In all the approaches reviewed, subcategorisation involves an argument structure providing a specification of the participant structure of a predicate, and a linking specification which determines the functional/syntactic realisation of participant roles of the predicate.
In GB, LFG and DUG, argument structure is defined as a level of
syntactic description; in HPSG and UCG, argument structure is part of
the semantic representation of predicates. Thematic role annotations
are used in GB, LFG and UCG to rank participants; this ranking
determines the syntactic functionality of participants. In GB, the
external/internal argument distinction is crucial in the determination
of the structural realisation of participant roles. This distinction
is not semantically motivated; it is simply assumed as a lexical
specification. In LFG, a Universal Thematic Hierarchy is assumed. In
the 1994 version of HPSG, there is no ranking of participant
roles. In UCG,
ranking is
semantically expressed in terms of Dowty's prototypical roles
augmented with the notion of semantic restrictness used in LFG. No
ranking of thematic roles is used in DUG.
In GB, LFG, HPSG and UCG, grammatical relations are viewed as links
between (syntactically) selected constituents and roles. In
GB, the thematic functionality of argument structure participants is
directly projected in syntactic structure. Grammatical relations such
as subject and objects are expressed as predication and government
relations between nodes in tree structures. In LFG, HPSG and UCG,
grammatical relations are specified in the lexicon. In LFG, this
specification is given by the Lexical Mapping Principles which governs
the linking of
roles to primitive grammatical function
features in lexical forms. In both HPSG and UCG, arguments are
syntactically ranked according to the obliqueness hierarchy which
reproposes the functions subject, object and indirect object in terms
of relative position in the subcategorisation list. In DUG, linking is
expressed lexically as elementary form-function units; grammatical
functions are viewed as primitives.
Further differences concern the treatment of control and extraposition structures with clausal complements. In GB, both equi and raising verbs subcategorise for a sentence. With equi and subject raising verbs, control is expressed syntactically. The complement subject of equi verbs is an empty category which is syntactically controlled according to the binding theory. With subject raising verbs, the complement subject moves into the matrix subject position to receive case leaving behind a coindexed empty category. Object raising verbs are treated as a case of straight sentential complementation, i.e. no control is involved, since the complement subject is allowed to receive case in situ by the matrix verb. In LFG, HPSG, UCG and DUG equi and raising verbs subcategorise for a verb phrase; control is lexically specified for all, including object raising verbs. A common trait to all is the need to specific the control propensity of a verb (e.g. subject or object control), which is done lexically. Finally, extraposition verbs include lexical specification of the pleonastic subject / object NP in LFG, HPSG, UCG and DUG, but not in GB where the pleonastic NP is inserted in syntax following movement of the argment which is amenable to extraposition.
Table 3.2 provides a schematic summary of the comparison drawn above.