T-CONTRACTION IN A PHASE-BASED THEORY OF GRAMMAR[1]

Andrew Radford (radford@essex.ac.uk)

              

 

It is well-known that (under specific conditions) finite auxiliaries and infinitival to can contract onto an immediately preceding word in sentences such as:

 

(1)(a)      He’s lying

    (b)       They wanna win the race

    (c)       They are gonna win the race

 

If finite auxiliaries are T(ense) constituents and infinitival to is a nonfinite tense particle of some kind, sentences like those in (1) have in common that they involve contraction of a (finite or infinitival) T constituent onto a preceding host. Accordingly, sentences like (1) and (2) can be seen as instances of a more general phenomenon of T-contraction whereby an affixal (or clitic) T contracts onto a preceding constituent.

 

T-contraction appears to be a phonological operation (involving local PF-cliticisation), since a contracted auxiliary cannot be pied-piped along with its host when its host undergoes syntactic movement, as the sentences in (2) below illustrate:

 

(2)(a)      You think who’s been lying?

    (b)      Who do you think’s been lying?

    (c)     *Who’s do you think been lying

 

So, when the host who undergoes wh-movement in (2), the clitic auxiliary 's cannot be pied-piped along with it, as the ungrammaticality of (2c) shows.

 

A general assumption which has been made about T-contraction is that T can only contract onto a preceding host H if there is no overt constituent intervening between the two. More controversial is the question of whether an intervening empty category blocks T from contracting onto H. There have been numerous claims in the earlier literature to the effect that some kinds of intervening empty category block contraction whereas others do not. For example, Jaeggli (1980) argued that case-marked traces block contraction but other traces do not. This assumption was motivated by contrasts like that in (3) below:

(3)(a)      Who do you wanna beat?

    (b)     *Who do you wanna win the race?

 

The two sentences in (3a/b) would have the respective superficial syntactic structures (4a/b) below[2], if we follow Bresnan (1972) in positing that want always selects a CP complement:

 

(4)(a)      Who do you want [CP t [C ø] [TP PRO [T to] beat t]]

 

    (b)      Who do you want [CP t [C ø] [TP t [T to] win the race]]

 

In (4), the italicised traces of who in spec-CP are caseless[3] and the bold-printed traces carry accusative case, assigned by the transitive verb beat in (4a) and by the transitive null complementiser ø  in (4b). Since the only trace which intervenes between want and to in (4a) is the italicised caseless trace in spec-CP, contraction is correctly predicted to be possible in (3a/4a): but since an additional bold-printed case-marked trace in spec-TP intervenes in (4b), contraction is correctly predicted to be ruled out in (3b/4b).

 

However, there are a number of aspects of Jaeggli’s analysis which prove problematic. One of these is that his claim that case-marked traces block contraction but caseless traces do not amounts to an ad hoc stipulation which lacks explanatory force. The second is that (as Schachter 1984 noted) data such as the following cast doubt on the descriptive adequacy of the claim:

 

(5)      Who do you think’s telling the truth?

 

Whether the complement of think is a TP (as in (6a) below) or a CP (as in (6b) below), the (bold-printed) case-marked trace which is the subject of is will intervene between think and is, as we see from the partial superficial structures for (5) shown in (6) below:

 

 

(6)(a)      Who do you think [TP t [T is] telling the truth]

 

    (b)      Who do you think [CP t [C ø] [TP t [T is] telling the truth]]

 

Hence, contraction will wrongly be predicted to be ruled out in (5) if case-marked traces block contraction.

 

One way of attempting to circumvent this problem is to posit that who in (5) originates (and is case-marked) in spec-VP rather than in spec-TP. On this view, (5) would have the structure (7a) or (7b) below (where bold print indicates case-marked traces):

 

(7)(a)      Who you think [TP t [T is] [VP t [V telling] the truth]]

 

    (b)      Who do you think [CP t [C ø] [TP t [T is] [VP t [V telling] the truth]]]

 

Since there would be no case-marked trace intervening between is and think, contraction would then correctly be predicted to be possible in (5).

 

However, the assumption that subjects are case-marked in spec-VP raises additional problems. For one thing, it leaves us with no account of the ungrammaticality of (3b) *Who do you wanna win the race?, since this will have the simplified structure (8) below (with the subject in spec-VP being case-marked in situ by the null transitive complementiser ø):

 

(8)      Who do you want [CP t [C ø] [TP t [T to] [VP t [V win] the race]]]

 

Since no case-marked trace of who intervenes between to and want, contraction is then wrongly predicted to be permissible in (3b) *Who do you wanna win the race?

 

In addition, the assumption that case-marked traces block contraction but caseless traces do not is problematic from a theoretical point of view. Under the copy theory of movement, traces and their antecedent form a chain in which each link of the chain is a copy of the constituent which originates at the foot of the chain. Hence, if subjects are case-marked in situ in spec-VP, all traces of the moved subject will likewise be case-marked. This is turn means that it is problematic to rely on the distinction between case-marked and caseless traces to determine whether contraction is permissible or impermissible.

 

Given the problems associated with the case-marked trace analysis, it is not surprising that Boskovic (1997) rejects the assumption that intervening traces of any kind block T contraction. Instead, he posits that intervening null lexical items (e.g. a null complementiser or null PRO subject) block T-contraction. This requires him to assume that where the head T of a complement clause contracts onto a matrix V, the complement clause has the status of TP (and not of a CP headed by a null complementiser). Hence (6) Who do you think’s telling the truth? for Boskovic would have the simplified structure (9a) below, not (9b):

 

(9)(a)      Who do you think [TP t [T 's] telling the truth]

 

    (b)      Who do you think [CP [C ø] [TP t [T 's] telling the truth]]

 

If (as Boskovic assumes) intervening traces do not block contraction but null lexical items do, it follows that the intervening trace t in (9a) will not block contraction of 's onto think, but that the intervening null complementiser ø in (9b) will. Since the resulting sentence (6) Who do you think’s telling the truth is grammatical, Boskovic concludes that the complement of think must be a TP rather than a CP.

 

In relation to sentences like (1b) They wanna do it, Boskovic’s assumptions lead him to conclude that the complement of want is a TP with a PRO subject in spec-VP[4] as in (10a) below, rather than a CP with a PRO subject in spec-TP as in (10b):

 

(10)(a)      They want [TP [T to] [VP PRO do it]]

 

      (b)      They want [CP [C ø] [TP PRO [T to] do it]]

 

Under the analysis in (10a), there is no null lexical item intervening between want and to, hence nothing to prevent to from contracting onto want. By contrast, in (10b) both the PRO subject in spec-TP and the null complementiser ø in C block contraction of to onto want, under Boskovic’s assumptions. Since the resulting sentence (2a) They wanna do it is grammatical, Boskovic is forced to opt for analysis (10a).

 

However, there are a number of aspects of Boskovic’s analysis which are potentially problematic. For one thing, no explanation is given for why null lexical items should block contraction, but not null traces: this appears to be an arbitrary stipulation, in that if both are null we should expect neither (or both) to block contraction. Moreover, his assumption that the complement clauses in (6a) and (7a) are TPs is at variance with the assumption in Chomsky (1998, 1999) that all finite clauses are CPs headed by a force-encoding complementiser, and that all control infinitivals are CPs headed by a null C. Likewise, Boskovic’s assumption that control clauses are TPs whose subject remains in situ in spec-VP entails positing that control to lacks an EPP-feature even though finite auxiliaries have an EPP-feature. The assumption that control to has no EPP-feature becomes even more problematic if EPP is associated with agreement (so that e.g. T has an EPP-feature if it agrees with its subject). There is some evidence which suggests that control to does indeed have agreement properties (and agrees with its PRO subject). Theoretical considerations suggest this: if (as in Chomsky 1998, 1999) case is assigned under agreement, then the null case assigned to PRO by control to suggests that to has abstract agreement properties.

 

Moreover, facts about complement ellipsis may lead to a similar conclusion. Lobeck (1990) and Murusagi and Saito (1994) argue that only heads which agree with their specifiers allow ellipsis of their complement: some evidence in support of their claim comes from contrasts such as the following (from Boskovic, 1997, p.12):

 

(11)(a)      John liked Mary and [Peter did ø] too

      (b)     *John believed Mary to know French but Bill believed [Jane to ø]

 

(12)(a)      John’s talk about the economy was interesting, but [Bill’s ø] was boring

      (b)     *A single student came to the class because [the ø] thought it was

                 important

 

(13)(a)      John met someone, but I don’t know [who comp ø]

      (b)     *John thinks that Peter met someone, but I don’t believe [that ø]

 

In (11a), the italicised finite T did agrees with its specifier Peter and hence allows ellipsis of its complement (denoted as ø); in the bracketed ECM clause in (11b), infinitival to has no case/agreement properties, and so does not allow complement ellipsis. In (12a) possessive 's is a determiner (as in Chomsky 1995, p. 263) which agrees with its specifier Bill and hence allows ellipsis of its complement; in (12b) the determiner the has no specifier and hence does not allow complement ellipsis. In (13a) the null complementiser comp agrees (in respect of wh-hood) with its specifier who and allows ellipsis of its complement; but in (13b) the complementiser that has no specifier and so does not allow complement ellipsis.

 

In the light of the claim that only a head which agrees with its specifier allows complement ellipsis, consider sentences such as the following:

 

(14)      I couldn’t marry you even if I wanted to ø

 

Here, control to allows complement-ellipsis. If only a head which agrees with its specifier allows complement ellipsis, this suggests that control to agrees with the PRO subject of its clause, and that PRO is in spec-TP (rather than in spec-VP): and this in turn means that Boskovic’s analysis  of wanna contraction (represented in skeletal form in (10a) above) is untenable.

 

Having looked at two earlier attempts to handle T-contraction, let me now turn to consider how contraction might be handled within the phase-based theory of grammar developed in Chomsky (1998, 1999). Before doing so, I will outline three key theoretical assumptions made by Chomsky in the relevant papers.

 

The first is that syntactic and phonological operations are local: more specifically, they are phase-bound and hence subject to the Phase Impenetrability Condition (which specifies that the domain of a subordinate (strong) phase is not accessible to the head of an immediately superordinate phase); the phases relevant to our discussion here are CP and (transitive) vP. The second is that syntactic movement operations involve two sub-operations of copying and deletion: a moved constituent is first copied into the position to which it moves, and then the original is deleted (i.e. given a null phonetic spellout) at the point in the derivation at which spellout occurs. The third is that ‘Spell-Out is cyclic, at the phase level’ (Chomsky 1999, p.9). More specifically, Chomsky (1999, p.10) posits that in a structure of the form [ZP Z ...[HP a [H YP]]] where HP and ZP are phases and ZP is the phase immediately superordinate to ZP, ‘YP is spelled out at the level HP’, whereas ‘H and its edge a...belong to ZP for the purposes of Spell-Out.’ In addition to these theoretical assumptions, Chomsky makes the further descriptive assumption that all finite and control clauses are CPs, but raising and ECM infinitivals are TPs.

 

Although Chomsky does not discuss the conditions governing phonological contraction phenomena, it seems clear from his overall view of language as a perfect system that any attempt to arbitrarily stipulate that only a subset of empty categories block contraction (e.g. case-marked traces but not caseless traces, or null lexical items but not null traces) would amount to an imperfection. From a strict minimalist perspective, the optimal account of contraction would seem to be to posit that contraction is blocked by overt constituents, but not by null constituents: this would seem to be the minimal conceptually necessary condition on contraction.

 

Given the background assumptions above, let’s turn to consider how we handle a sentence like (5) Who do you think’s telling the truth, which proved problematic for the view that case-marked traces block T-contraction. On the matrix vP phase, (5) will have the simplified structure (15) below[5], if the complement of think is a CP (in keeping with Chomsky’s seeming assumption that all finite clauses are CPs):

 

(15)    [vP who you [v ø] [VP [V think] [CP who [C ø] [TP who [T is] telling the  truth]]]]

 

Given Chomsky’s assumption that in a phase of the form [HP a [H YP]], YP is spelled out at HP, it follows that the trace of who in spec-TP will have been deleted (i.e. given a null spellout) on the complement clause CP phase, and that the trace of who in spec-CP will be deleted on the matrix vP phase. If we suppose that

T-contraction takes place at the point where the host is spelled out, it follows that contraction will apply on the vP phase (i.e. at the point where the VP containing think is spelled out[6]). At this point, both traces of who will be null; and since the head C of CP is also null, it follows that there are no overt constituents separating think from is, and hence that is can contract onto think and be spelled out as `s.  As should be obvious, contraction will also be predicted to be possible if the complement of think is a TP rather than a CP.

 

Now consider how we derive (1b) They wanna win the race. If (as Chomsky posits) control infinitivals are CPs with a PRO subject in spec-TP[7], (1b) will have the simplified structure (16) below on the matrix vP phase[8]:

 

(16)      [vP they [v ø] [VP [V want] [CP [C ø] [TP PRO [T to] win the race]]]]

 

Since there is no overt constituent intervening between want and to, an analysis such as (16) correctly predicts that wanna contraction can apply when VP is spelled out on the vP phase. It goes without saying that the same will be true if the complement of want is TP rather than CP, or if PRO is in spec-vP rather than spec-TP.

An interesting complication which arises in relation to (16) – and indeed (15) – is what happens if want raises from the head V position of VP to adjoin to the light-verb in the head v position of vP. Such raising seems to be possible (though optional), as we see from examples like:

 

(17)(a)      They desperately want to win the race

      (b)      They want desperately to win the race

 

If desperately can function as a VP-adverb, it would seem that want raises from V to v in (17b), though remains in situ in (17a). If want raises from V to v, (1b) will have the simplified structure (18) below on the matrix vP phase:

 

(18)    [vP they [v want+ø] [VP [V want] [CP [C ø] [TP PRO [T to] win the race]]]]

 

Wanna contraction cannot apply at this stage of derivation, since want has been moved out of VP and so cannot be spelled out at this point. On the contrary, want can only be spelled out at the end of the matrix CP phase, when (1b) has the simplified structure (19) below:

 

(19)     [CP [C ø] [TP They [T ø] [vP they [v want+ø] [VP [V want] [CP [C ø] [TP PRO [T to] win the race]]]]]]

 

But on the CP phase, material contained within the matrix VP is no longer accessible, since the Phase Impenetrability Condition specifies that the domain of a subordinate phase is inaccessible on a superordinate phase. Hence to can no longer contract onto want at this point. The more general conclusion would appear to be that wanna contraction can only apply if want remains in situ in the head V position of VP.

 

Now consider how we derive (3a) Who do you wanna beat? On the matrix vP phase, (3a) will have the simplified structure (20) below, if control clauses are CPs whose PRO subject is in spec-TP (as assumed by Bresnan (1972), Chomsky (1981) and Snyder and Rothstein (1992)):

 

(20)      [vP who you [V ø] [VP [V want] [CP who [C ø] [TP PRO [T to] beat]]]]

 

On the matrix vP phase, the trace of who will be deleted, and VP will be spelled out. Since there is no overt constituent between want and to, T-contraction can apply, ultimately deriving wanna. As should be obvious, the same will be true if the PRO subject is in spec-vP.

 

Now consider the more problematic question of how we account for the ungrammaticality of contraction in (3b) *Who do you wanna win the race? If we suppose that want always selects a CP complement and that infinitival to always has an EPP feature, (3b) will have the simplified structure (21) below on the matrix vP phase:

 

(21)      [vP who you [v ø] [VP [V want] [CP who [C ø] [TP who [T to] win the race]]]]

 

But the difficulty posed by the analysis in (21) is that on the vP phase, VP will be spelled out and all traces of who will be deleted. There will therefore be no overt constituent intervening between want and to, and we wrongly predict that wanna contraction is possible[9].

 

One way of overcoming the problems posed by the analysis in (17) is to suppose that structures of the form want someone to do something are ECM structures in which want selects a TP complement, and want case-marks the infinitive subject. Let’s suppose that in such structures, the infinitive subject who (by virtue of being an object) raises to spec-VP (an A-position) and then (by virtue of being a wh-expression) to spec-vP (an A-bar position), and that the verb want raises to v. Some evidence in support of the claim that the object raises to spec-VP in ECM structures comes from the distribution of adverbs in sentences such as:

 

(22)(a)      They wanted him desperately to win the race

      (b)     *They wanted desperately him to win the race

 

If desperately can function as a VP-adverb, the fact that both the infinitive subject him and the verb want in (22a) are positioned in front of desperately suggests that the subject raises to spec-VP and the verb to v[10].

 

Given the assumptions made here, (3b) will have the simplified structure (23) below on the matrix vP phase[11]:

 

(23)      [vP who you [v want+ø] [VP who [V want] [TP (who) [T to] win the race]]]

 

By this stage, all traces within VP will have been deleted, so that at first sight it might seem as if wanna contraction is wrongly predicted to be able to apply here. However, this is not the case, since want has been raised out of the complement VP into the head v position of VP, and so cannot be spelled out until the matrix CP phase – i.e. until we have derived the simplified structure in (24) below:

 

(24)    [CP Who [C do] [TP you [T do] [vP who you [v want+ø] [VP who [V want] [TP (who) [T to]

 

              win the race]]]]]

 

But by the time we reach the CP phase in the derivation, material contained within the matrix VP (i.e. within the domain of the light-verb v) is no longer accessible, given that the Phase Impenetrability Condition specifies that material within the domain of a subordinate phase is not accessible on a superordinate phase: consequently, to cannot contract onto want at this stage.

 

Under the analysis presented here, the verb want in structures such as want someone to do something selects a TP rather than a CP complement. Some evidence in support of the TP complement analysis comes from contrasts like that below:

 

(25)(a)      What John wants is [to win the race]

      (b)     *What John wants is [Mary to win the race]

 

As (25) shows, the infinitive complement of want can appear in focus position in a pseudo-cleft sentence when want has a control complement, but not when it has an ECM complement. If both bracketed complements in (25) are CPs, the contrast between them is inexplicable: for example, we can’t attribute the ungrammaticality of (25b) to case failure if the bracketed complement is a CP headed by a null transitive complementiser which assigns accusative case to Mary; nor can we attribute the ungrammaticality of (25b) to a requirement for a null C to be properly governed (as in Stowell 1981) or to cliticise a superordinate head which selects the relevant CP as its complement (as in Pesetsky 1995) if both bracketed complements in (25) are CPs headed by a null C. By contrast, if the bracketed complement in (25b) is an ECM complement and hence a TP, we can account for the ungrammaticality of (15b) in a number of ways. One possibility would be a structural constraint to the effect that CP (but not TP) can occur in focus position in pseudo-cleft sentences. Another possibility would be a case-failure account, under which the subject Mary in spec-TP lacks a case-assigner. A further possibility would be an EPP-violation, if to only has an EPP feature (and permits a subject in spec-TP) when it agrees with its subject (i.e. in control structures, but not in raising or ECM infinitives)[12],[13].

 

Now let’s consider how we derive a sentence like (1c) They are gonna win the race. If (as Chomsky assumes) an intransitive vP is not a phase, spellout of constituents within the matrix VP will be delayed until the matrix CP phase – i.e. until the point where we have formed the simplified structure (25) below:

 

(25)     [CP [C ø] [TP They [T are] [vP [v ø] [VP [V going] [TP (they) [T to] they

 

            win the race]]]]]

 

If to has no EPP-feature in raising structures, there will be no trace of they intervening between going and to to block gonna contraction; if to does indeed have an EPP feature, contraction will nonetheless not be blocked, since the trace they in spec-TP will be erased on the matrix CP phase and hence not block contraction.

 

The goal of this paper has been to show that we can arrive at a principled account of T-contraction in a phase-based theory of grammar in which we make the minimum conceptually necessary assumption that contraction is blocked only by an overt intervening constituent. Within such a framework, it is entirely unnecessary to make arbitrary stipulations about the kinds of empty categories which do and don’t block T-contraction[14].

References

Boskovic, Z (1997) The Syntax of Nonfinite Complementation: An Economy

   Approach, MIT Press

 

Bresnan, J W (1972) Theory of Complementation in English Syntax, MIT PhD diss,

 

Chomsky, N (1981) Lectures on Government and Binding, Foris, Dordrecht

 

Chomsky, N (1995) The Minimalist Program, MIT Press, Cambridge Mass

 

Chomsky, N (1998) Minimalist Inquiries, MIT Working Papers in Linguistics

   no. 15

 

Chomsky, N (1999) Derivation by Phase, MIT Working Papers in Linguistics

 

Jaeggli, O (1980) ‘Remarks on to contraction’, Linguistic Inquiry 11: 239-246

 

Kayne, R S (1984) Connectedness and Binary Branching, Foris, Dordrecht

 

Koizumi, M (1995) Phrase Structure in Minimalist Syntax, MIT PhD diss.

 

Lobeck, A (1990) ‘Functional heads as proper governors’, Proceedings of NELS

   20: 348-362

 

Murusagi, K & Saito, M (1994) ‘Adjunction and cyclicity’, Proceedings of the 13th

   West Coast Conference on Formal Linguistics, pp. 302-317

 

Pesetsky, D (1995) Zero Syntax: Experiencers and Cascades, MIT Press,

   Cambridge Mass

 

Radford, A (1997) Syntactic Theory and the Structure of English, Cambridge

   University Press

 

Schachter, P (1984) ‘Auxiliary Reduction: An argument for GPSG’, Linguistic 

   Inquiry 15: 514-523

 

Stowell, T (1981) Origins of Phrase Structure, PhD diss, MIT

 

 


[1] I am grateful to Bob Borsley for helpful comments on an earlier draft of this paper. This version appeared in Essex Research Reports in Linguistics (2000), vol 33, pp.21-32.

[2] Throughout this paper, only partial structural representations are shown: those aspects of structure which are not relevant to the discussion at hand are omitted in order to simplify exposition.

[3] A minor complication here is that Kayne (1984) argues that a transitive verb can assign case to a constituent in the specifier position of its CP complement, e.g. in structures like someone whom I believe is innocent. For me, at any rate, whom is also possible in structures where believe is replaced by an intransitive adjective like sure, as in

someone whom I am sure is innocent, so calling into question Kayne’s assumption about the nature of the who/whom contrast. For speakers like me, it would seem that whom is always an option when extracted out of the CP in which it originates (and hence, in some sense, whom can serve as a marker of long-distance movement).

[4] Boskovic assumes that verb phrases have a simple VP structure rather than the split vP/VP structure posited in Koizumi 1995 and Chomsky 1995.

[5] Simplified by ignoring the internal structure of the vP containing telling the truth. Here, as elsewhere, I do not show the structure of the complement of the clitic T, unless this is relevant to how contraction applies.

[6] A potential problem arises if the morphological form of think is not be determined until vP merges with T.

[7] Perhaps because control to agrees with its PRO subject, and agreement is an EPP-feature.

[8] Since want does not appear to case-mark any constituent here (unless it case-marks the null C heading CP), questions arise about whether the vP containing want is transitive, and hence whether it is a phase.

[9] For speakers of so-called ‘liberal’ dialects who allow contraction in cases like Who d’you wanna win the race, a CP analysis may be appropriate – though other possibilities also need to be explored.

[10] An apparent complication is posed by the possibility of saying They desperately wanted him to win. This suggests that desperately can modify vP if V raises to v.

[11] The parentheses around the trace of who in spec-TP indicate that this trace will only be present if to has an EPP-feature. If EPP is tied to agreement, then it seems likely that to will have an EPP-feature in control infinitives (by virtue of agreeing with its PRO subject), but not in raising or ECM infinitives.

[12] The analysis of want as an ECM predicate raises the question of why want (unlike believe) does not allow its complement subject to passivise: cf.

      (i)       He was believed by everyone to have resigned

      (ii)  ?*He was wanted by everyone to resign

It may be a lexical idiosyncrasy of want that it does not freely passivise when it has a clausal complement, even in non-ECM structures such as:

      (iii)  ?*It hadn’t been wanted by anyone for him to resign

[13] The assumption that to has no EPP-feature in clauses other that control clauses has interesting implications for the syntax of for-to infinitives, since it means that the subject cannot be in spec-TP. Adapting the split cP/CP analysis of for-to infinitives proposed in Radford (1997, pp.449-450), we might suppose that for originates in C and raises to c, and that the subject raises from spec-vP into spec-CP, so that the for clause in I want very much for him to win the race  would have the simplified structure:

      (i)     [cP [c for] [CP him [C for] [TP [T to] [vP him win the race]]]

[14] One residual problem which remains is accounting for the uncontractibility of has in sentences like:

     (i)      He has/*He’s his car serviced regularly

Since he occupied the matrix spec-TP position here, contraction cannot apply until the matrix CP phase, at which point we have the simplified structure (ii) below:

    (ii)      [CP [C ø] [TP he [T ø] [vP [v ø] [VP [V has] his car serviced regularly]]]]

If has remains in situ in V, it will be inaccessible to operations which apply on the matrix CP phase, and hence will not be able to contract onto he (on the assumption that contraction applies on the minimal phase containing the host). However, such an analysis raises the question of why has does not raise to v, and how the nominal his car is case-marked. A further residual problem relates to have contraction, which seems to be somewhat different from T contraction, in that the host must always be an immediately preceding pronoun with which have agrees in some respect. See Radford 1997, pp. 329-332 for one possible analysis.