On the Syntax of Headless XPs
Sun-Woong Kim
Kwangwoon University
1. Introduction
This study explores the syntax of headless/beheaded XPs which mean XPs whose head is eliminated either by movement or by ellipsis(/deletion). The aim of this study is twofold. First, this study tries to provide a deeper generalization of the syntactic constraints on movement and ellipsis of headless XPs in terms of the ban on X stranding. Second, this paper also tries to account for the ban in terms of the dynamic(/contextual) definition of phases (den Dikken 2007 and Bošković 2012a, 2012b, among others) in contrast to Chomsky's (1995 through 2008) rigid definition. To be concrete, this paper tries to answer the following question: Why is it that headless TPs, headless VPs, headless DPs, and other headless XPs in general cannot be moved or elided. To take an example, consider the following:
(1)a. Gesternhat [TPeinVertretertthatden Kai angerufen].
(German,Wurmbrand 2004)
yesterday has a salesman-nom the Kai-acccalled
'Yesterday a salesman called Kai.'
b.*Gestern[TPeinVertretertthatden Kai angerufen]
yesterdaya salesman-nom the Kai-acccalled hat wahrscheinlichtTP.
hasprobably
'Yesterday a salesman probably called Kai.'
(2)a. Mary saw someone but I don't know who <TP Mary saw twho>.
(Lasnik 1999)
b. * Mary saw someone but I don't know who did <TP Mary tT see twho>. (> for elided part)
In (1a), the embedded TP has its T moved from inside T to C. In other words, the embedded clause has become a headless TP. If this TP moves as shown in (1b), the whole string turns ungrammatical. (2a) is a typical example of sluicing in which the embedded wh-object of the second conjunct has been moved out of TP before ellipsis. If the head T is extracted out of TP, then the TP becomes literally headless. If headless XPs is banned to be elided, the ungrammaticality of (2b) comes from the ban on eliding headless XPs.
This paper proposes that the syntax of headless XPs can be taken care of under the dynamic definition of phases in terms of Phase Transfer, the Extended Phase Impenetrability Condition (EPIC), the Anti-locality Condition, and their cooperation. As consequences of the proposals, this study applies them to various headless XPs including NP, PP, VP, CP other than TP.
2. Headless XPs
2.1. Movement
As early as 1990's "beheaded" TPs were reportedly immobile (den Dikken and Naess 1993, Wurmbrand 2004). Consider the following regarding headless TPs, repeated from (1):
(3)a.Gesternhat [TPeinVertretertthatden Kai angerufen].
yesterday has a salesman-nom the Kai-acccalled
'Yesterday a salesman called Kai.'
b.*Gestern[TPeinVertretertthatden Kai angerufen]
yesterdaya salesman-nom the Kai-acccalled hat wahrscheinlichtTP.
hasprobably
'Yesterday a salesman probably called Kai.'
Once its head has moved out, TP cannot move to the front before gestern 'yesterday'.[1] Headless VPs also cannot move if their head is removed. Compare (4b) and (4c):
(4)a.John gave a book to Mary.
b.It's a book that John [VP gave ta book to Mary].
c.*It's a book to Mary that [TP John gave ta booktto Mary].
(5)*It's [VPa book tgave to Mary] that John gave tVP.
If only the object moves to the focus position, the sentence is of no problem as a cleft sentence (4b). If the object and the indirect object stay in the focus position as in (4c), the string turns out ungrammatical. If the representation of (4b) is (5), in which VP has moved with its head removed, then it is ungrammatical for the same reason why (3b) is ungrammatical.That is, it is ungrammatical because a headless XP has moved. Headless DPs show the same behavior with respect to preposition stranding.
(6)German
a.Von wasiredestdu ti?
fromwhat talk you
'What are you talking about?'
b.*Wasiredest duvon ti?
whattalk youfrom
'What are you talking about?'
(7)English
a.*Whichlakei did the boy scouts camp near ti?
b.*Whatcircumstancesi did the moon implode under ti?
c.*Whichcityi did George Washington sleep in ti?
d.cf. Whichbedi did George Washington sleep in ti? (Takami 2010)
As is well-known, German does not allow its preposition stranded as shown in (6b). Although English is one of a handful of languages which allow preposition stranding, it is not quite freely allowed as shown in (7a) through (7c). English does not allow preposition stranding if it is due to extraction out of adjunct PPs. If P attracts D, D raises up to P. If this is the case, the DP turns into a headless DP, which is expected not to be allowed to move. This is why (7c) is ungrammatical.[2]
(8)
2.2. Ellipsis
Let us get to ellipsis examples regarding headless XPs. As has been discussed, headless XPs are non-elidable in the same way as they are immobile. First consider headless TPs.
(9) a.Mary saw someone but I don't know who <TP Mary saw twho>.(Lasnik 1999) (=2)
b.* Mary saw someone but I don't know who did <TP Mary tT see twho>.
As was introduced in section 1, (9a) is a typical example of sluicing in which only a wh-phrase is extracted, followed by TP deletion. (9b) is bad since it involves ellipsis of a headless TP.
Headless VPs also demonstrate that headless XPs are non-elidable. Consider the following examples:
(10)a.John solved the problem, and Mary solved the problem, too.
b.John solved the problem, and Mary, too.
c.John solved the problem, and Mary did, too.
d.*John solved the problem, and Mary solved, too.
(11)a.John solved the problem, and Mary solved the problem, too.
b.John solved the problem, and Mary <vP solved the problem>, too.
c.John solved the problem, and Mary did <VP solve the problem>, too
d.*John solved the problem, and Mary solved <DP the problem>, too
Compare (11c) with (11d). (11c) is of no problem in that the VP of the second conjunct is deleted under identity with the first conjunct. What is wrong with (11d)? For this, this paper assumes that what is deleted is a remnant VP from which the head verb has been extracted. This is shown in (12) below:
(12)*John solved the problem, and Mary [vP solved <VPtsolve the problem>], too.
In (12), solved has moved out of VP. If this VP is later deleted with its head removed, the whole string turns ungrammatical. Headless DPs do not allow ellipsis, either. Consider the following examples of preposition stranding (P-stranding):
(13)a.?The girls talked to their teacher and the boys talked to, too.
b.*The girls camped around this lake and the boys camped around, too.
(14)a.?The girls talked to their teacher and the boys talked [PP to [DP their D teachers]], too.
b.*The girls camped around this lake and the boys camped [PParound+D<DP this tD lake>], too.
As is well-known, some PPs including adjuncts do not allow P-stranding in English. This is understood as another example of non-elidability of headless XPs if the structure is something like (14b). In (14b), the head D is extracted out of D to P resulting in a headless DP. If this DP is elided, then the whole string turns out ungrammatical in contrast to (14a) where no D raising occurred. In (14a), DP in the second conjunct is a headed DP.
The discussion so far leads us to a quite plausible generalization that headless XPs cannot be moved or elided. The real question is why they are immobile and non-elidable.
2.3. A Deeper Generalization
The question is why headless XPs arebanned in such a way? This study tries to capture a deeper generalization about the immobility and non-elidability of headless XPs from a new point of view: X0-stranding.[3] For this let us start with Law's (2006) proposal regarding P-stranding. He proposes the D to P incorporation analysis ofP-stranding. According to him, D can incorporate with P inside PP. If D is raisedup into P as shown in ①, then the extraction of NP is not possible, as shown in②.
(15) …[PPP+Di [DP …ti NP]]
①
②
D to P incorporation is morphologically supported by P+D suppletive forms in various languages.
(16)a.German: am = andem 'at/by the-mas/neu-dat', ans = an das'at/by the-neu-acc', etc.
b. French: au = a le 'to the', aux = a les 'to the', etc.
c. Italian: al = ail 'to the', alla = a la 'to the', etc.
d. English: *to-the,*at-the,*by-the…
Differently from German, French, and Italian, English does not have morphological forms of P+D as shown in (18d). This derivation leads to the failure of P-stranding in English. He does not, however, explainwhy D to P incorporation bleeds NP movement out of PP. The present study suggests that the deeper generalization has something to do with stranding of P(+D) (Kim 2010).
As is well-known, P-stranding is not allowed except in Modern English and a few otherlanguages. This study extends this to a more general ban on head (or X0-)stranding. To be more specific, the generalization is that if the head of an XP is removed, mainly by extraction, the XP cannot be moved or elided with its head left behind.
(17) FP
Xi+FXP
In (17), if X moves up to a higher head F, then the remnant XPbecomes headless. They cannot be moved or elided due to X+F stranding. The real question, however, is once again, why Xi+F stranding in (17) is banned? More generally, why is a moved head cannot be stranded?
3. Dynamically Defining Phases
To answer the question, this paper resorts to the notion of phases and their role in constraining syntactic operations.
3.1. Rigid Definition
One big assumption of the generative grammar is that movement and other grammatical operations as well are constrained by locality. The quintessence of locality is the notion of phase as originally proposed by Chomsky (1986, 1995, 2001, 2008). His notion of the phase is actually an evolved version of barriers which is in turn based on the notion of cyclicity. He still keeps his phase as vP, CP (strong phase) intact.
3.2. Dynamic Definition
The present study, in contrast, bases its theoretical apparatus on the dynamic/contextual definition of phases. The original notion of Chomsky (1995, 2008) has been criticized by many authors to be too rigid. They argue the notion of phases must be contextually defined. Two of such proposals are worth remarkable. One is den Dikken's (2006, 2007) phase extension proposal and the other is Bošković's (2012a, b) highest-phrase-as-a-phase proposal. Both proposals suggest that the rigid definition of phases is conceptually unsupported and empirically inadequate. Den Dikken (2006, 2007) suggests that the basic ingredient of a phase is not proposition but predication, which forms an inherent phase. This can be extended by raising the phase head up to a higher functional head. Bošković (2012a, b) takes a stronger stance in extending the possible range of phasehood. He assumes that every maximal projection is potentially a phase and claims that the highest phrase in the lexical domain counts as a phase.
(18)Different views to phasehood
Definition / Author / Basics / ExamplesRigid / Chomsky (1986) / Blocking category / Barrier
Chomsky (1995, 2008) / Proposition / vP, CP (strong phase)
Contextual / den Dikken (2006, 2007) / Predication / Inherent phase and its extensions
Bošković (2012a, b) / Every maximal projection / The highest phrase in the lexical domain
The ban on P-stranding in English can be explained in terms of phase impenetrability if the dynamic definition of phases. Take for example den Dikken's (2007, 2008) phase extension:
(19)Syntactic movement of the head H of a phase α up to the head X of the node β dominating α extends the phase up from α to β: α loses its phasehood in the process, and any constituent on the edge of α ends up in the domain phase β as a result of Phase Extension.
Now we are ready to answer the question why P-stranding is not allowed after D is incorporated into P. This is because PP becomes a phase due to phase extension in the sense of den Dikken(2006). This is schematically represented below in (20):
(20)a.[DPøSUBJECT D [øPREDICATE]](Φ = phase)
Φ
b.[PPP+Di [DPøSUBJECTti [øPREDICATE]]]
Φ (Φ)
Granting that DP is intrinsically predicational in the sense of Chomsky(1970), it is a phase in the sense of den Dikken(2006) (Bošković2002, Yang 2006, Chomsky 2006, for this possibility). By raising D, a phase head, to P, phasehood is extended to. If this is the case, no DP is predicted to be extractable out of PP since it is a phase. Since the DP is not at the edge of PP, the PIC would preventextraction from a non-edge position. There is no way for DP to get out of PP in (20b). If D does not incorporate into P, DP is free to move since PP above it is not a phase.
On a little different track, in a series of works, Bošković (2008, 2012a, 2012b, among others) argues that the highest projection XP in the extended projection of a major phrase counts as a phase. Being a phase, the XP enter into a couple of important syntactic operation: ellipsis and movement. Only a phase or the complement of the phase head can be elided or moved. What must be clarified in the present context is the notion of the highest phrase. Here is a set of examples that show the relevant point regarding the present paper.
(21)a.*Betsy must have been hassled by the police, and Peter must <have been being hassled …>.
b.Betsy must have been hassled by the police, and Peter must have <been being hassled …>.[4]
c.Betsy must have been hassled by the police, and Peter must have been <being hassled …>.
d.*Betsy must have been hassled by the police, and Peter must have been being <hassled …>.
Regarding this diverse pattern of VP-ellipsis, he proposes the following structure of the middle field (or the extended domain of VP):
(22)… [TP must [VPf1 have [AspP1bei+en [VPf2ti [AspP2ing [VPf3 be [VP hassle …[5]
Since AspP1 is counted as the highest projection within the VP domain, it works as a phase. The prediction is that AspP1 will be an eligible unit for ellipsis but VPf1 won't. This explains the contrast of (22a) and (22b). VPf2 is also predicted to be elided in that it is the complement of AspP1. This explains the grammaticality of (22c). (22d) is ungrammatical since the elided part is neither a phase nor its complement.
One more example concerns Serbo-Croation. Bošković (2012b) notices that genitive complements of nouns cannot be extracted in SC. This improves, interestingly, with an addition of a quantifier/numeral.
(23)?*Ovog studentaisampronašla[NP slikuti]
thisGENstudentGENam found pictureACC
‘Of this student I found the picture.
(24)Ovogstudentaisampronašla [mnogo/desetslikati]
thisGENstudentGENam found many/ten picturesGEN
Addition of QP over NP improves grammaticality of (23) to (24). This is because QP becomes a phase while NP loses its phasehood.
3.3. Proposed Analysis
This study looks for an answer to this question in the dynamic definition of phases. In a nut shell, this paper claims that headless XPs are neither mobile and nor elidable because of the condition which bans extraction or ellipsis of the complement of a phase head which is stated as follows:
(25)Extended Phase Impenetrability Condition (EPIC)
Syntactic operations (move and ellipsis) are constrained in such a way thatin phase α with head H, the domain is not accessible to operations outside α, only H and its edge are accessible to such operations.
In (17), XP which was a phase by definition, transfers its phasehood to a higher functional projection in its domain, which is FP here. If FP is a phase, then XP becomes its complement, which is banned from being moved or elided by EPIC. This is why headless XPs cannot enter into any syntactic processes. Along with the EPIC, Anti-locality is assumed in this study.
(26)Anti-locality Condition
Movement must not be too local (Grohmann 2003) so that there is no such movement from complement to its spec (Abels 2003).
To recap: This study assumes den Dikken and Bošković’s spirit--dynamic definition of phases along with the EPIC and the Anti-locality. Furthermore, this paper also assumes movement as copy. That is, ellipsis is copy deletion, so that movement and ellipsis go together. However, den Dikken's terminology "extension" is no good in that an ex-phase loses its phasehood after extension andBošković's proposal is too strong in that it has no consistency in phase enlarging. See (24), for example. Instead this paper makes a new proposal: Phase Transfer.
(27)Phase Transfer
Phasehood is transferred to a higher FP if the head X of a phase moves up to F; The XP ceases to be a phase by transferring its phasehood to FP.
All in all, this paper claims that headless XPs are neither mobile and nor elidable because of the condition which bans extraction or ellipsis of the complement of a phase. By EPIC, XP which is a phase by definition, transfers its phasehood to a higher functional projection in its domain, which is FP here. If FP is a phase, then XP becomes its complement, which is banned from being moved or elided by EPIC. This is why headless XPs cannot enter into any syntactic processes.
(28)
FP
Xi+FXP
Headless XP cannot be moved or elided due to Xi+Fstranding. X0 (or more exactly Xi+F) cannot be stranded with its headless XP left behind due to a cooperation of EPIC and Anti-locality since the headless XP is the complement of a transferred phase.
As consequences of the phase transfer proposal, this study explains syntactic properties of other headless XPs including CP and NP as well as TP and PP.
4. Consequences
4.1. More XPs: NP and CP
Now let us consider more headless XPs. According to Abels(2003), CPs cannot move with its head left behind.
(29)Passivization
a.Everybody believes fervently that John is a fool.
b.*[John is a fool] is believed fervently [that] by everybody.
(30)Topicalization
a.That John is a fool, Mary told herself at least twice a day.
b.*[John is a fool], Mary told herself [that] at least twice a day.
The question is why C is banned from stranding though TP does not have its head removed. For this paper proposes that this can be explained under the dynamic definition of phases if the following structure is assumed for declaratives:
(31)[ForcePForce+that [CPtthat [TP…
If ForceP is postulated above CP, ForceP becomes a phase by raising that to Force head. Then CP loses its phasehood and becomes the complement of Force+that. It is predicted that headless CP is immobile due to EPIC and Anti-locality. This is borne out by examples given in (29) and (30).
Specific NPs renders supportive evidence for the proposal of this paper. As is widely known, extraction out of a nominal is banned if the nominal is definite/specific.
(32)a.Who did you read [a book about twho]?
b.*Who did you read [the book about twho]?
c.*Who did you read [that book about twho]?