Control verbs differ as to which argument controls the understood subject of an infinitive. For a verb like promise it is the subject, for a verb like persuade it is the (in)direct object, for a verb like signal to it is the prepositional object, etc.
It is proposed that this is represented in the slot realisation part of the syntax frame. The slot realisation part of an infinitival clause (or perhaps also, for certain languages, for finite clauses) can contain the attribute controlled-by. This attribute takes disjunctions of slot-identifiers as its possible values.
In the following examples a single frame is given. The slots in this frame are identified by non-negative integers. After each slot identifier, each slot realisation is contained in square brackets. Not all attributes are represented.
Certain verbs allow either the subject or the complement to control
an infinitival complement without a difference in meaning for the
verb. Some examples from Dutch are Ik stelde hem voor/
bood hem aan/ suggereerde hem weg te gaan `I
proposed/ offered /suggested to him to leave'.
For such cases we need disjunctive values such as controlledby =
0 
1, etc. Thus propose might be represented as follows:
1,...],
[cat=NP,gf=obj,...],...
If a difference in controller goes together with a different meaning of the verb, two different lexical items must be postulated. This is the case e.g. for French dire:
| Jean | a | dit | à | Paul | être | très | fatigueé |
| Jean | has | said | to | Peter | be | very | tired |
| `Jean told Paul that he(=Jean) was very tired' | |||||||
| Jean | a | dit | à | Paul | de | partir |
| Jean | has | said | to | Paul | to | leave |
| `Jean told Paul to leave' (i.e that Paul should leave)' | ||||||
An additional distinction must be made to distinguish a verb like try from a verb such as decide. For a verb such as try the controller must be the subject in all cases, but for a verb such as decide it can be the subject or the by-phrase (even when it is absent) or the understood subject can be interpreted as involving a group containing the referent of the subject or the by-phrase, or completely arbitrary. The difference can be illustrated in passivised structures, cf.
| * | It was tried (by the committee) to leave. |
| versus | |
| It was decided (by the committee) to leave | |
It is proposed that within EAGLES such theory-dependent notions are not introduced. Instead, the exact complementation properties and control properties of each individual lexical item are enumerated in full, at the cost of redundancy. However, if two frames which specify such complementation and control properties are related, they are put together in the same set of frames. In this manner, it is indicated that these frames are indeed related and not just two arbitrary variants.
The same approach is taken to Bach's Generalisation (i.e. with certain verbs, absence of the object when it functions as a controller is impossible), and Visser's Generalisation (certain verbs cannot be passivised if their subject functions as a controller). These generalisations are not incorporated themselves, but different frames resulting from them are put together in one set of frames to express their relatedness. Thus, though there is a need to stipulate different frames under these assumptions in certain cases, their relatedness is indicated.
It is proposed that if the understood subject can be interpreted as involving a group containing the referent of some argument, then this is specified by adding an asterisk after the relevant slot-identifier in the value of the controlled-by attribute.
It is also assumed that, if the controller is optional, it is specified as being an optional argument in the same frame. In this situation, it is not possible to describe the behaviour of the verb by means of two frames, since the controller cannot be specified correctly then.
If the choice of controller is not determined by properties of the lexical item itself, nothing is specified for the attribute controlled-by. French falloir (e.g. il faut faire ça) might be an example of such a case, since adding an overt controller yields ill-formed or marginal results: ?? il me faut faire ça.
It is well-known that a characterisation in terms of slot identifiers or
grammatical
functions in itself is insufficient to fully characterise all control
possibilities. In particular, cases of `switch of controller' cannot
be dealt with adequately. However, it appears to us that a lexical
specification as given above is sufficient. It is possible to formulate
general rules which account for `controller switch' provided that a
specification of the controller in terms slot identifiers is
available. Such general rules will determine whether the controller
switches and where it switches to by taking into account the lexical
specifications plus properties of the embedded infinitival. We have in
mind examples such as Kim promised Sandy to be allowed to
attend the party (control by Sandy instead of by Kim), Dana
asked Pat to be allowed to attend the party (control by Dana instead
of by Pat), Sandy was promised to be allowed to attend the party
(control by the subject of passivised promise, instead of
ill-formedness
as in
* Sandy was promised by John to take care of himself).
We thus suggest that nothing has to be specified lexically for
such cases.