We have argued that all finite clauses are CPs, and we went on to argue that for infinitives with accusative subjects and control infinitives with null PRO subjects are likewise CPs. These two assumptions lead us to the more general conclusion that:

And indeed this is an assumption made by Chomsky in recent work. However, there is one particular type of clause which is exceptional in that it lacks the CP layer found in canonical clauses – namely infinitival complement clauses like those bracketed in (76) below which have (italicized) accusative subjects:

Complement clauses like those bracketed in (76) are exceptional in that their subjects are assigned accusative case by the transitive verb (believe/intend) immediately preceding them: what’s exceptional about this is that the verb is in a different clause from the subject which it assigns accusative case to. For this reason, such clauses are known as exceptional case-marking clauses (or ECM clauses); and verbs (like believe) when used with an ECM clause as their complement are known as ECM verbs.

ECM complement clauses seem to be TPs which lack the CP layer found in canonical clauses, and for this reason Chomsky (1999) terms them defective clauses. One reason for thinking that the bracketed ECM clauses in sentences like (76) are not full CPs is that they cannot readily be coordinated with for-infinitives, as we see from the ungrammaticality of (77) below:

Although (for speakers like me) the verb intend can take either a bare ECM infinitive complement or a for infinitive complement, the fact that the two cannot be conjoined suggests that the bare ECM infinitive clauses have the status of TPs while for-to infinitive clauses have the status of CPs. By contrast, coordination is indeed possible in sentences like:

and this is because both bracketed clauses in (78) are infinitival TPs.

Further evidence that ECM infinitive clauses like those bracketed in (76) are TPs rather than CPs comes from the fact that they cannot occur in focus position in pseudo-clefts, as we see from the ungrammaticality of the sentences below:

If ECM clauses are TPs, this follows from the restriction noted in (75) that only CPs (not TPs) can occur in focus position in a pseudo-cleft sentence. Moreover, a further property of sentences like (76) which would be difficult to account for if the bracketed complement clause were a CP is the fact that its (italicized) subject can be passivized and thereby made into the subject of the main clause, as in (80) below:

This is because it is a property of the subject of an infinitival CP complement clause like that bracketed in (81a) below that its subject cannot be passivized – as we see from the ungrammaticality of (81b):

Likewise, the subject of the infinitival CP complement of a for-deletion verb like want cannot be passivized either:

– and indeed this is precisely what we expect if the subjects of CPs cannot passivize, and if the bracketed complement clauses in (82) are CPs headed by a null counterpart of for. However, the fact that the passive sentences in (80) are grammatical suggests that the bracketed complement clauses in (76) are TPs rather than CPs (since the subject of an infinitival TP can be passivized, but not the subject of an infinitival CP). Hence, complement clauses like those bracketed in (76) above are defective clauses which have no CP layer, and (76a) They believe him to be innocent accordingly has the structure (83) below:

The particular aspect of the analysis in (83) most relevant to our discussion is the claim that the complement clause him to be innocent is an infinitival TP headed by to, and its subject him is assigned accusative case by the transitive verb believe: how this happens.

We can extend the analysis of ECM predicates like believe which select a bare infinitive complement. On this view, a sentence like I have never known him be rude to anyone would be analyzed as containing a transitive perfect participle known which selects a TP complement headed by a null counterpart of infinitival to – as shown in skeletal form in (84) below:

Since the subject of a TP complement can passivize, the analysis in (84) predicts that the subject of the bracketed infinitive complement in (84) can passivize, and this is indeed the case as we see from examples like (85) below:

Because infinitival to can only have a null spellout when the TP complement it heads is the complement of an active transitive verb-form like the perfect participle known in (84) and not when the relevant TP is the complement of a passive participle like known in (85), it follows that infinitival to must be given an overt spellout in sentences like (85).

Under the analysis proposed here, verbs which take a bare infinitive complement with an accusative subject are analyzed as ECM predicates which select a TP complement headed by an infinitival T which has an overt spell out as to in passive structures like (85) and a null spell out in active structures such as (84). However, one predicate which is problematic to classify in such terms is let, since it allows a bare infinitive complement in active structures like (86a) below but doesn’t allow the subject of the infinitive to passivize, as we see from the ungrammaticality of sentences like (86b):

We can’t describe the relevant facts by saying that let is a defective verb which has no passive participle form, since let is used as a passive participle in sentences like The prisoners were let out of jail. An alternative analysis is to suppose that whereas typical ECM predicates select an infinitival TP complement in both active and passive uses, let is irregular in that it only selects an infinitival TP complement in active uses, not when used as a passive participle. Similar lexical idiosyncrasies are found with a number of other verbs: for example, know only allows a bare infinitival complement with an accusative subject when used as a perfect participle in structures like (84) above. (An alternative way of accounting for the impossibility of passivisation in sentences like (86b) which we won’t adopt here is to take let to be a verb selecting a CP complement headed by an inherently null complementizer which in turn selects an infinitival TP complement headed by a null counterpart of infinitival to: the ungrammaticality of (86b) then follows from the impossibility of passivizing the subject of a CP complement.)