Head Parameter as Encoded in Functional Categories

Jun Abe

May, 2015

E-mail: [email protected]

Head Parameter as Encoded in Functional Categories

Abstract: In this article, I argue that linear ordering should be dealt with on a par with labeling in the sense of Chomsky (2013) and hence that the linear ordering algorithm (LOA) applies when the syntactic objects in a given phase domain are transferred. I propose a certain version of LOA, arguing that the head parameter is encoded to resolve the underspecification of the head-complement order as a result of applying the LOA. Further, I argue for a hybrid position in which lexical categories have only asymmetrical structures while functional categories allow symmetrical structures with the head parameter encoded. Keywords: head parameter, labeling, linear ordering, lexical vs. functional category, symmetrical vs. asymmetrical structure

1

Introduction

This article aims to examine how to characterize the head parameter under the minimalist framework. I argue, contrary to current efforts to eliminate it (see Kayne (1994) and Fukui and Takano (1998), among others), that the head parameter actually exists and that it is encoded to resolve the underspecification of the head-complement order as a result of applying what I call linear ordering algorithm, which operates upon Transfer on a par with the labeling algorithm proposed by Chomsky (2013). Further, I argue that the head parameter is encoded only in functional categories, taking a hybrid position in which lexical categories have only asymmetrical structures

1

a la Kayne (1994) and Fukui and Takano (1998) while functional categories allow symmetrical structures with the head parameter encoded. This hybrid approach will give support to Fukui’s (1995) Functional Parameterization Hypothesis (FPH), according to which lexical projections are uniform among languages and parameterization is attributed to functional categories. The article is organized as follows: Section 2 addresses the question how to deal with linear order under the minimalist framework. It is argued that linear ordering should be dealt with on a par with labeling in the sense of Chomsky (2013). Accordingly, Section 3 discusses how the linear ordering algorithm should be formulated. It is argued, following Abe (2001), that it is formulated in terms of visibility of labels and that the head parameter comes in for resolving the underspecification of the head-complement order. In Section 4, I argue that the head parameter is encoded only in functional categories. I argue for a hybrid approach according to which VP uniformly has an asymmetrical structure in which what precedes is structurally higher than what follows while functional categories above VP allow structures in which what precedes is structurally lower than what follows. I propose that the latter property follows from the possibility of right adjunction allowed by the head-initial value of the head parameter. Section 5 discusses two consequences of our theory of labeling and linear ordering: one has to do with heavy NP shift and the standard analysis of right adjunction of heavy NPs is defended by addressing the question why shifted heavy NPs behave as if they are lowered in some cases. The other has to do with the internal structure of DP and it is argued that

2

postnominal adjectives in English are incorporated into nominal structures by right adjunction in accordance with the head-initial value of the head parameter. Section 6 concludes with a brief summary.

2

How to Deal with Linear Order?

Chomsky (2013) holds the following thesis regarding linear ordering: (1) “Order and other arrangements are a peripheral part of language, related solely to externalization at the SM [=sensorimoter] interface, where of course they are necessary.”

(Chomsky 2013:36)

Accordingly, Chomsky claims that Aristotle’s dictum should be modified in such a way that “language is not sound with meaning, but rather meaning with sound,” since “language is primarily an instrument of thought.” This thesis is supported by the standard observation that those principles and constraints that regulate long dependency among syntactic objects are properly characterized in hierarchical terms rather than in terms of linear order. Chomsky (2013) notes one such case from binding theory, in which it was once assumed that linear order was essential in capturing anaphoric relations but this assumption was refuted by Reinhart (1976), who proposes that such a hierarchical notion as c-command suffices in capturing core data concerning anaphoric relations. Given this observation, it may be reasonable to claim that “order is a peripheral part of language,” relevant only for externalization at the SM interface, and hence that it plays no role in the computational system of narrow syntax.

3

Along this line of thought about meaning and sound, Chomsky (2013) seems to assume, though he does not explicitly mention it, that linearization of syntactic objects should be conducted outside narrow syntax, somewhere along the course to the SM interface (in the phonological component, according to Chomsky (2008)). On the other hand, Chomsky (2013) takes a different position on projection/labeling:1 (2) “Projection is a theory-internal notion, part of the computational process GP [= generative procedure]. For a syntactic object SO to be interpreted, some information is necessary about it: what kind of object is it? Labeling is the process of providing that information. Under PSG [= phrase structure grammar] and its offshoots, labeling is part of the process of forming a syntactic object SO. But that is no longer true when the stipulations of these systems are eliminated in the simpler Merge-based conception of UG. We assume, then, that there is a fixed labeling algorithm LA that licenses SOs so that they can be interpreted at the interfaces, operating at the phase level along with other operations.” (Chomsky 2013:43) According to this conception, labeling is necessary for an SO to be properly interpreted at the conceptual-intentional (CI) interface rather than to play any role in syntactic computation, hence conducted by LA upon Transfer. However, given that Transfer is a symmetrical operation in that it sends an SO constructed in narrow syntax with relevant information not only to the semantic but also to the phonological component, it is most natural to claim that linear ordering is also conducted by what may be called linear ordering algorithm (LOA) upon Transfer. As support for this

4

conjecture, let us replace the underlined “labeling” in the above passage in (2) by “linear ordering” to see if the resulting passage still makes sense: (3) “… For a syntactic object SO to be interpreted [at the SM interface], some information is necessary about it: … Linear ordering is the process of providing that information. Under PSG and its offshoots, linear ordering is part of the process of forming a syntactic object SO. But that is no longer true when the stipulations of these systems are eliminated in the simpler Merge-based conception of UG. We assume, then, that there is a fixed linear ordering algorithm LOA that licenses SOs so that they can be interpreted at the interfaces, operating at the phase level along with other operations.” According to this conception, linear ordering is necessary for an SO to be properly interpreted at the SM interface, just as labeling is at the CI interface, hence conducted by LOA upon Transfer.2 Crucially, linear ordering is not required to play any role in syntactic computation, a view in accord with the standard observation that linear order is irrelevant for formalizing principles and constraints that regulate long dependency such as anaphoric relations captured by binding theory. By the same logic, it follows that labels play no more role in syntactic computation than linear order, which is likely to be the case.3 This conception, if correct, may lead to the thesis alternative to Chomsky’s (2013) according to which language is simply a medium of sound and meaning, treating the two interfaces on a par. It seems, however, that there is one crucial respect in which labeling is more fundamental than linear ordering: whereas LA is

5

simply minimal search for a head in a given syntactic structure, LOA needs to be formulated in terms of labeling, as we see in the next section. In this sense, linear order may be claimed to be a more “peripheral part of language” than labels. Adopting the symmetrical view on labeling and linear ordering, I consider in the next section how LOA is formulated and how the head parameter plays a role in this formulation.

3

Linear Ordering Algorithm and Head Parameter

Abe (2001) proposes one version of LOA in terms of the visibility of a labeled SO: (4) When α and β merge to make K, so that K dominates α and β, α precedes β (α > β) if α is visible and β is invisible. Here the visibility of an SO is determined on the basis of relevance to interpretation at the CI interface. Chomsky (1995:242) assumes that “bare output conditions [= interface conditions] make the concepts ‘minimal and maximal projection’ available [=visible] to CHL,” since only these projections are relevant to interpretation at the CI interface. Under this assumption, (4), in effect, produces structures in which maximal and minimal categories precede intermediate categories. Thus, consider the following schematic structure, expressed by X’-theoretic notations:

6

(5)

XP YP1

X’1 YP2

X’2 X

ZP

In the merged pairs {YP1, X’1} and {YP2, X’2}, LOA in (4) determines that YP1 and YP2 precede X’1 and X’2, respectively, since the latter phrases are intermediate projections of X. On the other hand, this algorithm is silent about the merged pair {X, ZP} since both categories are visible in the sense relevant for (4). Thus, this algorithm guarantees that specifiers are always on the left side of given nodes, producing

strictly

right

branching

structures,

except

for

one

case:

the

head-complement relation. Given the minimalist thesis that the faculty of language (FL) is an optimal solution to interface conditions, Abe (2001) claims that LOA in (4) is an optimal solution to the determination of linear ordering and hence that to keep this optimality, the linear ordering of a head to its complement is determined by implanting a parameter. This is one illustration of the underspecification view on parameters. According to this view, there may exist a situation in FL where a solution to a requirement imposed by the interfaces is more optimal if a piece of that solution is left undecided, or left to the decision by experience. This is where a parameter comes in for dealing with such a situation in an optimal way. Assuming LOA in (4) and Abe’s (2001) proposal about how the head parameter plays a role in this algorithm, let us consider more closely how labeled SOs are

7

assigned linear order under Chomsky’s (2013) LA, which applies in terms of minimal search for a head. Thus, when a given SO consists of {X, YP}, then the head X is picked out for the labeling of SO, as in (6a), whereas when it consists of {XP, YP}, as in (6b), then the label of SO remains undetermined, hence uninterpretable at the CI interface. (6) a.

Label({X, YP}) = X

b.

Label({XP, YP}) = ?

This labeling algorithm causes a problem for determining the labels of the merged pairs {YP1, X’1} and {YP2, X’2} in (5), since these pairs belong to the case of (6b). This in turn causes a problem for LOA in (4) in determining the linear order of these pairs. To take a concrete case for illustration, let us consider the internal structure of vP/VP in Japanese. A relevant example is given below: (7) John-ga

Mary-ni

hon-o

ageta.

John-NOM Mary-DAT book-ACC gave ‘John gave Mary a book.’ Hoji (1985) convincingly demonstrates that Japanese has strictly binary structures and

further

that

Sub(ject)-I(ndirect)O(bject)-D(irect)O(bject)-V

reflects

the

underlying word order. Thus, adopting the vP-VP layered structure, we can assign the following vP structure to (7):

8

(8)

vP Sub

v’ VP

IO

v V’

DO

V

If we assume that this is the right vP structure for such a double object construction as in (7) in Japanese, then it causes a {XP, YP} problem for labeling; though the {Sub, v’} pair will not cause a problem under the assumption that Sub is raised into Spec-TP, so that its copy left behind is invisible for labeling, as assumed by Chomsky (2013), the {IO, V’} pair remains problematic unless IO is raised to the Spec of a functional category below TP. Given that no evidence for such raising of IO is attested, I suggest that the {XP, YP} problem that arises in the internal structure of vP is resolved by adding a special way of labeling in terms of θ-relations to the original LA:4 (9) Given K = {XP, YP}, if XP is an argument of YP, then YP is the label of K. This in effect weakens the role of minimal search for labeling in that the structure building of a category that involves predicate-argument relations is conducted in such a way that a predicate projects all the way up to the point at which it accommodates all its arguments (except for its external argument). Given this assumption, the SOs formed by merging the pairs {Sub, v’} and {IO, V’} are properly labeled vP and VP, respectively. Accordingly, LOA in (4) assigns Sub > v’ and IO > V’ to these pairs.

9

Further, given that V obligatorily raises to v, we need not determine the linear order of the {DO, V} pair in (8). As for the {VP, v+V} pair, I follow Abe (2001) in assuming that the head parameter decides the linear order of this pair: Since Japanese has the head-final value for this parameter, this pair is linearized as VP > v+V. In this way, we correctly obtain the right word order Sub-IO-DO-V from structure (8).

4

Functional Heads as Locus of the Head Parameter

In this section, I argue, following Abe (2001), that the way the head-parameter is implanted in LOA obeys Fukui’s (1995) FPH, so that this parameter is encoded only in functional categories. In Section 4.1, I demonstrate, by comparing VP structures of English

and

Japanese,

that

though

these

two

languages

have

opposite

head-complement orders, their VP structures basically have strictly right branching structures, so that what precedes is structurally higher than what follows, as expected by asymmetrical structure proponents such as Kayne (1994) and Fukui and Takano (1998). In Section 4.2, on the other hand, I demonstrate that adjunction structures of such a head-initial language as English allow a case in which what precedes is structurally lower than what follows. I argue that adjunction is allowed only in functional categories and that, following Fukui (1993), the value of the head parameter should be preserved in derived structures, so that an adjunction operation should create a structure that is consistent with the value of the head parameter in a given language. This in effect allows a head-initial language such as English to have right-adjoined structures in a functional category above VP.

10

4.1

Argument for the Asymmetrical Structure Thesis

Kayne (1994) proposes that hierarchical structures unambiguously determine the linear order of terminal symbols, thereby denying the existence of symmetrical structures; let us call this the asymmetrical structure thesis. According to Kayne, hierarchy and precedence correspond roughly as follows: what precedes is structurally higher than what follows. He further argues that the universal basic word order is S(pecifier)-H(ead)-C(omplement) and that various word orders exhibited by different languages are derived from this basic word order by applying movement operations to heads and phrases. Thus, this thesis denies the existence of the head-parameter. Takano (1996, 1998) and Fukui and Takano (1998) also argue for this thesis, though they claim, contrary to Kayne, that the universal basic word order is S-C-H. An initial motivation for the asymmetrical structure thesis comes from the following observation, described by Fukui and Takano (1998:34): “It seems almost universally true, as Kayne (1994) claims, that a Spec of the head H precedes rather than follows H and its complement, regardless of the relative order between H and its complement.” Along this line of observation, what I think is the strongest empirical argument for the asymmetrical structure thesis is found in Takano (1996, 1998), who argues for this thesis, following Kayne (1994), though he claims that the universal basic word order is S-C-H, contrary to Kayne. Further, he derives the effects of the head parameter from the (un)availability of movement of a lexical head to a higher

11

functional projection. Takano (1996, 1998) argues for the asymmetrical structure thesis based upon the observation that despite the difference in word order, English and Japanese show the same results with respect to the binding tests that are intended to examine the hierarchical structures of VPs; that is, within VP, what precedes is structurally higher than what follows in both languages. Recall that we have seen, following Hoji (1985), that Japanese has the following underlying v-VP structure: (10)

vP Sub

v’ VP

IO

v V’

DO

V

That subject is structurally higher than other arguments is already shown in Saito (1985), whose main purpose is to argue against the claim that Japanese has a flat clausal structure. Though I refer the reader to Saito (1985) and Hoji (1985) for a vast amount of evidence to support the structure given in (10), I will show one empirical argument for this structure, relying on Takano’s (1996, 1998) accounts of the following paradigm of data: (11) a.

Mary-ga

[subete-no

gakusei]1-ni

soitu1-no

Mary-NOM every-GEN

student-DAT

his/her-GEN teacher-ACC

12

sensei-o

syookaisita. introduced ‘Mary introduced his/her1 teacher to every student1.’ b.

*Mary-ga

soitu1-no

sensei-ni

[subete-no gakusei]1-o

Mary-NOM his/her-GEN teacher-DAT every-GEN student-ACC syookaisita. introduced ‘Mary introduced every student1 to his/her1 teacher.’ c.

Mary-ga [subete-no gakusei]1-o soitu1-no sensei-ni syookaisita.

d.

?Mary-ga soitu1-no sensei-o [subete-no gakusei]1-ni syookaisita.

The contrast between (11a) and (11b) shows that IO binds DO in the IO-DO order, but not conversely. This indicates that what precedes (IO, in this case) is structurally higher than what follows (DO). A different pattern shows up with the DO-IO order, however, as shown in (11c, d). (11c) shows that DO binds IO in the DO-IO order, and the fair acceptability of (11d) indicates that IO can also bind DO somehow in this order. If what follows does not bind what precedes, as is the case with (11b), we will expect that (11d) is unacceptable, contrary to fact. This fact, then, indicates that the basic word order in Japanese is IO-DO and that the DO-IO order is derived by moving DO to the left of IO, so that the acceptability of (11d) can be regarded as a result of connectivity effects of binding. Thus, in (11d), subete-no gakusei ‘every student’ successfully binds soitu ‘his/her’ to its left because the former c-commands the latter in the underlying structure.

13

Based upon Barss and Lasnik’s (1986) observation, Larson (1988) makes essentially the same claim as Hoji’s (1985) for the internal structure of VP in English: what precedes is structurally higher than what follows. The following examples illustrate this point: (12) a. b. (13) a. b.

I showed Mary herself. *I showed herself Mary. I showed Mary to herself. *I showed herself to Mary.

For ease of exposition, let us call double object constructions such as in (12) the IO-DO frame and their counterparts with reverse order such as in (13) the DO-IO frame, adapting Takano’s (1996, 1998) terminology. The contrast in acceptability between the (a) and (b) examples in (12) and (13) clearly shows that an internal argument asymmetrically binds another argument to its right. Larson (1988) proposes that the DO-IO frame reflects the basic word order and the IO-DO frame is derived from the DO-IO frame by a passive-like movement rule that moves IO to the left of DO. Aoun and Li (1989), Kitagawa (1994) and Pesetsky (1995) argue, however, that the opposite is true; that is, the IO-DO frame reflects the basic word order while the DO-IO frame is a derived word order.5 Let us first consider the following paradigm of data: (14) a.

I gave the mothers each other’s babies.

b.

I showed the parents each other’s boys.

(15) a.

*I gave each other’s mothers the babies.

14

b.

*I showed each other’s parents the boys.

These examples show that in the IO-DO frame, IO asymmetrically c-commands DO. Let us now compare the above paradigm with the following: (16) a.

I gave the babies to each other’s mothers.

b.

I showed the parents to each other’s boys.

(17) a.

?I gave each other’s babies to the mothers.

b.

?I showed each other’s parents to the boys.

The examples in (16) show that DO c-commands IO in the DO-IO frame. The fair acceptability of the examples in (17) is unexpected under Larson’s (1988) analysis, since the DO-IO frame reflects the basic word order and hence DO should asymmetrically c-command IO to its right. On the other hand, under the assumption that it is the IO-DO frame that reflects the basic word order, the acceptability of the examples in (17) is accounted for as a result of connectivity effects of binding; that is, in the DO-IO frame, IO c-commands DO in the underlying structure, hence licensing backward binding in the DO-IO order. The above facts fall into place if we assume that English has the following vP internal structure underlyingly:

15

(18)

vP Sub

v’ v

VP IO

V’ DO

V

Notice that (18) has exactly the same hierarchical structure as its Japanese counterpart (10) and hence that the binding facts shown in (12)-(17) are accounted for in exactly the same way as the Japanese facts shown in (11).6 On the other hand, in order to derive the difference in word order between English and Japanese, Takano (1996, 1998) proposes that in English, V in (18) moves up to Spec-vP whereas in Japanese, V stays in situ.7 Given that subject is moved to Spec-TP, it follows that English has Sub-V-IO-DO and Japanese has Sub-IO-DO-V as their base orders and that in both languages, the DO-IO order is derived by moving DO above IO. Under this account, the head parameter is replaced by another parameter concerning head movement in accordance with the asymmetrical structure thesis: if head movement takes place, it induces a head-initial structure whereas if it does not, it induces a head-final structure. Under the present theory of linear ordering that incorporates LOA in (4) and the head parameter, on the other hand, the English vP internal structure given in (18) is mapped into the correct word order in the following way: the linear orders of {Sub, v’} and {IO, V’} are determined by LOA as Sub > v’ and IO > V’, respectively. Though LOA is silent about the linear order of {DO, V}, this does not cause a

16

problem on the assumption that V is obligatorily raised to v. Since the head parameter encoded in this functional category is valued as head-initial in English, the linear order of {v+V, VP} is determined as v+V > VP. In this way, we can derive the correct word order from (18). In the following subsections, I argue that the present theory of linear ordering that incorporates the head parameter is superior to a theory such as Takano’s (1996, 1998) that adopts the asymmetrical structure thesis.

4.2

Argument against the Asymmetrical Structure Thesis

In the preceding subsection, we have seen Takano’s (1996, 1998) argument for the asymmetrical structure thesis, in which it is shown that what precedes is structurally higher than what follows in both English and Japanese despite their difference in the head-complement order. In this subsection, I will show that even though Takano’s claim might be correct for vP internal structure, it faces serious empirical problems if we consider the whole clausal structure. I will demonstrate that there are cases where what follows is structurally higher than what precedes in English. These phenomena strongly suggest the invalidity of the asymmetrical structure thesis as it stands. Recall that the thrust of this thesis is that hierarchical structures unambiguously determine the linear order of terminal strings. According to Kayne (1994), the relationship between structure and linear order is determined in such a way that when a category A asymmetrically c-commands another category B, all terminal symbols dominated by A must precede those dominated by B. This is the essence of Kayne’s (1994) Linear Correspondence Axiom (LCA), which in effect licenses only syntactic

17

structures that preserve the property that what precedes is structurally higher than what follows. Fukui and Takano (1998) propose a different way of determining the relationship between hierarchical structure and linear order, and yet their mechanism has essentially the same effects as Kayne’s. It is interesting in this context to consider how the notion of c-command, proposed by Reinhart (1976), plays a role in regulating binding relations. She argues that a notion in terms of hierarchy rather than precedence is relevant for constraining such relations. In order to show this, Reinhart provides some evidence to the effect that what follows can be structurally higher than what precedes and that this affects binding possibilities. One case in point is found in the discussion of two types of PP, sentential PPs and verb phrasal PPs. Reinhart observes that an object pronoun can take as its antecedent the DP contained in a sentential PP that follows it, but not the DP contained in a verb phrasal PP. Let us compare the examples with sentential PPs in (19) with those with verb phrasal PPs in (20). (19) a.

We sent him1 to West Point in order to please Ben1’s mother.

b.

We’ll just have to fire him1 whether McIntosh1 likes it or not.

c.

Rosa won’t like him1 anymore, with Ben1’s mother hanging around all the time.

(20) a. b.

*Rosa tickled him1 with Ben1’s feather. *It’s time to put him1 in the baby1’s bed.

(Reinhart 1976:60)

Given that the PPs in (20) are within VP, the unacceptability of the examples in (20) is straightforwardly explained as a Condition C violation under Takano’s (1996,

18

1998) analysis of English VP structure, which is incorporated into the present proposal about the internal structure of vP, as shown in (18). Thus, (20a) will have the following vP structure: (21)

vP Rosa

v’ v

VP him

V’

with Ben’s feather

tickle

For Takano, the right word order of (20a) will be derived from this structure by moving the V tickle to Spec-vP as well as moving Rosa to Spec-TP, whereas under our theory of linear ordering, the amalgamated v+tickle precedes VP in accordance with the head parameter. A problem arises for the asymmetrical structure approach with respect to the acceptability of the examples in (19). If the sentential PPs were within VP, then these examples would be predicted to be unacceptable, just like those in (20). If these PPs are hanging somewhere outside VP, the question is exactly which positions they are located in. Notice that we are in a paradoxical situation in answering this question under the asymmetrical structure approach. The fact that the examples in (19) are grammatical indicates that the sentential PPs are structurally higher than the object pronoun him. On the other hand, the fact that him precedes these PPs indicates that there is some category dominating him which asymmetrically c-commands those PPs,

19

according to Kayne’s (1994) LCA. We can think of at least two positions for sentential PPs, PP1 and PP2 in the structure given below (order is irrelevant here): (22)

vP vP

PP1

v

VP VP

PP2

... him ... There will be a good sense in which him is claimed not to c-command either PP in (22), hence accounting for the acceptability of the sentences in (19). LCA requires that there be some category dominating him which asymmetrically c-commands PP1 or PP2. There are candidates that may satisfy this condition: the lower vP for PP1 and the lower VP for PP2. Suppose that this is true. Then, it implies that in the adjunction structure given below: (23)

α α

β

The lower segment α asymmetrically c-commands β. Though it might be possible to invent a definition of c-command that gives rise to this consequence, it induces an obviously wrong consequence under the asymmetrical structure thesis: when β is adjoined to α, β always follows α. This flatly opposes Kayne’s theory of phrase structure, according to which c-command is defined in such a way that only “left adjunction” is possible. Further, this will be problematic for analyzing head-final

20

languages such as Japanese, since the standard claim is that Japanese is a strictly head-final language and hence only left adjunction is possible.8 It is thus safe to conclude from the above discussion that there is no natural place to accommodate allegedly right-adjoined elements such as sentential PPs under the asymmetrical structure approach. The same sort of argument can be given by using cases of extraposition from subject position. Reinhart (1976) provides the following examples: (24) a.

Nobody would ever call her1 before noon who knows anything about Rosa1’s weird sleeping habits.

b.

So many people wrote to him1 that Brando1 couldn’t answer them all. (Reinhart 1976:43)

The acceptability of these examples shows that the extraposed clauses are adjoined to some functional category outside VP (or is adjoined to VP). In such a case, however, it is expected in Kayne’s theory of phrase structure that these clauses should appear to the left of VP, the wrong result. Hence, there is no position available to the extraposed clauses that satisfies both Condition C and LCA simultaneously.

4.3

Adjunction as Regulated by the Head Parameter

Under our hybrid approach in which the head parameter is encoded in functional categories, there is a natural way to accommodate cases such as (19) and (24). Let us first assume, following Fukui and Speas (1986) and Fukui (1986), that the distinction between adjunction and substitution does not exist in lexical categories and hence that

21

adjunction structure is only relevant for functional categories. Let us further assume, following Saito (1985), Fukui (1993), and Saito and Fukui (1998), among others, that the direction of adjunction is sensitive to the head parameter. Fukui (1993) claims that the value of the head parameter should be preserved in derived structures, so that an adjunction operation should create a structure that is consistent with the value of the head parameter in a given language. Thus, head-initial languages such as English must conform to the X’-schemata given in (25a), so that only right-adjunction is possible, whereas head-final languages such as Japanese must conform to (25b), so that only left-adjunction is permitted. (25) a.

X’ -> X/X’ YP

b.

X’ -> YP X/X’

To illustrate, let us take a case where a phrase is right-adjoined to T’, as shown below: (26)

TP Sub

T’ T’

T

XP vP

Here XP is allowed to be right-adjoined to T’, since this adjunction respects the head-first structure demanded by the head parameter in such a language as English. Under the present theory of labeling and linear ordering, we need to address how LA deals with adjoined phrases and how the direction of adjunction is determined by

22

LOA. Given the standard assumption that adjunction of XP to YP does not fully project YP, only creating two segments of YP, it is natural to claim that adjoining XP does not change the status of the labeled YP. Let us assume, following Chomsky (2004), that adjunction is an operation of pair Merge, as opposed to set Merge, so that when α is pair-merged with β, <α, β> is formed. We then assume the following: (27) The label of <α, β> is β. With this assumption, the label of in (26) is determined as T’. Then, how is the linear order of <α, β> determined? We can incorporate Fukui’s (1993) proposal about the relationship between the head parameter and the directionality of adjunction into LOA in the following way: (28) a. b.

<α, β> is linearized as β > α if the head of β is initial. <α, β> is linearized as α > β if the head of β is final.

in (26), then, is linearized as T’ > XP according to (28a). We need to say something about the linear ordering of {Sub, T’} in (26). Chomsky (2013) proposes a way of labeling {XP, YP} other than moving XP or YP, which arises from the situation where XP and YP share some relevant feature(s). This way of labeling may be called labeling by agreeing features, in which these features serve as labels for their mother nodes. Thus, {Sub, T’} in (26) is labeled φ by the agreeing features. How is this set ordered with this labeling, then? One plausible way to answer this question is to claim, following Saito and Fukui (1998), that the linear order of the two phrases that enter into a “Spec-head agreement” is inherently determined. Notice that there is an asymmetrical property in labeling by agreeing

23

features: one relevant feature is interpretable whereas the other is uninterpretable. Hence there is a good sense in which a category bearing an uninterpretable feature is invisible for interpretation at the CI interface. Recall our LOA formulated in (4), reproduced below: (29) When α and β merge to make K, so that K dominates α and β, α precedes β (α > β) if α is visible and β is invisible. where the visibility of an SO is determined on the basis of relevance to interpretation at the CI interface. Thus, for the pair {Sub, T’} in (26), it is reasonable to claim that Sub is visible in the sense that it carries interpretable φ-features while T’ is invisible in the sense that it carries uninterpretable ones. Given (29), it follows that Sub precedes T’. Given the present assumptions, it is by now obvious how we can assign the correct structures for the sentences in (19) and (24), which were problematic for the asymmetrical structure approach. The sentential PPs in (19) and the extraposed clauses in (24) can right-adjoin to T’, as shown in (26), hence high enough to escape from the c-command domain of the object pronouns, thereby immune from Condition C violations.9 Further, there is a prediction arising from the characterization of how pair-merged elements are linearized, as given in (28), which requires that pair-merged β must be a projection of a head. Notice that Sub and T’ in (26) do not constitute an endocentric structure since their mother node is labeled φ, determined by their agreeing features, hence lacking a head. It is then predicted that adjunction cannot be applied to a category whose label is determined by agreeing features. That this is the

24

case is indicated by the fact that unlike object pronoun, subject pronoun induces a Condition C violation even if its antecedent is contained in sentential PPs, as observed by Reinhart (1976): (30) a. b. (31) a. b.

Rosa is kissing him1 passionately in Ben1’s high school picture. *She1 is riding a horse in Rosa1’s high school picture. People worship him1 in Kissinger1’s native country. *He1 is considered a genius in Kissinger1’s home town. (Reinhart 1976:68-69)

The acceptability of (30a) and (31a) indicates that the in-phrases put at the end of the sentences can be right-adjoined to T’, functioning as sentential PPs. The unacceptability of (30b) and (31b) follows if we assume that in these cases too, the in-phrases are right-adjoined to T’, since the subject pronouns, located in Spec-TP, c-command their antecedents contained in the in-phrases, inducing Condition C violations. However, it would become unclear if we allowed sentential PPs to be right-adjoined to TP, since there is a good sense in which such PPs escape from the c-command domain of the subject sitting in Spec-TP; witness the definition of c-command in which α c-commands β iff the first branching node of α dominates β, irrespective of whether it is segment or not (see the discussion on (37) below). This possibility is excluded under the present assumptions about adjunction, since “TP” is actually labeled φ, hence lacking a head, which makes it impossible to linearize a pair <α, TP> according to (28). The present assumptions about labeling and linear ordering in relation to

25

adjunction entails that modifiers are not necessarily incorporated into structures by way of adjunction or pair Merge. Let us consider a case in which a sentential PP is put at the top of a sentence, such as the following: (32) In Ben1’s high school picture, Rosa is kissing him1 passionately. In this case, the in-phrase cannot be pair-merged with TP for the reason just mentioned. Thus, we assume that in such a case, the PP occupies the second Spec of TP, as schematically shown below: (33)

φP in Ben’s high

φ’

school picture Sub

TP

Given that it is generally the case that modifiers cannot label their mother nodes, let us assume the following:10 (34) Given K = {XP, YP}, if XP is a modifier of YP, then YP is the label of K. Thus, the pair {in…, φ’} is labeled φ and, according to LOA in (29), is correctly linearized as in… > φ’. Andrews (1983) observes that when two adverbs are stacked in a seemingly left peripheral position, the preceding one takes scope over the other, as shown below: (35) a. ?John twice intentionally knocked on the door. b.??John intentionally twice knocked on the door. In (35a), twice takes scope over intentionally, so that this sentence expresses two events of intentional knocking, while in (35b), the opposite scope order holds, so that

26

this sentence expresses only one intention of knocking on the door twice. This fact is straightforwardly captured with the standard assumption (cf. Ernst (1991, 1994)) that c-command is a relevant notion for determining the scope order of adverbs. Under the present assumptions, the sentences in (35) have the following structures: (36) a.

[φP John1 [TP twice [T’ intentionally T [vP t1 v+knock [VP on the door tV]]]]]

b.

[φP John1 [TP intentionally [T’ twice T [vP t1 v+knock [VP on the door tV]]]]]

In these structures, twice and intentionally are put in Spec-TP; since they are modifiers, their labels are not projected to their mother nodes according to (34). Further, they precede their sisters according to LOA (29), since they are visible for the CI interface whereas their sisters are not due to their intermediate projection status. Since the preceding adverb asymmetrically c-commands the other in these structures, this c-command relation correctly captures the fact that what precedes is scopally wider than what follows. Andrews further observes that when two adverbs are stacked at the right edge, the following one takes scope over the other, as shown below: (37) a.

John knocked on the door twice intentionally.

b.

John knocked on the door intentionally twice.

This fact is straightforwardly captured under our theory of phrase structure by assuming that these two adverbs are right-adjoined to T’, as shown below: (38) a.

[φP John1 [T’ [T’ [T’ T [vP t1 v+knock [VP on the door tV]]] twice]

27

intentionally]] b.

[φP John1 [T’ [T’ [T’ T [vP t1 v+knock [VP on the door tV]]] intentionally] twice]]

In these structures, twice and intentionally are pair-merged with T’, hence linearized as following the latter, according to (28a). Thus, the hierarchical relation of the two adverbs exhibits the mirror image of that of the two adverbs in (35). Since the following adverbs asymmetrically c-command the preceding ones, it follows that what follows takes scope over what precedes.11 Note that this fact is unexpected under the asymmetrical structure approach, since this approach admits only structures in which what precedes is structurally higher than what follows. Pesetsky (1989) further examines the patterns of the scope order of adverbs and finds out that when two adverbs are put between a verb and its internal argument, the preceding one takes scope over the other, as shown below: (39) As for Mary, Bill relied intentionally twice on her. This is predicted under our hybrid theory of phrase structure, since VP has uniformly right branching structures, as shown below:

28

(40)

vP Bill

v’ v+rely

VP

intentionally

V’ twice

V’ on her

tV

Since intentionally asymmetrically c-commands twice in this structure, this c-command relation correctly captures the scope order of these two adverbs. Pesetsky (1989) further observes that when the PP put at the end in (39) is changed into a “heavier” PP, the sentence becomes ambiguous, as shown below: (41) John relied intentionally twice on the person you told me about. The reading in which intentionally takes scope over twice is derived from the same structure as given in (40). The other reading available to this sentence is derived under the assumption that the heavy PP undergoes focus movement, pair-merged with T’. In that case, the two adverbs can also be pair-merged with T’, thereby producing the structure in which the following adverb is structurally higher than the preceding one, as schematically shown below:

29

(42)

φP John

T’ T’

on the person you told me about

T’ T’ T

twice intentionally

vP

This structure represents the reading in which twice is scopally wider than intentionally. Again, the contrast between (39) and (41) with respect to the availability of this reading will not be given a natural account under the asymmetrical structure approach.

5 5.1

Further Consequences Heavy NP Shift

As briefly discussed toward the end of the last section, the present theory of adjunction can accommodate the standardly observed fact that English allows rightward movement such as heavy NP shift, as illustrated below: (43) Mary kissed t1 yesterday [the man she met in the park]1. Under the present assumptions, we can analyze the shifted DP the man she met in the park as being pair-merged with T’, as shown below:

30

(44) [φP Mary [TP [T’ kissed t yesterday] [the man she met in the park]]] In this structure, the pair-merged DP is linearized as following T’, since the head of this latter phrase is initial. Notice that given the present assumptions, the man she met in the park cannot be pair-merged with φP, since the latter phrase lacks a head. That this must be the case is indicated by the fact that subject is structurally higher than a phrase shifted to the right edge, as Abe and Hoshi (1997) observe: (45) a. b.

*He1 has not read [the letter that Mary sent to John1] until now. *He1 has not read t2 until now [the letter that Mary sent to John1]2. (Abe and Hoshi 1997:107)

(45a) is a typical instance of Condition C violation. The unacceptability of (45b) indicates that he c-commands John even if the DP containing the latter undergoes heavy NP shift. This fact will be unexpected if the shifted DP is allowed to be pair-merged with φP (see the discussion on (30) and (31)). Under the present assumptions, on the other hand, this DP is only allowed to be pair-merged with T’, as in (44), so that John cannot escape from the c-command domain of he even after the heavy NP shift takes place. Contrary to the above claim about heavy NP shift, Pesetsky (1995:265) notes that “binding phenomena with heavy shift constructions behave as if the construction involved lowering,” providing the following examples: (46) a.

*We gave ___ to him1 on Friday [John1’s brand-new toy].

b.

We gave ___ to them1 at the interviews [copies of reports on each other1].

c.

Bill heard ___ from each committee member1 on Friday [a report on his1

31

activities]. d.

Tom threw ___ to none of these people on Tuesday [any set of keys that had “Do Not Copy” stamped on them].

The acceptability of the sentences in (46b-d) indicates that the shifted DPs are c-commanded by the antecedents of an anaphor (46b), a variable pronoun (46c) and a negative polarity item (46d). Likewise, the unacceptability of (46a) suggests that him c-commands John, which is inside the shifted heavy DP, thereby inducing a Condition C violation. All these phenomena might give support to the asymmetrical structure approach, since they appear to show that what precedes is structurally higher than what follows even when “heavy NP shift” is involved. Given our theory of phrase structure, we need to explain these phenomena under the assumption that shifted heavy DPs are pair-merged with T’, which apparently involves upward movement. Takano (1996, 1998) proposes, under the asymmetrical structure thesis, that heavy NP shift is just a reflex of base-generated word order. Under the assumption that the IO-DO order reflects the underlying word order,12 the sentences in (46) do reflect this order (aside from the issue of where adjuncts are base-generated), and hence no movement is involved in deriving these sentences.13 Given this assumption, the acceptability of the sentences in (46) is straightforwardly accounted for; since the arguments of the verbs are all within VP reflecting the underlying word order, what precedes is structurally higher than what follows. Though Takano’s proposal handles the sentences in (46) very well, it

32

immediately faces a problem if we consider other data that show that shifted heavy DPs are structurally higher than what precedes them: (47) a.

John mistakenly returned the two babies that he had taken care of to each other’s mothers.

b.

?John mistakenly returned to each other’s mothers the two babies that he had taken care of.

(48) a. b.

John introduced the students he met yesterday to each other’s teachers. ?John introduced to each other’s teachers the students he met yesterday.

The acceptability of (47b) and (48b) is unexpected under Takano’s proposal, since these sentences reflect the underlying IO-DO order and hence the “shifted” heavy DPs should be structurally lower than the to-phrases including each other. Note that under the assumption that shifted heavy DPs are pair-merged with T’, the acceptability of (47b) and (48b) is straightforwardly accounted for, since in that case, the shifted heavy DPs c-command each other, thereby satisfying Condition A.14 The following pair of examples, provided by Saito (1994), shows the same point: (49) a.

*Mary wanted [PRO to meet [the men who had been accused of the crime]] until each other’s trials.

b.

?Mary wanted [PRO to meet t1] until each other’s trials [the men who had been accused of the crime]1.

(Saito 1994:267)

The significant improvement of acceptability in (49b) shows that the shifted heavy DP is moved to a position higher than the PP until each other’s trials. This is expected under the standard analysis of heavy NP shift, contrary to Takano’s.15

33

Having given support to the standard analysis of heavy NP shift as an instance of movement to T’, we now need to turn to the sentences in (46) to explain why these particular instances of heavy NP shift appear to involve “lowering”. Note first that we are assuming, following Takano (1996, 1998), that the IO-DO order is an underlying order and hence the underlying word orders of the sentences in (46) are the same as their surface orders, as given below: (50) a.

*we gave to him1 on Friday [John1’s brand-new toy]

b.

we gave to them1 at the interviews [copies of reports on each other1]

c.

Bill heard from each committee member1 on Friday [a report on his1 activities]

d.

Tom threw to none of these people on Tuesday [any set of keys that had “Do Not Copy” stamped on them]

Now we need to examine what happens when we pair-merge the heavy DPs to T’. The derived structures are schematically given below:

34

(51)

TP Sub

T’ T’

T

DO1 vP

tSub

v’ v+V

VP IO

V’ Adjunct

V’ t1

tV

Given this structure, the sentences in (46b-d) are ruled grammatical as a result of reconstruction effects with respect to binding, since the antecedents or licensers c-command their licensees in the underlying structures, as shown in (51). Likewise, the unacceptability of (46a) can also be accounted for as a result of reconstruction effects of Condition C. These effects are independently observed by Reinhart (1976) with such examples as the following: (52) a. b.

*With Zelda1’s feather, she1 tickled Dr. Levin. *In Ben1’s box, he1 put his cigars.

(Reinhart 1976:160)

Adapting Saito’s (1985) generalization, the effects are captured in the following way: if an R-expression is c-commanded by a pronoun coreferential to it in the underlying

35

structure and a phrase that dominates the R-expression escapes the c-command domain of the pronoun by movement, then the resulting structure is free from a Condition C violation only if the R-expression is “deeply embedded” in the moved phrase. Here, a typical case of “deep embedding” is one in which an element is embedded within a relative clause. Since the R-expressions Zelda and Ben are not “deeply embedded” within the moved phrases in (52), Condition C effects emerge. These effects disappear when Zelda and Ben are “deeply embedded,” as shown below: (53) a.

With the feather that Zelda1 inherited from her late peacock, she1 tickled Dr. Levin.

b.

In the ivory box that Ben1 brought from China, he1 put his cigars. (Reinhart 1976:160)

Reinhart (1976) further observes that the same pattern of facts obtains with rightward movement, as shown below: (54) a. b.

*After days of search, they finally found him1 in Dr. Levin1’s hotel room. After days of search, they finally found him1 in a sleazy hotel room that Dr. Levin1 had rented under a false name.

(55) a. b.

*Zelda sent him1 back all Dr. Levin1’s flowers. Zelda sent him1 back all the flowers which Dr. Levin1 had bought for her. (Reinhart 1976:160-161)

Saito (1991) argues that the same generalization as given above regarding Condition C reconstruction effects under leftward movement holds true for these cases as well.

36

Thus, in (54a) and (55a), him c-commands Dr. Levin in the underlying structures and then the phrases dominating this R-expression, in Dr. Levin’s hotel room in (54a) and all Dr. Levin’s flowers in (55a), undergo string-vacuous rightward movement, thereby escaping the c-command domain of him. Since Dr. Levin is not “deeply embedded” within these moved phrases, Condition C effects emerge. In (54b) and (55b), in contrast, these effects are suppressed because Dr. Levin is “deeply embedded” within the phrases that undergo string-vacuous rightward movement.

16

Given this

assumption, (46a) is correctly ruled ungrammatical as a Condition C violation, since John is not “deeply embedded” within the phrase that is right-adjoined to T’.

5.2

An Extension to DP Structure

In this subsection, I argue that the present theory of labeling and linear ordering can be extended to accommodate possible internal structures of DP. In so doing, I show that postnominal modifiers are right-adjoined to D’. Larson (2000a, b) makes the interesting observation that a sentence such as the following is ambiguous, depending upon how to interpret possible, as paraphrased in (56a) and (56b). (56) Mary interviewed every possible candidate (on her recent press tour). a.

‘Mary interviewed everyone that was a possible candidate.’

b.

‘Mary interviewed every (actual) candidate that it was possible for her to interview.’

Larson names the (a) reading direct modification reading (DMR) and the (b) reading

37

implicit relative reading (IRR). Characterizing D-modification as intersective and N-modification as non-intersective, Larson (2000a) proposes that the DMR is derived from N-modification whereas the IRR is derived from D-modification. Thus, the DMR of (56) is derived from the following structure of every possible candidate:17 (57)

DP every

D’ D

NP possible

candidate

Extending Chierchia’s (1995) analysis of predicate nominals, Larson (2000a) claims that NPs are interpreted in a Davidsonian way as events bound by generic quantifiers and that adjectives inside NPs function as predicates of events denoted by Ns. Possible candidate in (57) is, then, interpreted roughly as follows: (58) λx[Γe[e is running candidate & Agent(e, x) & possible(e)]] where Γ represents a generic quantifier. On the other hand, when (56) is interpreted under the IRR, possible is taken as intersective, so that the sentence means that Mary interviewed everyone that is a candidate and that is possible for her to interview. Note that this reading is paraphrased as (59a) or its elliptical version (59b). (59) a. b.

Mary interviewed every candidate possible for her to interview. Mary interviewed every candidate possible.

Given this, Larson (2000b) proposes that the IRR of (56) is derived from the

38

underlying structure equivalent to (59a) by moving possible before candidate and deleting the infinitival clause under identity with the matrix clause.18 Larson (2000a) further claims that the position to which possible is moved must be outside NP, since when candidate is modified by two instances of possible, as shown below, there is strong intuition that indicates that the first possible is an intersective modifier whereas the second is a non-intersective modifier. (60) every possible possible candidate Based upon this observation, Larson reaches the conclusion that intersective modifiers must be D-modifiers. Note that (59b) does not have a DMR; hence by parity of reasoning, postnominal modifiers should also be D-modifiers. To summarize, we have found the following distribution of possible and its available readings: (61) ... every (possible) [NP possible candidate] (possible) IRR

DMR

IRR

Larson (2000a) further observes that the positional contrast with respect to the two readings of possible seems to be just one instance of more general phenomena. He provides the following examples: (62) a.

the responsible individuals vs. the individuals responsible

b.

the navigable rivers vs. the rivers navigable

c.

the visible stars vs. the stars visible

As noted by Bolinger (1967), the postnominal adjectives attribute temporal properties whereas the prenominal adjectives can attribute not only temporal properties but also enduring or stable properties. Thus, to take (62c) for illustration, the visible stars is

39

ambiguous between the reading in which the stars are visible in their inherent properties relating to the distance from the earth and the one in which the stars are visible in a given place at a given time. The stars visible, in contrast, only has the latter temporal reading. Contra Bolinger, Larson claims that this is not simply a matter of prenominal versus postnominal modifiers, since when stars is modified by two instances of visible, as shown below, intuition strongly indicates that the first visible denotes its temporal property and the second denotes its enduring property. (63) the visible visible stars To summarize, we have the following distribution of visible and accompanying meanings: (64) the (visible) [visible

stars] (visible)

Temporary Enduring

Temporary

Taking into consideration the similarity between the distribution of visible and that of possible, summarized in (61), it is natural to regard the temporal use of visible as a D-modifier having intersective interpretation (the ones that are stars and that are visible) and to regard the enduring use of visible as an N-modifier having non-intersective interpretation. There is a natural way to capture these facts under our theory of labeling and linear ordering. First, let us follow Fukui and Takano (1998) in assuming that there is a functional projection between NP and DP, called nP, parallel to vP in clausal structure. Further, we assume, following Larson (2000a), that N-modifiers must move to Spec-nP to check their Case and/or φ-features with the n head. Then, (61), for

40

instance, can be given the following structure: (65)

DP every

D’ (possible)

D’ D’

D

(possible) nP

possible1

n’

n+candidate

NP t1

tN

Under our adjunction theory, D, as a functional category, takes the head-initial value and hence possible, when pair-merged with D’, follows the latter according to (28a). This explains the fact that possible, when interpreted as a D-modifier, can appear not only prenominally but also postnominally. Although the linear order of the N head candidate and the N-modifier possible is undetermined in their underlying positions, raising N to n as well as raising possible to the n projection for feature checking makes it possible to linearize these items as possible > candidate. Given that raising of N-modifiers to the above n projections for feature checking is mandatory, it follows that these modifiers always appear prenominally. Thus, (65) illustrates another case where adjunction (or pair Merge) is possible in a functional category and the way adjunction takes place respects the value of the head parameter.

41

6

Concluding Remarks

In this article, I have argued that linear ordering should be dealt with on a par with labeling in the sense of Chomsky (2013) and hence that LOA applies when the syntactic objects in a given phase domain are transferred. Based upon this conception, I have argued, following Abe (2001), that LOA is formulated in terms of visibility of labels for the CI interface in such a way that a visible SO precedes an invisible SO, and that the head parameter is encoded to resolve the underspecification of the head-complement order as a result of applying LOA. Further, I have argued that the head parameter is encoded only in functional categories, taking a hybrid position in which lexical categories have only asymmetrical structures a la Kayne (1994) and Fukui and Takano (1998) while functional categories allow symmetrical structures with the head parameter encoded. This approach thus gives support to Fukui’s (1995) FPH, according to which lexical projections are uniform among languages and parameterization is attributed to functional categories. The approach is mainly supported by the fact that the possibility of right adjunction correlates with the head-initial value of the head parameter. Thus, in a head-initial language such as English, what follows behaves as if it is structurally higher than what precedes in the domain of functional categories, and this fact is straightforwardly captured on the assumption that right adjunction is possible in functional categories.

References

42

Abe, Jun. 1993. Binding conditions and scrambling without A/A’ distinction. Doctoral dissertation, University of Connecticut, Storrs, CT. Abe, Jun. 2001. Relativized X’-theory with symmetrical and asymmetrical structure. In Minimalization of each module in generative grammar, Report for Grants-in-Aid for Scientific Research (B)(2), 1-38. Graduate School of Humanities and Informatics, Nagoya University. Abe, Jun and Hiroto Hoshi. 1997. Gapping and P-stranding. Journal of East Asian Linguistics 6:101-136. Andrews, Avery. 1983. A note on the constituent structure of modifiers. Linguistic Inquiry 14:695-697. Aoun, Joseph and Yen-hui Audrey Li. 1989. Scope and constituency. Linguistic Inquiry 20:141-172. Barss, Andy and Howard Lasnik. 1986. A note on anaphora and double objects. Linguistic Inquiry 17:347-354. Bolinger, Dwight. 1967. Adjectives in English: Attribution and predication. Lingua 18:1-34. Chierchia, Gennaro. 1995. Individual-level predicates as inherent generics. In The Generic book, ed. by Gregory N. Carlson and Francis Jeffry Pelletier, 176- 223. Chicago: University of Chicago Press. Chomsky, Noam. 1981. Lectures on government and binding. Dordrecht: Foris. Chomsky, Noam. 1995. The Minimalist Program. Cambridge, MA: MIT Press. Chomsky, Noam. 2008. On phases. In Foundational issues in linguistic theory:

43

Essays in honor of Jean-Roger Vergnaud, ed. by Robert Freiden, Carlos P. Otero, and Maria Luisa Zubizarreta, 133-166. Cambridge, MA: MIT Press. Chomsky, Noam. 2013. Problems of projection. Lingua 130:33-49. Ernst, Thomas. 1991. On the scope principle. Linguistic Inquiry 22:750-756. Ernst, Thomas. 1994. M-command and precedence. Linguistic Inquiry 25:327-335. Fox, Danny and David Pesetsky. 2005. Cyclic linearization of syntactic structure. Theoretical Linguistics 31:1-45. Fukui, Naoki. 1993. Parameters and optionality. Linguistic Inquiry 24:399-420. Fukui, Naoki. 1995. The principles-and-parameters approach: A comparative syntax of English and Japanese. In Approaches to language typology, ed. by Masayoshi Shibatani and Theodora Bynon, 327-371. Oxford: Oxford University Press. Fukui, Naoki and Yuji Takano. 1998. Symmetry in syntax: Merge and demerge. Journal of East Asian Linguistics 7:27-86. Goto, Nobu. 2015. Restricting n to two: When Merge requires Search. Ms., Toyo University. Hoji, Hajime. 1985. Logical Form constraints and configurational structures in Japanese. Doctoral dissertation, University of Washington, Seattle, WA. Kayne, Richard. 1994. The antisymmetry of syntax. Cambridge, MA: MIT Press. Kitagawa, Yoshihisa. 1994. Shells, yolks, and scrambled e.g.s. In Proceedings of the North East Linguistic Society 24, ed. Mercè Gonzàlez, 221-239. Amherst, MA: GLSA, University of Massachusetts. Larson, Richard K. 1988. On the double object construction. Linguistic Inquiry

44

19:335-391. Larson, Richard K. 2000a. Adjective position and interpretation. Paper presented at Nanzan University. Larson, Richard K. 2000b. ACD in AP? Paper presented at the 19th West Coast Conference of Formal Linguistics, Los Angeles, CA. Narita, Hiroki. 2014. Endocentric structuring of projection-free syntax. Amsterdam: John Benjamins. Pesetsky, David. 1989. Language-particular process and the Earliness Principle. Ms., MIT. Pesetsky, David. 1995. Zero syntax: Experiencers and cascades. Cambridge, MA: MIT Press. Reinhart, Tanya. 1976. The syntactic domain of anaphora. Doctoral dissertation, MIT, Cambridge, MA. Saito, Mamoru. 1985. Some asymmetries in Japanese and their theoretical implications. Doctoral dissertation, MIT, Cambridge, MA. Saito, Mamoru. 1991. Extraposition and parasitic gaps. In Interdisciplinary approaches to language: Essays in honor of S.-Y. Kuroda, ed. by Carol Georgopoulos and Roberta Ishihara, 467-486. Dordrecht: Kluwer Academic Publishers. Saito, Mamoru. 1994. Improper adjunction. In Formal Approaches to Japanese Linguistics I, MIT Working Papers in Linguistics 24, ed. by Masatoshi Koizumi and Hiroyuki Ura, 263-293. Cambridge, MA: MITWPL.

45

Saito, Mamoru and Naoki Fukui. 1998. Order in phrase structure and movement. Linguistic Inquiry 29:439-474. Takano, Yuji. 1996. Movement and parametric variation in syntax. Doctoral dissertation, University of California, Irvine, CA. Takano, Yuji. 1998. Object shift and scrambling. Natural Language and Linguistic Theory 16:817-889.

Notes 1

In this quote, underlining is added by the author for ease of reference in the

discussion that follows. 2

This view accords with Fox and Pesetsky’s (2005) “cyclic linearization”

mechanism, according to which “the relative ordering of words is fixed at the end of each phase, or ‘Spell-out domain’.” (p. 1) 3

See Narita (2014) for postulating label-free structures under a minimalist

conception, which may be made compatible with the current view taken in the text. But see Goto (2015) for the claim that labeling plays an important role in narrow syntax, so that unlabeled SOs block searching an SO for internal Merge. 4

Given the reasonable assumption that LA does not care for the θ-relational status of

merged SOs, it will be more appropriate to claim that when XP is merged with YP, the resulting {XP, YP} is allowed to freely pick up either head X or Y as the label of this set, and that the wrong labeling is ruled out by something like the θ-criterion in the sense of Chomsky (1981), which applies at the CI interface. Thus, if the {IO, V’} pair in (8) were labeled D, the label of IO, then the argument status of IO would be wrongly extended so that the predicate-argument relation holding between V and IO would be lost. Since it is not a main purpose of this article to provide a precise formulation of LA, I will use the formulation given in (9) in what follows, just for ease of presentation.

46

5

Kitagawa (1994) and Pesetsky (1995) claim, contrary to Larson (1988), that the

IO-DO and DO-IO frames are not in the relation of one deriving from the other. Thus, strictly speaking, the DO-IO frame is derived from the underlying to-DP-DP order. 6

Under the present analysis, the unacceptability of (13b) is not a result of Condition

A violation, but rather follows from a violation of Condition C, since herself c-commands Mary. 7

Strictly, Takano (1996, 1998) assumes that the basic word order of a clause is

S-C-H universally, as in (10). Thus, Takano’s account presupposes the unusual assumption that a head can move to a specifier position. See Fukui and Takano (1998) on this point. 8

Fukui and Takano (1998) do not discuss how adjunction structure is linearized. If

they do not allow adjunction structure in their theory of phrase structure, then PP1 and PP2 in (22) need to be identified as the Specs of vP and VP, respectively, so that they should be linearized as preceding their sisters, according to their mechanism of linearization. 9

The present system of adjunction also allows the sentential PPs and the extraposed

clauses to be right-adjoined to v’, but we abstract away from this option in the text simply for ease of exposition. 10

See fn. 4 for relevant discussion. Here again, we could claim that there is a choice

regarding which of the maximal projections is taken as the label of K, and that the wrong choice is excluded at the CI interface by an interpretive rule that reads off a modifier-modifiee relation from a given structure. 11

In this case, “the first branching node” used to define the notion of c-command

must be taken in a strict sense, so that a segment can also be counted as a first branching node relevant for the definition. 12

DO corresponds to Theme and IO to non-Theme in Takano’s terminology.

13

Takano must assume that “shifted” heavy DPs are somehow exempt from the

requirement that DPs carrying accusative Case be adjacent to verbs, as shown below: (i) a. John gave a book to Mary. b. *John gave to Mary a book. Takano (1998) suggests the possibility that “shifted” heavy DPs in fact involve

47

movement to positions adjacent to the verbs and that the tail of the chain so created rather than the head is pronounced due to its focus effects. He also claims that in such a chain, only the tail position is visible for LF interpretation, hence accounting for the acceptability of the sentences in (46). 14

The standard formulation of Condition A requires that the antecedent of an

anaphor be in an A position. See Saito (1994) for the claim that heavy NP shift is an instance of A-movement. See also Abe (1993) for the argument that the A/A’ distinction is unnecessary for the formulation of Condition A. 15

One may wonder how to create a structure of such a construction as (49b) that

involves heavy NP shift across an infinitival clause under Takano’s hypothesis that heavy NP shift is just a reflex of base-generated word order. He would rely on scrambling to derive the word order given in (49b); for example, scrambling the heavy object first and then scrambling the remnant clause to the left of that object. If this option is available for Takano’s analysis, it might account for the acceptability of (49b), since there is a stage in this derivation at which the heavy DP c-commands each other. Takano (1996) exploits data of NPI licensing to support his analysis of heavy NP shift: (i) a. I showed none of the pictures of John’s mother to anyone. b. *I showed to anyone none of the pictures of John’s mother. (ii) a.

I believe none of the students in the cafeteria to have kissed anyone.

b. *I believe to have kissed anyone none of the students in the cafeteria. According to his analysis, (ib) reflects the basic word order, hence anyone not c-commanded by none. He analyzes (iib) as derived by moving the embedded T’ to the left of its subject, hence none not c-commanding anyone. Under the standard analysis of heavy NP shift, in contrast, (ib) and (iib) are expected to be grammatical, since heavy NP shift is in fact an upward movement to T’. I suspect that NPI licensing is different from other binding phenomena such as anaphor binding and variable pronoun binding in that it involves more than just the c-command requirement. Note that even Takano’s analysis may not predict the ungrammaticality of sentence (iib), if NPI licensing is treated in the same way as other binding phenomena, since none c-commands anyone in the underlying structure and hence

48

reconstruction effects will be expected to emerge. 16

Note that if this generalization is correct, then this in itself constitutes an empirical

argument against such asymmetrical structure approaches as Kayne’s (1994), Takano’s (1996, 1998) and Fukui and Takano’s (1998). 17

We are now abstracting away from the question how the right linear order is

derived from a hierarchical structure such as (57). 18

Thus, this is a case of antecedent-contained deletion. We abstract away from this

part of the derivation in the following discussion just to concentrate on the relevant point, so that I assume that the IRR of (56) is derived from (59b) by moving possible before candidate. See Larson (2000b) for exactly how (56) is derived from (59a).

49