with adjectival adjuncts, which, as demon-
strated throughout, are not limited to modifying non-overt arguments.
Moreover, the dependency of the external argument (and the distribution
of implicit arguments) on the internal argument (as stated in (12) and
derived in Sections 3 and 4) is unexplained if the PGNP is only an
adjunct.
All of these classes of facts follow immediately if the PGNP and the
by-object are both arguments of destruction in (43), because the Theta
Criterion can be invoked. Moreover, in order to account for the extrac-
tion facts in (52) and (53), it would be necessary to distinguish 'adjunct
argument' PGNPs from PGN;Ps that are not interpreted thematically with
respect to the head nominal, another distinction that renders the idea of
'adjunct arguments' dubious.
Recent work by Lesnik (1986) provides additional evidence against
the Zubizarreta/Grimshaw approach with respect to the supposed ad-
junct status of by-phrase objects and some PGNPs. Lasnik notes that
instruments and agents are distinct thematic roles, as indicated by the
contrast in (63).
(63)a. *John will serve to sink the ship.
b. The missile will serve to sink the ship.
Yet the presence of a subject is incompatible with a by-phrase, even
though the instrument thematic role can be successfully expressed with
an instrumental with-phrase.
(64)a. John sank the ship with/"'by a missile.
b. The ship was sunk by John/a missile.
Lasnik points out that the incompatibility of the by-phrase with the
subject would be explicable if both positions represented the same
grammatical function - namely, the external argument. Naturally the
by-phrase will not have to compete with a dethematized subject position,
and so it is expected to be available in passive sentences realizing either
role, as in (64b).2s Lasnik shows that exactly the same facts hold of
nominals.

25 As Lasnik points out, examples like The Army sank the boat by missile are irrelevant, as
missile is still not a possible subject, e.g., • Missile sank the boat.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 599

(65)a. John's destruction of the ship with/*by the missile.


b. The ship's destruction by John/the missile.
Thus it appears that this complex of facts makes sense only under the
assumption that the by-object and the PGNP are arguments in the
relevant cases, and not adjuncts in any standard sense of the latter term.
To put it another way, any approach treating the PGNP and the by-
phrase as always being adjuncts would have to fashion the notion
'adjunct argument' into something quite like an argument, so much so
that one might wonder what the notion 'adjunct' is worth altogether.

8. CONCLUSION

What emerges from Sections 7.2 and 7.3 is that alternative theories of
linking in nominals, in so far as they are correct, are not incompatible
with GF Relativity. Moreover, assuming as much uniformity as possible
across morphologically related stems is 9till a strong motivation for
adopting the most general view of GF Relativity, namely one which
assumes it applies to sentential as well as nominal linking.
I conclude that the foregoing analysis of argument linking and pro-
jection in nominals establishes the importance of GF Relativity for
grammatical theory. Moreover, only a theory that permits some gram-
matical functions to be constructed from other notions, such as lexical
structure, can derive the effects of GF Relativity; among the major
generativist theories available, only Government-Binding theory, as it is
elaborated here, has this property.

REFERENCES

Anderson, M.: 1979, Noun Phrase Structure, unpublished University of Connecticut, Ph.D.
dissertation.
Anderson, M.: 1983, 'Prenominal Genitive NPs', The Linguistic Review 3, 1-24.
Aoun, J. and D. Sportiche: 1982, 'On the Formal Theory of Government', The Unguistic
Review 1, 211-236.
Bresnan, J. (cd.): 1982, The Mental Representation of Grammatical Relations, MIT Press,
Cambridge.
Bumo, L.: 1986, Italian Syntax, A Government-Binding Approach, Reidel, Dordrecht.
Carrier-Duncan, J. and J. Randall: 1986, 'Derived Argument Structure: Evidence from
Resultatives', manuscript, Harvard University and Northeastern University.
Chomsky, N.: 1970, 'Remarks on Nominalizations', in R. Jacobs and P. Rosenbaum, (eds.),
Readings in Transformational Grammar, Ginn, Waltham, MA.
Chomsky, N.: 1981, Lectures on Government and Binding, Foris, Dordrecht.
Chomsky, N.: 1986a, Knowledge of Language: lIS Nature, Origin and Use, Praeger, New
York.
Chomsky, N.: 1986b, Barriers, MIT Press, Cambridge.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

600 KEN SAFIR

Cinque, G.: 1980, 'On Extraction from NP in Italian', Journal of Italian Linguistics S,
47-99.
Clark, R.: 1985, Boundaries and the Treatment of Control, unpublished UCLA Ph.D.
dissertation.
Curme, G.: (1922), A Grammar of the German Language, Macmillan, London.
Fabb, N.: 1984, Syntactic Affixation, unpublished MIT Ph.D. dissertation.
Finer, D. and T. Roeper: 1983, 'From Cognition to Thematic Roles: the Role of the
Projection Principle in Language Acquisition', to appear in Proceedings of the Ontario
Conference on Learnability. R. Matthews and R. May (eds.).
Giorgi, A.: 1985, 'On C-command and Binding within NPs', manuscript, Istituto de
Psicologia del CNR.
Giorgi, A.: 1986, On the Italian Anaphoric System, Scuola Normale, Pisa.
Grimshaw, J.: 1986, 'Nouns, Arguments and Adjuncts', manuscript, Brandeis University,
Waltham.
Hale, K.: 1983, "Warlpiri and the Grammar of Non-configurational Languages', Natural
Language and Linguistic Theory 1, 5-47.
Hale, K. and S. J. Keyser: 1985a, 'Some Transitivity Alternations in English', manuscript,
MIT, Cambridge.
Higginbotham, J.: 1983, 'Logical Form, Binding, and Nominals', Linguistic Inquiry 14,
395-420.
Jackendoff, R.:·1972, Semantic Interpretation in Generative Grammar, MIT Press, Cam-
bridge.
Jaeggli,O.: 1985, 'Passive', manuscript, University of Southern California.
Kayne, R.: 1984, Connectedness and Binary Branching, Foris, Dordrecht.
Lasnik, H.: 1986, 'Subjects and the Theta Criterion', manuscript, University of Con-
necticut.
Lebeaux, D.: 1984, 'Normalizations, Argument Stucture and the Organization of the
Grammar', manuscript, University of Massachusetts at Amherst.
Levin, B. and M. Rappoport: 1985, 'The Formation of Adjectival Passives', Linguistic
Inquiry 17, 623-662.
Perlmutter, D. (ed.): 1983, Relational Grammar 1, University of Chicago Press, Chicago
and London.
Perlmutter, D. and C. Rosen: 1984, Relational Grammar 2, University of Chicago Press,
Chicago and London.
Pesetsky, D.: 1982, Paths and Categories, unpublished MIT Ph.D. dissertation.
Pesetsky, D.: 1985, 'Morphology and Logical Form', Linguistic Inquiry 16, 193-246.
Pustejovsky, J.: 1985, Studies in Generalized Binding, unpublished, University of Mass.
Ph.D. dissertation.
Randall, J.: 1984, 'Morphological Complementation', in M. Speas and R. Sproat (eds.),
MIT Working Papers in Linguistics Vol. 7, MIT Department of Linguistics and Philoso-
phy, Cambridge.
Rappoport, M.: (1983), 'On the Nature of Derived Nominals', in L. Levin et al. (eds.),
Papers in Lexical-Funcational Grammar, IULC, Bloomington.
Rizzi, L.: 1986, 'Null Objects in Italian and the Theory of PRO', Linguistic Inquiry 17,
501-558.
Roberts, I.: 1985, The Representation of Implicit and Dethematized Subjects, unpublished
U.S.C. Ph.D. dissertation.
Roeper, T.: 1984, 'Implicit Arguments', manuscript, University of Massachusetts, Amherst.
Roeper, T.: 1986a, 'Implicit Arguments and the Head-Complement Relation', Linguistic
Inquiry 18, 267-310.
Roeper, T.: 1986b, 'Implicit Arguments, Implicit Roles, and Subject/Object Asymmetry in
Morphological Rules', manuscript, University of Massachusetts, Amherst.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
SAFIR, KEN, The Syntactic Projection of Lexical Thematic Structure , Natural Language and
Linguistic Theory, 5:4 (1987:Dec.) p.561

PROJECTION OF LEXICAL STRUCTURE 601

Roeper, T. and D. Finer: 1983, 'From Cognition to Thematic Roles: The Role of the
Projection Principle in Language Acquisition', manuscript, University of MlISliachusetts,
Amherst.
Safir, K.: 1984, 'On Small Clauses as Constituents', Linguistic Inquiry 14, 73~735.
Safir, K.: 1985, Synt4Ctic Chains, Cambridge University Press, New York and Cambridge.
Safir, K.: 1986a, 'On Implicit Arguments and Thematic Structure', in NELS 16.
Safir, K.: 1986b, 'Evaluative Predicates', manuscript, Rutgers University.
Selkirk, E.: 1982, Word Syntax, MIT Press, Cambridge.
Sproat, R.: 1985, On Deriving the Lexicon, unpublished MIT Ph.D. dissertation.
Stowell, T.: 1981, Origins of Phrase Structure, unpublished MIT Ph.D. dissertation.
Torrego, E.: 1986, 'On Empty Categories in Nominals', manuscript, University of Mas-
sachusetts, Boston.
Weinberg, A., J. Aoun, N. Hornstein and D. Lightfoot: 1985, 'Two Types of Locality',
manuscript, USC and University of Maryland.
Williams, E.: 1981, 'Argument Structure and Morphology', The Linguistic Review I,
81-114.
Williams, E.: 1982, 'The NP Cycle', Linguistic Inquiry 13, 277-296.
Williams, E.: 1985, 'PRO and Subject of NP', Natural Language and Linguistic Theory 3,
297-315.
Zubizarreta, M.-L.: 1986, Levels of Representation in the Lexicon and in the Syntax,
manuscript, University of Tilburg and University of Maryland.

Received 6 October 1986


Revised 27 May 1987

Linguistics Program
Rutgers University
43 Mine Street
New Brunswick, NJ 08903
U.S.A.

Copyright (c) 2004 ProQuest Information and Learning Company


Copyright (c) Kluwer Academic Publishers Group
View publication stats

You might also like