The dog tried to bark.
Abrams persuaded the dog to bark.
It is easy for the dog to bark.

‘Control’ (also know as ‘equi’) refers to a class of predicates that take at least one individual type argument and at least one clausal argument and furthermore have the property that the individual is also an argument of the predicate inside the clausal argument. This phenomenon lexically-specific, that is, it is mediated by a class of lexical items that, in the composition, build the link between the individual and the embedded clause. The class of lexical items has many sub-classes, characterized by part of speech (verbs and adjectives) and argument structure patterns. While most of the argument structure variation is only syntactic (e.g., PP v. NP complements; base form VP complements v. to-marked VP complement v. progressive VP complements), there is still some variation at the semantic level. Control is distinguished from other configurations in which the same individual appears as the argument of multiple predicates by the relationship between the predicates themselves.

Shared argument is subject/ARG1 of matrix clause:

• Kim tried to leave.

• Kim refrained from laughing.

• Kim will try and find it. ;; Note that this use of ‘and’ is treated as semantically empty.

Shared argument is non-subject of matrix clause:

• Kim invited Sandy to stay.

• Kim imagined Sandy taller.

Control adjectives:

• Kim is anxious to leave. ;; Shared argument is ARG1

• It is easy for Kim to leave. ;; Shared argument is ARG2, clausal argument is ARG1

Because the shared argument and clausal argument can each fill different argument positions of the control predicate, and similarly the shared argument can have different roles in the embedded clause, if the fingerprints had to specify precise argument roles (ARGn), we would have many different sub-phenomena here. Instead, we make use of underspecified argument positions (see ErgSemantics/Conventions).

[ARG0 e1, ARG1+ x2, ARG1+ h3]
h4:[ARG0 e5, ARG1+ x2]
{ h3 =q h4 }

• v_np-vpslnp_oeq_le is a lexical type that does both control and lexical binding of a long-distance dependency (akin to tough-adjectives).

  • I have boxes to get rid of.

At first glance, control relations are not obviously a semantic phenomenon in their own right, but rather a sub-case of predicate-argument linking (i.e. just part of the basic components of semantic analysis). What distinguishes control from other types of predicate argument linking is the particular configuration of argument sharing: the ‘matrix’ or ‘upstairs’ predicate takes both the ‘downstairs’ predicate and one of its arguments as arguments. This is in contrast to the syntactically similar relation called ‘raising’. Raising predicates also relate one of their syntactic arguments to an argument position in another, but in this case the shared argument has no semantic role to play with respect to the matrix verb:

• Kim seems to like sushi.

• Sandy expected Kim to like sushi.

In these examples the shared argument (Kim) is semantically only an argument of the embedded verb (like). In our MRS representations, there is no connection between Kim and seems or expected; from a semantic point of view, raising is unremarkable.

Because raising and control are syntactically very similar, however, it is often not obvious which is in play in any given example. Sag et al (2003, Ch 12) list four tests for distinguishing them, all of which turn on the semantic difference (whether or not the shared argument has a role to play with respect to the matrix verb): (i, ii) Raising, but not control, predicates can accept the expletives it and there in the shared argument position provided the embedded predicate selects for that expletive; (iii) Raising, but not control, predicates can accept idiom chunks in the shared argument position, provided the embedded contains the rest of the idiom in an appropriate configuration; and (iv) It is possible to create paraphrases by passivizing the embedded predicate (and thus promoting a different argument into the shared position) with raising, but not control predicates. There is the added complication that any given lexical form might in fact represent more than one lexical entry (and thus be ambiguously both raising and control), although such cases of ambiguity appear to be rare.

The ERG recognizes a very broad range of lexical types which involve control phenomena. The shared argument may appear as a subject, direct object or object of a preposition. The embedded predicate may be verbal or not, if verbal a to-infinitive, a base form VP, an -ing form VP, or a passive, and may likewise fill different argument positions. Finally, both adjectives and verbs instantiate control relations.

• Are there any three-place adjectives where the clausal argument is ARG3 and the shared argument is ARG2?
• Are there any two-place verbs where the clausal argument is ARG1 and the shared argument is ARG2?
• Controlled ARG4?
  • A team member tried to be traded to the Giants for a pitcher.

• 1212

• Sag, I. A., Wasow, T., and Bender, E. M. (2003). Syntactic theory: A formal introduction (Second edition). Stanford: Center for the Study of Language and Information.

• ErgSemantics main page

• Inventory of semantic phenomena (to be) documented

• How to cite this work