Professional Documents
Culture Documents
Eliminating D Structure
Eliminating D Structure
ON ELIMINATING D-STRUCTURE:
THE CASE OF BINOMINAL EACH
Cedric Boeckx and Norbert Hornstein
Abstract. This paper provides a minimalist analysis of the binominal each phenomenon.
The analysis incorporates key ideas behind Sar and Stowells (1988) seminal paper but
avoids the complications that this approach entails. Our proposal provides one more
empirical argument for movement into h-position, sideward movement, the primacy of
movement over binding in matters of construal, and the virtues of a very derivational view
of syntax. It is also consistent with a framework that dispenses with LF movement entirely.
a.
b.
c.
d.
The faculty of language (FL) generates objects that pair sounds and
meaning.
Thus, FL objects must interface at least with the interpretive components that ascribe a sound and meaning to each generated object.
By assumption, interfacing takes place through levels of representation.1 Call the interface level for sound PF (Phonetic Form) and the
interface level for meaning LF (Logical Form). Both PF and LF
follow from virtual conceptual necessity.
Given the necessity of the interface levels LF and PF, minimalist
reasoning suggests that UG has no more than these. In fact, minimalist reasoning leads to the conclusion that grammar-internal levels
should be eliminated, leaving only interface levels like LF and PF.
If so, D-Structure (DS), a phrase marker formed by phrase structure
operations and lexical insertion and that feeds movement operations,
should not exist.
This is by no means a necessary conclusion. It could be that UG is not organized into levels, in
which case the grammar will still interface with AP and CI but not necessarily via levels. Epstein,
Groat, Kawashima, and Kitahara (1998), Uriagereka (1999), and Chomsky (2000, 2001) explore
this possibility. For present purposes, we keep to the more traditional assumption of interface
levels, which by and large seem well motivated empirically.
Blackwell Publishing Ltd, 2005. Published by Blackwell Publishing, 9600 Garsington Road, Oxford OX4 2DQ, UK and 350
Main Street, Malden, MA 02148, USA
As discussed extensively in Burzio 1986 and Sar and Stowell 1988 (S&S),
the phrase containing each in (2) (one book each) (henceforth the BE phrase)
must be licensed by an appropriate plural antecedent for the sentence to be
acceptable. (The same requirement holds for apiece phrases.4)
The problem posed by BE for the MP is that the evidence suggests that the
licensing of BE holds at DS. In fact, as discussed in this section, if we assume
that BE phrases must be locally bound at DS (i.e., they are DS anaphors), we
derive virtually all of their salient properties.
Consider rst the salient properties of BE.
2.1 BEs Require Antecedents
That BE requires an antecedent is illustrated by the unacceptability of (3).
(3)
The generalization illustrated by (2) and (3) is that BEs only occur in clauses
with multiple nominals, one of which is plural. This follows if one assumes
2
The reader should note that we focus exclusively on properties of English binominal each in
this paper. We leave a crosslinguistic investigation for future research. For work on binominal each
in other languages, see Gil 1982, and Zimmermann 2002, which draws in part on Choe 1987 (on
Korean) and Link 1998 (on German).
3
The name comes from Sar and Stowell 1988. Our discussion heavily leans on this earlier
analysis. We steal shamelessly from their earlier efforts. Their work, in turn, was based on a prior
analysis and discussion in Burzio 1986, where the BEs were rst extensively discussed.
4
Very often in this paper we exemplify core generalization by means of apiece. We do so for
two reasons. First, the apiece data is often clearer than the each facts (for reasons that we do not
understand). Second, there is little temptation to analyze apiece as a quantier rather than an
adverb. The fact that BEs with apiece have all the same properties as those with each suggests that
whatever is going on here should not be traced to the QP status of each.
a.
b.
a.
b.
c.
d.
e.
The contrast between (6a) and (6b) follows on the assumption that the subject
asymmetrically c-commands the object in simple transitive clauses. To
accommodate the contrast in (6c,d) we need to assume that the prepositions
that accompany indirect objects and commitatives do not block binding. This
seems necessary even in simple cases of reciprocal binding.
(7)
Indirect objects and commitatives are argument-like adjuncts (quasiarguments) in that they are typically thought to be generated lower than
those headed by other prepositions (like before/after). In many languages they
affect the morphology of the verb and they readily allow for preposition
stranding, as shown in (8).
(8)
a.
b.
It is plausible that these prepositions function more like Case-markers than like
true PPs. If this is so, then the men can bind the BEs in (6c) but not in (6d,e).
2.3 BE Must Be Locally Bound
The plural antecedent of BE cannot be in a different domain than BE. If we
assume that being in the same minimal clause puts two expressions in the same
5
See S&S for a fuller description than the one we provide here.
a.
The BE in both examples is in the embedded clause and its antecedent, the men,
is in the matrix. Consequently, BE cannot be locally bound and thereby licensed.
So far none of the requirements listed in this section require the assumption
that the licensing of BEs be at DS (though they are consistent with it). The
explanations would remain the same if we assumed that the binding
requirement was an LF condition. However, two additional sets of data
support the assumption that BEs are DS-anaphors.
2.4 ECM Contexts
The rst set of facts shows that BEs are not licensed when in the subject
position of nonnite clauses. As is well known, subjects of nonnite clauses
act with respect to binding as if they occupied the next higher clause. Thus,
ECM subjects can be bound by an antecedent in the next higher clause and
pronouns must be obviative with respect to elements in this clause.
(10)
a.
b.
c.
This follows if we assume that the ECM subject actually raises to the clause of
its case checker/assigner in overt syntax or LF (see Postal 1974 and Lasnik &
Saito 1991). If so, then movement renders the ECM subject a part of the upper
clause.
With this in mind, consider the following BE data.6
(11) a.
b.
c.
The sentences in (11) show that BEs are not acceptable as ECM subjects. This
follows if BE is not part of the same clause as its antecedent at the point where
6
(i)
a.
b.
(13)
a.
b.
(14)
These sorts of cleft sentences display all the connectedness effects found in standard reconstruction
cases. As such, we assume that they too would be formed via movement (see Kim 1998 and Boeckx
1999 for possible analyses). Note, moreover, that if reconstruction is relevant here, then cleft and
tough-constructions like (14) cannot be formed via null-operator movement (see Hornstein 2001
and Kiguchi 2002 for independent arguments that this is the case for tough-constructions). The
reason is that null operators will not provide an opportunity for reconstruction. We need an actual
copy, and given current assumptions, only movement of the actual expression will sufce.
Blackwell Publishing Ltd, 2005
a.
b.
The relevant reading of these sentences is the one in which the PP headed by
with modies the matrix verb. Under this reading, the PP is situated in the
matrix clause and so is not c-commanded by the small clause subject (the
prisoners, Sam and Tom) at DS. The BEs are, however, c-commanded by LF if
ECM subjects must raise to the matrix for case reasons (see Lasnik & Saito
1991). Thus, examples like (15) are problematic if one assumes that BEs are
licensed at DS via anaphoric binding.8
8
S&S were working under the GB assumption that ECM subjects do not raise to the domain of
their Case checker and so concluded that there was no c-command in such cases. Subsequent
research (see Lasnik & Saito 1991), however, has shown that a return to Postals (1974) analysis of
ECM as subject-to-object raising is superior.
a.
b.
The three arguments in this section suggest that an account like the one we
seemed to be led to in section 2 is not adequate. S&S offer an alternative
analysis, which we review in section 4. We point out problems for their
analysis in section 5 and propose an alternative in section 6. Section 7 is the
conclusion.
4. Sar and Stowell 1988:10 LF Movement
The conclusion we can draw from sections 2 and 3 is that a binding approach is
empirically problematic. What then accounts for the distribution and interpretation of BEs? S&S propose to rethink the basic facts in terms of LF movement
rather than binding.11 Their analysis rests on the following assumptions.
(18)
a.
b.
c.
d.
e.
f.
g.
If control is accounted for in terms of A-movement (Hornstein 1999, 2001; Manzini &
Roussou 2000; ONeil 1997; Kayne 2002), this argument is neutralized. However, to get off the
ground, a movement approach to control must dispense with notions like DS.
10
Our focus on S&S should not imply that S&Ss analysis is the only one worth considering.
We decided to concentrate on S&S because we think that its spirit can be easily accommodated to
the Minimalist Program, as we will show.
11
S&S conclude that a binding approach is inadequate. They do not fully discuss the notion that
BEs are DS anaphors. Thus, our discussionin particular, our emphasis on the role of DSis a
reconstruction of their paper.
Blackwell Publishing Ltd, 2005
(20)
[TP The men1 [TP [apiece EC1]2 [TP t1 read [two books t2]]]]
In (21) the each-phrase moves out of an embedded subject on its way to matrix TP.
There is one more assumption that S&S need to accommodate all of the relevant
data. They need to assume that it is possible to move from reconstructed
positions. This is required to handle the facts in section 2.5. Recall that it is
possible to have subject BEs so long as these are derived subjects formed by
raising or passive. We have just observed that S&S prevent LF movement out of
subjects to explain the ECM facts. This leaves cases like (22) a problem.
(22)
The LF in (22), shown in (23), is illicit as the each-phrase has moved from the
subject.
(23)
[TP The diplomats1 [TP [each/apiece EC]2 [TP [one interpreter t2] was
assigned t2 to t1]]]
To nesse this problem, S&S assume that one can reconstruct the derived
subject to its object position and move from there. The correct LF, therefore, is
(24), with the LF movement taking place from the underlying object position.
(24)
[TP The diplomats1 [TP [each/apiece EC]2 [TP t1 was assigned [one
interpreter t2] to t1]]]
Note that the movement has taken place from the second copy and so should
not violate the Subject Condition. The problem is that there is evidence that
such movement is not generally licit. If it were, we would expect to nd no
Subject Condition effects even for overt movement. Consider (26).
(26)
*What2 does John believe [[books about t2]1 to have been burned t1?
a.
b.
c.
d.
a.
b.
c.
d.
(30)
[TP [The men]1 T0 [vP t0t [VP read [NP [NP one book] [BE-P each/apiece t1]]]]]
The overt syntax structure in (30) is achieved by assuming that the R-NP the
men begins its derivational life as a complement of each/apiece, which in turns
heads the BE phrase. This phrase is adjoined to and modies the D-NP one
book (see (28b)). The D-NP and R-NP are interpretively related to one another
along the lines of (28) in this conguration. The anaphoric dependence in
(28c) translates as a structural dependence akin to the relation assumed to hold
between a clitic and its double in clitic-doubling structures (Kayne 1972,
Torrego 1986, Uriagereka 1995) and between a resumptive pronoun and its
antecedent (Boeckx 2001, 2003a)12 and indeed between an anaphor and its
antecedent (Kayne 2002, Zwart 2002).
12
A doubling structure has also been suggested to capture the relationship between an anaphor
and its antecedent (Kayne 2002, Zwart 2002). We do not adopt this analysis for anaphors, for
reasons developed in Hornstein 2004, and Boeckx, Hornstein, and Nunes 2004. (We favor a view
that takes anaphors to be [possibly morphologically modied] copies of their antecedents.)
With so much as background, lets quickly consider the facts concerning BEs
reviewed at the outset and see how they follow given this proposal.
The rst fact is that BEs require antecedents. In particular, if a sentence
contains a nominal and a BE, then it must contain at least one other nominal as
well. We illustrated this (following S&S) by contrasting BEs with clauses
involving oated quantiers.
(32)
13
Locality conditions such as Minimality/Shortest Move are observed, as the R-NP and D-NP
are plausibly equidistant from Spec,vP.
14
See also Hornstein 2001 and Nunes 1995, 2001, 2004.
15
Let us note that several authors have independently argued that adjuncts contained inside VP
are porous; see Uriagereka 1988 and Landau 1999, among others.
16
See Yoshida 2003 for a related point.
17
We include subjects with numeral quantiers. As S&S note, the R-NP typically has this form.
[Three men apiece DP] have arrived [[three men] apiece DP]
(33) assumes that there is some DP in the complement position of the BE. It
needs a h-role to be licensed.19 But three men apiece DP occupies the only
thematic position, so there is no way for it to get one. Another derivational
option is (34).
(34)
[Three men apiece t1] have arrived [[three men]1 apiece t1]
Here we move three men from the complement position of apiece and
promote it in a kind of relativization strategy. This too leads to
unacceptability, given the interpretive requirements of BEs outlined in (28).
As S&S observe, (28) requires that no two R-individuals map to the same set.
However, given the derivation in (34), there is only one set, that provided by
the moved three men. As such, the requirement in (28) cannot be met and the
sentence is thus interpretive gibberish. As these are the only two possible
derivations for the sentence, its unacceptability follows.20
The second fact that our analysis captures is that the R-NP cannot be an
external argument. For example, sentences like (35), with a derivation like
(36), are unacceptable. (The sentence has the intended interpretation in which
the books is the R-NP and three men apiece the D-NP.)
(35)
(36)
[[Three men apiece t1] T0 [[three men apiece t1][read [the books]]]]
"j
Under (36), the book moves from the complement of apiece to the internal
argument position of read, thereby getting a h-role. Three men apiece then
raises to Spec,TP for Case-checking purposes. The problem is that the rst
movement involves lowering in overt syntax, and this is prohibited. Lowering
18
A reviewer wonders how adjoined material like the each phrase can function as a predicate
taking the element it is adjoined to as its argument. We refer the reader to Hornstein and
Uriagereka 2002 and Boeckx 2003b for discussion of how quantiers function as predicates at LF
via a mechanism of reprojection.
19
More exactly, we assume that it needs a verbal/h-role. Even if it has some interpretive role
in virtue of being the complement of apiece, we assume this is not enough. We follow Baker 2003
in taking adjectives (/adjuncts) to be unable to assign h-roles of their own.
20
S&S note that there may also be a Case Filter violation here.
a. The men believed [that [three books each the men] were exciting]
"j
b. The men believed Bill to be reading one book apiece the men.
"j
(38a) moves the men out of a subject of an embedded nite clause, whereas
in (38b) we move the nominal the men over an intervening c-commanding
nominal Bill in violation of minimality. Both movements are illicit and so the
derivations are blocked.22
Our analysis is also able to derive the reconstruction cases that turned out
to be problematic for S&S.23 Take, for example, (39).
(39) How many books each did the boys read?
(39) has the derivation in (40).
#j
(40) [CP [TP the boys T [the boys [read [how many books each the boys]]]]
"
21
There are various other potential violations. Lowering plausibly prevents linearization of
copies if these must be within a chain, as Nunes (1995, 2004) proposes. Additionally, current
Agree-based theories (Chomsky 2000, 2001) require the targets of movement to c-command the
object moved. Either or all of these conditions sufce to prevent overt lowering.
22
Recall that movement across one book does not violate anything, as the BE is adjoined to this
nominal and so minimality does not come into play.
23
Tim Stowell (p.c.) observes that the absence of intermediate-trace reconstruction effects
with BE (unlike with regular anaphors), illustrated in (i), follows from our account, if the Amovement of [the boys] has to precede remnant movement of the wh-phrase and cannot skip over
the intervening subject. (S&S derived (i) in a similar but not identical wayby requiring QR of
each from reconstructed wh-phrase in the original object position [CED] and from the [tensed]
clause-boundedness of QR.)
(i) *How many books eachi did the boysi say the girls bought t?
Blackwell Publishing Ltd, 2005
S&S accounted for these by assuming that they involved movement out of
embedded subjects at LF, which was illicit. Setting aside the issue of island
violation for covert movement, S&Ss account does not go through since S&S
independently allowed movement from reconstructed positions. However, we
do not assume movement out of a reconstructed position, and all our relevant
movements are overt, so we can appeal to an island effect to rule out (41).
More precisely, we assume that (overt) movement out of an element that has
raised to matrix Spec,AgrOP is disallowed. This is Lasniks (2001) conclusion,
based on cases like (42).24
(42)
a.
The men threw out three bags each.
b. *?The men threw three bags each out.
a.
b.
That (44) closely parallels (43) supports the proposal that (overt) extraction
out the BE takes place.26
Having accounted for the impossibility of BEs as ECMed subjects/raised
objects, let us now turn to situations in which the R-NP is an ECMed subject,
as in (45).
(45)
a.
b.
Given the logic of our analysis, the derivation for, say, (45a) would be as in (46).
(46)
a.
b.
c.
d.
26
(i) Two women got the prize money and a silver medal each.
This would appear to violate the Coordinate Structure Constraint with two women moving to
Spec,vP from the right conjunct. We believe this example to be somewhat misleading. Structurally
similar cases are considerably less acceptable.
(ii) Two women swallowed a/*?this peanut and a raisin each/apiece.
(iii) Two women broke a/*?this plate and a cup each/apiece.
The only acceptable this-reading of (iii) is one where there are two separate events where one and
the same peanut is swallowed twice. As this is implausible, the sentence is odd. Similarly with (iv).
Why then is (i) acceptable? We suspect that this is due to the fact that two people can each receive
a share of the prize money whereas shares of peanuts and plates are less common.
(iv) Two women got a share of the prize money and a silver medal each.
When the prize is made difcult to share, the oddity resurfaces.
(v) Two women got a/*?the silver dollar and a medal each.
Blackwell Publishing Ltd, 2005
f.
The crucial aspect of this derivation is that we delay the adjunction of the PP
containing BE until the matrix vP is introduced. This is possible if we
capitalize on the possibility of sideward movement (movement across
unconnected subtrees).27 But we want to make sure that the derivation in
(53) hasnt opened the door to unwanted derivations.
In this context, consider again (35), repeated here as (47).
(47)
The derivation produced in (36) relied on the fact that the books must move
from the subject nominal three men apiece the books. This forces the books to
lower, and as this is not permissible, the derivation is unsuccessful. But what
prevents the derivation in (48)?
(48)
a.
b.
c.
d.
e.
f.
g.
So far as we can tell, (48) would render (47) grammatical. The derivation can
be blocked if we require that the BE merge to the nominal it modies as soon
as it is able to do so. Recall that we followed S&S in essentially considering
BEs as headed by transitive predicates. Thus, apiece/each have internal and
external arguments that are satised or discharged via Merge. Lets assume
that once one begins checking or assigning the h-roles of a predicate, that this
must continue until all the h-roles of that predicate are saturated. This has been
independently assumed by Frampton and Gutmann (1999, 2002), who claim
27
Since this possibility has been amply justied empirically and conceptually in Nunes 1995,
2001, 2004, and Hornstein 2001, we simply adopt it here without further justication.
Say that something like the Locus Principle is correct, and say that S&S are
right in thinking that BEs are species of transitive predicates. If so, once we
begin to form the BE, we must nish with it before moving onto something
else. In effect, this will block the movement in (48b) as we have not yet
discharged the external argument of the BE at the point where we move its
complement. Thus, the net effect of assuming (49) is to force the BE into
constituency with the nominal it modies. This will necessarily end by
yielding a structure like (36), which, we have seen, is sufcient to prevent the
derivation of (35).
There is a nal derivation to consider for (47), one in which we build the
whole BE (saturating each/apiece), sideward move the R-NP the books to the
object of read, build up the tree up to v, Merge the entire BE in Spec,vP, and raise
three men to Spec,TP.28 This derivation is excluded as it fails to be linearized.
The books would have to both precede and follow read according to Nuness
(2004) linearization of chains procedure based on Kaynes (1994) LCA.
All in all, an overt movement analysis of the licensing of BE successfully
covers the relevant data. We are further able to explain three additional
properties of BEs.
First, why is it that they follow the D-NP? We can, once again, adopt the
S&S proposal. They note that adjectives with complements must succeed the
nominals they modify. Witness (50).
(50)
a.
b.
If BEs have complements, as assumed here, then the same restriction can
account for why they are barred from prenominal positions.
(51)
*The men read most books each/the books apiece/every book each.
This follows once we note that movement out of nominals with strong
determiners is generally prohibited.
(53)
Cedric Boeckx
Harvard University
Department of Linguistics
Boylston Hall, Third Floor
Cambridge, MA 02138
USA
cboeckx@fas.harvard.edu
Norbert Hornstein
University of Maryland
Department of Linguistics
1401 Marie Mount Hall
College Park, MD 20742-7505
USA
nh10@umail.umd.edu