topics: Arbitrary PRO | Goal Control | Subject Control
Control is a semantic and grammatical relationship that holds between a NP and an understood but invisible anaphor The invisible anaphor is called PRO and it has special properties.Consider the following sentence which is repeated in bracketed clause form:
(1) John tried to go.
(2) [ John tried [ to go ]].
To go is a clause--an infinitival clause. Infinitives mark several types of mood. The exact mood of the infinitival complement of try is uncertain.
Control is the relationship between a specific theta role and the null subject of an infinitive or gerund. The theta role acts like an antecedent, and the null subject is an anaphor.
The verb go assigns the theta role agent to animate subjects. However, there is no over subject in the infinitival clause. Note the following sentence:
(3) [ John tried [ to shave himself ]].
The anaphor is bound within its own clause (the so-called clausemate condition). It has been proposed that the missing subject is the form PRO:
(4) [ John tried [PRO to shave himself ]].
Here, PRO is an agent. It turns out that it is also an antecedent, the antecedent of the anaphor himself. One of the unique properties of PRO is that is functions either as an anaphor or as a pro nominal, but always as one of them.
Here, it is an anaphor. The clausemate condition does not pertain to subject anaphors, but the anaphor must be bound in the next highest clause:
(4) Bill thinks John tried PRO to shave himself.
Here, John is the only antecedent for PRO; Bill is not a possible antecedent.
Example (3) is reanalyzed with PRO functioning as the subject of the infinitival clause:
(5) [ John tried [ PRO to shave himself ]].
The anaphor himself is bound to PRO, but PRO is an anaphor and has no reference. Hence, PRO must be bound to an anaphor which is John..
Note that (5) is ungrammatical:
(5) *John said that Mary tried to shave himself.
(6) *John said [ that Mary tried [ PRO to shave himself ]].
Here himself is bound to PRO.
However, it not quite as simple as it appears above. Mary controls PRO. PRO cannot agree both with its controller and with the anaphor that is bound to it if they have different agreement features.
There are constructions where PRO is not controlled. In this case it acts like a pronominal. PRO is this case is called arbitrary PRO:
(7) It is unclear what to buy for dinner.
(8) It is unclear [ PRO what to buy for dinner ].
Here there is no antecedent for PRO. PRO is interpreted as someone indefinite, though the context of the sentence may determine an antecedent. PRO is not arbitrary here--it has an antecedent:
(9) The lost hikers know that in order to feed themselves they may have to eat unfamiliar things.
(10) The lost hikers know that [ in order [ PRO to feed themselves ]] they may ...
Here, the anaphor themselves is bound to PRO, but know is not a controlling verb. Although PRO is bound to hikers it is two clauses down from its antecedent--a property of pronominals, not anaphors.
Control is partially syntactic in nature and partially semantic--theta roles (see theta.roles) play a role. The following type called object control is semantic only. A better name is goal control since it is the goal that controls PRO:
In both of these sentences Mary is the controller--that is, Mary is the goal, and it is the goal that controls PRO. Mary is the direct object in (1) and the subject in (2). The same holds in the following sentences:
Chomsky is the goal and hence it controls PRO; it is Chomsky who is to abandon Move Alpha.
Notice the difference between control constructions and raising constructions:
Try assigns two arguments: an agent JOHN, and the infinitival clause "to write an essay".
Seem assigns only one argument: the infinitival clause "to write an essay." Semantically, JOHN is an agent assigned by WRITE, and AN ESSAY is a theme assigned by WRITE. The logical structure of the control clause is the following ignoring the operators under TENSE:
The logical structure of the raising clause is:
|first argument (an infinitive)|
<constructum: AN ESSAY>
<constructum: AN ESSAY>
|first argument of WRITE|
|second argument of WRITE|
|second argument of TRY|
In natural languages lexical predicates, which Chomsky calls R-expressions, cannot bound to an antecedent. If the second JOHN has the same referential index as the first one, a violation occurs. If the second JOHN does not have the same referential index as the first one, the sentence passes, though one does not know which JOHN each clause refers to. In ordinary speech, the appropriate thing to do is to identify both JOHNs: John Smith, John Jones. However, the meaning of the TRY construction is that it is the same JOHN. Note if the embedded agent is another person, the sentence fails:
To get around this many languages have developed a strategy whereby a proform is generated in the place of the second <JOHN>. The proform is PRO, which has special properties:
PRO is technically a grammatical form; it not a lexical form. PRO has no inherent meaning, though it does have features. PRO cannot be marked for Case. And PRO cannot be governed. PRO here is an anaphor that requires an antecedent. The antecedent of PRO must c-command PRO and there must be no potential antecedent that would be 'closer' to PRO.The the above sentence, JOHN c-commands PRO. PRO is coindexed with JOHN, giving the intended reading.
In the raising construction, JOHN occurs only in the embedded sentence. if PRO occurred there, it would have no antecedent.Without an antecedent PRO fails.In the raising construction JOHN needs Case and raises first to the subject NP in he embedded clause. But the unnamed mood cannot assign Case to John. John must raised again where it will be assigned Case by the indicative mood ([-Irreal]).
The next type of control is called subject control.
To return to course outline Click here.