Abstract

I make two proposals in this article: (a) an economy condition on the operation Copy, which states that Copy should apply to as small an element as possible, and (b) the “two types of head movement” hypothesis, which states that Universal Grammar allows head movement via substitution as well as head movement via adjunction. I argue that with these proposals, we can not only explain two generalizations about what I call headless XPs, but also attribute crosslinguistic variation in the applicability of these generalizations to parameters that are responsible for the availability of multiple specifiers.

1 Introduction

This article is concerned with what I call headless XPs. By this term, I mean a phrase whose head has moved out of it. For example, the XP in (1) is a headless XP.

(1) [YP[Y X-Y] [XPtX ZP]]

Notice that it does not matter whether the lower copy of the head, which is expressed as tX in (1), is actually pronounced or not. Even if the lower copy is pronounced, the XP in (1) is headless because the head X has moved out of it.

In particular, I am concerned with the following two generalizations about headless XPs: Takano’s Generalization (TG; Takano 2000) and Lasnik’s Generalization (LG; Lasnik 1999). The former states that headless XP-movement is prohibited, and the latter states that headless XP-ellipsis is prohibited.1 TG and LG are illustrated by the unacceptability of English examples like (2a) and (2b), respectively.

(2)

  • a.

    *It’s [VP a book t1 to Mary]2 that John gave1t2.

  • b.

    *Meg gave Ken advice, and Ken will give Sam [VPtgivetSam advice].

The cleft sentence in (2a) involves headless VP-movement; and in (2b), a sort of pseudogapping sentence involves headless VP-ellipsis (sections 2.12.2).

The aim of this article is to explain TG and LG and crosslinguistic variation in the applicability of these generalizations. In order to account for TG, first I propose an economy condition on Copy requiring Copy to apply to as small an element as possible (section 3.1). In section 3.2, I argue that the same account can be applied to LG if we adopt a modified version of Johnson’s (2001) movement approach to VP-ellipsis.

Furthermore, I point out that there are languages where TG and/or LG do not hold (section 4), as illustrated by a Japanese example in (3) and a Hebrew example in (4).2

(3)

  • [CP John-ga t1 ageta2 no]-wa [vP Mary-ni hon-o ni-satu t2]1 da.

  • John-NOM gave COMP-TOP Mary-DAT book-ACC two-CL be

  • (Lit.) ‘It is Mary two books that John gave.’

  • ‘John gave Mary two books.’

(4)

  • a.

    graphic

    ‘(Did) Miryam drive Dvora to the grocery store?’

  • b.

    graphic

    ‘Yes, she drove Dvora to the grocery store.’

    (Goldberg 2005:53)

According to Koizumi (1995), Japanese cleft sentences like (3) are instances of headless VPmovement. Goldberg (2005) argues that Hebrew sentences like (4b), in which elements within the VP are elided, are derived by headless VP-ellipsis. I argue that this crosslinguistic variation in the applicability of TG and LG can be explained and attributed to parameters that are responsible for the availability of multiple specifiers if the “two types of head movement” hypothesis holds (section 5). The hypothesis states that Universal Grammar (UG) allows head movement via substitution as well as head movement via adjunction. Although this hypothesis runs contrary to the widely held assumption that head movement is always performed via adjunction, this is the null hypothesis, as I will argue in section 5.1.

2 Constraints on Headless XPs

2.1 Headless XP-Movement: Takano’s Generalization

Takano (2000) argues that the following generalization holds:

(5)

  • Takano’s Generalization

  • XP-movement is prohibited if the head of XP has moved out of XP.

He attributes examples like (6a–c) to the generalization in (5).

(6)

  • a.

    *[Ihr ein Buch t1]2 gab1 Hans t2.

    her a book gave Hans

    ‘Hans gave her a book.’

  • b.

    *[Het boek aan Marie t1]2 gaf1 Jan (waarschijnlijk) t2.

    the book to Marie gave Jan probably

    ‘Jan (probably) gave the book to Marie.’

  • c.

    *It’s [a book t1 to Mary]2 that John gave1t2.

    (Takano 2000:145)

The German example in (6a) and the Dutch example in (6b) are derived by headless VP-movement, in which the finite verb moves out of the VP. The English example in (6c) also involves movement of a headless VP whose head has undergone head movement out of the VP, given the Larsonian shell analysis of ditransitive clauses (Larson 1988, Chomsky 1995c).3 All of these examples involve headless XP-movement. Therefore, the unacceptability of these examples is consistent with TG as stated in (5).4

Moreover, Wurmbrand (2004) argues that the following paradigm in German can be attributed to TG (the Headless Fronting Constraint in her terms):

(7)

  • a.

    weil Hans seinen Bruder angerufen hat

    since Hans his brother up.called has

    ‘since Hans phoned his brother’

  • b.

    graphic

    ‘Hans phoned his brother yesterday.’

  • c.

    [vP Seinen Bruder angerufen] hat nur der Hans tvP/VP.

    his.ACC brother up.called has only the Hans

    ‘Only Hans phoned his brother.’

  • d.

    *[vP Seinen Bruder an tV] rief der Hans gestern tvP/VP.

    his.ACC brother up called the Hans yesterday

    ‘It was yesterday that Hans phoned his brother.’

    (Wurmbrand 2004:7)

All of these examples involve particle-verb constructions. The acceptability of (7b) indicates that a particle can be stranded when a verb moves to C via verb-second movement. The acceptability of (7c) illustrates that fronting of a V/vP that includes the object, the verb, and the particle is allowed. Since V-movement stranding a particle is generally allowed and fronting of V/vP that includes a particle is allowed, as shown in (7b) and (7c), respectively, the unacceptability of (7d) must be attributed to the fact that the headless V/vP is fronted. Thus, the unacceptability of (7d) suggests that TG holds in German.5

2.2 Headless XP-Ellipsis: Lasnik’s Generalization

Lasnik (1999) considers the following generalization on headless XPs, which is similar to TG:6

(8)

  • Lasnik’s Generalization

  • XP-ellipsis is prohibited if the head of XP has moved out of XP.

LG amounts to saying that headless XP-ellipsis is prohibited. Some facts concerning English pseudogapping are attributed to the generalization. Lasnik argues that pseudogapping sentences like (9a) involve VP-ellipsis. According to this analysis, the sentence in (9a) has the underlying representation in (9b).7

(9)

  • a.

    Mary hasn’t dated Bill, but she has Harry.

  • b.

    Mary hasn’t dated Bill, but she has Harry [VP dated t].

    (Lasnik 1999:147)

In (9a), the object Harry (A-)moves out of the VP while the V dated stays in situ. Thus, VPellipsis yields the pseudogapping sentence. Lasnik argues that the reason why V can stay in situ in the pseudogapping sentence is that V’s strong feature that drives V-movement is eliminated by VP-ellipsis, avoiding a PF crash. This explains why V can stay in situ in the pseudogapping sentence.8 However, it does not prohibit Vs from moving. Thus, this analysis cannot account for the sentence in (10a), where the direct object of a double object construction is elided. This sentence can be derived by VP-ellipsis if the V give and Sam move out of the VP and advice does not, as shown in (10b). Notice that if the V stays in situ, a legitimate pseudogapping sentence is obtained, as shown in (10c).

(10)

  • a.

    *Meg gave Ken advice, and Ken will give Sam advice.

  • b.

    *Meg gave Ken advice, and Ken will give Sam [VP tgive tSam advice].

  • c.

    ?Meg gave Ken advice, and Ken will Sam [VP give tSam advice].

If LG holds in English, the unacceptability of (10a) is unproblematic for Lasnik’s analysis of pseudogapping, since the headless VP is elided in (10a).

Furthermore, the fact that null objects are not allowed in English can be partially attributed to LG. As shown in (11), English does not allow null objects.

(11) *John solved the problem, and Mary solved, too.

One of the possible representations for (11) is (12).

(12) *John solved the problem, and Mary solved pro, too.

The second conjunct in (12) contains pro. This representation is ruled out by the standard assumption that English does not have pro. However, this assumption is not sufficient to rule out all apparent null objects in English since the null object sentence in (11) can have another possible representation without pro. Given that V moves to v and an object can stay in situ in English, headless VP-ellipsis yields a null object sentence, as shown in (13).

(13) *John solved the problem, and Mary [vP solved [VPtsolve the problem]], too.

In the second conjunct in (13), the verb moves to v and the object stays in the VP, resulting in a headless VP. If the headless VP is elided, the null object sentence obtains. Thus, in order to capture the fact that null objects are not allowed in English, we need to rule out VP-ellipsis in English. Notice that vP-ellipsis is possible in English, as shown in (14).

(14) John solved the problem, and Mary did [vP solve the problem], too.

It is unclear why vP can be elided but VP cannot. This difference between vP and VP can be attributed to LG since VP is headless while vP is “headed.” Under LG, “headed” vPs can be elided but headless VPs cannot.

As pointed out by Lasnik (1999), LG is also consistent with the fact that sluicing is prohibited when I moves out of IP, as shown in (15b).

(15)

  • a.

    Mary saw someone.

  • b.

    Who (*did)?

  • c.

    Who did [IP Mary t1 see]?

    (Lasnik 1999:158)

Given that sluicing involves IP-ellipsis, this fact falls under LG. This is because, as illustrated in (15c), the headless IP must be elided in order to derive the sluicing example with the remnant auxiliary.

3 Explanation

In this section, I propose an economy condition on Copy, according to which Copy must apply to as small an element as possible.9 In section 3.1, I will show that this condition can explain TG. In section 3.2, I will argue that the same condition can also explain LG if we adopt a revised version of Johnson’s (2001) movement approach to VP-ellipsis.

3.1 Explanation for Takano’s Generalization: Economy Condition on Copy

TG follows from an economy condition on Copy, which states, in effect, that Copy should apply to as small an element as possible. Before formulating a precise definition of this condition, I will outline informally how this idea can account for TG.

3.1.1 Informal Explanation

Consider the structure in (16), where a probe α and a potential goal X have a feature [F], which is the driving force for Agree, and XP also has [F] since XP is a projection of X.

(16)

graphic

In (16), X undergoes head movement to Y, yielding a headless XP. Crucially, X adjoins to Y, where Y projects a segment, not a category. Thus, given the distinction between categories and segments and the definition of c-command based on that distinction (X c-commands Y if and only if every category that dominates X also dominates Y and X does not dominate Y), X c-commands XP and conversely. That is, a mutual c-command relation holds between X and XP.

As a result, both X and XP can be candidates for Agree in terms of minimality. Given the copy theory of movement and the widely held view that Agree is a prerequisite for Move, this means that both X and XP are candidates for Move. However, X is smaller than XP in some sense. Thus, the economy condition on Copy prevents the bigger element, XP, from moving. In this way, headless XP-movement is prohibited by the economy condition on Copy: headless XP-movement is impossible since a smaller element X is a candidate for Move.10

A crucial assumption in this explanation is that the computational system does not discriminate head movement from phrasal movement in the relevant respect since they compete against each other with regard to which movement is performed. In the case of headless XPs, the possibility of head movement of X prohibits XP from moving. I take this assumption to be desirable, in the context of a research program that aims to assimilate head movement as far as possible to phrasal movement. An interesting issue under this research program is the question of how differences between head movement and phrasal movement can be derived from independent factors. One important difference is concerned with locality: head movement seems to be restricted to the next higher head (the Head Movement Constraint), while phrasal movement can target a position farther away. In the next section, I will show that this difference is derived from an economy condition on Copy and structural intervention.

3.1.2 Formal Explanation

Now, in order to implement this general idea within the formal theory of grammar, we need to formulate several definitions. The following is one of the plausible implementations.

First, I define Agree domain (AD), based on Agree. AD is a set whose members can be candidates for other operations such as Copy (and maybe Value).

(17) Agree domain

  • a.

    At a given stage of a derivation where α is a probe, AD(α) = {x: Agree(α, x)}.

  • b.

    Agree(α, β) iff

    • i.

      α c-commands β and

    • ii.

      α matches β and

    • iii.

      there is no γ that structurally intervenes between α and β.

The statement in (17a) amounts to saying that the AD of α is a set of potential goals of Agree. As shown in (17biii), a locality condition is incorporated into the definition of Agree. Structural intervention is defined as follows:

(18) Structural intervention

γ structurally intervenes between α and β iff

  • a.

    every category that dominates γ also dominates β and

  • b.

    some category that dominates β does not dominate γ and

  • c.

    α c-commands γ and

  • d.

    γ matches α and β and

  • e.

    γ is not a complement of α.

The statement in (18a) (adopted from Tanaka 2004:925) amounts to saying that either a c-commanding element or a dominating element can be an intervener. Thus, this definition has the combined effect of Relativized Minimality or the Minimal Link Condition (MLC) (Rizzi 1990, Chomsky 1995c) and the A-over-A Principle (Chomsky 1964). Take the following schematic structures, for example:

(19)

  • a.

    [XP α [YP γ [ZP β . . . ]]]

  • b.

    [XP α [YP . . . [γ β . . . ]]]

The structure in (19a) illustrates MLC effects. In (19a), γ could structurally intervene between α and β since every category that dominates γ also dominates β (XP and YP). (19b) illustrates A-over-A effects. γ could structurally intervene between α and β in this configuration too: every category that dominates γ also dominates β.

I need to incorporate the A-over-A Principle into this definition because if the MLC were the only locality condition, an economy condition on Copy would prohibit all phrasal movement.

We will see why this is so after I introduce the formal definition of an economy condition on Copy. (See Fukui 1997, 1999 for detailed discussion of the A-over-A Principle from the minimalist perspective.)

The statement in (18b) is required to prevent an element γ from intervening between a probe and γ itself. For example, in (19a), it is trivially true that every category that dominates γ (i.e., XP and YP) dominates γ. Thus, without (18b), γ structurally intervenes between α and γ. This means that nothing can be a goal of Agree because everything intervenes between a probe and itself. The statement in (18b) overcomes this self-intervening problem since there is no category that simultaneously dominates and does not dominate something, a clear contradiction.

The statement in (18e) realizes some kind of notion of antilocality (see Abels 2003, Grohmann 2003), saying that an element that is too close to the probe is not an intervener. We will see why this condition is required after the economy condition on Copy has been formally defined.11

Given these assumptions, consider schematic structures like (20a) and (20b).

(20)

  • a.

    [α(F) [γ(F) [β(F ) . . . ]]]

  • b.

    [α(F) [γ(F) β(F)]]

In (20a), AD(α) is {γ}. β is not in AD(α) since γ structurally intervenes between α and β. Thus, operations like Copy can apply to γ but not β. On the other hand, in (20b), AD(α) is {β, γ} since β does not intervene between α and γ nor does γ intervene between α and β: although every category that dominates γ/β also dominates β/γ, there is no category that dominates γ/β and does not dominate β/γ. As a result, in (20b), unlike in (20a), Copy can apply to either β or γ. The goal that undergoes Copy is determined by other factors including economy conditions. I propose one such economy condition, the Economy Condition on Copy (ECC), which is formulated as follows:12

(21) Economy Condition on Copy

At a given stage of a derivation where α is a probe:

Copy(β) only if

  • a.

    β ∈ AD(α) and

  • b.

    β ∉ AD(α) such that γ ≺ β.

This condition says that the copying of an element is prohibited if there is another smaller element in the relevant AD. For example, in (20b), if γ is smaller than β, copying of β is impossible. In order to make this condition precise, we need to define smallness. The following is one possible definition, based on Chomsky’s (1995c) notion term:

(22) Smallness

  • a.

    α ≺ β iff α ∈ subterm of β (ST(β)).

  • b.

    ST(β) = {x: x is a term of β and x ≠ β}

(23)

  • a.

    K is a term of K.

  • b.

    If L is a term of K, then the members of the members of L are terms of K.

    (Chomsky 1995c:247)

According to these definitions, α is smaller than β if and only if α is a term of β unless α = β.

With these definitions in mind, let us return to TG. Consider again the schematic structure in (16), repeated here.

(24)

graphic

In (24), AD(α) is {X, XP}, since X does not structurally intervene between α and XP nor does XP structurally intervene between α and X: although every category that dominates X/XP also dominates XP/X, there is no category that dominates X/XP and does not dominate XP/X. Thus, Copy can apply to either X or XP. However, X is smaller than XP according to the definition of smallness in (22) because X is a term of XP and X ≠ XP. Therefore, Copy(XP) is prohibited by the ECC. This is why headless phrases cannot be moved.13

Notice in passing that although Copy(X) and movement of X in (24) (i.e., excorporation of X) are allowed in principle, the resulting representation is filtered out by some morphological constraint such as Baker’s (1988) that prohibits word-internal traces. That is, the movement of X itself is a legitimate operation in syntax but the resulting head Y violates Baker’s morphological condition since Y contains the trace of X.

Before closing this section, let us see why the A-over-A Principle must be incorporated into the definition of structural intervention (see (18a)) and why it must be assumed that the complement of a probe is not an intervener (the condition in (18e)). First, consider the following configuration, which illustrates movement of subjects to Spec,TP:

(25) [TP T(uφ) [vP[DP(φ) Spec [D′(φ) D(φ) Compl]] [v′ v VP]]]

In (25), T is a probe because it has an uninterpretable φ-feature, [uφ]. Suppose that we have only the MLC as a locality condition. Then, DP, D′, and D are contained in AD(T) since D has the matching feature [φ] and D′ and DP are projections of D. However, given that intermediate projections are syntactically invisible, following Chomsky (1995a) (see also Travis 1984, Speas 1990), only DP and D are contained in AD(T).14 Notice that D is in AD(T). This is because T c-commands D and there is no intervener between them with respect to the MLC (there is no c-command relation between DP and D). Thus, Copy(DP) is prohibited in terms of the ECC since D is smaller than DP (D is a subterm of DP). Therefore, the ECC in (21) predicts that there is no phrasal movement, which is obviously an undesirable consequence.

On the other hand, with the definition of structural intervention in (18), in which both the A-over-A Principle and the MLC are incorporated, D is not contained in AD(T). The DP structurally intervenes between T and D according to this definition: (a) every category that dominates DP also dominates D, (b) some category that dominates D does not dominate DP, (c) T c-commands DP, (d) DP matches T and D with respect to the φ-feature, and (e) DP is not a complement of T. Therefore, Copy can apply to DP even if DP is bigger than D.

Next, let us consider the noncomplement restriction on the interveners. We need this restriction since head movement would never occur otherwise. To see this, let us consider the following configuration, which illustrates V-to-v movement:

(26) [vP v(F) [VP(F) V(F) DP]]

In (26), V cannot move to v without (18e): VP intervenes between v and V. On the other hand, with (18e), VP is not considered an intervener since VP is a complement of the probe v. Then AD(v) is {VP, V}. V is smaller than VP. Therefore, V rather than VP moves to v.

Notice that the Head Movement Constraint (HMC, Travis 1984) and the Antilocality Principle can be deduced from the proposed system thanks to this noncomplement restriction. The HMC states that head movement of X to Y cannot skip an intervening head Z. For example, in the following schematic structure, X cannot move to Y:

(27) [YP Y(F) [ZP Z [XP(F ) X(F) . . . ]]]

Under the proposed system, this fact can be explained as an A-over-A effect. AD(Y) is {XP}. Crucially, X is not contained in AD(Y) because XP, which is not a complement of Y, structurally intervenes between Y and X. The definition of structural intervention in (18) predicts that head movement occurs only if the projection of the goal is a complement of the probe, thus deriving the HMC.

The most standard version of the Antilocality Principle states that a complement of a head X cannot move to Spec,XP. Thus, in (26), VP cannot move to Spec,vP. This can be explained by the noncomplement restriction in the definition of structural intervention and the ECC. That is, movement of VP (actually, Copy of VP) is prohibited by the ECC because AD(v) contains a smaller element V (VP does not intervene between v and V because VP is a complement of v). Thus, Antilocality effects can be reduced to the ECC.

To sum up, in this section, I proposed a formal system that implements the idea of an economy condition on Copy, and I illustrated that TG can be reduced to the ECC under this system. Furthermore, I showed that HMC effects and Antilocality effects can also be deduced under the proposed system.

3.2 Explanation for Lasnik’s Generalization: Movement Approach to VP-Ellipsis

TG and LG look similar. Thus, a natural question is whether they are rooted in the same principle (i.e., whether they can be uniformly explained). In this section, I answer this question in the affirmative, at least in the domain of VP-ellipsis. In particular, I argue, in the spirit of Johnson’s (2001) movement approach to VP-ellipsis and an idea proposed by Aelbrecht and Haegeman (2012), that VP-ellipsis involves VP-movement to either Spec,TopP in the CP periphery or Spec, TopP in the vP periphery.15 If the idea that ellipsis somehow reduces to movement is correct, LG can be explained by the ECC in the same manner as TG.

3.2.1 VP-Ellipsis Is Derived through VP-Topicalization: Johnson 2001

Johnson (2001) argues, in the spirit of Lobeck 1995, that sentences involving VP-ellipsis like the second conjunct in (28a) are derived by topicalizing VP and eliding the moved VP, as illustrated in (28b). Thus, according to this analysis, in order for a VP to elide, it must first topicalize.

(28)

  • a.

    John ate a cake, and Bill did, too.

  • b.

    [VP eat a cake]1 Bill did t1, too

This analysis is conceptually appealing since it does not require us to assume an additional operation specific to ellipsis: the same operation that is used for copy deletion can be used for ellipsis. This analysis also has an empirical motivation, because it can straightforwardly capture similarities in the licensing conditions of VP-ellipsis and VP-topicalization. VP-ellipsis is licensed by auxiliary verbs like do, have, and be, and infinitival to, as illustrated in the following sentences, which are drawn from Johnson 2001:440:

(29)

  • a.

    José Ybarra-Jaegger likes rutabagas, and Holly does Δ too.

  • b.

    José Ybarra-Jaegger ate rutabagas, and Holly has Δ too.

  • c.

    José Ybarra-Jaegger is eating rutabagas, and Holly is Δ too.

  • d.

    Mag Wildwood wants to read Fred’s story, and I also want to Δ.

In the absence of such auxiliary verbs, VP-ellipsis is prohibited, as the following examples indicate (drawn from Johnson 2001:439, 440):

(30)

  • a.

    I can’t believe Holly Golightly won’t eat rutabagas. I can’t believe Fred *(won’t), either.

  • b.

    Sally Tomato started running down the street, but only after José started *(to).

Johnson shows that the same licensing condition is applied to VP-topicalization, as the following examples illustrate (drawn from Johnson 2001:444):

(31) Madame Spanella claimed that . . .

  • a.

    eat rutabagas, Holly wouldn’t t.

  • b.

    eaten rutabagas, Holly hasn’t t.

  • c.

    eating rutabagas, Holly should be t.

  • d.

    eating rutabagas, Holly wants to t.

(32) Madame Spanella claimed that . . .

  • a.

    *would eat rutabagas, Holly t.

  • b.

    *hasn’t eaten rutabagas, Holly t.

  • c.

    ?*eating rutabagas, Holly started t.

This distributional similarity between VP-ellipsis and VP-topicalization can be straightforwardly captured by Johnson’s movement approach to VP-ellipsis since the former is derived through the latter under this approach. Whatever rules out (32a–c) also rules out (30a–b).

Despite its advantages, however, we cannot adopt Johnson’s analysis in its original form because there are a number of counterexamples to it, as Aelbrecht and Haegeman (2012) point out. These are reviewed in the next section.16

3.2.2 TopP in a vP Periphery: Aelbrecht and Haegeman 2012

Aelbrecht and Haegeman (2012) argue against Johnson’s analysis, showing that the distribution of VP-topicalization is much more restrictive than that of VP-ellipsis (i.e., there are a number of cases where VP-ellipsis is allowed even if VP-topicalization is not). Some counterexamples are listed in (33)–(35).

(33) Wh-complements

  • a.

    *I knew that one student presented this article in my class but I can’t recall now

    [which of the students [present this article] did t].

  • b.

    I knew that some students presented this article in my class but I couldn’t recall

    [which of the students didn’t ∅].

    (Aelbrecht and Haegeman 2012:599)

(34) Adverbial clauses

  • a.

    *Mary wanted to move to London, and after [move to London] she did t, her life changed entirely.

  • b.

    Mary wanted to move to London, and after she did, her life changed entirely.

    (Aelbrecht and Haegeman 2012:600, 601)

(35) Factive complements

  • a.

    *John intended to make a table, and we were glad that [make one] he did t.

  • b.

    John intended to make a table, and we were glad that he did.

    (Aelbrecht and Haegeman 2012:602)

As these examples illustrate, VP-ellipsis is allowed in the environments where VP-topicalization is not, such as wh-complements, adverbial clauses, and factive complements (see Aelbrecht and Haegeman 2012 for other counterexamples to Johnson’s analysis). This discrepancy between VPellipsis and VP-topicalization is unexpected under Johnson’s analysis, in which VP-ellipsis is derived through VP-topicalization: again, under his analysis, whatever rules out VP-topicalization should also rule out VP-ellipsis.

Aelbrecht and Haegeman (2012) suggest that we can overcome the problem with Johnson’s approach while maintaining its core idea if we take a low vP periphery into consideration.17 Assuming that there is a TopP in the vP periphery, along lines proposed by Jayaseelan (2001) and Belletti (2001, 2004, 2009), they argue that VP-ellipsis involves VP-movement to Spec,TopP in the vP periphery rather than in the CP periphery. Then the counterexamples to Johnson’s original analysis lose their force under this alternative analysis if we assume that what causes the ungrammaticality in the counterexamples is not VP-movement itself but VP-movement to the CP periphery.

Given this discussion, I propose the following hypothesis:

(36)

  • a.

    VP-topicalization is derived through verbal phrase movement to Spec,TopP in the CP periphery.

  • b.

    VP-ellipsis is derived through verbal phrase movement to Spec,TopP in either the CP periphery or the vP periphery.

This hypothesis accounts for the fact that the distribution of VP-topicalization is more limited than that of VP-ellipsis since VP-ellipsis can be derived in two ways while VP-topicalization can only be derived in one.18 For example, the second conjunct clause in (37a) can be derived in two ways, as illustrated in (37b) and (37c).

(37)

  • a.

    John solved the problem, and Mary did too.

  • b.

    [TopP[solve the problem] [TP Mary did t]]

  • c.

    [TP Mary did [TopP[solve the problem] [t]]]

In (37b), the elided verbal phrase is moved to Spec,TopP in the CP periphery, while in (37c), it is moved to Spec,TopP in the vP periphery. On the other hand, the embedded clause of the second conjunct in (38a) can only be derived in one way, as shown in (38b).

(38)

  • a.

    John solved the problem, and I think that [solve the problem] Mary will too.

  • b.

    [TopP [solve the problem] [TP Mary will t]]

The difference between VP-ellipsis and VP-topicalization with respect to possible landing site can be explained by the ECC (Antilocality effects) and Lasnik’s (1999) assumptions about V-movement, which are (a) that V has a strong feature that must be eliminated before the derivation reaches PF and (b) that the strong feature of V can be eliminated either by moving to v or by being elided by deletion. First, VP-topicalization cannot be derived through VP-movement to Spec,TopP in the vP periphery. This is because in the derivation illustrated in (39), V does not move to v.

(39) *[TP Mary will [TopP[VP solve(F ) the problem] [Top′ Top [vP v tVP]]]]

This is an illicit PF representation since the strong feature of V (F ) is not eliminated. On the other hand, in the derivation in which VP-topicalization is derived through vP-movement to Spec,TopP in the vP periphery, the strong feature of V is eliminated since V can move to v. This is illustrated in (40).

(40) *[TP Mary will [TopP[vP solve(F)-v the problem] [Top′ Top tvP ]]]

However, notice that vP-movement to Spec,TopP in the vP periphery is a typical movement that violates the Antilocality Principle. As we saw in section 3.1.2, this kind of superlocal movement is prohibited by the ECC. Therefore, there is no licit way to derive VP-topicalization through verbal phrase movement to Spec,TopP in the vP periphery.

On the other hand, the derivation in which vP moves to Spec,TopP in the CP periphery is unproblematic with respect to both V’s strong feature and the Antilocality Principle, as illustrated in (41).

(41) [TopP[vP solve(F)-v the problem] [Top′ Top [TP Mary will tvP ]]]

The derivation in which VP moves to Spec,TopP in the CP periphery is ruled out for the same reason as the case of VP-movement to Spec,TopP in the vP periphery (i.e., V’s strong feature fails to be eliminated). Therefore, the only possible derivation for VP-topicalization is the one in which vP moves to Spec,TopP in the CP periphery. This is why VP-topicalization can be derived only through verbal phrase movement to Spec,TopP in the CP periphery.

In contrast, VP-ellipsis can be derived through verbal phrase movement to Spec,TopP in either the CP periphery or the vP periphery. This is because V’s strong feature can be eliminated along with the deletion of VP even if V does not move to v. VP-ellipsis cannot be derived through vP-movement to Spec,TopP in the vP periphery as in the case of VP-topicalization. This is due to the ECC (Antilocality effects), as illustrated in (42).

(42) *[TP Mary will [TopP[vPsolve(F )-v the problem] [Top′ Top tvP ]]]

In contrast to VP-topicalization, VP-ellipsis can be derived through VP-movement to Spec,TopP in the vP periphery even if V does not move to v to eliminate its strong feature F. This is because the strong feature can be eliminated along with the deletion of VP, as (43) shows.

(43) [TP Mary will [TopP[VPsolve(F ) the problem] [Top′ Top [vP v tVP]]]]

Therefore, VP-ellipsis, unlike VP-topicalization, can be derived through verbal phrase movement to Spec,TopP in the vP periphery. As for movement to Spec,TopP in the CP periphery, both VP and vP can be moved. Both derivations are unproblematic in terms of the ECC and V’s strong feature, as illustrated in (44).

(44)

  • a.

    [TopP[VPsolve(F ) the problem] [Top′ Top [TP Mary will [vP v tVP]]]]

  • b.

    [TopP[vPsolve(F )-v the problem] [Top′ Top [TP Mary will tvP ]]]

Thus, VP-ellipsis can be derived through verbal phrase movement to Spec,TopP in either the CP or the vP periphery.

In this way, the hypothesis in (36) can be derived under the present system, in which the ECC and Lasnik’s assumptions about V-movement are adopted. Furthermore, in Funakoshi 2011b, I provide empirical evidence that favors this hypothesis over Johnson’s original movement analysis of VP-ellipsis and other nonmovement approaches to VP-ellipsis.

Given this analysis of VP-ellipsis, we can explain LG in the case of V/vP-ellipsis by means of the ECC and structural intervention (MLC effects). To do so, we need to show that both headless V/vP-movement to the CP periphery and headless V/vP-movement to the vP periphery are ruled out.

First, let us see how V/vP-movement to Spec,TopP in the CP periphery is ruled out when the moved phrase is headless. There are five possible configurations to be considered, depending on (a) what is elided (VP or vP) and (b) where V moves to (v, T, or Top). Suppose that V moves to v and stops there. Then the configuration in (45) results after the higher Top is merged with TP.

(45)

graphic

In this configuration, a headless VP obtains. This headless VP cannot be moved to Spec,TopP since AD(Top) contains both V and VP (V does not intervene between Top and VP, nor does VP intervene between Top and V). According to the ECC, Copy cannot apply to VP because of the presence of a smaller element V in AD(Top).19

If the [V-v] complex further moves to T, a headless VP and vP obtain, as shown in (46).

(46)

graphic

Headless vP-movement to Spec,TopP is ruled out since AD(Top) is {v, vP}, prohibiting Copy from applying to vP by the ECC. Furthermore, the headless VP cannot be moved to Spec,TopP because VP is not contained in AD(Top) owing to structural intervention by V (every category that dominates V also dominates VP (TopP and TP) and some category that dominates VP does not dominate V (vP)).

Finally, if the [V-v-T] complex further moves to Top, the configuration in (47) obtains.

(47)

graphic

Here, the headless vP cannot be moved since AD(Top) does not contain vP owing to structural intervention by v: Top c-commands v,20 every category that dominates v also dominates vP (TopP), and some category that dominates vP does not dominate v (TP). The headless VP cannot be moved either, for the same reason: V structurally intervenes between Top and VP.

In this way, both headless VP-movement and headless vP-movement to Spec,TopP in the CP periphery are ruled out either by the ECC or by structural intervention (MLC effects), regardless of the final landing site of V.

Let us next consider whether headless V/vP-movement to Spec,TopP in the vP periphery is ruled out. In this case, headless V/vP-ellipsis can be derived in three ways, depending on (a) what is elided (VP or vP) and (b) where V moves to (v or Top). When V moves to v and stops there, a headless VP results, as illustrated in (48).

(48)

graphic

In this configuration, the headless VP cannot be moved since AD(Top) is {V, VP}: V does not intervene between Top and VP, nor does VP intervene between Top and V, since there is no category that dominates V/VP and does not dominate VP/V. The ECC requires that Copy apply to V rather than VP. Therefore, the headless VP cannot be moved in this case.

Next, let us consider the case in which the [V-v] complex moves to Top. In this case, there are two headless XPs, vP and VP, as shown in (49).

(49)

graphic

The headless vP cannot be moved because AD(Top) is {v, vP}, and the ECC prohibits Copy from applying to vP. The headless VP cannot be moved because of structural intervention by V: AD(Top) does not contain VP since V structurally intervenes between Top and VP (Top c-commands V, every category that dominates V also dominates VP (TopP), and some category that dominates VP does not dominate V (vP)).

Therefore, headless V/vP-movement to Spec,TopP in the vP periphery is prohibited either by the ECC or by structural intervention (MLC effects) regardless of the final landing site of V, as in the case of headless V/vP-movement to Spec,TopP in the CP periphery. Under the hypothesis in (36), then, we can explain why headless V/vP-ellipsis is prohibited: it is because V/vP-ellipsis is derived through V/vP-movement to Spec,TopP either in the vP periphery or in the CP periphery.

If this line of analysis can be extended to the ellipsis of other categories such as IP and NP, and if movement is required to license ellipsis in general, as Johnson (2001) suggests, then LG is fully explained. However, the explanation of the prohibition of headless V/vP-ellipsis might be sufficient because almost all the data that LG accounted for in section 2.2 involve headless VP-ellipsis except the matrix sluicing data in (15). Notice that the matrix sluicing data are consistent not only with LG but also with Merchant’s (2001:62) sluicing-COMP generalization that “in sluicing, no non-operator material may appear in COMP.” Thus, it is conceivable that the prohibition against headless IP-ellipsis will be explained by an independent principle that explains the sluicing-COMP generalization.

3.3 Interim Summary

To summarize the discussion so far, I proposed an economy condition on Copy (the ECC) and showed that it can explain TG (i.e., why headless XPs cannot be moved). Furthermore, I argued that if we adopt a movement analysis of VP-ellipsis, which is a revised version of Johnson’s (2001) analysis based on Aelbrecht and Haegeman’s (2012) idea, LG (a subpart of it, at least) can be reduced to the ECC.

In what follows, I turn to crosslinguistic variation in the applicability of TG and LG and argue that if UG allows two types of head movement, the crosslinguistic variation can be reduced to parameters that are responsible for the availability of multiple specifiers.

4 Exceptions to Takano’s and Lasnik’s Generalizations

If the account given in the last section is correct, it is expected that TG and LG should hold in all languages. In this section, however, I show that there are languages where TG and/or LG do not hold. That is, languages vary in the availability of headless XP-movement/ellipsis.

4.1 Exceptions to Takano’s Generalization

Takano (2000) points out that TG seems not to hold in Japanese, given Koizumi’s (1995) analysis of Japanese multiple-cleft sentences (see Hoji 1987, Koizumi 1995, Kuwabara 1997). In multiplecleft sentences in Japanese, exemplified in (50), more than one element can appear in the focus position, which is the position right before the copula verb da.

(50)

  • [CP John-ga t1 ageta2 no]-wa [vP Mary-ni hon-o ni-satu t2]1 da.

  • John-NOM gave COMP-TOP Mary-DAT book-ACC two-CL be

  • (Lit.) ‘It is Mary two books that John gave.’

  • ‘John gave Mary two books.’

Koizumi (1995) argues that sentences like (50) involve headless verbal phrase movement, which I assume is vP-movement. Thus, as illustrated in (51) (with English vocabulary), the verb age ‘give’ moves to v, and v containing the verb moves to T. This movement yields the headless vP containing the indirect object Mary-ni ‘Mary-DAT’ and the direct object hon-o ni-satu ‘book-ACC two-CL’. The multiple-cleft sentence in (50) obtains if the resulting headless vP moves to the focus position.

(51) [TP John-NOM [vPtJohn [VP Mary-DAT books-ACC two-CLtV] tv] [T give-v-T]]

If this analysis is correct, it indicates that headless vP-movement is possible in Japanese. That is, TG does not hold in Japanese.

Takano (2000) also points out (attributing the observation to an LI reviewer) that according to Huang (1997), certain action sentences in Chinese involve headless verbal phrase movement. If this is correct, Chinese does not obey TG either.21

Furthermore, Landau (2006) argues that verbal-phrase-fronting sentences like (52a) in Hebrew involve headless verbal phrase movement, which I assume is vP-movement.

(52)

  • a.

    Liknot et ha-praxim, hi kanta.

    to.buy ACC the-flowers she bought

    ‘As for buying the flowers, she bought.’

  • b.

    [vP Liknot1 et ha-praxim]2 [TP hi kanta1t2].

    to.buy ACC the-flowers she bought

    ‘As for buying the flowers, she bought.’

    (Landau 2006:37)

In (52), liknot ‘to.buy’ moves to v and v moves to T, as illustrated in (53), yielding a headless vP. If the resulting headless vP moves to Spec,CP and the copies of liknot ‘to.buy’ are pronounced both in the head and in the tail of the chain, the sentence in (52) is derived. Both copies of the verb are pronounced for phonological reasons (see Landau 2006 for details).

(53) [CP[vPtshe [tv [VPtbuy the-flowers]]] [C′ C [TP she [T′[T buy-v-T] tvP ]]]]

Although the head of the fronted vP can be “seen,” this is an instance of headless vP-movement in the sense that the head moves out of the vP. Thus, TG does not hold in Hebrew.

To sum up, Japanese and Hebrew (and possibly Chinese) have constructions that involve headless vP-movement. This means that TG does not hold in these languages. This conclusion appears problematic for the present account of TG since headless vP-movement should be prohibited because of the ECC.

4.2 Exceptions to Lasnik’s Generalization

Recall the discussion about null objects in English in section 2.2, where I argued that in order to rule out null object sentences, headless VP-ellipsis must be prohibited. Thus, the ECC (combined with the assumption that pro is not available in English) can explain the fact that null objects are not allowed in English.

However, there are languages that allow null objects, unlike English. For example, the following languages allow null objects (and/or null VP-internal elements): Japanese (Otani and Whitman 1991), Chinese (Huang 1988, 1991), Korean (Otani and Whitman 1991), Hebrew (Doron and Heycock 1999, Goldberg 2005), Swahili (Goldberg 2005), Persian (Sailor 2010), Serbo-Croatian (Lasnik 1997), and Irish (McCloskey 1991, Goldberg 2005). Some examples are given in (54)–(59).

(54)

  • John-wa [zibun-no tegami-o] suteta. Mary-mo [e] suteta.

  • John-TOP self-GEN letter-ACC discarded Mary-also discarded

  • (Lit.) ‘John threw out letters of himself. Mary also threw out.’

  • (Japanese; Otani and Whitman 1991:346–347)

(55)

  • a.

    graphic

    ‘(Did) Miryam drive Dvora to the grocery store?’

  • b.

    graphic

    yes she drive.PAST.3SG.FEM

    ‘Yes, she drove Dvora to the grocery store.’

    (Hebrew; Goldberg 2005:53)

(56)

  • graphic

  • ‘Juma carried a child, and Kamau carried a child too.’

  • (Swahili; Goldberg 2005:57)

(57)

  • Naysan ketaab-ro ba deghat khoond, Nasim ham khoond [e].

  • Naysan book-OBJ with caution read Nasim also read

  • ‘Naysan read the book carefully, and Nasim also did (read the book carefully).’

  • (Persian; Sailor 2010:11)

(58)

  • Ivan piše rad pažljivo, a njegov asistent čita [e].

  • Ivan writes paper carefully and his assistant reads

  • ‘Ivan is writing the paper carefully and his assistant is reading it carefully.’

  • (Serbo-Croatian; Lasnik 1997:180)

(59)

  • a.

    graphic

    ‘Did you apply for it?’

  • b.

    graphic

    ‘Yes, I applied for it.’

    (Irish; McCloskey 1991:272, slightly modified)

These languages are radical pro-drop languages, in which objects as well as subjects can be dropped without any agreement on predicates. The authors listed above argue that these languages utilize what I call headless V/vP-ellipsis in order to derive some instances of null object (and/or VP-internal element) sentences. For instance, Goldberg (2005) analyzes the Hebrew sentence in (55b) as having the following underlying structure:

(60) [TP she [T drive-v-T] [vPtshe [tv [VPACC-Dvora tV to.the-grocery.store]]]]

In (60), V moves to v and v moves to T. As a result of this head movement, a headless vP and a headless VP are derived. If either of these headless phrases is elided, the elliptical sentence in (55b), in which both the direct and the indirect object are null, is derived.

If the headless V/vP-ellipsis analysis of null object constructions in these languages is correct, it means that LG does not hold in these languages. However, null object constructions could be analyzed in multiple ways. They could at least be analyzed as constructions involving noun phrase ellipsis, pro, or headless V/vP-ellipsis. Goldberg (2005) provides evidence that some null object sentences in Hebrew, Swahili, and Irish should be derived by headless V/vP-ellipsis. For example, she argues that the Hebrew sentence in (55) cannot be analyzed as involving either pro or noun phrase ellipsis. Consider (61).

(61)

  • graphic

  • et ha-kadur.

  • ACC the-ball

  • ‘Karmela gave the book to Chagit, and Yosef threw the ball.’

  • (Goldberg 2005:45)

The second conjunct can only mean that Yosef threw the ball—not that Yosef threw the ball to Chagit. This suggests that locative arguments can neither elide nor be instances of pro by themselves in Hebrew. On the other hand, as shown in (55b), a locative argument can be null when a direct object is also null. Thus, the sentence in (55b) cannot be analyzed as either a sentence in which the direct object and the locative phrase are independently elided or a sentence in which two instances of pro appear. Only a headless V/vP-ellipsis analysis can capture the acceptability of this sentence. See Goldberg 2005 for similar arguments for the availability of headless V/vPellipsis in Swahili and Irish. Furthermore, see Sailor 2010 and Lasnik 1997 on Persian and Serbo-Croatian, respectively. If these authors’ arguments are correct, headless V/vP-ellipsis is allowed in Hebrew, Swahili, Irish, Persian, and Serbo-Croatian.

The analysis of null object constructions in Japanese is controversial because a number of researchers (among many others, Oku (1998), Saito (2004, 2007), Goldberg (2005), and Takahashi (2008a,b)) argue that null object (or null VP-internal element) constructions in Japanese should be analyzed as noun phrase (or argument) ellipsis. However, in Funakoshi 2011a I provide arguments, based on empirical evidence, that some instances of the null object (or null VP-internal element) construction in Japanese involve headless V/vP-ellipsis. In what follows, I review one of those arguments.

In Japanese, the focus particle -dake ‘only’ can attach to a PP. It has been observed that -dake can appear either after a postposition or between the complement of a postposition and the postposition, as shown in (62).

(62)

  • a.

    • i.

      Mary-to-dake

      Mary-with-only

    • ii.

      Mary-dake-to

      Mary-only-with

  • b.

    • i.

      Mary-ni-dake

      Mary-to-only

    • ii.

      Mary-dake-ni

      Mary-only-to

  • c.

    • i.

      basu-de-dake

      bus-with-only

    • ii.

      basu-dake-de

      bus-only-with

Shoji (1986) observes that the linear order of -dake and a postposition affects the scope pattern of -dake. When -dake precedes a postposition (hereafter the internal -dake), -dake can take narrow scope under -(rar)e ‘can’, as shown in (63a). In contrast, when -dake follows a postposition (hereafter the external -dake), -dake can only take wide scope, as shown in (63b).22

(63)

  • a.

    John-wa Mary-dake-to asob-e-ru.

    John-TOP Mary-only-with play-can-PRES

    ‘John can play only with Mary.’ (only > can, can > only)

  • b.

    John-wa Mary-to-dake asob-e-ru.

    John-TOP Mary-with-only play-can-PRES

    ‘John can play only with Mary.’ (only > can, *can > only)

    (Funakoshi 2011a:66)

Sentence (63a) can mean either that the only person who John can play with is Mary (he cannot play with others: the only > can reading) or that John can play with Mary alone (without playing with others: the can > only reading). Only the former reading is available for sentence (63b).

Shoji (1986) accounts for this difference between the internal -dake and the external -dake by assuming that a PP with the external -dake is forced to move to a focus position outside vP, leading to the wide scope reading of -dake, while a PP with the internal -dake can stay in situ, resulting in the narrow scope reading of -dake.

With this in mind, consider (64).

(64)

  • John-wa Mary-dake-to asob-e-ru. Bill-mo [e] asob-e-ru.

  • John-TOP Mary-only-with play-can-PRES Bill-also play-can-PRES

  • (Lit.) ‘John can play only with Mary. Bill can play, too.’ (*only > can, can > only)

  • (Funakoshi 2011a:67)

Sentence (64) involves a null PP whose antecedent is a PP with the internal -dake. Crucially, this sentence only allows the narrow scope reading of -dake. More strikingly, sentences involving null PPs that take as their antecedents PPs with the external -dake are totally unacceptable, as illustrated in (65).23

(65)

  • John-wa Mary-to-dake asob-e-ru. *Bill-mo [e] asob-e-ru.

  • John-TOP Mary-with-only play-can-PRES Bill-also play-can-PRES

  • (Lit.) ‘John can play only with Mary. Bill can play, too.’

  • (Funakoshi 2011a:67)

These null PP sentences cannot be considered to involve pro because pro is a DP, not a PP. Therefore, these sentences should be analyzed either as PP-ellipsis sentences or as headless V/vP-ellipsis sentences. Suppose, following Shoji (1986), that PPs with the external -dake must move out of vP, while PPs with the internal -dake can stay within vP. Then the headless V/vPellipsis analysis can account for the fact that null PPs with the internal -dake must take scope under ‘can’ and null PPs with the external -dake are unacceptable.

Suppose that movement of -dake-phrases is overt movement to Spec,FocP in the vP periphery (see Miyoshi and Hoshi 2007).24 Then sentence (64) with the internal -dake could have the following two representations, since PPs with the internal -dake can either stay within vP or move out of vP:

(66)

  • a.

    Bill-also [v/VP Mary-only-with tplay] play-can

  • b.

    Bill-also [FocP Mary-only-with [v/VPtPPtplay]] play-can

As shown in (66a) and (66b), headless V/vP-ellipsis yields a null PP sentence only when the PP stays within vP. Therefore, under the headless V/vP-ellipsis analysis, the lack of the wide scope reading of a null PP with the internal -dake can be explained. On the other hand, this fact remains a mystery under the PP-ellipsis analysis.

Under the headless V/vP-ellipsis analysis, the null PP sentence with the external -dake in (65) only has a structure like (67), since a PP with the external -dake must move out of vP.

(67) Bill-also [FocP Mary-with-only [vPtPPtplay]] play-can

Thus, headless V/vP-ellipsis cannot derive the null PP sentence. On the other hand, if PP-ellipsis is possible, this fact remains a mystery. In this way, these null PP sentences suggest that headless V/vP-ellipsis is available in Japanese. See Huang 1988, 1991, Otani and Whitman 1991, and Takahashi 2008a for arguments supporting the headless V/vP-ellipsis analysis of null objects in Korean and Chinese.

In addition to the above languages, Chickasaw (Munro 1999), Thai (Huang 2000, Simpson 2005), Indonesian (Simpson 2005), Vietnamese (Simpson 2005), and Hungarian (Huang 2000) are also radical pro-drop languages (in particular, they allow null object sentences). If at least some null object sentences in these languages must be analyzed as involving headless V/vPellipsis, like the languages mentioned above, LG does not hold in these languages either.

To summarize this section, we have seen that TG does not hold in Japanese and Hebrew (and possibly Chinese) and that LG does not hold in Japanese, Chinese, Korean, Hebrew, Swahili, Persian, Serbo-Croatian, and Irish (and possibly Chickasaw, Thai, Indonesian, Vietnamese, and Hungarian, too). Interestingly, Japanese and Hebrew (and Chinese) do not obey either TG or LG. This fact suggests that these generalizations are rooted in the same principle, supporting the unified account of these generalizations that I propose. Whether the other languages in which LG does not hold also do not obey TG remains a topic for future research.

The question naturally arises, why do the generalizations hold in some languages like English but not in others like Hebrew? In the next section, I answer this question by arguing that TG and LG hold only in languages that do not allow head movement via substitution (they only allow head movement via adjunction).

5 Explanation for the Crosslinguistic Variation

5.1 The “Two Types of Head Movement” Hypothesis

It has been generally assumed (exceptions are Rizzi and Roberts 1989, Roberts 1991, 1994) that head movement, unlike phrasal movement, has only one mode: adjunction. However, I argue that there are two types of head movement, and I propose the following hypothesis:

(68)

  • “Two types of head movement” hypothesis

  • UG allows head movement via substitution (HMS) as well as head movement via adjunction (HMA).

More specifically, I claim that although HMA is basically always available, HMS is also an available option under certain circumstances. Thus, either HMA or HMS can occur when HMS can occur.

The two types of head movement are illustrated in the following schematic structures:

(69)

graphic

(70)

graphic

In (69), X moves to Y by adjunction, which is standard head movement. On the other hand, in (70), X moves to Y but via substitution, projecting Y into Y′.25 HMS as illustrated in (70) corresponds to Pesetsky’s (2007) Undermerge and Uriagereka’s (2010) Submerge. Takano (2007) also proposes the same type of operation for phrasal movement (complement-forming movement, in his terms).

I argue that (68) is the null hypothesis. UG has two independently motivated structurebuilding operations, as Chomsky (1995b, 2000, 2004) argues, adjunction/substitution in Government-Binding terms, Pair-Merge/Set-Merge in Chomsky’s (2000, 2004) terms. Thus, unless we make stipulations to the contrary, we should postulate the two types of operations not only for phrasal movement but also for head movement.

However, one might wonder if the configuration that HMS creates would violate any principles. Before showing how the hypothesis in (68) can capture the crosslinguistic difference, I would like to address this question. First, as a result of HMS of X to Y in (70), the complement of Y is a head (X) rather than a maximal projection. This is problematic under X-bar theory since it requires that every complement and specifier be a maximal projection (see Chomsky 1986). However, under the bare phrase structure theory, this ceases to be problematic since under this theory, minimal projection (head) and maximal projection (phrase) are considered entirely relational notions.

(71)

  • Given a phrase marker, a category that does not project any further is a maximal projection XP and one that does not project at all is a minimal projection X0; any other is an X′.

  • (Chomsky 1995a:396)

In (70), the higher copy of X is simultaneously minimal and maximal since it does not project further and does not project at all even if the lower copy of X is a minimal projection.26 Therefore, (70) is unproblematic under the bare phrase structure theory. Although I have used X-bar-theoretic notions like adjunction, substitution, categories, and segments, the present system is compatible with the bare phrase structure theory, as we will see in section 5.3.

Second, in (70), HMS of X leads to the situation where the original complement of Y, XP, becomes a specifier of Y, given a relational definition of complement and specifier.

(72)

  • The head-complement relation is the “most local” relation of XP to a terminal head Y, all others within YP being head-specifier (apart from adjunction . . . ).

  • (Chomsky 1995a:397)

Thus, the potential problem with the structure in (70) is that XP becomes a specifier of Y even though Y subcategorizes for XP. However, as Takano (2007) argues, this is unproblematic if we adopt the derivational approach to syntactic relations (see Epstein et al. 1998). Under the derivational approach, Y’s subcategorization requirement is derivationally satisfied when Y is merged with XP before HMS of X to Y. Therefore, HMS of X to Y does not affect the relation already derivationally satisfied between Y and XP.27 In fact, as I will show in section 5.2, the fact that HMS turns an original complement into a derived specifier plays a crucial role in reducing the crosslinguistic variation in the applicability of TG and LG to a multiple-specifier parameter.

Finally, HMS creates a configuration in which the moved head does not c-command its original position, violating both the Proper Binding Condition and the Extension Condition. The former requires that traces be bound (Fiengo 1977, May 1977, Saito 1989), and the latter states that Merge must always expand the tree. As for the Proper Binding Condition, Takano (2007) points out that its empirical validity has been questioned, given examples like this:28

(73) [How t1 likely to win]2 is John1t2?

As for the Extension Condition, as Bošković and Lasnik (1999), Richards (1999, 2004), and Lasnik (2006) point out, all the derivations that the Extension Condition rules out can also be ruled out by Featural Cyclicity, which requires that a strong feature be checked as soon as possible after being introduced into the derivation (Richards 1999:127). Given this redundancy and the empirical evidence for Featural Cyclicity over the Extension Condition, these authors argue that we should adopt the former, dispensing with the latter. Notice that HMS of X to Y in (70) does not violate Featural Cyclicity if we, following Chomsky (1995b), interpret “as soon as possible” in its definition as meaning ‘before the maximal projection of the head bearing the strong feature gets embedded’. Therefore, HMS is unproblematic under the model in which Featural Cyclicity is adopted instead of the Extension Condition.

Given these discussions, I conclude that (a) the “two types of head movement” hypothesis is conceptually well-motivated since it is the null hypothesis and (b) it is also theoretically tenable under a certain model of syntax in which the bare phrase structure theory, the derivational approach to syntactic relations, and Featural Cyclicity are adopted. Furthermore, in what follows, I argue that this hypothesis is also empirically motivated by illustrating that it can explain why some languages like English do not allow headless XP-movement/ellipsis while other languages like Hebrew do.

Notice that the explanation for TG and LG given in section 3 was based solely on the structure in which HMA occurs (see (16), repeated here as (74)).

(74)

graphic

In this configuration, XP does not structurally intervene between α and X nor does X structurally intervene between α and XP since there is no category that dominates XP/X but does not dominate X/XP. Thus, AD(α) contains both XP and X. Then headless XP-movement is ruled out by the ECC since the smaller element X blocks Copy(XP).

Given the “two types of head movement” hypothesis in (68), can the explanation for TG and LG also apply to the structure in which HMS occurs? In fact, it cannot. To see this, consider the schematic structure in (75), where X head-moves to Y but the head movement takes place via substitution.

(75)

graphic

Notice that in this case, X is not in AD(α) since XP intervenes between α and X: (a) α c-commands XP, (b) XP matches α and X with respect to [F], (c) every category that dominates XP also dominates X, and (d) and some category (Y′) that dominates X does not dominate XP. Therefore, Copy(XP) is allowed even if X is smaller than XP. Given this consideration, the condition in (76) follows.

(76) XP-movement/ellipsis can be headless only if X head-moves to Y, the next higher head, via substitution.

To see how (76) works, let us look at more concrete examples. For example, if headless vP-movement/ellipsis is allowed in a language, v should be able to head-move to T via substitution in the language, as illustrated in (77).

(77)

graphic

In (77), V head-moves to v via adjunction and then the [V-v] complex head-moves to T via substitution. In this configuration, vP intervenes between v and a probe α. Thus, v cannot be a member of AD(α), allowing headless vP-movement/ellipsis even if v is smaller than vP.29

Likewise, if a language allows headless VP-movement/ellipsis, HMS of V to v is allowed in the language, as illustrated in (78). In (78), VP structurally intervenes between α and V. Thus, V is not in AD(α) and headless VP-movement/ellipsis is allowed.

(78)

graphic

Before we proceed to the next section, a remark is in order with regard to the Phase Impenetrability Condition (PIC) (Chomsky 2000). I do not assume the “standard” version of the PIC that states that a complement domain of a phase head is inaccessible to operations outside the phase. This version of the PIC prohibits the headless VP from moving in (78) since VP is the complement of v, a phase head. Therefore, I adopt a slightly weaker version of the PIC that is proposed by Chomsky (2001). According to this version, the complement domain of a phase head is inaccessible to operations at the next higher phase. Given this, VP-movement in (78) is unproblematic since the next higher phase, CP, has not been completed. See Asarina 2011 for independent evidence in favor of this weaker version of the PIC over the stronger version.

5.2 Multiple Specifiers

Given the condition in (76), the question immediately arises, when is HMS allowed? I argue that the (im)possibility of HMS of X to Y is reduced to the possible number of specifiers that Y can take. To see this, recall that HMS turns the original complement into a derived specifier, as mentioned in section 5.1. Given this, HMS of X to Y creates the multiple-specifier configuration if Y has another (original) specifier. This is illustrated in (79).

(79)

graphic

In (79), X head-moves to Y via substitution, turning the former complement XP into a specifier of Y. Note that in (79), Y has another specifier, WP. Thus, Y has multiple specifiers in this structure, WP and XP. If Y does not allow multiple specifiers, the configuration in (79) should be ruled out. Given these considerations, condition (80) follows.30

(80) Head movement of X to Y, the next higher head, can be performed via substitution only if either

  • a.

    Y allows multiple specifiers or

  • b.

    Y does not have an element in its specifier.

For example, let us consider when HMS of v to T is possible. When a language allows multiple TP specifiers, v can head-move to T via substitution in the language even if T has a subject in its specifier, as illustrated in (81).

(81)

graphic

On the other hand, if the subject does not move to Spec,TP in a language, HMS of v to T is possible in the language even if T does not allow multiple specifiers. This is illustrated in (82).

(82)

graphic

Combining the condition in (80) with the condition on headless XP-movement/ellipsis in (76), the following condition obtains by transitivity:

(83) XP-movement/ellipsis can be headless only if either

  • a.

    Y, the next higher head, allows multiple specifiers or

  • b.

    Y does not have an element in its specifier.

Given this condition, we expect (a) that in a language in which headless V/vP-movement/ellipsis is allowed, either multiple TP/vP specifiers are allowed or subjects do not move to Spec,TP and (b) that in a language in which multiple TP/vP specifiers are not allowed and Spec,TP cannot be empty, headless V/vP-movement/ellipsis are not allowed.

The latter prediction is borne out in languages like English. English allows neither multiple TP/vP specifiers nor empty Spec,TP and does not allow headless V/vP-movement/ellipsis, as we saw in section 2.31

The former prediction is confirmed in the languages where TG and LG do not hold. Following Ura (1994), I assume that T allows multiple specifiers in languages that have multiple-subject constructions. Japanese, Korean, Chinese, Persian, Indonesian, Vietnamese (Ura 1994); Hebrew (Doron and Heycock 1999); Thai (Kumashiro and Langacker 2003); and Chickasaw (Munro 1999) have multiple-subject constructions. Some examples are given here:

(84)

  • Zoo-ga hana-ga nagai.

  • elephant-NOM nose-NOM long

  • ‘Elephants’ noses are long.’

  • (Japanese; Ura 1994:34)

(85)

(86)

  • Mahmud ketâb-aš gom šod.

  • Mahmud.NOM book-3SG.NOM got lost

  • ‘It is Mahmud that his book got lost.’

  • (Persian; Ura 1994:33)

(87)

(88)

  • Hattak-at ofi’-a im-alhkaniya.

  • man-NOM dog-NOMDAT-forget

  • ‘The man forgets the dog.’

  • (Chickasaw; Munro 1999:254)

Furthermore, according to Richards (1997), T allows multiple specifiers in Serbo-Croatian and Hungarian as well (IP-absorption languages in his terms). Finally, on the basis of his analysis of possessor raising, Ura (1994) argues that in Swahili and Korean, V (v under the phrase structure adopted here) allows multiple specifiers.

If this is correct, these languages, which allow headless V/vP-movement/ellipsis (as I argued in section 4), allow multiple vP specifiers and/or multiple TP specifiers. This is what my analysis predicts.

Irish, which allows headless vP-ellipsis, can be considered a subject-in-situ language in accord with what (83) predicts. Irish has a VSO basic order in finite clauses. McCloskey (1991) argues that VSO order in Irish is derived by V-movement to T and leaving the subject in situ.32

Furthermore, it is expected that in Irish, subjects are obligatorily elided along with objects (i.e., an object cannot be elided by itself ). To see this, let us consider the structure in (82). There, v head-moves to T via substitution and Subj stays in Spec,vP. Thus, headless vP-ellipsis is allowed. Notice that VP-ellipsis in (82) is impossible owing to the ECC since VP does not intervene between the probe (Top above TP) and V. Thus, if pro is not available for null objects in Irish, a headless vP must be elided in order to derive a null object sentence. The vP contains an in-situ subject. Therefore, it is predicted that subjects are obligatorily elided along with objects in Irish. This is indeed the case (see McCloskey 1991, Goldberg 2005). In the Irish example in (59), the subject as well as the object is elided.

The present analysis makes a prediction about island effects. When a head X moves to Y via substitution, the complement XP becomes a specifier of Y. This means that elided headless XPs are always in the specifier position, since XP-ellipsis can be headless only if X head-moves to Y via substitution (see (76)). Given that specifiers are islands (see Huang 1982, Nunes and Uriagereka 2000), it is predicted that elements in elided headless XPs cannot be extracted. This prediction is indeed borne out in Japanese.

In Japanese, a complement CP can be elided, as shown in (89).

(89)

  • Mary-wa [John-ga ringo-o tabeta to] omotteiru, kedo Sue-wa [John-ga

  • Mary-TOP John-NOM apple-ACC ate COMP think but Sue-TOP John-NOM

  • ringo-otabeta to] omottei-nai.

  • apple-ACC ate COMP think-NEG

  • ‘Mary thinks John ate an apple, but Sue does not think John ate an apple.’

In the second conjunct of (89), the complement clause is elided. According to the headless vPellipsis analysis of null objects, this is derived by applying headless ellipsis to the matrix vP that contains the complement clause. As shown in (90), the matrix verb moves out of the matrix vP.

(90) Sue-TOP [vP[CP John-NOM apple-ACC ate COMP] tthink] think-NEG

If the resulting headless vP is elided, sentence (89) is derived.

Moreover, Japanese has long-distance scrambling, as shown in (91).

(91)

  • Ringo1-o Mary-wa [John-ga t1 tabeta to] omotteiru.

  • apple-ACC Mary-NOM John-NOM ate COMP think

  • ‘Mary thinks John ate an apple.’

In (91), the object in the embedded clause, ringo-o ‘apple-ACC’, is scrambled out of that clause. However, as Tanaka (2008) observes, scrambling out of an elided complement CP is prohibited, as shown in (92).

(92)

  • *Ringo-o1 Mary-wa [vP [John-ga t1 tabeta to] tv] omotteiru, kedo orenzi-o2

  • apple-ACC Mary-TOP John-NOM ate COMP think but orange-ACC

  • Sue-wa [vP [John-ga t2 tabeta to] tv] omottei-nai.

  • Sue-TOP John-NOM ate COMP think-NEG

  • ‘Mary thinks John ate an apple, but Sue does not think John ate an orange.’

In the second conjunct in (92), the object in the embedded clause, orenzi-o ‘orange-ACC’, is scrambled out of the headless vP, which is elided, yielding an unacceptable sentence. This is what the the present analysis of headless XP-ellipsis predicts. Notice that the island effect does not obtain without ellipsis in (92). This is expected under the present analysis because HMA is always possible when HMS is available. HMS is forced only when headless XP-movement/ellipsis has taken place as in (92).33

To sum up, from the “two types of head movement” hypothesis and my analysis of headless XP-movement/ellipsis, it follows that the availability of headless XP-movement/ellipsis is related to the availability of multiple specifiers: the former entails the latter. Japanese, Korean, Chinese, Persian, Hebrew, Swahili, Serbo-Croatian, and Irish allow some instances of headless XP-movement and/or ellipsis. Indonesian, Vietnamese, Thai, Hungarian, and Chickasaw are radical prodrop languages, which might be analyzed as involving headless V/vP-ellipsis. All of these languages except Irish allow some instances of multiple specifiers (multiple vP specifiers and/or multiple TP specifiers). Irish allows Spec,TP to be empty. On the other hand, languages like English allow neither multiple TP/vP specifiers nor empty Spec,TP. Thus, they do not allow headless V/vP-movement/ellipsis either. In this way, we can attribute the crosslinguistic variation in the availability of headless XP-movement/ellipsis to the parameters that are responsible for the availability of multiple specifiers.34

5.3 Bare Phrase Structure Theory and Unbounded Merge

Before I close this section, a remark is in order with regard to X-bar-theoretic notions like adjunction, substitution, segments, and categories, which I have used through this article. It is not clear how to define the distinction between segments and categories and that between substitution and adjunction under the current theory of bare phrase structure (BPS). However, this does not mean that my analysis is incompatible with BPS theory. Here, I suggest one possible way to reconcile the two.

The crucial assumption adopted in this article is that a node that is created as a result of adjunction of X to Y does not dominate X, while a node that is created as a result of substitution of X for Y does. The first one is May’s theorem that “adjuncts are not dominated by the categories to which they are adjoined” (May 1989:92). This is a defining property of adjunction. If there are at least two modes for the structure-building operation, that is, substitution/Set-Merge and adjunction/Pair-Merge, as Chomsky (1995a, 2000, 2004) argues, then any phrase structure theory must be able to capture this property in one way or another.

One promising way of capturing this property of adjunction under BPS theory is to adopt a proposal developed by Chametzky (2000), Uriagereka (2002), Hornstein and Nunes (2008), and Hornstein (2009). A core idea of this proposal is that substitution yields a labeled constituent while adjunction yields a nonlabeled constituent. I call the former operation L(abeled)-Merge and the latter operation Concatenate. Following Hornstein and Nunes (2008) and Hornstein (2009), I assume that Concatenate (adjunction) is a simple operation that combines two syntactic objects, while L-Merge, which is the same operation as Chomsky’s Set-Merge, is a composite operation consisting of Concatenate and Label. Label is an operation whereby one of the two inputs to Concatenate names the resulting constituents. This is illustrated in (93) (the underlined symbol is a label).35

(93)

  • a.

    Concatenate(X, Y) = {X, Y}

  • b.

    L-Merge(X, Y) = Label(Concatenate(X, Y)) = {X, {X, Y}}

Under this version of BPS theory, the structures created by the two types of head movement can be derived in the following way, using Concatenate and L-Merge (the subscript numbers are just for expository purposes):

(94) Head movement of X to Y via Concatenate (adjunction)

  • a.

    L-Merge(X, ZP): {X, {X, ZP}} (= XP)

  • b.

    L-Merge(Y, XP): {Y, {Y, {X, {X, ZP}}}}

  • c.

    Copy(X)

  • d.

    Concatenate(Y, X): {Y, {{Y, X2}, {X, {X1, ZP}}}} (= YP)

  • e.

    L-Merge(α, YP): {α, {α, {Y, {{Y, X2}, {X, {X1, ZP}}}}}}

(95) Head movement of X to Y via L-Merge (substitution)

  • a.

    L-Merge(X, ZP): {X, {X, ZP}} (= XP)

  • b.

    L-Merge(Y, XP): {Y, {Y, {X, {X, ZP}}}}

  • c.

    Copy(X)

  • d.

    L-Merge(Y, X): {Y2, {{Y1, {Y, X2}}, {X, {X1, ZP}}}} (= YP)

  • e.

    L-Merge(α, YP): {α, {α, {Y2, {{Y1, {Y, X2}}, {X, {X1, ZP}}}}}}

Under this framework, the difference between HMA and HMS is that in the former, head movement is realized by Concatenate, combining two heads without applying Label to the result, while in the latter, head movement is realized by L-Merge, combining two heads followed by Label.

Given this, let us consider how we can define the notion of dominance based on set-theoretic notations of phrase structure, which BPS theory assumes. In a tree diagram, dominators are always nodes (or a certain kind of node—namely, a category) and dominatees are everything under the dominator node. What corresponds to nodes in set-theoretic notations like that used in (94) and (95) is a label. Given this consideration, it seems natural to define the dominance relation as follows:

(96) (Irreflexive) dominance

α dominates β iff α is a label and

  • a.

    β is a sister of α or

  • b.

    β is a descendant of a sister of α.

The crucial point is that only a label can be a dominator in this definition. Sisters and descendants are defined as follows:

(97)

  • Sister

  • β is a sister of α iff β is another member of a set A such that α is a member of A.

(98) Descendant

  • a.

    A member of γ is a descendant of γ.

  • b.

    If δ is a descendant of γ, then a member of δ is also a descendant of γ.

Given these definitions, it follows that Copy of a headless XP is ruled out by the ECC when HMA of X to Y occurs, while it is not when HMS of X to Y occurs. In (94), X2, which underwent HMA, does not intervene between the probe α and {X, {X1, ZP}}, which is a headless XP, nor does {X, {X1, ZP}} intervene between α and X2. What dominates X2 are α and Y (X2 is a descendant of both the sister of α and the sister of Y). What dominates {X, {X1, ZP}} are also α and Y. Therefore, there is nothing that dominates one of them and does not dominate the other. This means that AD(α) contains both X2 and {X, {X1, ZP}}. Then the ECC prevents Copy from applying to {X, {X1, ZP}} since it is bigger than X2.

In the representation in (95), derived by HMS of X to Y, the dominators for X2 are α, Y2, and Y1, while those for {X, {X1, ZP}} are only α and Y2; {X, {X1, ZP}} is neither a sister nor a descendant of a sister of Y1. Therefore, {X, {X1, ZP}} structurally intervenes between the probe α and X2. As a result, AD(α) does not contain X2, and Copy is able to apply to {X, {X1, ZP}}.

In this way, it is possible to make the present analysis compatible with BPS theory without using X-bar-theoretic notions or notations. Although the specific implementation might turn out to be wrong, the important point I would like to make here is that the present analysis is not inherently incompatible with BPS theory.

Finally, I would like to point out that under this version of BPS theory, we can reduce the multiple-specifier parameter to a parameter that has to do with the restriction on the number of applications of L-Merge, without resorting to the notion of specifiers. Suppose that UG has the following parameter:

(99)

  • a.

    Languages like English: L-Merge can apply to a single head at most two times.

  • b.

    Languages like Japanese: L-Merge can apply to a single head without limit.

Given this parameter, both English and Japanese allow the following structure, in which X takes a complement, a specifier, and an adjunct:

(100) {WP, {X, {YP, {X, {X, ZP}}}}} (Specifier = YP, Complement = ZP, Adjunct = WP)

  • a.

    L-Merge(X, ZP): {X, {X, ZP}} (= A)

  • b.

    L-Merge(A, YP): {X, {YP, {X, {X, ZP}}}} (= B)

  • c.

    Concatenate(B, WP): {WP, {X, {YP, {X, {X, ZP}}}}}

This is because L-Merge applies to a single head X only two times: the third operation is Concatenate rather than L-Merge. On the other hand, only languages like Japanese allow the following structure, where X takes a complement and two specifiers:

(101) {X, {WP, {X, {YP, {X, {X, ZP}}}}}} (Specifiers = WP and YP, Complement = ZP)

  • a.

    L-Merge(X, ZP): {X, {X, ZP}} (= A)

  • b.

    L-Merge(A, YP): {X, {YP, {X, {X, ZP}}}} (= B)

  • c.

    L-Merge(B, WP): {X, {WP, {X, {YP, {X, {X, ZP}}}}}}

This is because L-Merge applies to X more than two times. The difference between L-Merge and Concatenate in this respect (i.e., the application of the former is restricted while that of the latter is free) seems natural since Concatenate is a general cognitive operation not specific to language while L-Merge involves Label, which is a language-specific operation. In a sense, the application of L-Merge seems more costly than that of Concatenate.

Under this conception of the multiple-specifier parameter, when a head Y has an element in its specifier, HMA of X to Y is allowed both in languages like English and in languages like Japanese, while HMS of X to Y is allowed only in languages like Japanese, as illustrated in (102).

(102)

  • a.

    {Y, {WP, {Y, {Y, {X, {X, ZP}}}}}}

  • b.

    graphic

    Y, {WP, {Y, {{Y, X}}, {X, {X, ZP}}}}}

  • c.

    graphic

    {Y, {WP, {Y, {{Y, {Y, X}}, {X, {X, ZP}}}}}}

Before head movement takes place, as shown in (102a), L-Merge applies to Y two times, leaving Y with a specifier, WP, and a complement, {X, {X, ZP}}. Therefore, HMA of X to Y can take place in both types of languages, while HMS of X to Y can take place only in the Japanese-type languages: HMS of X to Y requires that a third L-Merge apply to Y.

This view of the multiple-specifier parameter fits well in Fukui’s (2011) conception of multiple specifiers or multiple subjects (see footnote 30). Fukui (2011) argues that a number of differences between English and Japanese, including the availability of multiple subjects, can be reduced to a more fundamental difference between them: namely, that “‘unbounded Merge’ is in full force in the syntax of Japanese” (Fukui 2011:90) while Merge is bounded somehow in English. If this is correct, we can conclude that the crosslinguistic difference in the applicability of TG and LG is eventually reduced to the availability of unbounded (L-)Merge.

6 Concluding Remarks

In this article, I have proposed two things: the Economy Condition on Copy and the “two types of head movement” hypothesis. I have argued that (a) the former explains TG and LG and (b) with the former and the latter, the crosslinguistic variation in the applicability of these generalizations can be attributed to parameters that are responsible for the availability of multiple specifiers. Furthermore, I have suggested that the availability of multiple specifiers can be reduced to the availability of unbounded (L-)Merge if we adopt a certain model of BPS theory in which Concatenate and L-Merge are available as basic structure-building operations.

Notes

An earlier version of the material in this article was presented to audiences at Kwansei Gakuin University, Osaka University, and the University of Maryland. I would like to express my thanks for insightful comments and suggestions by Tonia Bleam, Chris LaTerza, Masao Ochi, Masahiko Takahashi, Hiroyuki Tanaka, and Hiroyuki Ura. I am especially grateful to Alex Drummond, Sayaka Goto, Norbert Hornstein, Bradley Larson, Howard Lasnik, Jeffrey Lidz, Terje Lohndal, Yuji Takano, and two LI reviewers for reading earlier versions and making valuable suggestions for improvement. This work was partially supported by a Fulbright grant.

1Lasnik (1999), however, immediately rejects the generalization on the basis of the fact that there are languages in which V is moved out of VP, while VP-ellipsis is allowed. I will account for exceptions to LG in section 4.

2 Abbreviations: ACC = accusative, CL = classifier, COMP = complementizer, DAT = dative, FEM = feminine, FV = final vowel, GEN = genitive, MASC = masculine, NEG = negation, NOM = nominative, Obj = object, PAST = past, PRES = present, Q = question particle, SG = singular, Subj, SUBJ = subject, TOP = topic marker, 3 = 3rd person.

3 As a reviewer correctly points out, although the English example in (6c) is consistent with TG, it does not support TG. This is because this sentence might be ruled out independently of TG: clefting of VP is not allowed in English, as the following sentence shows:

(i) *It’s give a book to Mary that I might.

Given this interfering factor, the following example might be more appropriate to illustrate TG:

(ii) *The book to Mary, John gave.

One of the possible representations of this sentence, in which the book and to Mary independently undergo topicalization, is illicit because multiple topicalization is prohibited in English.

(iii) *[DP The book] [PP to Mary], John gave tDPtPP.

However, in order to account for the unacceptability of the sentence in (ii), we need to rule out another possible representation for the sentence, which is given in (iva).

(iv)

  • a.

    *[VP The book t1 to Mary]2, John gave1t2.

  • b.

    [vP Give the book to Mary], John did t.

As shown in (iva), sentences like (ii) can be derived by headless VP-movement. Given that vP-topicalization is allowed in English, as shown in (ivb), it is unclear why (ivb) is acceptable while (iva) is not. TG can accommodate this fact since VP-topicalization in (iva) involves headless XP-movement.

4Takano (2000) gives an explanation for his generalization, which is based on Chomsky’s (1995b:304) condition that only the head of a chain enters into the operation Attract/Move. Although this can account for TG, the status of Chomsky’s condition is unclear under the copy theory of movement. If a trace is indeed a copy of the moved element, it is not clear why a trace (a copy) is invisible for the operation Attract/Move. But more importantly, this approach remains silent about exceptions to TG, which I will discuss in section 4.

5Kasai (2001:101), attributing the observation to Akira Watanabe, points out that there is a counterexample in English to TG.

(i)

  • a.

    They all said that John was shrewd and shrewd he was.

  • b.

    *They all said that John was shrewd and shrewd he looks.

The unacceptability of (ib) indicates that an AP cannot be fronted by itself. Thus, it must be the case that in (ia), the moved category is a VP rather than an AP. The fronted category in (ia) can be analyzed as a VP if the verb was moves out of the VP and the resulting headless VP is fronted. Notice that the same kind of derivation is impossible in (ib) since V, unlike be, does not move to T. Thus, the acceptability of (ia) could be a counterexample to TG.

However, a reviewer suggests that the acceptability of (ib) improves if the copula verb in the first conjunct clause

is replaced by a verb like look or seem. Thus, sentences like (ii) are much better than (ib).

(ii) ?They all said that John looked/seemed shrewd and shrewd he looks/seems.

This suggests that AP can be fronted. Furthermore, the reviewer notes that (iii) makes the same point.

(iii) They all said that John might be late and late he probably will be.

Unless nonfinite be undergoes short head movement, it seems to be the AP that is fronted in (iii). If this is correct, we cannot conclude that it is a headless VP rather than an AP that is fronted in (ia) since the AP is a possible candidate for fronting.

6Roberts (1998) and Potsdam (1997) propose a similar generalization, which amounts to saying that headless VPs cannot antecede VP-ellipsis. Lasnik (1997) sets aside their generalization because of the exceptions to it.

7Lasnik (1999) adopts the split-VP hypothesis of Koizumi (1995), although I omit the detailed structures for simplicity.

8Lasnik (2001) proposes an alternative explanation based on the Move F approach (Chomsky 1995c, Ochi 1999).

9 This proposal has a precursor in Chomsky’s (1995b) principle concerning pied-piping:

(i) F carries along just enough material for convergence.

(Chomsky 1995b:262)

This principle minimizes the amount of material that is pied-piped (copied in my terms), just as my economy condition does.

10 This informal explanation and a formal one that I will propose in section 3.1.2 crucially rely on X-bar-theoretic notions like categories, segments, and adjunction. This, however, does not mean that I adopt X-bar theory as a phrase structure theory. As is obvious from seeing that I assumed the copy theory in the explanation, I adopt a bare phrase structure theory in this article. In section 5.3, I will show that we can reconcile these X-bar-theoretic notions with a bare phrase structure theory if we adopt an idea developed by Chametzky (2000), Uriagereka (2002), Hornstein and Nunes (2008), and Hornstein (2009) to the effect that adjunction is a kind of unlabeled Merge. However, until then I use the X-bar-theoretic notions and notations just for expository purposes.

11 A reviewer criticizes the notions of AD and structural intervention in minimalist terms, pointing out that I bring in additional operations. However, notice that the definitions of AD and structural intervention do not use any operations other than standardly assumed operations or relations such as Agree, c-command, domination, and match. Moreover, AD is just a formal implementation of a tacitly assumed view that Agree is a prerequisite for Move. Thus, Agree-as-aprerequisite-for-Move presupposes some formal apparatus like this.

12 The economy condition in (21) is compatible with Ura’s (2001) view on economy that (Generalized) Pied-Piping, like other syntactic operations, is subject to the general economy condition. This runs counter to Tanaka 2004, where it is argued that (Generalized) Pied-Piping (Copy in my terms) is cost-free, meaning that (Generalized) Pied-Piping is not constrained by any economy conditions.

13 As Yuji Takano (pers. comm.) points out, in order to maintain the present explanation of TG, it must be guaranteed that in (24), XP cannot move to the edge of YP. This is because if XP were able to move to the edge of YP, XP would structurally intervene between α and X. As a result, AD(α) only contains XP. Then, Copy should be able to apply to XP because there is no smaller element than XP in AD(α). This undesirable, superlocal movement of XP to the edge of YP can be ruled out as an Antilocality effect. At the end of this section, I will show that this kind of Antilocality effect can be reduced to the ECC.

14 As Howard Lasnik ( pers. comm.) points out, I cannot say that intermediate projections are completely invisible in syntax because intermediate projections must be visible for computation of the dominance relation, which will be crucial to my explanation of the crosslinguistic variation in applicability of TG and LG (see section 5). Thus, I should say instead that intermediate projections cannot be candidates for syntactic operations such as Agree.

15Donati (2003) also explores the hypothesis that ellipsis can be reduced to movement. See also Kayne 2005, Rizzi 2005, Szczegielniak 2005, Fitzpatrick 2006, and Hornstein 2009 for relevant discussion.

16 I thank a reviewer for drawing my attention to Aelbrecht and Haegeman 2012.

17 Actually, they suggest two other alternatives in addition to the vP periphery analysis. See Aelbrecht and Haegeman 2012 for details.

18 This analysis is also compatible with the following examples that a reviewer provides as potential counterexamples to Johnson’s analysis:

(i)

  • a.

    Did someone call 911? Yes, John did [[call 911] and [(then) waited patiently]].

  • b.

    John’s having been arrested surprised me but [Mary’s having been [arrested]] was completely expected.

In (ia), the elided VP is contained in the VP-coordination structure. If VP-ellipsis is derived through VP-movement to the CP periphery, as Johnson argues, it is unclear why this sentence is not ruled out by the Coordinate Structure Constraint. On the other hand, under the vP periphery analysis, this sentence can be derived without extracting VP out of the coordinate structure. Suppose that the coordinated categories are TopPs in a vP periphery in (ia). Then, as the structure in (ii) illustrates, VP-movement to the lower Spec,TopP does not violate the Coordinate Structure Constraint.

(ii) [&P[TopP[VP call 911] [Top [vP v tVP]]] and [TopP Top [vP waited patiently]]]

Likewise, sentence (ib) would violate the Subject Island Constraint under Johnson’s analysis but not under the vP periphery analysis.

19 The Phase Impenetrability Condition (Chomsky 2000) could redundantly rule out VP-movement in (45) since VP is the complement of v, a phase head. As I will discuss in section 5.1, however, I do not adopt this version of the Phase Impenetrability Condition.

20 Top c-commands v since every category that dominates Top also dominates v (TopP) and Top does not dominate v. Notice that Top does not dominate v, because only one segment of Top dominates v.

21Takano (2000) does not give an explanation for these exceptions, although he suggests that the constructions in question should be analyzed in a different way. Indeed, he later (Takano 2002) provides an alternative analysis of multiplecleft sentences in Japanese, in which the moved phrase in sentences like (50) is not a headless vP.

22Shoji (1986) judges sentences like (63a) as unambiguous, allowing only the narrow scope of -dake. However, Futagi (2004) finds such sentences ambiguous, allowing both the narrow scope and the wide scope of -dake. My judgment agrees with Futagi’s.

23 The undeletability of the external -dake is observed by Futagi (2004). However, she argues that the internal -dake cannot be null either. Thus, according to her judgment, sentences like (64) are also unacceptable. However, I have a robust intuition that the sentences are acceptable under the narrow scope reading of -dake. All Japanese speakers that I have consulted agreed with my judgment.

24 I need to assume that -dake-phrase movement is overt because this article presupposes the PF deletion analysis of ellipsis rather than the LF copy analysis (see Merchant 2001).

25 In section 5.3, I will demonstrate how to implement this idea in a bare phrase structure theory without using X-bar-theoretic notions like substitution and adjunction. There, substitution and adjunction are reinterpreted as labeled Merge (L-Merge) and unlabeled Merge (Concatenate).

26 The higher copy does not inherit the minimal/maximal status from the lower copy since minimal projections and maximal projections, as just noted, are relational notions rather than inherent properties that categories have.

27 As a reviewer points out, this argument does not hold if head movement is derived through what Bobaljik and Brown (1997) call interarboreal movement. This is because under this approach, X has to move to Y before Y is merged with XP, whereby Y’s subcategorization requirement is satisfied, in order to satisfy the Extension Condition (Chomsky 1993). However, as I discuss in a moment, I adopt Featural Cyclicity (Chomsky 1995c, Richards 1999) as a way of deriving cyclicity, discarding the Extension Condition. Therefore, I do not have to assume that head movement is derived through interarboreal movement.

28Lasnik and Saito (1992), however, argue that (73) does not necessarily constitute a counterexample to the Proper Binding Condition because likely can ambiguously take a raising complement and a control complement. They provide the following example as evidence for their claim, attributing the observation to Anthony Kroch:

(i) *[How likely t1 to be a riot]2 is there1t2?

In this sentence, the matrix subject is expletive there. Thus, likely in this sentence, unlike the instance in (73), is unambiguously a raising verb. See Müller 1996 and Takano 1994 for relevant discussion.

29 Although I am agnostic about linearization mechanisms in this article, it might be the case that I cannot adopt the Linear Correspondence Axiom (LCA; Kayne 1994) or other mechanism which predicts that specifiers always precede heads. According to the present analysis, SVO languages like Hebrew allow structures like (77), in which v head-moves to T via substitution. The choice between HMA and HMS is completely optional when HMS is available. It is then wrongly predicted that objects in Hebrew can optionally precede or follow verbs if specifiers always precede heads. That is, vP containing an object becomes a specifier of T after HMS of v to T applies. I thank Alex Drummond, Yuji Takano, and a reviewer for bringing this issue to my attention.

30 A reviewer correctly points out that this does not follow if we adopt a certain view of a multiple-specifier parameter according to which a ban on multiple specifiers for a head H in a language L is reduced to the fact that H in L is not allowed to have the features that would introduce more than one specifier. For example, under Ura’s (1994, 2000) conception, a head that allows multiple specifiers, unlike a head that does not, has the ability to check a certain feature of elements in its specifiers multiple times. Under this conception of multiple specifiers, the condition in (80) does not follow. This is because the derived specifier does not have to be licensed as a specifier by the relevant feature since it is introduced as a complement prior to HMS. Given this, the reviewer argues that in order to maintain my argument, I need to adopt a representational view of a ban on multiple specifiers. However, as I will discuss in detail in section 5.3, in addition to fitting in with the representational view, the present analysis fits in with Fukui’s (2011) view of multiple specifiers (or multiple subjects), according to which in languages like Japanese, unlike English, the number of applications of Merge (substitution or Set-Merge) is unbounded.

31 Here, I assume that in successive-cyclic movement, Ā-movement to an intermediate landing site is performed via adjunction rather than substitution. For example, when an object wh-phrase moves to Spec,CP, the intermediate movement to the edge of vP takes place via adjunction. Therefore, even if a subject occupies Spec,vP, the successive-cyclic movement of the object wh-phrase does not yield a multiple-specifier configuration. I thank Yuji Takano and a reviewer for bringing this issue to my attention.

32 As a reviewer points out, more recently McCloskey (1996) argues that subjects do not stay in situ in Irish but move to a position somewhere between TP and vP. Even if this is correct, however, my prediction is not obviated because Spec,TP is empty under the new analysis as well.

33 As Tanaka (2008) points out, extraction out of elided material is not prohibited in general since in English, extraction out of an elided vP is possible, as illustrated in (i).

(i) I know which book John read, but I don’t know which one1 Bill did [vP read t1].

A sentence structurally more similar to the Japanese one is also acceptable, as shown in (ii).

(ii) Bob’s book, I argued John read, but Bill’s book1, I didn’t [vP argue [John read t1]].

This is because the elided vPs in (i) and (ii) are not headless but “headed.”

34 However, there are several other languages, not discussed here, that have been argued to have headless verbal phrase ellipsis. Examples are Chingoni and Kikuyu (Ngonyani and Githinji 2006), Finnish (Holmberg 2001), European and Brazilian Portuguese (Martins 1994, Cyrino and Matos 2002), and Tagalog (Richards 2003). It is not clear to me whether they can be analyzed either as languages allowing multiple specifiers or as languages in which subjects do not move to Spec,TP. If they cannot be analyzed either way, I need to explain why they have headless verbal phrase ellipsis. One possible approach to this problem is to reject the headless verbal phrase ellipsis analysis of the relevant constructions in these languages, arguing that the alleged headless verbal phrase ellipsis should be analyzed in some other way. I leave this issue for future research.

35 I assume that some independent principle or algorithm, such as the one proposed in Chomsky 2008, determines what becomes a label.

References

References
Abels
,
Klaus
.
2003
.
Successive cyclicity, anti-locality, and adposition stranding
.
Doctoral dissertation, University of Connecticut, Storrs
.
Aelbrecht
,
Lobke
, and
Liliane
Haegeman
.
2012
.
VP-ellipsis is not licensed by VP-topicalization
.
Linguistic Inquiry
43
:
591
614
.
Asarina
,
Alevtina
.
2011
.
Case in Uyghur and beyond
.
Doctoral dissertation, MIT, Cambridge, MA
.
Baker
,
Mark
.
1988
.
Incorporation
.
Chicago
:
University of Chicago Press
.
Belletti
,
Adriana
.
2001
.
‘‘Inversion’’ as focalization
. In
Subject inversion in Romance and the theory of Universal Grammar
, ed. by
Aafke C.
Hulk
and
Jean-Yves
Pollock
,
60
90
.
Oxford
:
Oxford University Press
.
Belletti
,
Adriana
.
2004
.
Aspects of the low IP area
. In
The structure of CP and IP
, ed. by
Luigi
Rizzi
,
16
51
.
Oxford
:
Oxford University Press
.
Belletti
,
Adriana
.
2009
.
Structures and strategies
.
New York
:
Routledge
.
Bobaljik
,
Jonathan David
, and
Samuel
Brown
.
1997
.
Interarboreal operations: Head movement and the Extension Requirement
.
Linguistic Inquiry
28
:
345
356
.
Bošković
,
Željko
, and
Howard
Lasnik
.
1999
.
How strict is the cycle?
Linguistic Inquiry
30
:
691
703
.
Chametzky
,
Robert A
.
2000
.
Phrase structure: From GB to minimalism
.
Oxford
:
Blackwell
.
Chomsky
,
Noam
.
1964
.
Current issues in linguistic theory
.
The Hague
:
Mouton
.
Chomsky
,
Noam
.
1986
.
Barriers
.
Cambridge, MA
:
MIT Press
.
Chomsky
,
Noam
.
1993
.
A minimalist program for linguistic theory
. In
The view from Building 20
, ed. by
Kenneth
Hale
and
Samuel Jay
Keyser
,
1
52
.
Cambridge, MA
:
MIT Press
.
Chomsky
,
Noam
.
1995a
.
Bare phrase structure
. In
Government and Binding Theory and the Minimalist Program
, ed. by
Gert
Webelhuth
,
383
439
.
Oxford
:
Blackwell
.
Chomsky
,
Noam
.
1995b
.
Categories and transformations
. In
The Minimalist Program
,
219
394
.
Cambridge, MA
:
MIT Press
.
Chomsky
,
Noam
.
1995c
.
The Minimalist Program
.
Cambridge, MA
:
MIT Press
.
Chomsky
,
Noam
.
2000
.
Minimalist inquiries: The framework
. In
Step by step: Essays on minimalist syntax in honor of Howard Lasnik
, ed. by
Roger
Martin
,
David
Michaels
, and
Juan
Uriagereka
,
89
155
.
Cambridge, MA
:
MIT Press
.
Chomsky
,
Noam
.
2001
.
Derivation by phase
. In
Ken Hale: A life in linguistics
, ed. by
Michael
Kenstowicz
,
1
52
.
Cambridge, MA
:
MIT Press
.
Chomsky
,
Noam
.
2004
.
Beyond explanatory adequacy
. In
The cartography of syntactic structures. Vol. 3, Structures and beyond
, ed. by
Adriana
Belletti
,
104
131
.
Oxford
:
Oxford University Press
.
Chomsky
,
Noam
.
2008
.
On phases
. In
Foundational issues in linguistic theory: Essays in honor of Jean-Roger Vergnaud
, ed. by
Robert
Freidin
,
Carlos P.
Otero
, and
Maria Luisa
Zubizaretta
,
133
166
.
Cambridge, MA
:
MIT Press
.
Cyrino
,
Sonia M. L
., and
Gabriela
Matos
.
2002
.
VP ellipsis in European and Brazilian Portuguese
.
Journal of Portuguese Linguistics
1
:
177
195
.
Donati
,
Caterina
.
2003
.
Merge copy
. In
The interfaces: Deriving and interpreting omitted structures
, ed. by
Kerstin
Schwabe
and
Susanne
Winkler
,
155
175
.
Amsterdam
:
John Benjamins
.
Doron
,
Edit
, and
Caroline
Heycock
.
1999
.
Filling and licensing multiple specifiers
. In
Specifiers
, ed. by
David
Adger
,
Bernadette
Plunkett
,
George
Tsoulas
, and
Susan
Pintzuk
,
69
92
.
Oxford
:
Oxford University Press
.
Epstein
,
Samuel David
,
Erich M.
Groat
,
Ruriko
Kawashima
, and
Hisatsugu
Kitahara
.
1998
.
A derivational approach to syntactic relations
.
Oxford
:
Oxford University Press
.
Fiengo
,
Robert
.
1977
.
On trace theory
.
Linguistic Inquiry
8
:
35
61
.
Fitzpatrick
,
Justin M
.
2006
.
Deletion through movement
.
Natural Language and Linguistic Theory
24
:
399
431
.
Fukui
,
Naoki
.
1997
.
Attract and the A-over-A Principle
. In
UCI working papers in linguistics 3
, ed. by
Luther Chen-Sheng
Liu
and
Kazue
Takeda
,
51
67
.
Irvine
:
University of California, Irvine, Department of Linguistics
.
Fukui
,
Naoki
.
1999
.
An A-over-A perspective on locality
. In
Linguistics: In search of the human mind
, ed. by
Masatake
Muraki
and
Enoch
Iwamoto
,
109
129
.
Tokyo
:
Kaitakusha
.
Fukui
,
Naoki
.
2011
.
Merge and bare phrase structure
. In
The Oxford handbook of linguistic minimalism
, ed. by
Cedric
Boeckx
,
73
95
.
Oxford
:
Oxford University Press
.
Funakoshi
,
Kenshi
.
2011a
.
Object drop in Japanese: Evidence for the VP-ellipsis analysis
. In
The project on language and culture 2010: Theoretical approaches to natural language
,
61
70
.
Osaka
:
Osaka University, Graduate School of Language and Culture
.
Funakoshi
,
Kenshi
.
2011b
.
VP-ellipsis, vP-ellipsis, and vP-fronting
.
Ms., University of Maryland, College Park
.
Futagi
,
Yoko
.
2004
.
Japanese focus particles at the syntax-semantics interface
.
Doctoral dissertation, Rutgers University, New Brunswick, NJ
.
Goldberg
,
Lotus Madelyn
.
2005
.
Verb-stranding VP ellipsis: A cross-linguistic study
.
Doctoral dissertation, McGill University, Montreal
.
Grohmann
,
Kleanthes K
.
2003
.
Prolific domains: On the anti-locality of movement dependencies
.
Amsterdam
:
John Benjamins
.
Hoji
,
Hajime
.
1987
.
Weak crossover and Japanese phrase structure
. In
Issues in Japanese linguistics
,
163
202
.
Dordrecht
:
Foris
.
Holmberg
,
Anders
.
2001
.
The syntax of yes and no in Finnish
.
Studia Linguistica
55
:
140
174
.
Hornstein
,
Norbert
.
2009
.
A theory of syntax: Minimalist operations and Universal Grammar
.
New York
:
Cambridge University Press
.
Hornstein
,
Norbert
, and
Jairo
Nunes
.
2008
.
Some thoughts on adjunction
.
Biolinguistics
2
:
57
86
.
Huang
,
C.-T. James
.
1982
.
Logical relations in Chinese and the theory of grammar
.
Doctoral dissertation, MIT, Cambridge, MA
.
Huang
,
C.-T. James
.
1988
.
Comments on Hasegawa’s paper
. In
Proceedings of Japanese Syntax Workshop: Issues on empty categories
, ed. by
Tawa
Wako
and
Mineharu
Nakayama
,
77
93
.
New London
:
Connecticut College
.
Huang
,
C.-T. James
.
1991
.
Remarks on the status of the null object
. In
Principles and parameters in comparative grammar
, ed. by
Robert
Freidin
,
56
76
.
Cambridge, MA
:
MIT Press
.
Huang
,
C.-T. James
.
1997
.
On lexical structure and syntactic projection
.
Chinese Languages and Linguistics
3
:
45
89
.
Huang
,
Yan
.
2000
.
Anaphora: A cross-linguistic study
.
New York
:
Oxford University Press
.
Jayaseelan
,
K. A
.
2001
.
IP-internal topic and focus phrases
,
Studia Linguistica
55
:
39
75
.
Johnson
,
Kyle
.
2001
.
What VP ellipsis can do, and what it can’t, but not why
. In
The handbook of contemporary syntactic theory
, ed. by
Mark
Baltin
and
Chris
Collins
,
439
479
.
Oxford
:
Blackwell
.
Kasai
,
Hironobu
.
2001
.
Headedness and scope rigidity
. In
Proceedings of ConSOLE X
, ed. by
Marjo
van Koppen
,
Joanna
Sio
, and
Mark
de Vos
,
95
109
.
Leiden
:
Leiden University
.
Kayne
,
Richard
.
1994
.
The antisymmetry of syntax
.
Cambridge, MA
:
MIT Press
.
Kayne
,
Richard
.
2005
.
On parameters and on principles of pronunciation
. In
Organizing grammar: Studies in honor of Henk van Riemsdijk
, ed. by
Hans
Broekhuis
,
Norbert
Corver
,
Riny
Huybregts
,
Ursula
Kleinhenz
, and
Jan
Koster
,
289
299
.
Berlin
:
Mouton de Gruyter
.
Koizumi
,
Masatoshi
.
1995
.
Phrase structure in minimalist syntax
.
Doctoral dissertation, MIT, Cambridge, MA
.
Kumashiro
,
Toshiyuki
, and
Ronald W.
Langacker
.
2003
.
Double-subject and complex-predicate constructions
.
Cognitive Linguistics
14
:
1
45
.
Kuwabara
,
Kazuki
.
1997
.
On the properties of truncated clauses in Japanese
. In
Researching and verifying an advanced theory of human language
, ed. by
Kazuo
Inoue
,
61
84
.
Chiba
:
Kanda University of International Studies
.
Landau
,
Idan
.
2006
.
Chain resolution in Hebrew V( P)-fronting
.
Syntax
9
:
32
66
.
Larson
,
Richard
.
1988
.
On the double object construction
.
Linguistic Inquiry
19
:
335
391
.
Lasnik
,
Howard
.
1997
.
A gap in an ellipsis paradigm: Some theoretical implications
.
Linguistic Analysis
27
:
166
185
.
Lasnik
,
Howard
.
1999
.
Pseudogapping puzzles
. In
Fragments: Studies in ellipsis and gapping
, ed. by
Shalom
Lappin
and
Elabbas
Benmamoun
,
141
174
.
Oxford
:
Oxford University Press
.
Lasnik
,
Howard
.
2001
.
When can you save a structure by destroying it?
In
NELS 31
, ed. by
Minjoo
Kim
and
Uri
Strauss
,
301
320
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Lasnik
,
Howard
.
2006
.
Conceptions of the cycle
. In
Wh-movement: Moving on
, ed. by
Lisa Lai-Shen
Cheng
and
Norbert
Corver
,
197
216
.
Cambridge, MA
:
MIT Press
.
Lasnik
,
Howard
, and
Mamoru
Saito
.
1992
.
Move α
.
Cambridge, MA
:
MIT Press
.
Lobeck
,
Anne
.
1995
.
Ellipsis: Functional heads, licensing, and identification
.
Oxford
:
Oxford University Press
.
Martins
,
Ana-Maria
.
1994
.
Enclisis, VP-deletion and the nature of sigma
.
Probus
6
:
173
206
.
May
,
Robert
.
1977
.
The grammar of quantification
.
Doctoral dissertation, MIT, Cambridge, MA
.
May
,
Robert
.
1989
.
Bound variable anaphora
. In
Mental representations
, ed. by
Ruth
Kempson
,
85
104
.
Cambridge
:
Cambridge University Press
.
McCloskey
,
James
.
1991
.
Clause structure, ellipsis and proper government in Irish
.
Lingua
85
:
259
302
.
McCloskey
,
James
.
1996
.
On the scope of verb movement in Irish
.
Natural Language and Linguistic Theory
14
:
47
104
.
Merchant
,
Jason
.
2001
.
The syntax of silence: Sluicing, islands, and the theory of ellipsis
.
Oxford
:
Oxford University Press
.
Miyoshi
,
Nobuhiro
, and
Koji
Hoshi
.
2007
.
A derivational approach to dake: A preliminary sketch of association with focus
.
Language, Culture and Communication
39
:
25
48
.
Müller
,
Gereon
.
1996
.
A constraint on remnant movement
.
Natural Language and Linguistic Theory
14
:
355
407
.
Munro
,
Pamela
.
1999
.
Chickasaw subjecthood
. In
External possession
, ed. by
L. Doris
Payne
and
Immanuel
Barshi
,
251
289
.
Amsterdam
:
John Benjamins
.
Ngonyani
,
Deo
, and
Peter
Githinji
.
2006
.
The asymmetric nature of Bantu applicative constructions
.
Lingua
116
:
31
63
.
Nunes
,
Jairo
, and
Juan
Uriagereka
.
2000
.
Cyclicity and extraction domains
.
Syntax
3
:
20
43
.
Ochi
,
Masao
.
1999
.
Some consequences of Attract F
.
Lingua
109
:
81
107
.
Oku
,
Satoshi
.
1998
.
A theory of selection and reconstruction in the minimalist perspective
.
Doctoral dissertation, University of Connecticut, Storrs
.
Otani
,
Kazuyo
, and
John
Whitman
.
1991
.
V-raising and VP-ellipsis
.
Linguistic Inquiry
22
:
345
358
.
Pesetsky
,
David
.
2007
.
Undermerge . . . and the secret genitive inside every Russian noun
.
Handout of a talk at Formal Approaches to Slavic Linguistics 16
.
Potsdam
,
Eric
.
1997
.
English verbal morphology and VP ellipsis
. In
NELS 27
, ed. by
Kiyomi
Kusumoto
,
353
368
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Richards
,
Norvin
.
1997
.
What moves where when in which language
?
Doctoral dissertation, MIT, Cambridge, MA
.
Richards
,
Norvin
.
1999
.
Featural cyclicity and the ordering of multiple specifiers
. In
Working minimalism
, ed. by
Samuel David
Epstein
and
Norbert
Hornstein
,
127
158
.
Cambridge, MA
:
MIT Press
.
Richards
,
Norvin
.
2003
.
Why there is an EPP
.
Gengo Kenkyu
123
:
221
256
.
Richards
,
Norvin
.
2004
.
Against bans on lowering
.
Linguistic Inquiry
35
:
453
463
.
Rizzi
,
Luigi
.
1990
.
Relativized Minimality
.
Cambridge, MA
:
MIT Press
.
Rizzi
,
Luigi
.
2005
.
Phase theory and the privilege of the root
. In
Organizing grammar: Studies in honor of Henk van Riemsdijk
, ed. by
Hans
Broekhuis
,
Norbert
Corver
,
Riny
Huybregts
,
Ursula
Kleinhenz
, and
Jan
Koster
,
529
537
.
Berlin
:
Mouton de Gruyter
.
Rizzi
,
Luigi
, and
Ian
Roberts
.
1989
.
Complex inversion in French
.
Probus
1
:
1
30
.
Roberts
,
Ian
.
1991
.
Excorporation and minimality
.
Linguistic Inquiry
22
:
209
218
.
Roberts
,
Ian
.
1994
.
Two types of head movement in Romance
. In
Verb movement
, ed. by
David
Lightfoot
and
Norbert
Hornstein
,
207
242
.
Cambridge
:
Cambridge University Press
.
Roberts
,
Ian
.
1998
.
Have/Be raising, Move F, and Procrastinate
.
Linguistic Inquiry
29
:
113
125
.
Sailor
,
Craig
.
2010
.
VP ellipsis in tag questions: A typological approach
.
Paper presented at the 46th meeting of the Chicago Linguistic Society. To appear in the proceedings. Available at http://cwsailor.bol.ucla.edu/papers/CLS46_tagQs.pdf
.
Saito
,
Mamoru
.
1989
.
Scrambling as semantically vacuous A′-movement
. In
Alternative conceptions of phrase structure
, ed. by
Mark R.
Baltin
and
Anthony S.
Kroch
,
182
200
.
Chicago
:
University of Chicago Press
.
Saito
,
Mamoru
.
2004
.
Japanese scrambling in a comparative perspective
. In
Peripheries: Syntactic edges and their effects
, ed. by
David
Adger
,
Cécile
De Cat
, and
George
Tsoulas
,
143
163
.
Dordrecht
:
Kluwer
.
Saito
,
Mamoru
.
2007
.
Notes on East Asian argument ellipsis
.
Language Research
43
:
203
227
.
Shoji
,
Atsuko
.
1986
.
Dake and sika in Japanese: Syntax, semantics and pragmatics
.
Doctoral dissertation, Cornell University, Ithaca, NY
.
Simpson
,
Andrew
.
2005
.
Pro drop patterns and analyticity
.
Handout distributed at course
LSA.222, Syntactic Analyticity: The Other End of the Parameter
.
LSA Institute 2005, Cambridge, MA
. .
Speas
,
Margaret
.
1990
.
Phrase structure in natural language
.
Dordrecht
:
Kluwer
.
Szczegielniak
,
Adam
.
2005
.
VP ellipsis and topicalization
. In
NELS 35
, ed. by
Leah
Bateman
and
Cherlon
Ussery
,
603
614
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Takahashi
,
Daiko
.
2008a
.
Noun phrase ellipsis
. In
The Oxford handbook of Japanese linguistics
, ed. by
Shigeru
Miyagawa
and
Mamoru
Saito
,
394
422
.
New York
:
Oxford University Press
.
Takahashi
,
Daiko
.
2008b
.
Quantificational null objects and argument ellipsis
.
Linguistic Inquiry
39
:
307
326
.
Takano
,
Yuji
.
1994
.
Unbound traces and indeterminacy of derivation
. In
Current topics in English and Japanese
, ed. by
Masaru
Nakamura
,
229
253
.
Tokyo
:
Hituzi Syobo
.
Takano
,
Yuji
.
2000
.
Illicit remnant movement: An argument for feature-driven movement
.
Linguistic Inquiry
31
:
141
156
.
Takano
,
Yuji
.
2002
.
Surprising constituents
.
Journal of East Asian Linguistics
11
:
243
301
.
Takano
,
Yuji
.
2007
.
Making rightward scrambling possible
.
Kinjo Gakuin Daigaku Ronshu, Studies in Humanities
3
:
17
58
.
Tanaka
,
Hidekazu
.
2008
.
Clausal complement ellipsis
.
Ms., University of York
.
Tanaka
,
Hiroyuki
.
2004
.
On the minimality of Generalized Pied-Piping
.
Lingua
114
:
911
933
.
Travis
,
Lisa
.
1984
.
Parameters and effects of word order variation
.
Doctoral dissertation, MIT, Cambridge, MA
.
Ura
,
Hiroyuki
.
1994
.
Varieties of raising and the feature-based bare phrase structure theory
.
MIT Occasional Papers in Lingustics 7. Cambridge, MA: MIT, MIT Working Papers in Linguistics
.
Ura
,
Hiroyuki
.
2000
.
Checking theory and grammatical functions in Universal Grammar
.
Oxford
:
Oxford University Press
.
Ura
,
Hiroyuki
.
2001
.
Local economy and Generalized Pied-Piping
.
The Linguistic Review
18
:
169
191
.
Uriagereka
,
Juan
.
2002
.
Pure adjuncts
.
Ms., University of Maryland, College Park
.
Uriagereka
,
Juan
.
2010
.
To escape an island: Submerge and resurface
.
Handout of a talk at Formal Approaches to Slavic Linguistics 19
.
Wurmbrand
,
Susi
.
2004
.
No TP-fronting meets Nearly Headless Nick
. .