Abstract

To account for several asymmetries between A- and Ā-movement, Takahashi and Hulsey (2009) generalize the late merger option (Lebeaux 1988, Chomsky 1995) as wholesale late merger (WLM). In particular, allowing an NP to merge with a head D as late as (but no later than) its Case position explains why Ā but not A-movement displays Principle C reconstruction effects. In this article, I claim that WLM is also responsible for pervasive asymmetries within the class of Ā-extractions. The evidence comes from restrictions on English preposition stranding. I document a correlation between a preposition’s complementation properties and its ability to be stranded: prepositions that disallow pronominal complements can only be stranded by a subset of Ā-extractions. I argue that the extractions allowing pronoun-rejecting prepositions to be stranded disallow WLM, while those that disallow the stranding allow (and require) WLM.

1 Introduction

It is well-known that A- and Ā-movement differ in the reconstruction properties that they exhibit. For example, while both types of movement allow reconstructed bound variable readings, Principle C reconstruction effects are usually obligatory for Ā-movement but absent for A-movement.1 As noted by many authors (e.g., Sauerland 1998, Fox 1999), this pattern poses a challenge for existing theories of movement: while it appears that Ā-movement must leave behind a contentful copy, A-movement can act as if it leaves behind a contentless trace.

Takahashi (2006) and Takahashi and Hulsey (2009) argue that this asymmetry arises as a consequence of the possibility for limited countercyclic merger, known as late merger (Lebeaux 1988, Chomsky 1993). Building on work by Fox (2002), they propose that late merger is possible whenever an output representation can be interpreted by the semantics (as is the case for the late-merged relative clauses posited by Lebeaux and Chomsky). It is therefore possible for an NP to merge countercyclically with its head D, as the result is semantically interpretable (see Takahashi and Hulsey 2009 for details). They name this optional operation wholesale late merger, or WLM.

An additional proposal, that WLM is constrained by the Case Filter (applying to NP as well as DP), explains the differences between A and Ā reconstruction properties. Since A-movement is movement to a Case position, an NP can merge with its head D in the derived position and bleed Principle C. Because Ā-movement is generally movement to a non-Case position, this option is not available. In short, independent differences between A- and Ā-movement control the applicability of WLM, which in turn leads to distinct patterns of reconstruction effects.

Since WLM offers a principled reason for the reconstruction contrast between A- and Ā-movement, it can be seen as an independent argument for the somewhat controversial proposal that late merger is possible in the first place (see Chomsky 2004). In this article, I strengthen this argument by offering an independent set of arguments for WLM itself (see also Bhatt and Pancheva 2004, 2007, Nikolaeva 2014). I argue that WLM not only explains the well-known contrasts between A- and Ā-extractions investigated by Takahashi (2006) and Takahashi and Hulsey (2009), but also is responsible for lesser-known but pervasive differences within the class of Ā-extractions: in particular, a set of asymmetries in English preposition stranding (P-stranding). In section 2, I document a correlation between two seemingly independent facts about prepositions: their ability to take a pronominal complement (see also Kuroda 1964, 1968), and their ability to be stranded (see Ross 1967:119; also see Postal 1998, for closely related work focusing on similar constraints on movement out of other antipronominal contexts). A preposition that cannot take a pronominal complement, like temporal in in (1), allows us to distinguish two types of Ā-extraction: the type allowing a pronoun-rejecting preposition to be stranded (e.g., wh-movement, restrictive relatives; (2a–b)), and the type forbidding a pronoun-rejecting preposition from being stranded (e.g., topicalization, tough-movement; (2c–d)).

(1) Temporal in is pronoun-rejecting

  • a.

    *I went swimming in December, and John went swimming in it, too.

  • b.

    *I will be arriving in a few days, and John will be arriving in them, too.

(2)

  • Stranding temporal in: Restricted2

  • graphic

By contrast, a preposition that takes a pronominal complement, like locative in (3), can be stranded by any type of Ā-extraction (4).

(3) Locative in is pronoun-accepting

  • a.

    Michelle’s cat hid in the cardboard box, and my cat hid in it, too.

  • b.

    I stored my cereal in the pantry, and Chris stored his cereal in it, too.

(4)

graphic

I propose that these two types of Ā-extraction differ according to whether or not WLM applies. An extraction like wh-movement, which can strand a pronoun-rejecting preposition, leaves behind a fully constructed copy. An extraction like topicalization, which cannot strand a pronoun-rejecting preposition, leaves behind only a determiner; the rest of the DP merges after movement. In section 7, I attribute these differences to interactions among the Case Filter, a violable constraint favoring WLM, and restrictions on the timing of countercyclic merger (see also Sauerland 1998: sec. 2.2).

Assuming that pronouns are bare determiners (e.g., Postal 1966, Abney 1987), I argue that the link between a preposition’s ability to take a pronominal complement and its ability to be stranded is structural. I propose that a pronoun-rejecting preposition requires its complement to contain a certain kind of DP: temporal in, for example, requires its complement DP to contain an NP denoting an interval of time (section 3). Because a pronominal DP does not contain an NP, it is an unacceptable complement for a preposition like temporal in. The base position of extractions like topicalization likewise lacks an NP, explaining the inability of these extractions to strand pronoun-rejecting prepositions.

I provide further support for the proposed analysis by showing that it correctly predicts not only contrasts among extractions, but also contrasts internal to individual extractions. For example, P-stranding with wh-pronouns, but not full wh-phrases, is sensitive to whether or not a given preposition is pronoun-rejecting (section 8). Contrasts of this kind show that the (in)ability to strand pronoun-rejecting prepositions should not be linked to intrinsic properties of individual extractions (cf. Postal 1998): all that matters is the structural configuration of the gap site.

2 P-Stranding Asymmetries

In this section, I focus on P-stranding asymmetries in temporal (section 2.1) and locative (section 2.2) PPs. I show that if a given preposition cannot accept a pronoun as its complement, it can be stranded by only a subset of Ā-extractions (see Postal 1998 for similar phenomena).

2.1 Temporals

It is possible to divide temporal DPs into two classes. There are interval DPs, referring to points or spans in time (e.g., Monday, the last five hours). There are also event DPs, referring to events that occupy certain portions of time (e.g., John’s party, Sue’s talk). A basic difference between these two types of temporal DP is that while an interval DP is defined by its location and extent along a timeline, an event DP is defined by other properties. (5) gives some further examples of interval and event DPs.

(5) Interval vs. event DPs

  • a.

    Intervals: Monday, 5:00, Christmas, Chris’s youth, Mary’s birthday, . . .

  • b.

    Events: John’s party, Sue’s talk, Mary’s progress meeting, Christmas dinner, . . .

In temporal PPs, interval and event DPs differ in a fundamental way: an event but not an interval complement of a preposition can be pronominalized with it, as in (6)–(7).

(6) Event DP complements:it

  • a.

    I left after John’s party, and Mary left after it, too.

  • b.

    I left before Christmas dinner, and John left before it, too.

(7) Interval DP complements: *it3

  • a.

    *I left after 5:00, and Mary left after it, too.

  • b.

    *My family eats lamb on Easter, and John’s family eats lamb on it, too.4

An appealing but incorrect analysis of (6)–(7) could claim that an interval DP, unlike an event DP, has some property that renders it unable to be pronominalized. If this were the case, the split between the event and interval complements in (6)–(7) would not be surprising: if an interval complement cannot be pronominalized in general, then its behavior in a PP does not require an independent explanation. Note however that June, an interval, can be pronominalized as a subject (8a) or as the complement of spend (8b),5 but not as the complement of in (8c).

(8) Interval DPs can be pronominalized

  • a.

    I spent June at the pool. It is my favorite month.

  • b.

    I spent June at the pool, but John spent it in his office.

  • c.

    *John visited his family in June, and Mary visited her family in it, too.

The contrast between (8a–b) and (8c) shows that the potential for a given DP to be pronominalized is, at least in part, dependent on the head that selects for it. An interval can be pronominalized as the complement of spend, for example, because spend accepts a pronominal complement. An interval cannot be pronominalized as the complement of temporal in because temporal in does not accept a pronominal complement.6

The pronominalization contrasts in (6)–(7), then, do not point to a difference between event and interval DPs per se; instead, they localize the difference to properties of the prepositions that select for them. A preposition selecting for an event DP is a pronoun-accepting preposition (a PA): it can accept either a pronoun or a lexical noun as its complement. A preposition selecting for an interval DP, by contrast, is a pronoun-rejecting preposition (a PR): it accepts a lexical noun, but not a pronoun, as its complement. Examples of pronoun-accepting and pronoun-rejecting temporal prepositions are given in (9). Note that many prepositions in English, like before and after in (9), have PA and PR uses.

(9)

graphic

PAs are the group of prepositions that can be stranded under any type of Ā-extraction. I illustrate with the PAs before (10) and after (11).

(10)

graphic

(11)

graphic

PRs are the group of prepositions that can be stranded in only a subset of these extractions. I illustrate with the PRs on (12) and in (13).

(12)

  • Stranding on (a PR): Severely constrained7

  • graphic

(13)

graphic

The contrast between (10)–(11) and (12)–(13) documents a correlation between the ability of a given preposition to take a pronominal complement and the ability of that preposition to be stranded. Some Ā-extractions allow a PR to be stranded, but others do not. Following Postal’s (1998) terminology for extraction types, based on observations of other antipronominal contexts,8 I refer to an extraction allowing a PR to be stranded as an A-type extraction (not to be confused with A-movement). An extraction forbidding a PR from being stranded is a B-type extraction. The full partition is given in (14), and examples are provided in (15)–(16). Throughout the rest of the article, I use wh-movement and restrictive relatives as representative examples of A-types, and topicalization and tough-movement as examples of B-types.

(14)

graphic

(15)

graphic

(16)

graphic

The partition in (14) makes explicit an implicational generalization. A PA can be stranded under A-type and B-type extractions, but a PR can only be stranded under A-type extractions. Therefore, any preposition that can be stranded under a B-type extraction must be able to be stranded under all A-type extractions, but not vice versa.

(17)

  • Generalizing over extractions

  • If a given preposition can be stranded by a B-type extraction, that preposition can also be stranded by all A-type extractions.

A counterexample to the generalization in (17) would be a preposition that could be stranded by a B-type extraction (e.g., topicalization) but not an A-type extraction (e.g., wh-movement). To the best of my knowledge, such a preposition does not exist.

2.2 Locatives

Like temporal DPs, locative DPs can be divided into two classes. There are location DPs, which refer to points or regions of space (e.g., the second balcony, the fourth floor). There are also entity DPs, referring to physical entities that occupy certain portions of space (e.g., the box, the car). Like the difference between an interval and an event (temporal) DP, the difference between a location and an entity (locative) DP is semantic. A location DP like the fourth floor is defined by its spatial coordinates, while an entity DP like the box is defined by other properties. Further examples of location and entity DPs are shown in (18).

(18) Location vs. entity DPs

  • a.

    Locations: the ground floor, the fourth floor, 10,000 feet, the sky, . . .

  • b.

    Entities: the box, the forest, the hut, the television, the car, . . .

The distinction between an entity and a location DP is seen clearly in locative PPs. A preposition that selects for an entity complement is pronoun-accepting (19), while a preposition that selects for a location complement is pronoun-rejecting (20).

(19) Entity DP complement:it

  • a.

    I ate dinner on the wooden table, and John ate dinner on it, too.

  • b.

    I climbed to the summit of the mountain, and John climbed to it, too.

(20) Location DP complement: *it9

  • a.

    *I ate dinner on the fourth floor, and John ate dinner on it, too.

  • b.

    *My airplane climbed to 10,000 feet, and John’s airplane climbed to them, too.

Note that the sentences in (20) are only ungrammatical under the location readings of the fourth floor and 10,000 feet. (20a) is grammatical if John ate dinner on the surface of some floor, which happens to be fourth in a series of floors; (20b) is grammatical if 10,000 feet is a visible destination that can be climbed to, like the top of a mountain. In other words, (20a–b) are grammatical if the DP is treated as an entity. This is consistent with the generalization that an entity DP can be pronominalized as a preposition’s complement, but a location DP cannot be.

The stranding facts parallel the observations concerning temporals in section 2.1. A preposition selecting for an entity DP can be stranded under any type of Ā-extraction; examples are given with in (21) and on (22), in their PA uses.

(21)

graphic

(22)

graphic
graphic

A preposition selecting for a location DP, however, can only be stranded under A-type extractions. I illustrate with to (23) and on (24), in their PR uses.

(23)

graphic

(24)

graphic

Place names, a type of locative DP, are ambiguous between locations and entities. A place name like Boston can be interpreted in two ways, as a set of coordinates (either geographical, or a point on a cognitive map) or as a physical entity, as Chomsky (1999:42) notes: ‘‘if London is reduced to dust, it...can be re-built elsewhere and still be the same city.’’ We can say that London was reduced to dust because place names, in one sense, refer to a collection of entities that occupy specific points in space. We can also say that London can be rebuilt elsewhere with different buildings because, in another sense, place names refer to a set of abstract coordinates, not necessarily linked to their physical contents.

To the extent that a place name can be pronominalized as the complement of a preposition, the entity reading is preferred (25).

(25)

graphic

The first conjunct in (25b) has a location reading under which I sat on the banks of the Charles River, near the water, and ate my lunch. This reading is absent in the second conjunct: the only licit interpretation here is one under which John sat on the surface of the Charles River and ate his lunch. In other words, the sentence is only licit if the Charles River is treated as an entity.

In (26), we see that the location reading of on the Charles River is preserved when on is stranded by A-type extractions (26a–b), but not when it is stranded by B-type extractions (26c–d).

(26)

graphic

The lack of an available location reading in (26c–d) is consistent with the generalization that a preposition selecting a location DP cannot be stranded by B-type extractions.10

2.3 Summary

Sections 2.1 and 2.2 have demonstrated a link between (a) whether or not a given preposition can accept a pronoun as its complement and (b) whether or not that preposition can be stranded by B-type extractions. The relevant generalizations are as follows. If a preposition is pronoun-accepting, then it can be stranded by both A-type and B-type extractions.

(27) If [preposition pronoun], then A-type and B-type.

If a preposition is pronoun-rejecting, then it can be stranded under A-type extractions only.

(28) If *[preposition pronoun], then A-type but *B-type.

Although the empirical focus of this article is on temporal and locative PPs, the generalizations in (27)–(28) hold more widely. In (29), for example, instrumental with is a PA and can be stranded under both A-type and B-type extractions. In (30), by contrast, circumstantial under is a PR and can be stranded under A-type extractions only.

(29)

graphic

(30)

graphic

The remainder of the article seeks to explain why some prepositions are PRs, and why this should have anything at all to do with their ability to be stranded.

3 On Antipronominal Contexts

In section 2, I established that temporal and locative DP complements of a preposition can be divided into two classes, according to their ability to be pronominalized with it (or them). As the complement of a temporal preposition, an event but not an interval DP can be pronominalized. As the complement of a locative preposition, an entity but not a location DP can be pronominalized.

Interval and location DPs—the types of DPs that cannot be pronominalized as prepositional complements—are coordinate-denoting DPs. A coordinate-denoting DP is a predicate ranging over either spatial or temporal dimensions; such a DP is defined by the amount of time or space that it occupies. Event and entity DPs—the types of DPs that can be pronominalized as prepositional complements—are concrete DPs, in the sense that they are predicates ranging over entities that occupy certain portions of temporal or spatial dimensions. The particular type of space that a concrete DP occupies determines whether it is treated as a locative or a temporal, but it is defined by properties other than its coordinates. The event DP John’s party, in (31), illustrates: John’s party takes up a portion of Monday, but its length does not define it.

(31)

graphic

Prepositions that select for coordinate-denoting DPs are PRs: they can take DPs containing lexical nouns (i.e., the box, John’s party), but not pronouns, as their complements. Prepositions that select for concrete DPs are PAs: they can take either pronouns or DPs containing lexical nouns as their complements.

What does it mean, though, for a preposition to be pronoun-rejecting? What is the relevant difference between a pronoun and a coordinate-denoting DP, such that a preposition can discriminate between the two? One potentially relevant difference is that pronouns and lexical nouns are not members of the same syntactic category: pronouns are determiners that do not have an NP complement (Postal 1966, Abney 1987, Uriagereka 1995,though cf. Elbourne 2005),while lexical nouns are Ns. One argument for this difference in category membership comes from the fact that pronouns and determiners are in complementary distribution (32a), but lexical nouns and determiners are not (32b–c). (Examples (32a–c) are from Abney 1987:281.)

(32) Pronouns and determiners are in complementary distribution

  • a.

    *The she that I talked to was nice.

  • b.

    The Mary that I talked to was nice.

  • c.

    The woman that I talked to was nice.

The fact that determiners are in complementary distribution with pronouns (but not lexical nouns) suggests that pronouns (but not lexical nouns) have the same structural position as determiners. Further evidence for this comes from the fact that pronouns can act as overt determiners under certain circumstances: for example, we religious ones (see Postal 1966 11). If the analysis of pronouns as determiners is on the right track, then a pronoun and a phrasal DP are structurally distinct: a phrasal DP like the box contains an NP, but a pronoun like it does not. Returning to the properties of PRs, the basic proposal here is that a PR is a preposition requiring its complement DP to contain something more than just a bare D. In essence, it appears that the complement of a PR must contain an NP. But when we look more closely at the selectional restrictions of PRs like temporal in and on, it becomes clear that it isn’t just any sort of NP that a PR requires: it’s a certain semantic type of NP. Note, for example, that temporal in (33) and on (34) require interval DP complements and forbid event DP complements.

(33)

graphic

(34)

graphic

The only situation in which (33b) and (34b) are marginally acceptable is one in which the speaker is using John’s party and Sue’s talk to denote units of time—in other words, if the DPs are treated as intervals. Thus, we cannot simply say that a PR is a preposition requiring its complement to contain an NP, because sentences like (33b) and (34b), where a temporal PR takes an event complement, are ungrammatical.

Understanding why the complements of temporal PRs are restricted to intervals requires us to find a property of intervals that both events and pronouns lack. Following Kayne (2005b), I propose that an interval DP contains a nominal category belonging to a particular semantic cate-gory, abbreviated here as TIME.TIME can be either null (as in December the 25th (DAY)) or pronounced (as in the MONTHof June, Avery’s wedding DAY, etc.), and this TIME element is what lends the DP its intervalic denotation. For evidence that TIME resides in N, consider the alternation in (35).12

(35) TIME is a noun

  • a.

    the 25th (DAY) of December

  • b.

    December 25th (DAY)

In (35a), we see that the null DAY acts as a noun: it has a determiner, an ordinal modifier, and a PP modifier. (35b) is derivationally related to (35a): December has moved into a position structurally higher than D and lost its preposition (perhaps because December is now in Spec,DP and no longer needs to be Case-licensed by of). Note that even though pronouncing the TIME element in (35b) is marginally acceptable at best, the presence of the ordinal modifier 25th strongly suggests that something occupies N.

We can say something similar for location DPs. Following Kayne (2005a), I propose that a location DP contains the noun PLACE. Like TIME,PLACE can be either null (as in Boston (CITY)) or overt (as in the Charles RIVER, New York CITY). PLACE lends a locative DP a location denotation by signifying that the DP occupies a span of coordinates along a spatial plane. Some evidence that PLACE occupies N comes from the alternation in (36), structurally identical to (35).

(36) PLACE is a noun

  • a.

    the CITY of New York

  • b.

    New York CITY13

Returning now to P-stranding, we can say that an interval-selecting temporal PR is a preposition that requires its complement to contain TIME, and a location-selecting locative PR is a preposi-tion that requires its complement to contain PLACE. Presumably this is a semantic requirement, as there is no reason to believe that an interval DP that contains TIME is syntactically distinct from an event DP.14

In sum, we can now say that a preposition requiring its complement to contain either TIME or PLACE is a coordinate-selecting preposition. When a coordinate-selecting preposition takes a pronominal complement, its need for a TIME or PLACE element is not met. This is because a pronominal DP does not contain an NP and therefore cannot possibly contain TIME or PLACE.15

4 Linking Complementation and Stranding

We can now recharacterize the link between complementation and stranding in light of the previous section. A temporal or locative PR, whose semantics require its complement to contain TIME or PLACE, can be stranded only under A-type extractions. A temporal or locative PA, which does not require its complement to contain TIME or PLACE, can be stranded under both A-type and B-type extractions. The question, then, is this: how can we distinguish the A-type extractions from the B-type extractions in a way that allows us to capitalize on the specific needs of PRs?

I propose that the A-type and B-type extractions differ in their sensitivity to the selectional restrictions of stranded prepositions because they leave behind copies of different sizes. A-type extractions (e.g., wh-movement, restrictive relatives) leave behind fully constructed copies (see Chomsky 1993, 1995). A-type extractions can strand PRs because the copy in the base position can contain an NP. If the copy in the base position can contain an NP, it can contain TIME or PLACE, and the selectional restrictions of PRs can be satisfied (37).

(37) A-type extractions leave full copies

  • a.

    Which day of the month did you eat lamb onR [which day of the month]?

  • b.

    Which city are we eating lunch inR [which city]?

B-type extractions, by contrast, leave behind something smaller than a full DP—crucially, something that does not contain an NP. Because TIME and PLACE reside within the NP, the selectional restrictions of PRs cannot be satisfied (38).

(38) B-type extractions do not leave NPs

  • a.

    *No holiday will I ever eat lamb onRt.16

  • b.

    *The fourth floor is fun to eat dinner onRt.

The proposed link between complementation and stranding, then, is structural. The reasons why a PR cannot take a pronoun as its complement, or be stranded under B-type extractions, are one and the same. Neither of these configurations allows the preposition’s complement to contain an N, where TIME and PLACE reside.17

To make sense of this idea, we need to allow for the possibility that some Ā-extractions leave behind full copies, while others do not. This is where the theory of wholesale late merger (WLM; Takahashi 2006, Takahashi and Hulsey 2009) comes in. In short, the idea behind WLM is that certain extractions (i.e., A-movement constructions) permit countercyclic merger of an NP to its head D, while other extractions (e.g., wh-movement) do not. I first provide some necessary background on WLM (section 5) and then extend the theory to account for the differences among Ā-extractions discussed here (sections 6 and 7).

5 Wholesale Late Merger

As noted in section 1, WLM addresses a well-known problem for the copy theory of movement (see Chomsky 1993, 1995): extractions appear to differ in whether or not they leave a contentful copy behind. Takahashi (2006) and Takahashi and Hulsey (2009) (hereafter T±H) focus on the difference between A- and Ā-movement: this basic distinction will be the focus of the discussion in this section, though later there will be reason to look more closely at differences within the class of Ā-extractions.

The presence of reconstructed bound variable readings in both A- and Ā-movement suggests that both types of movement can leave behind a full copy. In examples (39a–b), the QP every professor binds the variable his, suggesting that a copy of the variable is c-commanded by the QP at LF. (Example (39a) is from T&H 2009:391, (7a); and (39b) is adapted from T&H 2009: 390, (2a).)

(39)

graphic

The data in (39) show that leaving a full copy behind is at least optional for A- and Ā-movement. But is it obligatory? For Ā-movement, Principle C reconstruction effects suggest that the answer is yes. As shown in (40a), sentences where an R-expression in the moving NP is c-commanded by a coindexed pronoun at LF are ungrammatical. For A-movement, however, the answer appears to be no: Principle C reconstruction effects are absent (40b). (Examples (40a–b) are from T&H 2009:391 and Lebeaux 1998:23, respectively.)

(40)

graphic

The A-movement facts pose a problem for the copy theory of movement. While it appears that Ā-extractions must always leave a copy behind, A-extractions do so only optionally. The presence of reconstructed bound variable readings for A-extractions suggests that leaving a full copy in the base position is a possibility, but the absence of Principle C effects suggests that leaving a full copy behind is not obligatory.

To solve this problem, T±H propose that it is possible for an NP to countercyclically merge with its head D (see (41)). Their proposal builds on work by Fox (2002), who proposes that late merger is possible whenever the output representation is semantically interpretable. Fox’s (2002) proposal differs from Lebeaux’s (1988, 1998) theory of late merger because it does not limit the operation to adjunct phrases. It allows an NP to merge late with its head D, because the result is interpretable (see T±H for details).

(41)

graphic

In this diagram, the bare D is externally merged in Position D. It then moves to Position C, and again to Position B. Position B is the site of WLM: here, the NP merges with its head D. The entire DP then moves to Position A, where it is pronounced.

(41) shows how WLM allows for A-extractions to bleed Principle C. If an NP containing an R-expression does not merge with D until D is outside the c-command domain of a coindexed nominal, Principle C effects will be absent.

The simple mechanism of WLM, however, does not yet allow us to explain why the reconstruction properties of A- and Ā-movement should differ. Why, for example, is it possible for A-movement to bleed Principle C through WLM, while this option is not available to Ā-movement? In other words: if we grant that WLM is an option available to the grammar, what constrains it?

T±H’s answer to this question is that WLM is constrained by the Case Filter. To move toward an explanation of the differences between A- and Ā-movement, T±H advocate an approach in which ‘‘DPs as a whole demand Case because both determiners and nouns, which constitute DPs, must receive Case’’ (T&H 2009:401).18 Thus, WLM is blocked when an NP would merge outside the domain of its Case assigner.

The appeal to Case allows T±H to capture the difference between A- and Ā-movement. A-movement, which is movement from a non-Case to a Case position, permits the application of WLM (though cf. Nikolaeva 2014:sec. 5 for a discussion of reconstruction effects in Russian scrambling, a form of A-movement to a non-Case position). This means that a restrictor NP can merge outside of the c-command domain of a coindexed nominal. In (42), Principle C is bled because there is no R-expression in the base position. (Example (38b) is from T&H 2009:402.)

(42) WLM is permitted in A-movement

  • a.

    The corner of Johni’s room seems to himi [the] to be very dusty.

  • b.

    Every argument that Johni is a genius seems to himi [every] to be flawless.

Ā-movement, by contrast, is movement from a Case to a non-Case position. T±H propose that WLM is blocked in this context because the NP must merge in the base position to get Case. In the examples in (43), Principle C effects are present because he c-commands John at LF.

(43) WLM is blocked in Ā-movement

  • a.

    *Which corner of Johni’s room was hei sitting in [which corner ...]?

  • b.

    *Which proof that Johni is a genius did hei believe [which proof ...]?

In sum, the reconstruction properties of A- and Ā-movement differ because the two types of movement place differing restrictions on the application of WLM. A-movement permits WLM because an NP can merge outside of its base position and still receive Case. Ā-movement blocks WLM because the NP must merge in its base position to receive Case. What is important is that, depending on the extraction in question, the size of the copy that a DP leaves behind can differ.

6 Relating WLM to P-Stranding

Although the only Ā-extraction that T±H consider is wh-movement, they implicitly assume that all varieties of Ā-movement exhibit Principle C effects. Looking more broadly across the variety of Ā-extractions discussed in relation to P-stranding, we see that this assumption is mostly well-founded. A representative sample of Ā-extractions displaying Principle C effects is given in (44).

(44)

graphic

Parasitic gaps (45a), tough-movement (45b), and gapped degree phrases (45c) behave differently, however: as with A-extractions, Principle C effects are absent.

(45)

graphic

Crossing the presence/absence of Principle C effects with a given extraction’s ability to strand a PR yields the result in table 1. We can now distinguish among three different types of extractions. The A-type extractions (e.g., wh-movement) are extractions that exhibit Principle C effects and permit stranding of PRs. The B1-type extractions (e.g., topicalization) exhibit Principle C effects but ban stranding of PRs. The B2-type extractions (e.g., tough-movement) bleed Principle C and ban stranding of PRs.

Table 1

Principle C effects vs. stranding

Principle C effects: PresentPrinciple C effects: Absent
[PRtA-type extractions wh-movement restrictive relatives infinitival relatives free relatives  
*[PRtB1-type extractions appositive relatives topicalization negative inversion  B2-type extractions tough-movement gapped degree phrases parasitic gap formation 
Principle C effects: PresentPrinciple C effects: Absent
[PRtA-type extractions wh-movement restrictive relatives infinitival relatives free relatives  
*[PRtB1-type extractions appositive relatives topicalization negative inversion  B2-type extractions tough-movement gapped degree phrases parasitic gap formation 

If we allow for the possibility that WLM can occur in A-movement and Ā-movement alike, WLM can help us understand why A-type, B1-type, and B2-type extractions differ according to the properties in table 1. I sketch this analysis below.

For A-type extractions (e.g., wh-movement), the NP must merge maximally early, in the base position. Thus, in (46a) the pronoun she c-commands Mary at LF, and Principle C reconstruction effects result. Because merging the NP in the base position means that a full copy is left behind, a PR can be stranded (46b): because the preposition’s complement can contain an NP, it can contain TIME or PLACE.

(46) Properties of A-type extractions (here: wh-movement)

  • a.

    *Which criticism of Maryi’s proposal did shei reject [which criticism ...]?

  • b.

    Which holiday does your family eat lamb onR [which holiday]?

In B1-type extractions (e.g., topicalization), WLM must occur early, but not too early. We know from the presence of Principle C effects (47a) that the site of WLM cannot be maximally late, as in A-movement, because the R-expression Mary must be merged within the c-command domain of the coindexed pronoun she. We also know from the ban on stranding PRs (47b) that WLM is not blocked entirely: the fact that B1-type extractions are sensitive to the distinction between PAs and PRs suggests that their base positions do not contain NPs. Thus, the site of WLM must be above the base position, but within the c-command domain of a coindexed pronoun.

(47) Properties of B1-type extractions (here: topicalization)

  • a.

    *The latest criticism of Maryi’s proposal, shei rejected t.

  • b.

    *Easter, I would never eat lamb onRt.

The analysis of B2-type extractions adopted here (see section 8.1 for some justification) assumes that all B2-type extractions involve movement of a null operator (see, e.g., Chomsky 1977, 1981 for tough-movement; Brillman 2014 for gapped degree phrases; Nissenbaum 2000 and references there for parasitic gaps). Because null operators do not contain lexical material, this analysis correctly predicts that Principle C effects are absent (48a) and that PRs cannot be stranded (48b). The selectional restrictions of PRs cannot be satisfied because their complements do not contain TIME or PLACE.19

(48) Properties of B2-type extractions (here: tough-movement)

  • a.

    The criticism of Maryi’s proposal is tough for heri to accept [∅].

  • b.

    *Easter is easy to eat lamb onR [∅].

It is important to note that there is no extraction that can both bleed Principle C and strand a PR. This gap is significant, and is predicted from a WLM-style analysis of the patterns in table 1. An extraction that can bleed Principle C is by definition an extraction with no NP in the base position. An extraction that can strand a PR is by definition an extraction where the NP must merge in the base position, in order to satisfy the selectional restrictions of the PR. A WLM-style analysis of this type of extraction is impossible, because it leads to a contradiction: the NP needs to merge both very late (to bleed Principle C) and very early (to strand a PR).

7 Deriving the Patterns

7.1 The Basic Proposal

What we need, at this point, is an explanation for the fact that WLM is tightly constrained in B1-type extractions and blocked in A-types. What motivates WLM, and why does its application vary across Ā-extractions?

The logic of the facts we have observed suggests that WLM is not just an option, but in fact a violable preference, in the sense of Optimality Theory (Prince and Smolensky 2004). This principle lies at the heart of the proposal developed here. I claim, furthermore, that WLM is constrained in A-type and B1-type extractions because there are conflicting constraints that take priority over the preference for WLM. Following T±H, one such conflicting constraint is the Case Filter. To model this interaction, we need two constraints: one requiring NPs to receive Case (GETCASE, defined in (49)), and one enforcing the general preference for WLM (MERGELATE, defined in (50)).

(49)

  • GETCASE

  • Assign one * if NP is caseless.

(50)

  • MERGELATE

  • If NP is merged at position x, assign one * for each possible merger site x' whose mother node c-commands x.

The presence of Principle C reconstruction effects in Ā-movement shows that GETCASE dominates MERGELATE: it is more important for an NP to receive Case than it is for that NP to be merged as late as possible. The interaction between GETCASE and MERGELATE is modeled schematically in (51). A boxed letter (here, C) indicates the last site of WLM where NP can receive Case; the node where WLM occurs is boldfaced and italicized. I assume that successive-cyclic movement occurs from Positions D to A. Note that the constraints and tableaux in this section are presented only as a way of formalizing and visualizing the pressures governing the application and location of WLM. Presumably, there are deeper reasons why these constraints exist, as well as reasons why they interact as they do.

(51)

graphic
graphic

Both (51a) and (51b) incur fatal violations of GETCASE because the NP is merged above C, the last position where Case can be assigned. The decision between (51c) and (51d) is made by MERGELATE. (51d) incurs three violations of MERGELATE, as there are three potential landing sites (A, B, and C) that c-command D, the location where the NP is merged. (51c) incurs only two violations of MERGELATE, as there are two potential merge sites (A and B) that c-command the site of WLM. Between (51c) and (51d), (51c) better satisfies MERGELATE and is selected as optimal.

This analysis predicts that, all else being equal, WLM of NP with D will occur at the latest step in the derivation at which the NP can receive Case. Looking more globally at the position of MERGELATE in the grammar, however, it is clear that there are other constraints that outrank it. For example, the need to establish a variable binding dependency can force an NP to merge earlier than its Case position (see T±H; see also section 5 above). In addition to the necessity to bind a variable, it is likely that there are quite a few other semantic constraints that overrule the preference for WLM.20 An anonymous reviewer notes three such constraints, listed here. First, scope reconstruction is possible in A-movement, suggesting that the need to derive nonsurface scope can overrule MERGELATE; see (52a). Second, Fox (2000) and others have noted that objects of creation verbs undergo reconstruction for what are arguably semantic reasons; see (52b). Third, Sportiche (2006) argues that when A-movement applies to pieces of idiom chunks in French,

(52) Reconstruction in A-movement

  • a.

    Reconstruction for nonsurface scope

    Someone from New York is very likely to win the lottery. (some > likely; likely > some)

  • b.

    Reconstruction for objects of creation verbs

    #For those issues to be clarified, [many more/new papers about Quinei’s philosophy] seem to himi to be needed.

Although the global interaction of MERGELATE with other constraints is an area worthy of further study, in the remainder of this article I focus more narrowly on the interaction of only those constraints necessary to explain the P-stranding facts introduced in section 2: that is, for now, the interaction between MERGELATE and GETCASE. In section 7.2, I discuss Case assignment in PPs and show that the schematic analysis sketched in (51) is sufficient to derive the properties of B1-type extractions. In section 7.3, I suggest that the moving DP in A-type extractions is embedded within an additional layer of functional structure, Cable’s (2007, 2010) QP (see also Hagstrom 1998), and I propose that WLM is blocked in these extractions because it would be, in a sense, too countercyclic. While WLM is by definition a countercyclic operation, there are constraints on just how countercyclic it can be (see also Tada 1993:63–70, Sauerland 1998:sec. 2.2). In section 7.4, I show that the analysis presented here makes correct predictions regarding different types of extraction out of other domains.

7.2 Deriving B1-Type Extractions

Recall that, in order to derive both the P-stranding and reconstruction properties of B-type extractions, it must be the case that the site of WLM in these extractions is very low—but not maximally low, as a B-type extraction cannot strand a PR. One possible way to achieve this balance is to assume that PPs have a pP shell, and that p is what assigns Case to the preposition’s complement. Arguments for a little-p projection have been made by a number of different authors, on a number of different grounds (see Van Riemsdijk 1990, Manninen 2003:117 and references there, Svenonius 2003, 2008, 2010, Levinson 2011, Oxford 2011, and many others). In Ā-movement, assuming that pP is a phase (see, e.g., Abels 2003) and that Case can be assigned in a Spec-head relation-ship,21 the last possible landing site where an NP will be able to receive Case is Spec,pP. The analysis sketched in section 7.1 then predicts that, for Ā-movement out of pP, WLM should take place in Spec,p.22 This is because Spec,p is the last possible landing site at which an NP can receive Case; see (53).

(53)

graphic

This simple proposal is sufficient to derive the Principle C and P-stranding facts for B1-type extractions. No conceivable pronoun can intervene between the complement of the preposition and Spec,pP. Thus, Principle C effects are present, just as if the NP were merged in the base position. But because what is left in the base position is just a determiner, the preposition’s complement does not contain an NP, and a PR cannot be stranded.

An obvious question arises here: if Case considerations can force NP to merge with D earlier than otherwise required, why can’t the selectional restrictions of a PR have a similar effect, requiring NP to merge with D in its base position? The answer, I believe, lies in the differing nature of the Case Filter and the selectional requirements of particular lexical items. The requirement for an NP to receive Case is syntactic, but the requirement for a PR to have a TIME-or PLACE- containing complement is presumably semantic (though cf. footnote 14). A PR demands a DP complement endowed with a certain kind of meaning, not a certain structural configuration: there is no obvious reason to believe that an interval DP and an event DP, for example, are structurally distinct. It seems that while the grammar can regulate interactions among purely syntactic con-straints, it cannot regulate interactions between syntactic constraints and those that interface with the semantics.

This statement, though, will ultimately need to be qualified. While it is impossible to overrule WLM in order to satisfy the selectional restrictions of PRs, it is possible to overrule WLM to bind a variable. Note here that the definition of MERGELATE (50) predicts that, if an NP must merge early to bind a variable, it should merge in the highest position where the variable contained in the NP can be bound. This is consistent with the P-stranding facts: even when NP must be merged early to establish a variable binding dependency (54), it must still be merged above the base position (i.e., in Spec,p), because a PR cannot be stranded in a B-type extraction.

(54) Variable binding is possible in B-type extractions (* = P-stranding impossible)

  • a.

    *The river of hisi ancestors, [every boy]i ate lunch onR.

  • b.

    *Hisi mother’s favorite month is easy for [every boy]i to swim inR.

  • c.

    *None of hisi favorite holidays will I eat [any butcher’s]i lamb onR.

As we saw in section 5, WLM can also be overruled to obtain nonsurface scope or to satisfy the semantic properties of certain verbs. These facts suggest that some semantically relevant constraints, but not others, can overrule syntactic ones. I do not have an explanation for why this partition exists, but it is worth noting that a similar asymmetry—that some semantic constraints are more important than others—has been discovered in work by Fox (1998).23

7.3 Deriving A-Type Extractions

A-type extractions pose an interesting problem for the current analysis. The fact that A-type extractions permit a PR to be stranded, as seen in (55), suggests that some factor blocks WLM entirely. For a PR’s selectional restrictions to be satisfied, the NP must merge as part of the preposition’s complement.

(55) A-type extractions can strand PRs

  • a.

    Which floor are we eating dinner onR [which floor]?

  • b.

    Which month did you vacation inR [which month]?

To explain why the examples in (55) are acceptable, we must identify another constraint, specific to A-type extractions, that overrides the preference for WLM. To identify this constraint, we need to find some property that the A-types share. More than this, it must also be some property that all B-types lack.

One possibility is that the moving DP in A-type extractions is embedded within an additional layer of structure: a QP, in Cable’s (2007, 2010) sense. Though the bulk of his discussion focuses on wh-movement, Cable (2007, 2010) notes that restrictive relatives and free relatives in English abide by similar constraints, and he posits that the moving constituents in these extractions are embedded within a larger QP shell.24 He speculates (2007:369–375) that a Q-based analysis could be extended to all other types of Ā-extraction. My proposal is more restrictive: the moving DP in the A-type Ā-extractions, but not the B-types, is embedded within a QP shell.

Some evidence that this is a promising line of analysis comes from a difference in acceptable pied-piping size between the A-type and B-type extractions. English is an example of what Cable terms a limited pied-piping language:in wh-movement, very little material can be pied-piped along with the wh-phrase. This limitation is illustrated in (56). Example (56a), where the wh-phrase is embedded within a larger DP, is much less acceptable than (56b), where it is not.

(56) Pied-piping in wh-movement25

  • a.

    ??Pictures of who do you display in your home?

  • b.

    Who do you display pictures of in your home?

Cable (2007:279) argues that restrictions on pied-piping size, as in (56), are due to the fact that Q-wh agreement is obligatory in limited pied-piping languages (like English). In these languages, lexical categories, like N, cannot intervene between Q and a wh-word. This restriction on pied-piping size also holds for restrictive relatives: compare the ungrammatical (57a) with the grammatical (57b).

(57) Limited pied-piping in restrictive relatives

  • a.

    *The scientist pictures of whom I display in my home won a Nobel Prize.

  • b.

    The scientist who I display pictures of in my home won a Nobel Prize.

Appositive relatives and other B-type extractions, however, allow what Heck (2004) terms massive pied-piping (see also Bresnan 1976, Emonds 1976, Jackendoff 1977, and other more recent work cited in Cable 2007:341). While pied-piping in wh-movement and other A-type ex-tractions is limited, pied-piping in B-types is not. Examples of massive pied-piping in appositive relatives (58a) and negative inversion (58b) follow; in (58b), I assume that the negative phrase no famous scientist is targeted for movement, and pictures of is pied-piped along with it.

(58)

graphic

The grammatical sentences in (58) blatantly violate Q-wh agreement. Rather than trying to explain why the A-type extractions appear to require Q-wh agreement and the B-types do not, it is perhaps simpler to say that the moving DP in the A-types, but not the B-types, is embedded within a QP shell (though see Heck 2004, 2008 and Cable 2007:341–348 for alternatives). Under this account, massive pied-piping and apparent violations of Q-wh agreement can occur in B-type extractions because there is no Q. To be clear, this proposal entails that movement in A- and B-type extractions is triggered by different things. In A-type extractions, the probe carries uninterpretable Q-features and searches for interpretable Q-features on its goal (Cable 2007, 2010), meaning that a DP must be embedded in a QP shell for movement to occur.26 In B-type extractions, worked out under this approach is a theory of the constraints that can interact with, and dominate, AGREE (though cf. Preminger 2011:130ff. on violable constraint models). however, the probe looks for other kinds of interpretable features that are perhaps extraction-specific: a Topic feature for topicalization, for example, or a Neg feature for negative inversion.

Why, though, would the presence of a QP shell in A-type extractions block WLM? In line with Tada’s (1993:63–70) modification of Lebeaux’s (1988) theory of late merger, I propose that WLM cannot apply in A-type extractions because it would be too countercyclic. WLM is permitted when the NP merges with the daughter of the current root node, but prohibited when its application would modify the internal structure of the daughter’s complement. In (59a), WLM is permitted because NP merges with D, the daughter of the current root node. In (59b), WLM is forbidden, because NP merges inside DP embedded within a QP.

(59)

graphic

The proposed restriction on WLM bears a striking resemblance to restrictions on late adjunction described by Sauerland (1998:sec. 2.2). In brief, Sauerland shows that there is an ordering effect among relative clauses, predicted by Tada’s (1993:63–70) modification of Lebeaux’s (1988) proposal. While it is possible to force reconstruction of an inner modifier without forcing recon-struction of an outer modifier, the reverse is not possible: forcing reconstruction of an outer modifier forces reconstruction of an inner modifier as well. Compare, for example, (60a) and (60b) (from Sauerland 1998:52).

(60) Sauerland 1998 and the timing of late merger

  • a.

    [Which computer compatible with hisj that Maryk knew how to use]i did shek tell every boyj to buy ti?

  • b.

    *[Which computer compatible with Maryk’s that hej knew how to use]i did shek tell every boyj to buy ti?

In (60a), the inner relative clause (underlined) must merge at ti for every boy to bind his.It is not necessary for the outer relative clause (that Mary knew how to use) to merge at ti,so a Principle C violation is avoided through late adjunction. In (60b), however, reconstruction of the outer modifier is forced: the underlined clause must merge at t forievery boy to bind he. The fact that a Principle C violation is present in (60b) suggests that if an outer modifier is forced to reconstruct to bind a variable, the inner modifier must reconstruct as well. This suggests that it is impossible for an inner modifier to merge countercyclically between a noun and another modifier; that is, the inner modifier must merge derivationally prior to the outer modifier. See Sauerland 1998:sec. 2.2 for more examples and discussion.

Though the sources of these two effects might ultimately be unrelated,27 we could achieve a unified account by stating the constraint on excessive countercyclicity in relational, rather than absolute, terms. Consider again the difference between the structure in (59a), where WLM is permissible, and (59b), where it is not. In the impossible (59b), notice that the NP that has undergone WLM is asymmetrically c-commanded by an element, Q, that already stands in a complementation relation with DP. Now consider the difference between the structures described by Sauerland (1998): the acceptable configuration where an outer but not an inner modifier undergoes late merger (61a), and the ungrammatical configuration where an inner but not an outer modifier undergoes late merger (61b). Below, the CP that has undergone late merger is boxed.

(61)

graphic

In the acceptable (61a), the modification relation established by late merger is the structurally highest relationship of its type: CP1 is not asymmetrically c-commanded by another element that is either a modifier or sister to a modifier. In the unacceptable (61b), however, the modifier that has undergone late merger is c-commanded by an element, NP, already standing in a modification relationship with CP1. To unify the restrictions on WLM observed here with those on late adjunction observed by Sauerland, we could say that countercyclic merger is permitted only when the relationship established is the structurally highest of its type (i.e., complementation vs. modification) within the current root node. In other words, if a complement undergoes WLM, it cannot be asymmetrically c-commanded by another element in a complementation relationship; if a modifier undergoes late merger, it cannot be asymmetrically c-commanded by another element in a modification relationship. (59b) is unacceptable, then, because the late-merged complement is asymmetrically c-commanded by Q, which takes a DP complement; (61b) is unacceptable because the late-merged adjunct, CP2, is asymmetrically c-commanded by CP1, which is a modifier of NP.

Let us now integrate the above discussion into the analysis. The A-type extractions show that the constraint requiring WLM to create the structurally highest relationship of its type (*TOOLATE) dominates MERGELATE, as WLM of NP to D within a QP is impossible. A possible definition for *TOOLATE, based on the above discussion, is given in (62). The desired result is shown in (63): WLM is blocked when it would result in a violation of *TOOLATE.

(62) *TOOLATE

Assign one * if the relationship established by late merger is not the structurally highest of its type.

(63)

graphic

Candidate (63a), where WLM applies, incurs a violation of * TOOLATE: NP merges with D, creating a complementation relation that is not the structurally highest of its type. Even though candidate (63b), where WLM is blocked, incurs more violations of MERGELATE, it is selected as optimal. NP must merge with the D before the DP merges with Q; otherwise, the derivation fatally violates * TOOLATE.28 This modification to the analysis allows us to derive the Principle C and stranding properties of A-type extractions. WLM is blocked owing to the presence of a QP shell, so the NP merges in the base position. Because A-type extractions leave behind a full copy, Principle C effects are present, and PRs can be stranded.

Note that this modification to the analysis makes a broader prediction: all movement of wh-phrases (and other elements in the same category), not only A-extractions to Ā-positions, should exhibit reconstruction effects. This is not the case: when A-movement is followed by wh-move-ment, Principle C is bled (e.g., Which of Johni’s siblings seems to himi to be smart?). We can solve this problem by allowing for the possibility that Q, like NP, prefers to merge late, and by making two additional claims. First, the moving DP in an A-type Ā-extraction must be embedded within a QP shell before movement begins, as Q is targeted for movement in these extractions (see argumentation in Cable 2007, 2010). On the other hand, I propose that the moving DP in Ā-movement need not be embedded within a QP shell. The result is that while Q prefers to be merged late, it must be merged immediately prior to the first step of Ā-movement. This allows for the possibility that WLM is impossible in an A-type Ā-extraction, but possible when A-movement feeds Ā-movement.29

7.4 Further Predictions

The interaction of MERGELATE and GETCASE makes a very basic prediction: all else being equal, WLM should occur at the last position in which the NP can receive Case. Here I expand the body of evidence supporting this prediction by investigating properties of other extraction types out of different domains. I will begin with the English pseudopassive, a type of A-movement that can strand prepositions. Several examples are provided in (64).

(64) Examples of the pseudopassive

  • a.

    The bed was slept in.

  • b.

    I was talked about.

  • c.

    The desk was written on.

The only crucial assumption I will make regarding the mechanics of the pseudopassive is that, as is the case for the true passive, the moved DP receives Case not from P or p, but from the matrix T (see Abels 2003:234). This is evident in (64b) from the fact that the subject I bears nominative case, not accusative, as would be assigned by the preposition about (compare *Me was talked about with They talked about me). In addition, I assume that movement to Spec,T in the pseudopassive is driven by the same factors that drive movement in the true passive: the standard analysis is that the lower DP moves to check Case (e.g., Baker, Johnson, and Roberts 1989:222).

If T assigns Case to the moving DP, then the preference for WLM dictates that NP must merge with its head D in Spec,T. Because the NP undergoes WLM (i.e., is not present in the base position), this analysis predicts that the pseudopassive should behave like a B-type extraction: stranding a PR should be impossible. This is in fact the case. In the locative examples that follow,30 pseudopassivization is possible when the preposition is entity-selecting (65), but not when it is location-selecting (66).

(65) Pseudopassives are licit for entity-selecting prepositions

  • a.

    The house was lived in.

    (cf. We lived in {the house, it}.)

  • b.

    Boxes were peered into.

    (cf. We peered into {boxes, them}.)

  • c.

    Mice were stepped on.

    (cf. We stepped on {mice, them}.)

  • d.

    The wall was written on.

    (cf. We wrote on {the wall, it}.)

(66) Pseudopassives are banned for location-selecting prepositions (* = lack of a location reading)

  • a.

    *The second floor was worked on.

    (cf. We worked on {the second floor, *it}.)

  • b.

    *The sky was flown into.

    (cf. We flew into {the sky, *it}.)

  • c.

    *The Charles River was dined on.

    (cf. We dined on {the Charles River, *it}.)

  • d.

    *The fourth floor was celebrated on.

    (cf. We celebrated on {the fourth floor, *it}.)

This is the pattern observed for B-type extractions. Because the moving NP is not present in the base position, the complement of a location-selecting preposition does not contain an element of the semantic category PLACE. In B-type extractions and pseudopassivization alike, a PR cannot be stranded.31

I turn now to Ā-extraction of an object out of vP. Recall the claim in this section that Case assignment can be delayed until Spec,x, where x is the relevant Case assigner. Assuming that accusative case is assigned by v, the analysis predicts that WLM of an object NP can be delayed until its head D reaches Spec,v. This prediction has potential implications for the proposed analysis of movement out of double object constructions: if the NP report on John’s student undergoes WLM at Spec,v, we would expect (67a) to be grammatical but (67b) to be ungrammatical.

(67) Extraction out of double object constructions (judgments suppressed, for now)

  • a.

    That report on Johni’s student, Mary showed himi.

  • b.

    That report on Johni’s student, hei showed Mary.

We might expect to find a contrast here because the position of the coindexed pronominal differs. In (67a), the coindexed pronominal is VP-internal. If the NP report on John’s student undergoes WLM at Spec,vP, then him should never c-command John, and a Principle C violation should be absent (68a). If, however, the coindexed pronominal resides in Spec,vP, he will c-command John, and a Principle C violation should be present (68b). (I assume that movement of the object to an inner Spec,v must precede merger of the external argument to an outer Spec,v; see Müller 2010.)

(68) Positioning of the coindexed pronominal matters

  • a.

    Coindexed object: WLM applies above c-command domain

    [vP Mary [vP that report on Johni’s student [VP showed himithat]]]

  • b.

    Coindexed subject: WLM applies within c-command domain

    [vP hei [vP that report on Johni’s student [VP showed Mary that]]]

What we find, however, is that both sentences in (67) are ungrammatical: a Principle C violation results regardless of whether the coindexed pronominal is a subject or an object. Rather than posing a problem for the current analysis, however, the failure of (68a) to bleed Principle C has the potential to teach us something about the structure of double object constructions. Pesetsky (1996:chap. 5) argues that the second argument in a double object construction does not receive Case from any verbal projection; rather, it is introduced and Case-licensed by a null preposition, called here P (see also Harley 2002 for a similar proposal). Assuming this analysis, WLM of the NP report on John’s student can occur no later than Spec,PP—a position that is crucially within the c-command domain of either a coindexed object (69a) or subject (69b).

(69) Positioning of the coindexed pronominal does not matter

  • a.

    Coindexed object: WLM applies within c-command domain

    [vP Mary [vP that report . . . [VP showed himi [PPthat report on Johni’s student]]]]

  • b.

    Coindexed subject: WLM applies within c-command domain

    [vP hei [vP that report . . . [VP showed Mary [PPthat report on Johni’s student]]]]

We can therefore interpret the failure of (67a) to bleed Principle C not as a problem for the current analysis, but as an indication that this analysis, following Pesetsky and Harley, is on the right track. If we assume that the second object in a double object construction is introduced and assigned Case by a null prepositional element, then the failure of both examples in (67) to bleed Principle C is not surprising.

8 Further Support: B2-Type Extractions, and More Pronouns

The analysis proposed in section 7 claims that the link between a preposition’s selectional restric-tions and its potential to be stranded is structural. Interval and location PRs reject all DP comple-ments, pronounced or not, that do not contain TIME or PLACE. This approach makes a prediction: if the preposition’s complement does not contain an NP for other reasons, stranding a PR should be impossible, regardless of the extraction in question.

Here I verify this prediction with evidence from two domains. In section 8.1, I return to the properties of B2-type extractions, and argue that their behavior is best explained as the result of null operator movement. Because the preposition’s complement never contains an NP, a PR cannot be stranded. In section 8.2, I show that the analysis correctly predicts contrasts within individual extraction types. When wh-movement involves movement of wh-pronouns, wh-movement acts like a B-type extraction: it is sensitive to whether or not a given preposition is pronoun-rejecting.

8.1 B2-Type Extractions and Null Operators

If: we assume that the B2-type extractions (tough-movement, gapped degree phrases, and parasitic gaps) are structurally similar—that is, they all involve movement of a null operator—then the fact that they bleed Principle C and cannot strand PRs is predicted by the current analysis. Null operators do not contain lexical material (i.e., NPs), so there is nothing in their structure that could satisfy the selectional restrictions of PRs or drive a Principle C effect. But the assumption that these three extractions all involve null operator movement is not universally accepted. Below, I briefly sketch the basis of the controversy and provide some evidence in favor of the null-operator-style analyses.

While it is common to propose that parasitic gaps involve movement of a null operator or some other null pronominal element (see Nissenbaum 2000:24–29 for an overview), the analysis of tough-movement and gapped degree phrases is more controversial. The existence of defective intervention effects (see Hartman 2011, also Brillman 2014) and the presence of reconstructed bound variable readings support an improper-movement analysis, in which the moving constituent originates in the embedded clause. The absence of scope reconstruction from the subject into the embedded clause (see Postal 1974, Epstein 1989, Fleisher 2013) and the absence of Principle C effects, however, supports an account where the DP originates in the matrix clause, like Chomsky’s (1977, 1981) null operator analysis.

It is beyond the scope of this article to provide a conclusive argument for either improper movement or null operators, nor do I attempt to further differentiate within these two classes of analyses. When we look at interactions between the B2-type extractions and wh-movement, however, some evidence for a null operator analysis of the B2-type extractions emerges. Stranding PRs using a combination of wh-movement and a B2-type extraction is impossible (see (70a), (71a)) and these extractions bleed Principle C (see (70b), (71b)). In short, hybrids of A-type and B2-type extractions behave like B2-types.

(70) Hybrid type 1: Tough-and wh-movement

  • a.

    *Which holiday is easy to eat lamb on?

    (cf. Which holiday did we eat lamb on?)

  • b.

    Which criticism of Maryi’s proposal is hard for heri to accept?

    (cf. *Which criticism of Maryi’s proposal did shei reject?)

(71) Hybrid type 2: Gapped degree phrases and wh-movement

  • a.

    *Which holiday is too joyous to eat lamb on?

    (cf. Which holiday did we eat lamb on?)

  • b.

    Which criticism of Maryi’s proposal is too harsh for heri to accept?

    (cf. *Which criticism of Maryi’s proposal did shei reject?)

These differences between the hybrid extractions and wh-movement are predicted by the null operator analysis. If the matrix subject never occupies any position in the embedded clause, then it follows that a PR cannot be stranded, because its complement cannot contain TIME or PLACE. Similarly, in (70b) and (71b) Principle C is bled because Mary is never within the c-command domain of coindexed her.

An improper-movement-style analysis, where the QP originates in the embedded clause, would face some problems in accounting for (70) and (71). Because the hybrid extractions are wh-interrogatives, I assume that the fronted DP is embedded within a QP shell. An analysis assuming that the QP originates in the embedded clause would need to explain why it is apparently acceptable for excessively late WLM to occur in the hybrid extractions, but not in pure wh-movement.32

8.2 Wh-Pronouns

We saw in section 8.1 that if a given extraction never allows a PR to have a complement containing TIME or PLACE, that extraction will never allow a PR to be stranded. The analysis outlined in section 7, in turn, suggests that the ability to strand a PR is linked to the structural configuration of the gap, not to the extraction that creates it. Given this, we would expect, under certain conditions, to find contrasts within extractions. Extractions that are in principle capable of leaving behind full copies should be sensitive to PRs if the full copy is pronoun-sized (i.e., a D).

One way to assess this prediction is to compare the behavior of wh-movement with wh-pronouns and full wh-phrases. It is a plausible assumption that wh-pronouns like who and what do not contain NPs, like their non-wh counterparts her and it. Wh-movement with wh-pronouns should then be sensitive to whether a preposition is a PA or a PR, while wh-movement with full wh-phrases should not discriminate. This is because A-type extractions leave behind a full copy, but the full copy of a wh-pronoun does not contain an NP and therefore cannot contain TIME or PLACE.

This prediction holds. The examples in (72) show that movement of both wh-pronouns and wh-phrases is grammatical when stranding a PA; the examples in (73) show that movement of wh-phrases is grammatical when stranding a PR, but movement of wh-pronouns is not.

(72) Stranding a PA

  • a.

    My family eats dinner on the green table. John’s family eats dinner on it, too.

  • b.

    What table does your family eat dinner on?

  • c.

    What does your family eat dinner on?

(73) Stranding a PR

  • a.

    *My family eats turkey on Thanksgiving. John’s family eats turkey on it, too.

  • b.

    What holiday does your family eat turkey on?

  • c.

    *What does your family eat turkey on?

Note that the status of (73c) is not due to a restriction on movement: in a multiple question, the phrasal DP which holiday but not the pronoun what is an acceptable complement of the temporal PRon, as predicted.33

(74) What vs. which holiday in a multiple question

  • a.

    Which person thinks that John’s family eats turkey on which holiday?

  • b.

    *Which person thinks that John’s family eats turkey on what?

In addition, the ungrammaticality of (73c) cannot be attributed to a more general restriction on the types of DPs that wh-pronouns can replace. Thanksgiving can be pronominalized with it when it is the complement of celebrate (75a). In this context, movement of both wh-pronouns and wh-phrases is permitted (75b–c).

(75) Interval DPs and wh-pronouns

  • a.

    My family celebrates Thanksgiving, and John’s family celebrates it, too.

  • b.

    Which holiday does your family celebrate?

  • c.

    What does your family celebrate?

The analysis proposed in section 7 bears some resemblance to Postal’s (1998) account of antipronominal contexts, but the contrasts in (72)–(73) reveal that the WLM-style analysis makes accurate predictions that Postal’s analysis does not. Postal proposes that certain extractions are sensitive to antipronominal contexts because they leave behind null resumptive pronouns. Because his account proposes that extractions themselves are linked to certain types of null categories, it is unable to predict contrasts within extractions, like the contrast between wh-pronouns and wh-phrases documented in (72)–(73). However, under an analysis where the ability to strand a PR is linked not to intrinsic properties of individual extractions, but to the structural configuration of the gap, the pattern in (72)–(73) is accounted for. The ability to strand a temporal or locative PR depends entirely on whether or not its complement contains TIME or PLACE. Specific extractions only appear to play a role because their independent properties (e.g., presence of a QP shell) regulate the application and location of WLM.

9 Conclusion

The P-stranding asymmetries and Principle C facts discussed in this article demonstrate a need to divide Ā-extractions into two major groups. I have argued that the differences between these groups arise because they place differing constraints on the application of WLM. The A-type extractions leave behind fully constructed copies, as the application of WLM is blocked. The B-type extractions do not leave behind fully constructed copies, either because WLM is required (B1-type extractions) or because the extractions in question involve null operator movement (B2-type extractions). The extractions that leave behind fully constructed copies (A-types) can strand pronoun-rejecting prepositions; the extractions that do not leave behind fully constructed copies (B-types) cannot.

More broadly, all English Ā-extractions have the capability to strand prepositions, but differing properties of individual extractions lead to distinct constraints on which prepositions can be stranded. Thus, while it is true that all Ā-extractions share the same set of basic properties (e.g., Chomsky 1977, 1981, 1982, Pesetsky 1982, Engdahl 1983), asymmetries within a single property can point to subtle, but fundamental, differences among them.

Notes

Many thanks to David Pesetsky, Adam Albright, Danny Fox, Sabine Iatridou, Chris O’Brien, Norvin Richards, Donca Steriade, Coppe van Urk, and audiences at MIT Ling-Lunch, ECO-5 2014, LAGB 2014, and NELS 45 for helpful comments and moral support. I am also grateful to two anonymous LI reviewers, whose detailed and insightful feedback has led to many improvements in this article.

1 Various principled exceptions to this statement do, of course, exist. For example, Principle C can be bled in Ā-movement when the relevant R-expression is contained within an adjunct that has undergone late merger (Lebeaux 1988, Chomsky 1995). Principle C effects can be present in A-movement when the moving constituent is forced to reconstruct for other reasons, for example, for scope (Fox 1999).

2 The majority of speakers consulted find these contrasts extremely sharp, but some speakers find them subtler (or they detect the contrasts for some but not all predicates or extraction types). I have not found a pattern to this variation, and leave it as a matter for further investigation.

3 It is possible to replace an interval PP with then: for example, My family eats turkey on Thanksgiving, and John’s family eats turkey then, too. Since then has the distribution of a PP (i.e., it can appear in non-Case positions, as in John arrived then), I assume it is a pro-PP: a prepositional element that conflates a temporal preposition with a nominal whose reference is salient in the discourse. I hypothesize that the reason why then (a pro-PP) can replace an interval PP is that, although then can refer to a PP headed by an interval-selecting preposition, it is not itself an interval-selecting preposition. Some support for the idea that then is not interval-selecting comes from the fact that then can also refer to event PPs: for example, We left after John’s talk, and Mary left then, too.

4 Replacing Thanksgiving with a demonstrative, this or that, is similarly ungrammatical (*My family eats turkey on Thanksgiving, and John’s family eats turkey on that, too). Thus, the inability of it to replace a preposition’s interval complement also cannot be attributed to its status as a weak pronoun (see Cardinaletti and Starke 1994). For an interesting case where the weak vs. strong status of a pronoun does figure into a preposition’s complementation properties, see Zribi-Hertz 1984 on French.

5 Some other verbs that allow interval complements are waste, kill, and use.

6 It is worth asking whether or not there are other structural positions that display the same interval vs. event pronominalization asymmetry. To the best of my knowledge, there aren’t. We saw in (8) that both object and subject intervals can be pronominalized.

7 Note that pied-piping can rescue sentences like (12c) and (12d): compare *Not a single holiday will I ever eat lamb on with the more acceptable On not a single holiday will I ever eat lamb.

8Postal (1998) shows that some Ā-extractions are sensitive to other antipronominal contexts (name positions, changeof- color environments, etc.). His partition differs in several ways from mine; potential reasons for these differences are not explored here.

9 As with temporal PPs and then, it is possible to replace locative PPs (or at least, part of them; see Svenonius 2008 for some discussion) with there. Thus, I ate dinner on the fourth floor, and John ate dinner there, too is grammatical. I assume that there, similarly to then, is a pro-PP that takes as its antecedent any kind of locative PP.

10 The V+P combination live in is a counterexample to the generalizations established here. It is impossible to pronominalize Boston, for example, as the complement of in (*I live in Boston, and John lives in it, too), but for most speakers, stranding in with B-type extractions is possible (Boston is easy to live in). This is an isolated counterexample, and its status may have more to do with properties of live in, yet to be defined.

11 Others argue that examples like we religious ones should be analyzed as appositive modifiers to a pronominal head. See Delorme and Dougherty 1972 for discussion, and Pesetsky 1978 for a critique of their proposal.

12 For temporals, I assume that day, month, and other interval-denoting nouns are subspecies of TIME. Similarly, for locatives, I assume that city, state, and other place-denoting nouns are subspecies of PLACE.

13 An interesting difference between the temporal example in (35b) and the locative example in (36b) is that the presence of the is grammatical in the former (December 25th vs. December the 25th), but not in the latter (New York City vs. *New York the city). It is also worth noting that using the in temporal DPs is impossible for holidays named by their dates ( July 4th vs. *July the 4th).

14 An anonymous reviewer notes that the restriction could also be due to a syntactic constraint that penalizes a failure to meet a head’s complementation property. This is a possibility; what would need to be worked out is how it is possible for a preposition to see through the structure of its complement DP to discern the identity of the embedded NP.

15 Although the only types of PPs I discuss in this section and in what follows are locative and temporal PPs, I anticipate that the analysis proposed here could extend to circumstantial and other types of PPs (see section 2.3).

16t denotes that what is left behind is less than a full copy; exactly what is left behind will be discussed later.

17 The idea that some extraction sites behave like pronouns because they are in some sense pronouns builds on previous work drawing the same conclusion, such as Perlmutter 1972, Cinque 1990, and Postal 1998. While the idea has antecedents, the proposed formalization is novel. See section 8.2 for a brief comparison with Postal 1998.

18 Support for the idea that both Ds and Ns receive Case comes from many Indo-European languages, such as Greek (Ancient (Smyth 1956) and Modern (Sabine Iatridou, pers. comm.)), where Ds and Ns bear case morphology.

19 An alternative is to assume an improper-movement analysis of tough-movement and gapped degree phrases (e.g., Brody 1993, Hornstein 2000, Hicks 2003) and say that WLM occurs maximally late (this idea due to Norvin Richards, pers. comm.). See section 8.1 for a brief discussion of this alternative.

20 An anonymous reviewer notes that the effects of MERGELATE can only be seen under very restricted circumstances, as there are many other constraints that dominate it. While it is true that the effects of MERGELATE are often invisible, especially in the case of Ā-movement, other phenomena display the same split between A- and B-type extractions documented here. Take, for example, Lasnik and Stowell’s (1991) discussion of weakest crossover, where they demonstrate that certain Ā-extractions—but not all—behave like A-movement in that weak crossover violations appear to be absent entirely. The extractions where weak crossover is present include wh-movement and restrictive relatives (my A-types); the extractions where weak crossover is absent include topicalization, parasitic gap formation, and tough-movement (my B-types). Various authors (e.g., Koopman and Sportiche 1983, Lasnik and Stowell 1991, Sauerland 1998) have related the presence or absence of weak crossover to the contents of the gap site (i.e., pronominal vs. nonpronominal); perhaps weak(est) crossover is an example of another phenomenon in which the effects of MERGELATE are discernible in Ā -movement.

21 In this derivation, it is crucial that Case assignment can follow movement of the NP to Spec,p. Although we know from much recent work that Case does not have to be assigned in a Spec-head relationship, I assume here that it can be. More precisely, when a probe with an EPP property checks Case on its goal, Case assignment can either precede or follow movement of the goal. In derivations where WLM is involved (as in (53)), Case assignment will always follow movement, as this allows for maximal satisfaction of MERGELATE.

22 Note that positing a pP shell makes it possible for a DP to move successive-cyclically through Spec,pP without violating Abels’s (2003) stranding generalization.

23 The relevant examples are ones in which inverse scope is generally impossible for reasons of Scope Economy and parallelism, but permitted when necessary to bind a variable. The following contrast, from Fox 1998:39, demonstrates:

  • (i)

    Someone in the audience knows the capital of every country. The person who was invited to talk about it does, too. (someone > every, every > someone)

  • (ii)

    Someone in the audience knows the capital of every country. The person who was invited to talk about these countries does, too. (someone > every, *every > someone).

In (i), the inverse scope reading is licensed by the necessity for the pronoun it (in the second sentence) to be bound by every country (in the first). In (ii), the inverse scope reading is impossible when there is no need to bind a variable. See Fox 1998, especially pages 37–45, for more discussion.

24 The Qs involved in wh-questions and restrictive relatives are presumably different kinds of Q. Cable dubs the wh-variety QQ and the relative variety QREL. What matters here is only the presence of the extra layer of structure.

25 Some speakers report that the contrast in (56) is not strong. What’s important is that there is a contrast in (56) and that it disappears for the corresponding sentences with appositive relatives. Note also that in the most natural pronunciation of (56a), the wh-word is focused (Pictures of who . . .?) as if it were a wh-in-situ question (You displayed pictures of who . . .?). Without focusing the wh-word, as is usual for wh-questions with movement, (56a) is much less natural.

26 Concretely, I assume that there is a constraint requiring agreement between the probe and the goal (AGREE) and that AGREE dominates MERGELATE. We would expect that other constraints can dominate AGREE, causing agreement to fail (see Preminger 2011). What remains to be worked out under this approach is a theory of the constraints that can interact with, and dominate, AGREE (though cf. Preminger 2011:130ff. on violable constraint models).

27 If the speculation that follows is ultimately incorrect, we are left without an explanation for why late merger of adjuncts appears to be permitted to a greater degree than that of complements. I leave open the possibility that the potential for countercyclic merger of complements is just more limited than that of adjuncts. We know already that there is an asymmetry between late merger of complements and adjuncts: while late adjunction is always interpretable, late merger of a complement is not always (see Fox 2002:69). Perhaps they are also differentiated according to the degree of countercyclicity allowed.

28 There is an additional candidate, in which only Q merges as the preposition’s complement, and DP merges with Q after Q moves to Spec,pP. As an anonymous reviewer notes, this derivation does not incur a violation of *TOOLATE, and it incorrectly predicts that WLM should occur in A-type extractions. It is not immediately obvious, though, that the result of the operation just described would be semantically interpretable. WLM of NP is possible because what is left behind—a quantificational determiner—can be converted to a definite description through Trace Conversion. Cable’s Q serves as a variable over choice functions (see also Hagstrom 1998, Yatsushiro 2001); it is unclear that the operations of variable insertion and determiner replacement are even applicable here, or, if they are, that they would result in a semantically interpretable structure.

29 An anonymous reviewer notes a potential loophole in this proposal. Suppose it is possible for DP to move to Spec,p for Case reasons, and for (a) late merger of Q and (b) Ā-movement to follow. Given the preference for WLM, this would predict that late merger of Q should be the desired state of affairs, and furthermore that stranding a PR should be impossible for all extractions, even A-types. We can avoid this unwanted prediction by asserting that, at least in English, Case alone cannot trigger movement to Spec,p; p lacks an EPP feature. It just so happens, however, that if a DP lands in Spec,p on its way up, Case can be assigned there. Under this analysis, there is no reason for a DP to land in Spec,p unless Spec,p is an intermediate Ā -step—in which case, in A-type extractions, the Q element must be present. See Kayne 1984:2ff. for a similar argument from Case assignment into an embedded Comp: a DP cannot land in an embedded Comp unless it has other (non-Case-related) reasons for being there.

30 I focus on locatives because stranding a temporal preposition under A-movement is generally degraded, regardless of whether it is a PR (*Christmas was eaten lamb on) or a PA (*John’s party was left after).

31 An anonymous reviewer notes that we would expect pseudopassives of the sort in (66) to become grammatical when the subject contains a bound variable, as the constraint penalizing a failure to bind a variable will force NP to merge in the complement of a pronoun-rejecting preposition. However, this is not the case: even when the bound variable reading is possible, stranding a PR is impossible.

  • (i)

    *[The floor that hisi mother liked] was jumped on by [every child]i.

  • (ii)

    *[The river that hisi mother swam in] was dined on by [every politician]i.

If, however, the by-phrase is merged in Spec,vP (see Collins 2005:84–85), and the NP restrictor can undergo WLM at the VP-adjoined position (see Takahashi and Hulsey 2009:401), the data in (i)–(ii) are exactly what we would expect: reconstruction is forced to the matrix VP-adjoined position, where the variable his can be bound, but not all the way to the base position. Because the NP is not present in the base position, a PR cannot be stranded.

32 An anonymous reviewer suggests an analysis along these lines. As the initial step of Ā-movement in tough-movement may well not be Q-driven, late merger of Q could be possible—meaning that NP is free to merge after D has left the infinitival clause. For the purposes of this section, the two accounts make identical predictions ( Principle C is bled, stranding a PR is impossible), so the choice between them is not crucial here.

33 In addition, the what vs. what x contrast appears under sluicing: compare the ungrammatical *John’s family eats turkey on some holiday, but I don’t know what with the grammatical John’s family eats turkey on some holiday, but I don’t know which one. The emergence of this contrast under sluicing is consistent with the analysis of antipronominal contexts as a consequence of selectional restrictions, rather than island-like constraints on movement, which can in some cases be circumvented by sluicing (see Ross 1969 for the original observation, and Merchant 2001 for more recent work). Thanks to an anonymous reviewer for pointing this out.

References

Abels,
Klaus
.
2003
.
Successive cyclicity, anti-locality, and adposition stranding
. Doctoral dissertation, University of Connecticut, Storrs. Abney, Steven. 1987. The English noun phrase in its sentential aspect. Doctoral dissertation, MIT, Cambridge, MA.
Baker,
Mark
,
Kyle
Johnson
, and
Ian
Roberts
.
1989
.
Passive arguments raised
.
Linguistic Inquiry
20
:
219
251
.
Bhatt,
Rajesh
, and
Roumyana
Pancheva
.
2004
.
Late merger of degree clauses
.
Linguistic Inquiry
35
:
1
45
.
Bhatt,
Rajesh
, and
Roumyana
Pancheva
.
2007
.
Degree quantifiers, position of merger effects with their restrictors, and conservativity
. In
Direct compositionality
, ed. by
Chris
Barker
and
Pauline
Jacobson
,
306
335
.
Oxford
:
Oxford University Press
.
Bresnan,
Joan
.
1976
.
On the form and functioning of transformations
.
Linguistic Inquiry
7
:
31
40
.
Brillman,
Ruth
.
2014
.
Gapped degree phrases are improper movement constructions
. Ms., MIT, Cambridge, MA.
Brody,
Michael
.
1993
.
θ-theory and arguments
.
Linguistic Inquiry
24
:
1
23
.
Cable,
Seth
.
2007
.
The grammar of Q: Q-particles and the nature of wh-fronting, as revealed by the wh-questions of Tlingit
. Doctoral dissertation, MIT, Cambridge, MA.
Cable,
Seth
.
2010
.
The grammar of Q: Q-particles, wh-movement, and pied-piping
.
Oxford
:
Oxford University Press
.
Cardinaletti,
Anna
, and
Michal
Starke
.
1994
.
The typology of structural deficiency: On the three grammatical classes
. Ms., MIT, Cambridge, MA.
Chomsky,
Noam
.
1977
.
On wh-movement
. In
Formal syntax
, ed. by
Peter
Culicover
,
Thomas
Wasow
, and
Adrian
Akmajian
,
71
132
.
New York
:
Academic Press
.
Chomsky,
Noam
.
1981
.
Lectures on government and binding: The Pisa lectures
.
Dordrecht
:
Foris
.
Chomsky,
Noam
.
1982
.
Some concepts and consequences of the theory of government and binding
.
Cambridge, MA
:
MIT Press
.
Chomsky,
Noam
.
1993
.
A minimalist program for linguistic theory
. In
The view from Building 20: Essays in linguistics in honor of Sylvain Bromberger
, ed. by
Kenneth
Hale
and
Samuel Jay
Keyser
,
1
52
.
Cambridge, MA
:
MIT Press
.
Chomsky,
Noam
.
1995
.
The Minimalist Program
.
Cambridge, MA
:
MIT Press
.
Chomsky,
Noam
.
1999
.
Powers and prospects: Reflections on human nature and the social order
.
London
:
Electric Book Company
.
Chomsky,
Noam
.
2004
.
Beyond explanatory adequacy
. In
Structures and beyond: The cartography of syntactic structures
,
vol. 3
, ed. by
Adriana
Belletti
,
104
131
.
Oxford
:
Oxford University Press
.
Cinque,
Guglielmo
.
1990
.
Types of Ā dependencies
.
Cambridge, MA
:
MIT Press
.
Collins,
Chris
.
2005
.
A smuggling approach to the passive in English
.
Syntax
8
:
81
120
.
Delorme,
Evelyne
, and
Ray
C. Dougherty
.
1972
.
Appositive NP constructions: We, the men; we men; I, a man; etc
.
Foundations of Language
8
:
2
29
.
Elbourne,
Paul D
.
2005
.
Situations and individuals
.
Cambridge, MA
:
MIT Press
.
Emonds,
Joseph
.
1976
.
A transformational approach to English syntax: Root, structure-preserving, and local transformations
.
New York
:
Academic Press
.
Engdahl,
Elisabet
.
1983
.
Parasitic gaps
.
Linguistics and Philosophy
6
:
5
34
.
Epstein,
Samuel David
.
1989
.
Quantification in null operator constructions
.
Linguistic Inquiry
20
:
647
658
.
Fleisher,
Nicholas
.
2013
.
On the absence of scope reconstruction in tough-subject A-chains
.
Linguistic Inquiry
44
:
321
332
.
Fox,
Danny
.
1998
.
Economy and semantic interpretation: A study of scope and variable binding
. Doctoral dissertation, MIT, Cambridge, MA.
Fox,
Danny
.
1999
.
Reconstruction, binding theory, and the interpretation of chains
.
Linguistic Inquiry
30
:
157
196
.
Fox,
Danny
.
2000
.
Economy and semantic interpretation
.
Cambridge, MA
:
MIT Press
.
Fox,
Danny
.
2002
.
Antecedent-contained deletion and the copy theory of movement
.
Linguistic Inquiry
33
:
63
96
.
Hagstrom,
Paul
.
1998
.
Decomposing questions
. Doctoral dissertation, MIT, Cambridge, MA.
Harley,
Heidi
.
2002
.
Possession and the double object construction
.
Linguistic Variation Yearbook
2
:
31
70
.
Hartman,
Jeremy
.
2011
.
Intervention in ‘tough’ constructions
. In
NELS 39
, ed. by
Suzi
Lima
,
Kevin
Mullin
, and
Brian
Smith
,
387
396
.
Amherst
:
University of Massachusetts, Graduate Linguistic Student Association
.
Heck,
Fabian
.
2004
.
A theory of pied piping
. Doctoral dissertation, Universität Tübingen.
Heck,
Fabian
.
2008
.
On pied-piping: Wh-movement and beyond
.
Berlin
:
Mouton de Gruyter
.
Hicks,
Glyn
.
2003
.
So easy to look at, so hard to define: Tough movement in the Minimalist framework
. Master’s thesis, University of York.
Hornstein,
Norbert
.
2000
.
Move! A Minimalist theory of construal
.
Oxford
:
Blackwell
.
Jackendoff,
Ray
.
1977
.
{eto ung X overbar, ausin mo nlang, haha :) } syntax: A study of phrase structure
.
Cambridge, MA
:
MIT Press
.
Kayne,
Richard S
.
1984
.
Connectedness and binary branching
.
Dordrecht
:
Foris
.
Kayne,
Richard S
.
2005a
.
Here and there
. In
Movement and silence
,
65
84
.
Oxford
:
Oxford University Press
.
Kayne,
Richard S
.
2005b
.
Silent years, silent hours
. In
Movement and silence
,
241
260
.
Oxford
:
Oxford University Press
.
Koopman,
Hilda
, and
Dominique
Sportiche
.
1983
.
Variables and the Bijection Principle
.
The Linguistic Review
2
:
139
160
.
Kuroda,
S.-Y
.
1964
.
A note on English relativization
. Ms., MIT, Cambridge, MA.
Kuroda,
S.-Y
.
1968
.
English relativization and certain related problems
.
Language
44
:
244
266
.
Lasnik,
Howard
, and
Tim
Stowell
.
1991
.
Weakest Crossover
.
Linguistic Inquiry
22
:
687
720
.
Lebeaux,
David
.
1988
.
Language acquisition and the form of grammar
. Doctoral dissertation, University of Massachusetts, Amherst.
Lebeaux,
David
.
1998
.
Where does binding theory apply?
Technical report, NEC Research Institute, Princeton, NJ.
Levinson,
Lisa
.
2011
.
Possessive WITH in Germanic: HAVE and the role of P
.
Syntax
14
:
355
393
.
Manninen,
Satu Helena
.
2003
.
Small phrase layers: A study of Finnish manner adverbials
.
Amsterdam
:
John Benjamins
.
Merchant,
Jason
.
2001
.
The syntax of silence: Sluicing, islands, and the theory of ellipsis
.
Oxford
:
Oxford University Press
.
Müller,
Gereon
.
2010
.
On deriving CED effects from the PIC
.
Linguistic Inquiry
41
:
35
82
.
Nikolaeva,
Liudmila
.
2014
.
The secret life of pronouns
. Doctoral dissertation, MIT, Cambridge, MA.
Nissenbaum,
Jon
.
2000
.
Investigations of covert movement
. Doctoral dissertation, MIT, Cambridge, MA.
Oxford,
Will
.
2011
.
The syntax of Innu-aimun locatives
. In
UBCWPL Vol. 31: The Proceedings of WSCLA 16
, ed. by
Alexis
Black
and
Meagan
Louie
,
135
150
.
Vancouver, BC
:
University of British Columbia, Department of Linguistics
.
Perlmutter,
David M
.
1972
.
Evidence for shadow pronouns in French relativization
. In
The Chicago which hunt: Papers from the Relative Clause Festival, April 13, 1972; a paravolume to Papers from the Eighth Regional Meeting
, ed. by
Paul M.
Peranteau
Judith N.
Levi
, and
Gloria C.
Phares
,
73
105
.
Chicago
: University of Chicago, Chicago Linguistic Society.
Pesetsky,
David
.
1978
.
Category switching and so-called so-called pronouns
. In
Papers from the 14th Regional Meeting, Chicago Linguistic Society
, ed. by
Donka
Farkas
,
Wesley
Jacobsen
, and
Karol
Todrys
,
350
361
.
Chicago
:
University of Chicago, Chicago Linguistic Society
.
Pesetsky,
David
.
1982
.
Paths and categories
. Doctoral dissertation, MIT, Cambridge, MA.
Pesetsky,
David
.
1996
.
Zero Syntax
.
Cambridge, MA
:
MIT Press
.
Postal,
Paul M
.
1966
.
On so-called ‘‘pronouns’’ in English
. In
Report of the 17th Annual Round Table Meeting on Language and Linguistics
, ed. by
Francis P.
Dinneen
,
177
206
.
Washington, DC
:
Georgetown University Press
.
Postal,
Paul M
.
1974
.
On raising
.
Cambridge, MA
:
MIT Press
.
Postal,
Paul M
.
1998
.
Three investigations of extraction
.
Cambridge, MA
:
MIT Press
.
Preminger,
Omer
.
2011
.
Agreement as a fallible operation
. Doctoral dissertation, MIT, Cambridge, MA.
Prince,
Alan
, and
Paul
Smolensky
.
2004
.
Optimality Theory: Constraint interaction in generative grammar
.
Oxford
:
Blackwell
. (Revision of 1993 technical report, Rutgers University Center for Cognitive Science. Available at http://roa.rutgers.edu/article/view/547.)
Riemsdijk,
Henk van
.
1990
.
Functional prepositions
. In
Unity in diversity: Papers presented to Simon C. Dik on his 50th birthday
, ed. by
Harm
Pinkster
and
Inge
Green
,
229
242
.
Berlin
:
Walter de Gruyter
.
Ross,
John Robert
.
1967
.
Constraints on variables in syntax
. Doctoral dissertation, MIT, Cambridge, MA.
Ross,
John R
.
1969
.
Guess who?
In
Papers from the 5th Regional Meeting of the Chicago Linguistic Society
, ed. by
Robert
Binnick
,
Alice
Davison
,
Georgia
Green
, and
Jerry
Morgan
,
252
286
.
Chicago
:
University of Chicago, Chicago Linguistic Society
.
Sauerland,
Uli
.
1998
.
The meaning of chains
. Doctoral dissertation, MIT, Cambridge, MA.
Smyth,
Herbert Weir
.
1956
.
Greek grammar
.
Cambridge, MA
:
Harvard University Press
.
Sportiche,
Dominique
.
2006
.
Reconstruction, binding, and scope
. In
The Blackwell companion to syntax
,
vol. IV
, ed. by
Martin
Everaert
and
Henk van
Riemsdijk
,
35
93
.
Oxford
:
Blackwell
.
Svenonius,
Peter
.
2003
.
Limits on P: Filling in holes vs. falling in holes
.
Nordlyd
31
:
431
445
.
Svenonius,
Peter
.
2008
.
Projections of P
. In
Syntax and semantics of spatial P
, ed. by
Anna
Asbury
,
63
84
.
Amsterdam
:
John Benjamins
.
Svenonius,
Peter
.
2010
.
Spatial P in English
. In
Mapping spatial PPs: The cartography of syntactic structures
,
vol. 6
, ed. by
Guglielmo
Cinque
and
Luigi
Rizzi
,
127
160
.
Oxford
:
Oxford University Press
.
Tada,
Hiroaki
.
1993
.
A/A-bar partition in derivation
. Doctoral dissertation, MIT, Cambridge, MA.
Takahashi,
Shoichi
.
2006
.
Decompositionality and identity
. Doctoral dissertation, MIT, Cambridge, MA.
Takahashi,
Shoichi
, and
Sarah
Hulsey
.
2009
.
Wholesale late merger: Beyond the A/Ā distinction
.
Linguistic Inquiry
40
:
487
526
.
Uriagereka,
Juan
.
1995
.
Aspects of the syntax of clitic placement in Western Romance
.
Linguistic Inquiry
26
:
79
123
.
Yatsushiro,
Kazuko
.
2001
.
The distribution of mo and ka and its implications
. In
Formal Approaches to Japanese Linguistics 3
, ed. by
María Cristina
Cuervo
,
Daniel
Harbour
,
Ken
Hiraiwa
, and
Shinichiro
Ishihara
,
181
198
.
MIT Working Papers in Linguistics 41
.
Cambridge, MA
:
MIT, MIT Working Papers in Linguistics
.
Zribi-Hertz,
Anne
.
1984
.
Orphan prepositions in French and the concept of ‘null pronoun’
.
Bloomington
:
Indiana University Linguistics Club
.