(1) Mary defended herself/*yourself.
Principle A guarantees that a word like herself or yourself needs to have a c-commanding antecedent that's close enough to it. In (1), the only potential antecedent is Mary, and since Mary is 3rd person and feminine, the anaphor it binds will need to be herself, not yourself.
On the other hand, consider examples like (2):
(2) Defend yourself!
Where's the binder for the anaphor yourself? One possibility, of course, is that we've gone wrong somewhere, and anaphors don't always need binders. But another possibility is that there actually is a subject in (2)--in fact, we would need the subject to be you, since it's binding yourself. To say this, we'd need to be willing to say that subjects can sometimes be unpronounced.
In fact, positing an unpronounced subject in (2) would be useful, not only for binding theory, but also for the EPP; we know that the TP in (2) ought to have a specifier. So now we have two reasons to think that it does have one, though the specifier of TP happens to be unpronounced.
We find these kinds of null subjects in a variety of places. Consider the pair of examples in (3-4):
(3) [Defending yourself] would be a good idea
(4) [Defending you] would be a good idea
The sentences in (3-4) are both fine, but differ in who's going to do the defending. We can count on the binding theory to get us this difference in meaning, as long as we're willing to say that defending has an unpronounced subject in both examples. In (3), the unpronounced subject of defending needs to be you, so that yourself can have a binder, satisfying Principle A. In (4), the unpronounced subject can be anything other than you, so that the pronoun you can be free, satisfying Principle B. So your defender could be I, for example, or he.
These null subjects raise a lot of questions. For one thing, what determines when subjects can be null? English doesn't ordinarily let you fail to pronounce subjects, though there are certainly languages that do.
There is one class of infinitival clauses that very routinely seem to have null subjects, not just in English but in most languages. Consider the pair of examples in (5-6):
(5) John forgot that he ate the beans
(6) John forgot to eat the beans
In (5), there are three NPs, John. he, and the beans. These are selected, respectively, by matrix v, the v of the embedded clause, and the embedded verb ate.
(6) doesn't mean the same thing as (5), but the selection relations all ought to be the same. In other words, we expect there to be theta-roles for three NPs in (6), just as in (5). There are only two visible NPs. As usual, we could conclude that our theories need work--in this case, our theory of selection--but an alternative, which the discussion above might make us feel more comfortable with, is to posit a phonologically null subject for the embedded clause of (6), which is selected by the v of the embedded clause. The null subject is referred to as PRO. PRO is required to refer back, in this case, to John; we say that John controls PRO. The semantic relation between PRO and its controller is an interesting topic in its own right, but one that we'll mostly leave aside.
If we analyze (6) as control, we are left with a problem: how do we distinguish control from raising?
(7) John seems to like the beans
The two sentences in (6-7) look pretty similar on the surface. But the claim we're considering is that they have different syntactic structures. In (6), John gets its theta-role in the matrix clause, while in (7) John is raising out of the embedded clause into the matrix subject position:
(The trees above contain another claim: a control infinitive is a CP, and a raising infinitive is a TP. We'll see some evidence for that later on.)
There are various tests which allow you to determine whether a sentence involves control or raising. The four tests given below are all intended to let you determine whether a theta-role is being assigned in the matrix clause or not; if there's an upstairs theta-role, then you have the control structure in (8), and if there isn't, then the subject in the matrix clause got there via raising, as in (9).
|
Idioms We first encountered idioms as one of the arguments for the existence of A-movement. The parts of idioms seem to need to start out quite close to each other--but they can be separated by A-movement, as in examples like (11-12):
(10) The cat is out of the bag. The idea here is that the cat in (11-12) isn't in the position where it's pronounced because it's selected by seems, or by the v associated with seems--rather, the cat starts out in the mostly deeply embedded clause, and is moved away from the rest of the idiom (both so that the cat can be assigned case and so that the matrix TP will have a specifier, obeying the EPP). On this account, we expect the corresponding examples to be impossible if we switch from raising to control: (13) *The cat forgot to be out of the bag The sentence in (13) is grammatical, but not as an idiom; it can only be a sentence about a literal, forgetful cat. On the account of control being outlined here, then this is what we expect; in (13), the cat begins the derivation in the matrix clause, getting a theta-role from the v associated with forget. This means that the cat can't be a metaphorical cat; only real cats can forget things. The parts of the idiom are never anywhere near each other, unlike in (10-12), and the idiomatic reading is therefore impossible. Expletives Another way to convince ourselves that a position isn't associated with a theta-role is to see whether an expletive can appear there. As we already know, in sentences containing raising predicates, expletives are possible: (14) It seems [that John likes the beans] In (14) there's no raising going on, since the embedded clause is tensed, and we can see from the presence of the expletive that there's no theta-role for the subject of seem. This tells us, again, that in sentences like (15), John raised into the position where it was pronounced, after getting its theta-role in the embedded clause: (15) John seems to like the beans What distinguishes a control verb like forget from a raising verb like seem, we've been saying, is that forget does have a theta-role for its subject. So there shouldnŐt be anything corresponding to (14) if the verb of the matrix clause is forget, and that seems to be right: (16) It forgot [that John likes the beans] The sentence in (16) is grammatical, but it can't be interpreted as an expletive; it has to refer to some inanimate object. So here's another way of determining whether a theta-role is assigned in the main clause, and hence whether an infinitive is a raising infinitive or a control infinitive Complement drop It's sometimes possible for the clausal complement of a verb to be dropped:
(17) John forgot [that he's allergic to beans] The sentence in (18) is a complete sentence; it just means that John forgot something salient. We could think of forget as a verb that optionally selects a clausal complement (we've seen cases of optional selection before; eat, for example, optionally selects an NP object). If that's the right way to think about (18), then we can also see in (18) that John gets a theta-role in the matrix clause; even if the embedded clause is gone, John is still there. We wouldn't expect examples like (18) to be possible with raising verbs, and that seems to be right: (19) *John seems. In (19), John has no source for a theta-role, since the embedded clause is gone. Equivalence under passive Here are two sentences that mean more or less the same thing:
(20) A specialist seems to have examined John. Since seem is a raising verb, it makes sense that these sentences should have similar meanings. The only difference between them is whether the verb of the embedded clause has been passivized or not; the subject of the embedded clause, whether it's the agent of examine or the patient, then undergoes raising into the matrix clause. In other words, these sentences have similar meanings for the same reason that the sentences in (22-23) do:
(22) A specialist examined John. By contrast, consider the pair of examples in (24-25):
(24) A specialist forgot to examine John. These sentences don't mean the same thing at all (though they are alike in that the specialist didn't end up examining John). The difference between them has to do with who was forgetful, a specialist or John. And since (24-25) are examines of control, we can understand why this is: there's a theta-role being assigned in the main clause, that of the agent of forget, and this theta-role is going either to a specialist or to John. In both cases, the subject of the embedded clause is a null PRO, controlled by the matrix subject. |
Now, the only control verb we have seen so far is forget. But there are plenty of control verbs. Here are two more:
(26) I promised John to defend myself.
(27) I persuaded John to defend himself.
Both promise and persuade are control verbs, which can select for infinitival clauses with PRO subjects. PRO then refers either to the subject (as in (26)) or the object (as in (27)); how to determine what PRO will refer to is an open research question.
"Object control" examples like (27) look a lot like ECM:
(28) I expected John to defend himself
But in (27), John is getting a theta-role from persuade, while in (28), John gets Case but not a theta-role from expect. The tests developed above for distinguishing control from raising confirm this:
|
Idioms (29) *I persuaded the cat to be out of the bag. (has no idiomatic reading) (30) I expected the cat to be out of the bag.
Expletives
Complement drop
Equivalence under passive |
Finally, let's look at some evidence that control infinitives and raising infinitives are of different categories: raising infinitives (and ECM infinitives) are TP, while control infinitives are CP. Consider the following examples of control infinitives:
(37) John is eager to leave.
(38) I would like very much to leave.
(39) I arranged to leave early.
The infinitives in (37-39) are all control infinitives; spend some time making sure you can convince yourself of that. Now, consider (40-42):
(40) John is eager for Mary to leave.
(41) I would like very much for Mary to leave.
(42) I arranged for Mary to leave early.
These sentences are all grammatical, and they're loosely based on (37-39), though of course they don't have the same meanings. So now let's think about the boldfaced strings. What is for Mary?
We might be tempted at first to think that this string is a PP; it's a P followed by an object which is presumably getting both a theta-role and Case from P, or maybe from functional heads near P. But in fact, there are good reasons to think that Mary isn't getting a theta-role from for; her theta-role is from leave. Consider data like the following:
(43) I arranged for it to be obvious [that Syntax is fun].
(44) I would like very much for the cat to be out of the bag.
The boldfaced NPs in (43-44) can't be getting a theta-role from for. In (43), it is an expletive, so it isn't getting a theta-role at all. In (44), the cat is part of an idiom, so again it shouldn't be capable of getting a theta-role from for.
So the NPs following for in (40-44) are not the complements of for. Instead, we can analyze for as a C; it takes the infinitive as a complement, and is responsible for assigning Case to the NP after it (which doesn't have any other sources for Case, after all):
There are probably also cases where we want to posit a null version of for. Consider an example like (46):
(46) I would like Mary to leave.
In (46), it's tempting to think of Mary as getting Case from like; that is, we could think of this as an ECM construction, like (47):
(47) I believe Mary to be a genius.
But there's an important difference between (47) and (46). In (47), if you passivize the verb, Mary becomes the matrix subject:
(48) Mary is believed to be a genius.
That's what we expect; if you passivize the verb that was assigning Mary Case in (47), Mary will head elsewhere in search of Case.
But this isn't what happens in (46):
(49) *Mary would be liked to leave.
If like in (46) were the source of Case for Mary, like believe in (47), we'd expect passivizing like to be a way to deprive Mary of Case and force it to move elsewhere, which is what happened when we passivized believe in (48). But that's not what's happening in (49).
An alternative analysis: what's assigning Case to Mary in (46) is not like, but a null version of for. Notice that for can also be overt:
(50) I would like for Mary to leave.
Control infinitives like the ones in (37-39), then, have alternatives with either overt or null complementizers. A possibility is that control infinitives are just always CP (often with null C).
Interestingly, raising and ECM infinitives have no alternatives with for:
(51) *It seems for Chris to like Syntax.
(52) *I believe for Mary to be smart.
So here's a reason to think that raising and ECM infinitives are TPs, while control infinitives are CPs.
All of this does leave us with an unanswered question: what determines the distribution of PRO? We've seen that PRO can be the subject of a control infinitive. It can't just be anywhere:
(53) *John thinks that I like PRO.
We'll talk more about this in class.