Professional Documents
Culture Documents
Transparent Adjuncts
CLAUDIA BORGONOVO
Universite Laval
and
AD NEELEMAN
University College, London
1. INTRODUCTION
All theories of movement have as one of their central aims to explain the fact that
extraction is dependent on the argument-adjunct status of the category extracted
from. Since Barriers it is assumed that categories are islands for extraction
unless they are L-marked or selected (cf. Chomsky 1986, Rizzi 1990, Cinque
1990, and Manzini 1992). Hence, extraction out of adjuncts is ruled out. While
this is certainly true for temporal, purpose and cause adjuncts (the examples
typically used to show extraction patterns), there are unexpected exceptions when
predicative adjuncts1 are considered. In the contexts in (1) and (2), predicative
adjuncts are well-behaved: they do not allow extraction.
(1) a. John danced [dressed as Carmen Miranda].
b. *Whatj did John dance [dressed as y ?
c. John danced [imagining the Gobi Desert].
d. *Whatj did John dance [imagining t;]?
(2) a. Johnfinishedthe portrait [covered in blue paint],
b. *Whatj did Johnfinishthe portrait [covered in y ?
This article was presented at CONSOLE 1 (Utrecht University), HI Coloquio de
Gramatica Generativa in El Escorial, Rutgers University Colloquium Series, Brandeis
University Research Seminar and Utrecht University "Onderzoekers" Series in 1993-1994.
We would like to thank the audiences for valuable comments. We also thank Peter Ackema,
Piroska Csiiri, Jane Grimshaw, Ray Jackendoff, Joan Maling, Javier Ormazlbal, James
Pustejovsky, Tanya Reinhart, Ken Safir, Hubert Truckenbrodt, Myriam Uribe-Etxebarria,
and an anonymous reviewer for comments.
1
By predicative adjuncts we mean an XP with a 6-role to be discharged to a DP external
to it, the typical example being depictives. We exclude from this definition adjuncts that
are predicated of the Event position, typically temporal, manner, and location adjuncts.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
200 CJL/RCL 45(3/4), 2000
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 201
Second, such sentences are presumably not very salient in the learner's input,
due to their low frequency, which probably does not exceed that of parasitic gap
constructions. Nevertheless, native speakers have fairly robust judgements about
them. This means that the properties of transparent adjuncts must be explained by
independently established principles.
In this article we discuss the principles involved, as developped within the
Principles and Parameters approach. Generally, a category is an island unless
it holds a certain relation to a head. Chomsky (1986) refers to this relation as
L-marking. Given that the adjuncts in (3), (4c), and (5a) can be subextracted
from, the definition of L-marking becomes crucial. Taking the interconnectedness
of 0-theory and L-marking seriously, we propose that only categories that bind
a 9-role of the verb and are contained within V may count as L-marked. We
observe that only predicative adjuncts can be transparent. We show that, given a
standard theory of predication, predicative adjuncts do indeed bind a 8-role of the
verb. We also show that transparent adjuncts can appear inside V , and therefore
can be L-marked (section 2).
We show that if adjuncts are interpreted as being L-marked, a representation
obtains that has the defining characteristics of reflexivity. According to Rein-
hart and Reuland (1993), reflexive constructions have to be licensed by reflexive
marking. Therefore, in order ot interpret adjuncts as L-marked, reflexive marking
must be present; it is then predicted that only reflexive predicates allow extrac-
tion out of adjuncts. Given an analysis of ergatives as inherent reflexives (as in
Chierchia 1989), it now follows that ergative, but not transitive or unergative verbs
take transparent adjuncts; this accounts for the data in (3) and (4c-d). It also
follows that "traditional" reflexive verbs have the same property, which accounts
for (5a-b). In section 3 we discuss extraction out adjuncts with reflexive verbs; the
extraction patterns that are observed force us to refine extraction conditions: they
apply to a subtype of reflexives, the inherent reflexives of traditional grammar.
This further requirement, we show, follows from well-formedness conditions on
reflexive chains.
2. O N L-MARKING
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
202 CJL/RCL 45(3/4), 2000
V G; DPj
The theory that arguments bind 9-roles leads to the following modification of the
definition of L-marking:
(10) L-Marking (final version):
A head a L-marks a maximal projection (3 iff
(i) f3 binds a 9-role of a, and
(ii) P is contained in a'.
The stage is now set for the discussion of L-marked, hence transparent, adjuncts.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 203
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
204 CJL/RCL 45(3/4), 2000
in (13), where the DP in object position, John, binds XP, the secondary predicate,
under predication. We give evidence below for the low (under V ) attachment of
the gerundive predicate; we stipulate it in (13) given the transparency of the XP
in question.
(13) V
XRO;
The same DP is also the internal argument of the verb, and therefore binds its
internal 0-role, as in (14).
(14)
Given that DP; in (13) and (14) binds both the 9-role of the verb and the predicate
XP, transitivity of coindexation will obtain, resulting in the XP bearing the index
of the 0-role of the verb, as in (15).
(15) V
V 9; DP XPj 9j
The coindexation between the predicate and the verb's 8-role has direct conse-
quences for the status of the adjunct. Since that 0-role is both coindexed with
and c-commanded by the secondary predicate, a binding relation is established
by definition. This means that if the secondary predicate is contained in V , as
we have indicated in the structures above (see evidence below), it will count as
L-marked, because it satisfies the definition in (10).
Although we still have to show that transparent adjuncts are contained in V ,
it will already be clear that it is essential that the prospective L-marked adjunct be
a predicate. Were it not, no coindexation and hence no binding relation could be
established between it and a 0-role of the verb. Consequently, adjuncts like those
in (12) cannot be L-marked, regardless of their place of attachment, and extraction
out of them is ruled ungrammatical.2
What about the structural condition on L-marking, the one that requires con-
tainment in the V projection? The following evidence suggests that transparent
adjuncts are indeed in complement position, that is, under V :
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 205
Whenever a clear adjunct (such as the manner adjunct in (16a), the temporal
adjunct in (16b) or the comitative in (16e)) intervenes between the verb and the
predicative adjunct, extraction degrades. This evidence indicates that in order for
the adjunct to be transparent, that is, L-marked, it has to be lower in the tree than
adjuncts, thus suggesting that transparent adjuncts do indeed satisfy the structural
condition on L-marking.
A note on the position of depictives: the evidence shown so far indicates that
depictives of the type we are considering have double attachment possibilities.
They can be adjoined to VP, in the canonical position for depictives argued for in
Roberts (1988) and McNulty (1988); this is the case of the post-adjunct gerunds in
(16).3 The evidence also shows that transparent adjuncts can be generated lower,
as sisters to V. The next section will show that this possibility actually results in a
formally reflexive construction.
3. REFLEXIVITY
The defining property of reflexive structures is the coindexation between two ar-
guments within the domain of a single predicate. The following schemas illustrate
the point:
(17) a.
3
A reviewer points out that the "do so" test suggests that predicative adjuncts of the
kind studied here appear outside V':
(i) John died singing Mozart and Bill did so singing Wagner.
Given the double attachment possibilities that these adjuncts have, we expect this test to
give acceptable results: predicative adjuncts always have the option of attaching higher.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
206 CJL/RCL 45(3/4), 2000
Configurations such as those in (17) can hold at different levels, and the defini-
tion of what can constitute an argument will change accordingly. An inherently
reflexive verb such as shave instantiates (17) in the lexicon. It has the following
argument structure, in which two of its 6-roles are coindexed:4
(18) (Agent; (Theme;))
A syntactic instantiation of (17) is provided by structures in which an anaphoric
relation holds between two (or more) A-positions projected from the same predi-
cate. A straightforward example appears below:
(19) [John; [likes himself;]].
Finally, there is the possibility that at a semantic level, in which predicate
logic is relevant, the definition can be met. An example of this is provided by the
contrast in (20). The semantic representation of the sentences in (20) contains a
reflexive clause, namely the right conjunct in (21). The reflexivity of this clause is
licensed in (20a), but not in (20b) (see Reinhart and Reuland 1993 for discussion
of this point).
(20) a. The queen invited both Max and herself to our party,
b. *The queen invited both Max and her to our party.
(21) The queen (Ax (x invited Max & x invited x)).
The assumption we are making about the definition of reflexivity being met
at different levels underlies other theories of reflexivity as well. The reflexivity of
shave in John shaves has to be presyntactic, because from a syntactic point of view
the verb is monadic. The reflexivity of like in (19), on the other hand, must be
syntactic, because like is not necessarily self-oriented (witness John likes Mary).
Given that the definition of reflexivity applies across levels, with a concomitant
difference in the interpretation of the notion "argument", there is no reason to
assume that syntactic reflexivity entails argument structure or semantic reflexivity.
Rather, syntactic reflexivity can be a purely formal property, because syntactic
arguments are defined formally. For example, non-thematic subjects are arguments
in this sense. One might therefore expect the existence of constructions that meet
the definition of reflexivity in the syntax without reflecting it in their semantics or
argument structure. We will now show that transparent adjuncts entail this kind
of reflexivity.
Consider, for example, John died whistling. As we argued above, there is one
variant of this construction in which the predicative adjunct binds a 0-role of the
verb and is contained in V (in the other variant the predicative adjunct is attached
higher, as discussed in the preceding section). Binding of a 0-role and containment
4
In theories such as Grimshaw's (1990), the level at which this coindexation obtains is
lexical conceptual structure, not argument structure. For the purposes of this article, this
distinction can be ignored.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 207
V DP: XP:
I I
V Bj X 0;
If XPj binds 0j in (22), the structure contains two coindexed syntactic arguments,
DPj and XPj (as said, XP qualifies as a syntactic argument by virtue of binding the
verb's 9-role and being inside V , though it is not a semantic argument of the verb).
Consequently, the structure in (22) instantiates (17) and is therefore reflexive.
As we argued above, in order for the predicative adjunct to be L-marked (and
hence transparent), it has to bind a 0-role of the verb and be contained in V . This
means that a construction such as What did John die whistling? will always be
reflexive, since two arguments (one semantic, one formal) within the domain of
one predicate are coindexed. We will now argue that conditions on the licensing
of reflexivity determine felicitous extraction.
Reinhart and Reuland (1993) argue that binding theory is a theory about
predicates, and not about referential dependencies. The notion of reflexivity plays
a crucial role in defining their principles A and B:
(23) Principle A:
A reflexive-marked predicate is reflexive.
Principle B:
A reflexive predicate is reflexive-marked.
As we said above, transparent adjunct constructions are reflexive. Therefore,
reflexive marking has to be present by Reinhart and Reuland's principle B. Since
it is crucial to this article, let us spell out our line of reasoning here. If extraction
out of a predicative adjunct has taken place, we know this adjunct is L-marked.
If it is indeed L-marked, the resulting construction is reflexive, and by principle
B it requires reflexive marking. The presence of reflexive marking, then, is a
prerequisite for extraction out of adjuncts. This will be shown to derive the class
of verbs that allow transparent adjuncts.
According to Reinhart and Reuland, reflexive marking can be achieved in
two ways. Either the predicate is reflexive-marked by a SELF-anaphor, or it is
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
208 CJL/RCL 45(3/4), 2000
inherently reflexive (that is, reflexive-marked in the lexicon). Assuming for the
moment that these two ways of marking a predicate as reflexive are equivalent,
we predict that XP can be transparent if either the matrix verb is reflexive-marked
by a SELF-anaphor or if it is inherently reflexive. It can be argued that all the good
cases of subextraction presented so far exemplify these two options: an L-marked
adjunct as part of an inherently reflexive or a SELF-anaphor construction, whereas
all the ungrammatical cases lack reflexive marking of any kind.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 209
5
We thank Federica Busa (p.c.) for the Italian examples.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
210 CJL/RCL 45(3/4), 2000
(26) Whj... VP
[+REFL]
It now also becomes clear why the transitives and unergatives in (1) and (2) do
not allow transparent adjuncts. A sentence such as * Who did John write his third
novel thinking oft? is ungrammatical because the matrix verb is not reflexive and
therefore lacks reflexive marking of any kind. Consequently, the representation
in which the XP binds a 9-role of the verb and appears within the latter's V (the
necessary conditions for L-marking) is not licensed. So, either principle B is
violated, or there is no L-marking. Both situations lead to ungrammaticality:
(27) *Whj...
6
Some speakersfindthese examples less clear than the ones involving non-alternating
ergatives; an anonymous reviewer reports that he tends to prefer depictive gerunds that
can be interpreted agentively. We have nothing to say about this for the moment. As the
same reviewer points out, it has been claimed (Haegeman 1990 and others) that alternating
unaccusatives of the type shown in these examples are actually syntactically unergative,
i.e., generated with a DP in subject position. If this were true, it would explain why these
data are less acceptable, maybe on a par with other unergatives.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 211
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
212 CJL/RCL 45(3/4), 2000
NP V
I I
John shaves
Let us now look again at the definition of L-marking, repeated here as (37).
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 213
(37) L-Marking:
A head a L-marks a maximal projection fi iff
(i) 0 binds a 0-role of a
(ii) 13 is contained in a'.
If we consider the second clause in (37), it becomes apparent that verbs that do
not take an internal argument will not be able to L-mark, since L-marking requires
containment in a projection that by hypothesis does not exist in constructions
headed by an unergative. The ensuing representation is as follows:
(38) VP
VP XP
DP V
In (38), the gerund depictive (XP) is adjoined to VP, the unmarked position
for depictives. The V does not project a V , following the considerations discussed
in the preceding paragraph.
Note that the strict interpretation of the definition of L-marking is indepen-
dently justified. Given Chomsky's claim that unergatives do not project a V'-level,
in (38) the verb would L-mark the subject if L-marking were defined on the basis
of sisterhood (39). However, subjects of unergatives, like subjects of transitives,
are islands, witness (40).
(39) VP
DPj V 9j
(40) *Whati did [an author of tj] laugh?
In sum, it has been shown that the fact that inherent reflexives do not take
transparent adjuncts does not prove wrong the analysis based on reflexivity. The
constructions under discussion are excluded for independent reasons having to do
with the structural, rather than the thematic, component of L-marking.
We are still left with a problematic case: the subgroup of transitive reflexive
verbs exemplified in (34). These verbs are reflexive, hence should take transparent
adjuncts, and contrary to the previous type of exceptions, they are transitive. We
cannot appeal to their lack of L-marking capacity, since they take an object, and
hence project a V .
It turns out that these verbs share one characteristic: they all take the anaphor
zichzelf in Dutch whereas the grammatical cases of extraction involve verbs that
in Dutch take the anaphor zich. Let us first consider the reflexive system of Dutch,
and then show why the above mentioned distinction is relevant for English and
subextraction. Dutch has two anaphors. The first one, zichzelf 'SE + self, can
function as a SELF-marker, and thus license non-inherently reflexive constructions.
The second, zich 'SE', is not a SELF-marker, and can therefore only appear in
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
214 CJL/RCL 45(3/4), 2000
What is crucial to the present discussion is that the presence of zich signals
inherent reflexivity. Now compare the following data in English and Dutch:
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 215
A surprising correlation emerges from comparing the English and the Dutch
data; this correlation is given in (44).8
(44) Ackema 's Generalization:
The transitive reflexive verbs that take transparent adjuncts in English take zich in
Dutch.
Zich, as seen above, only shows up in inherently reflexive constructions.
Given that the Dutch verbs in (43) allow both zich and zichzelf (and a referential
NP), we must conclude that they have an inherent reflexive variant (besides a
variant that allows a referential internal argument or a SELF-anaphor). Given
Ackema's generalization, let us claim that verbs that are inherently reflexive in
Dutch are inherently reflexive in English, modulo some peripheral cases. The
conclusion that emerges is that inherent reflexivity, rather than reflexivity as such,
is the sine qua non condition for extraction out of predicative adjuncts. The
examples of felicitous extraction involve verbs that have an inherent reflexive
marking, such as ergatives and the zich verbs. Putting aside unergatives (out for
structural reasons having to do with L-marking), the examples of ungrammatical
extraction involve verbs that are either not reflexive at all, or that are not inherently
reflexive.
There is some evidence that it is the inherently reflexive variant of cut or the
other verbs in (43) that shows up with transparent adjuncts. The presence of zich
with these verbs in Dutch requires a specific interpretation: that the subject be
non-agentive. This is highlighted in the following examples:
(45) a. dat Jan zich/?zichzelf per ongeluk snijdt
'that John SE/SE-self accidentally cuts'
b. dat Jan??zich/zichzelf met opzet snijdt
'that John SE/SE-self with purpose cuts'
Our English informants independently report that in the cases of transpar-
ent adjuncts the subject also has to receive a non-agentive interpretation. If
non-agentivity is a property of the inherently reflexive variant of cut, speaker's
intuitions corroborate the connection between inherent reflexivity and the possi-
bility of subextraction.9 Given the impossibility of an adjunct intervening between
the transparent adjunct and V, the intuition cannot be corroborated by an appro-
priate manner adverb that would reinforce the non-agentive interpretation of the
subject:10
8
We thank Peter Ackema (p.c.) for pointing out this correlation.
9
It has been noticed that inherent reflexives in general express actions that are "un-
marked" when directed towards oneself (Jan Koster, class lectures, and traditional grammar
sources). Since it is not "unmarked" to inflict physical damage on oneself on purpose, it is
to be expected that the inherently reflexive variant of the verbs in (43) is non-agentive.
I0
A reviewer points out that our prediction can be tested by placing the manner adverbial
before the VP. The following data (provided by the same reviewer) seems to confirm our
predictions, though a bit awkwardly:
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
216 CJL/RCL 45(3/4), 2000
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 217
(49) V
[+REFL]
[DP]
x ik
The answer lies in the fact that there are two reflexive relations in (50), one
involving DP and DP+SELF, and the other involving DP and XP. Let us assume the
following: both relations need reflexive marking, and one SELF-marked anaphor
cannot license two seprarate relations. Let us see how this intuition can be
implemented.
Suppose that reflexivity is a property of chains rather than of predicates. A
chain is reflexive if two of its links are arguments of the same predicate. This
definition denotes the same set of constructions we labelled reflexive before.
However, it attributes the reflexivity of a construction to a chain, rather than to the
predicate. We can now redefine principles A and B of the binding theory in terms
of chains rather than in terms of predicates:"
(51) Principle A:
A reflexive-marked chain is reflexive.
Principle B:
A reflexive chain is reflexive-marked.
Notice that as a consequence of this proposal the existence of chains at the argument
structure and semantic levels has to be assumed. Since both levels allow for coindexation
and impose hierarchical relations (see Grimshaw 1990) this is not altogether surprising,
since chains are defined in terms of command and coindexation.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
218 CJL/RCL 45(3/4), 2000
(52) VP
DPk V'
V DPk
[+REFL]
In the second instance, the foot of the chain carries its own reflexive marking,
that is, it is a SELF-anaphor. The claim that the SELF-marking has to be on the foot
of the chain is illustrated by the Dutch examples in (53). The contrast between
(53a) and (53b) shows that in a simple transitive construction the SELF-marker
must be on the object, not on the subject. The examples in (53c-d) show that in
double object constructions the direct object, and not the indirect object must be
SELF-marked.
(53) a. Jan haat zichzelf.
'John hat^f SE+SELF.'
b. *[Jan zelf] haat zich.
'John himself hates SE.'
c. Jan raadde zich zichzelf aan.
'John recommended SE SE+SELF PRT.'
d. *Jan raadde zichzelf zich aan.
'John recommended SE+SELF SE PRT.'
We would like to point out that this way of looking at reflexivity is in many
ways a notational variant of Reinhart and Reuland's proposal. It has one advantage:
the two types of reflexive marking are realized in a single place, the foot of the
reflexive chain.
We can now come back to the question of why Ackema's generalization
holds, or to put it differently, what makes the structure in (50) ungrammatical.
The offending pattern contains a non-inherently reflexive verb followed by a SELF-
marked anaphor and a coindexed predicate. Since this construction contains one
reflexive-marked element, in principle it can only be grammatical if a single chain
is formed (remember that only one chain can be formed if there is only one SELF-
marked element). Given the c-command relations in the structure in (50), there are
two ways in which the three coindexed elements can form a single chain: either
as in (54a) or as in (54b), since the reflexive marking does not appear on the foot
of the chain.
(54) a. {DPj DP+SELFj XPJ
b. {DP XP; DP+SELFj
The chain in (54b) is well formed from the point of view of reflexive marking:
the SELF-anaphor is the foot of the chain. However, it is ungrammatical because
it contains an ill-fomed link, namely *{DPj XPJ. XP, because it is a predicate,
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 219
does not have the right features to be the local antecedent of an anaphor; it is not
referential, to begin with. The ill-formedness of this link renders the single-chain
interpretation of (50) ungrammatical.
The only remaining option, then, is that (50) contains two chains: a predication
chain consisting of the subject and the predicative adjunct (55a) and an ordinary
binding chain consisting of the subject and the SELF-anaphor (55b).
(55) a. {DP;XP;}
b. {DP; DP+SELFj}
The second obviously agrees with the principles of binding theory as stated
in (51). The predication chain, however, does not, since its foot is not reflexive-
marked (it is not SELF-marked). In sum, no matter how the three coindexed
elements in (54) combine into chains, the construction is ruled out.
As it turns out, constructions containing a SELF-marked anaphor and no
inherent reflexive marking on the verb do not allow an interpretation of predicative
adjuncts as L-marked. Such an interpretation violates principle B of the binding
theory: a reflexive chain {DPj PredJ is formed that is not reflexive-marked.
The situation changes dramatically if the verb is inherently reflexive, that is,
capable of reflexive marking the foot of the chain. In this construction, the same
chains re-occur. However, the first chain, {DPj DP+SELFj}, is reflexive-marked
by the SELF-anaphor. The second chain, {NPj XP;}, is reflexive-marked by the
inherently reflexive predicate. Furthermore, the two feet of the two chains bear
the crucial marking. The construction is therefore grammatical, and subextraction
is ruled in:
(56) Wh k ...
4. CONCLUSION
The problem of transparent adjuncts is now solved. We have argued that in order
for a verb to take transparent adjuncts it must take an object (otherwise it is not
able to function as an L-marker at all) and it must be inherently reflexive-marked
(otherwise the L-marking of the transparent adjunct will result in a representation
that violates principle B). So, the following lexical representation, first introduced
in (49), is required:
(57) V
[+REFL]
[_DP]
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
220 CJL/RCL 45(3/4), 2000
This requirement correctly derives the class of verbs that take transparent ad-
juncts. We have argued that transitive verbs like invite are not inherently reflexive,
and therefore do not allow extraction out of predicative adjuncts. Transitive verbs
like cut have an inherently reflexive variant, and hence they do allow subextraction.
The same is true of unaccusative verbs like die, which are inherently reflexive by
hypothesis. Verbs like wash, finally, are either inherently reflexive unergatives, or
normal transitive verbs. Consequently, they do not allow transparent adjuncts in
either variant.
To conclude, the existence and properties of transparent adjuncts now follow
from principles seemingly unrelated to the phenomenon, namely the principles
governing predication and binding. The latter, as a result, receive extra confirma-
tion from this extension of their empirical coverage.
Our account makes a last prediction: object depictives with non-reflexive transitive
verbs should not allow transparent adjuncts, since the reflexive representation
needed for extraction cannot be licenced. This prediction is only partially true:
(58) a. The meat was served [oozing blood].
b. *Whatk was the meat served [oozing tk]?
c. The meat was served [wrapped in foil].
d. ?Whatk was the meat served [wrapped in tjj?
Speakers, though they consistently reject (58b), tend to accept (58d). We
have no explanantion for this at this moment, though maybe categorial differences
between Adjective Phrases and Gerund Phrases play a bigger part than suggested
here. We leave the issue open.
The analysis proposed in this article has repercussions for the theory of reflexivity:
it affects both the definition of reflexivity and the kind of object reflexivity is a
property of. In this appendix, we will briefly discuss both issues, starting with the
latter.
We have proposed that reflexivity is a property of chains, rather than of pred-
icates. In many respects, this approach represents a notational variant of Reinhart
and Reuland's proposal. Consider, for instance, the issue of logophoricity. The
basic intuition is that a SELF-anaphor can be used as a logophor if the position it
appears in is not regulated by binding theory. So, if the structure a SELF-anaphor
is in cannot be reflexive, principle A may be ignored. Some examples are given in
(59). In these examples the heads that take the SELF-anaphor as their complement
do not have a subject. Hence principle A cannot, and therefore does not have to,
be obeyed.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 221
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
222 CJL/RCL 45(3/4), 2000
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
BORGONOVO and NEELEMAN 223
REFERENCES
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680
224 CJL/RCL 45(3/4), 2000
Evers, Arnold. 1975. The transformational cycle in Dutch and German. Doctoral disser-
tation, Utrecht University. Distributed by the Indiana University Linguistics Club.
Grimshaw, Jane. 1982. On the lexical representation of Romance reflexive clitics. In
The mental representation of grammatical relations, ed. Joan Bresnan, 87-148. Cam-
bridge, Mass.: MIT Press.
Grimshaw, Jane. 1989. Argument structure. Cambridge, Mass.: MIT Press.
Haegeman, Liliane. 1990. The syntax of motional goan in West Flemish. In Linguistics in
the Netherlands, ed. Reineke Bok Bennema and Peter Coopmans, 81-90. Dordrecht:
Foris.
Higginbotham, James. 1985. On semantics. Linguistic Inquiry 16:547-593.
Hoekstra, Teun. 1984. Transitivity: Grammatical relations in government-binding theory.
Dordrecht: Foris.
Levin, Beth and Tova Rapoport. 1988. Lexical subordination. In Papers from the Twenty-
fourth Regional Meeting, 27'5-289. Chicago Linguistic Society, University of Chicago.
Manzini, Maria. R. 1992. A theory of locality conditions. Cambridge, Mass.: MIT Press.
Marantz, Alec. 1984. On the nature of grammatical relations. Cambridge, Mass.: MIT
Press.
McNulty, Elaine. 1988. Secondary predicates in Spanish and English. Doctoral disserta-
tion, University of Connecticut.
Neeleman, Ad. 1993. Complex predicates: A comparative analysis of Dutch and English
verb-predicate constructions. Doctoral dissertation, Utrecht University.
Reinhart, Tanya, and Eric Reuland. 1993. Reflexivity. Linguistic Inquiry 24:657-720.
Rizzi, Luigi. 1990. Relativized minimality. Cambridge, Mass.: MIT Press.
Roberts, Ian. 1988. Predicative APs. Linguistic Inquiry 19:703-710.
Rothstein, Susan. 1983. The syntactic forms of predication. Doctoral dissertation,
Massachusetts Institute of Technology.
Williams, Edwin. 1980. Predication. Linguistic Inquiry 11:203-238.
Williams, Edwin. 1981. Argument structure and morphology. The Linguistic Review
1:81-114.
Williams, Edwin. 1989. The anaphoric nature of 9-roles. Linguistic Inquiry 20:425^-56.
Downloaded from https:/www.cambridge.org/core. New York University, on 18 May 2017 at 18:26:13, subject to the Cambridge Core
terms of use, available at https:/www.cambridge.org/core/terms. https://doi.org/10.1017/S0008413100017680