Subcategorisation is concerned with the lexical specification of a predicate's local phrasal context. In general, this specification involves reference to the semantic arguments of the predicate. For example, the verb devour subcategorises for two noun phrases insofar as, semantically, it can be characterised as the set of all participant pairs which stand in the devouring relation. Indeed, subcategorisation is often contrasted with collocation which describes the preferential or statistically salient local phrasal contexts of a word which are not readily expressible in terms of semantic selection -- e.g. nominal compounds, adjective-noun pairs.
Semantic selection, however, is neither a sufficient nor a necessary condition for subcategorisation. It is in fact customary to include, in the subcategorisation frame of a predicate, phrases whose occurrence is obligatory in the local phrasal context of the predicate but are not semantically selected by it. Extraposition as well as subject and object raising verbs are the most well-known cases of predicates subcategorising for phrases which they do not select semantically. For example, the subject of an extraposition verb such as bother in (1) is a pleonastic element and cannot therefore be regarded as being semantically selected:
(1) | It bothers Bill that Mary left |
Similarly, a raising verb cannot be regarded as semantically selecting its subject simply because the semantic restrictions on this subject are solely determined by the complement verb. For example, the subject of a raising verb such as seem is either contentful or pleonastic (2) according to the complement's subject subcategorisation, e.g.
|
(2) | Mary seems to be happy |
| It seems to bother Bill that Mary left |
Analogous remarks apply to object raising verbs, as in (3):
|
(3) | John believes Mary to be happy |
| John believes it to be possible that Mary left |
Subcategorisation is thus concerned with the specification of phrases which are salient to the local context of a predicate because they are selected by the predicate both syntactically and semantically or simply syntactically. Although all linguistic theories have means to express (morpho)syntactic aspects of subcategorisation, direct reference to semantic selection can only be expressed in those frameworks which include a level of semantic representation (e.g. UCG or HPSG). Nevertheless, because of the importance of semantic selection in subcategorisation, frameworks such as GB or LFG which do not include a level of semantic representation provide a level of linguistic description which expresses the semantic argument structure of predicates in syntactic terms.
Linguistic theories differ in the amount of information which is provided in the subcategorisation frame of a verb. This is mostly due to diverging dispositions to using syntactic rules and principles to express linguistic generalisations, with a consequent shift of emphasis away from or towards lexical specification. For example, control links which obtain with raising and equi verbs can be included in the subcategorisation frame of the relevant predicate as in HPSG/UCG/LFG, or handled in terms of syntactic principles or rules as in GB.