Preliminary Recommendations

The distinction between raising and control verbs

To our knowledge, all theories agree that raising and control verbs differ at least semantically. Though these verb types may perhaps have the same syntactic properties, a raising verb takes one complement or subject which is not semantically an argument of this verb.

This can be illustrated with the verbs try and seem: they may both take a subject and an infinitival complement with to; however, with try the subject phrase is semantically an argument of the verb, but with seem it is not.

This can also be illustrated with the verbs persuade and believe: they may both take a subject, a direct object and an infinitival complement with to; however, with persuade the direct object is semantically an argument of persuade, but with believe the direct object is not semantically an argument of this verb.

If this semantic difference were the only difference between control and raising verbs, there would be no need to distinguish them syntactically. Only their semantic specifications would differ.

However, many theories claim that the distinction between control and raising verbs is not only semantic in nature, but also syntactic. This is true for GB, for HPSG, for GPSG and perhaps even for LFG.

There is interesting evidence from Icelandic that the distinction between control and raising verbs must indeed be represented syntactically. The evidence is basically the following. Consider verbs which combine with a subject and an infinitival complement, and take as the infinitive a verb which requires quirky case on its subject. Here, the subject of the infinitive cannot occur. We then have structures of the form: NP V Vinf ....

The question is now

-- What is the case of the subject NP?

The answer is

-- The case of this NP is determined by V if V is a control verb, but it is determined by Vinf (the case its subject would get) if V is a raising verb.

A similar example can be constructed for object-control and object-raising verbs. Consider verbs which take a subject, a direct object and an infinitival complement, and again take an infinitival verb which requires `quirky' case on its subject. We then have structures of the form: NPsubj V NPobj Vinf ....

Now the question arises

-- What case does NPobj bear?

The answer is

-- Its case is determined by V if V is a control verb, but it is determined by Vinf if V is a raising verb.

Since many theories require a syntactic distinction between control and raising verbs, it is proposed that the distinction between control and raising verbs is obligatorily indicated in the syntactic representation in the EAGLES standard. Theories which do not require this distinction can simply ignore this information.

The next question is: how do we represent the difference between control and raising verbs syntactically. Each theory has different mechanisms for this, but EAGLES should restrict itself to a simple distinction between control verb and raising verb by means of an attribute which can take two values. Let us call the attribute CONTROL, and let it have two possible values (for the moment), viz: control-verb, and raising-verb. It will be slightly refined below.