As in GB, subcategorisation in LFG is based on a syntactic representation of predicate argument structure. In addition, the notion of grammatical function occupies a central role in determining which of the arguments semantically selected by a predicate are syntactically realised and how.
In Bresnan (1982), grammatical functions are defined as universal syntactic primitives of the grammar and classified according two main parameters: subcategorisability and semantic restrictedness. Subjects, objects and sentential/VP-complements are subcategorisable functions in that they can be assigned to the arguments of lexical items. Nonsubcategorisable functions correspond to adjunct phrases, and -- as the term suggests -- may not be associated with the arguments of lexical items. The subcategorisability of grammatical functions such as Topic and Focus is subject to parametric variation and distinguishes between subject-oriented and topic-oriented languages.
Subcategorised functions may differ with respect to the range of argument types with which they can be associated. Grammatical functions which are not inherently tied to specific selectional restrictions are semantically unrestricted, while those which can only be paired with arguments of specific semantic types are semantically restricted. For example, the subject function can be linked to any thematic role and can also occur as a non-thematic function when encoding the subject-subcategorisation of raising verbs such as `seem' and `appear', while oblique functions are always thematic (i.e. they are never associated with pleonastic elements) and are generally more sensitive to selectional restrictions (they are inherently tied to a thematic role (Rappaport, 1983)).
In addition, complements and adjuncts may occur as closed or open functions. Closed complement and closed adjunct functions (COMP, ADJ) are assigned to clausal expressions which have a controller of their own, e.g. the emboldened NPs in (13) and (14):
|(13)||John believes [that Bill is a genius ]|
|(14)||[John being angry], Mary left|
Clausal expressions which are assigned the open complement and open adjunct functions (XCOMP, XADJ) are instead controlled from without, as in (15) and (16):
|(15)||John wants [to be a genius ]|
|(16)||[Being angry at John], Mary left|
A schematic picture of the grammatical functions adopted in LFG is given in table 3.1.
|Unrestricted||Restricted||ADJ (controlled from within)|
|SUBJ||OBL||XADJ (controlled from without)|
|OBJ||COMP (controlled from within)|
|OBJ2||XCOMP (controlled from without)|
Grammatical relations are associations of grammatical functions with thematic roles or with non-thematic values. These associations are encoded in the lexicon, where each verb is represented as a lexical form consisting of a predicate argument structure and a grammatical function assignment. The predicate argument structure of a lexical form is a list of the arguments for which there are selectional restrictions. The grammatical function assignment of a lexical form is a list of its syntactically subcategorised functions.
|a.||predicate argument structure:||breakagent, theme|
|b.||grammatical function assignment:||((SUBJ), (OBJ))|
Biuniqueness of Function-Argument AssignmentsGrammatical function assignment lists serve as subcategorisation frames. Subcategorisation is checked in Functional Structure for Completeness and Coherence (Kaplan & Bresnan, 1982). Completeness ensures that all subcategorised arguments are present in functional structure (e.g. it rules out sentences like * John devours, * eats a cookie), while Coherence restricts the occurrence of subcategorisable grammatical functions to those listed in the verb's lexical form (e.g. it rules out sentences like * John arrives Bill).
G = g,...,g is a possible grammatical function assignment to P(1,...,m) if and only if the mapping from 1,...,m to G defined by i g is injective (one-to-one and into). (Bresnan (1982, 163))
More recently, the theory of grammatical relations in LFG has been revised to make grammatical relation assignment and changing monotonic (Levin, 1987; Bresnan & Kanerva, 1988; Bresnan & Moshi, 1989; Alsina & Mchombo, 1988). Grammatical functions such as SUBJ, OBL, etc., are no longer viewed as atomic specifications, but are defined in terms of more primitive functional features. The emerging theory, Lexical Mapping Theory, consists of four basic components, which we now explain.
The hierarchy includes the following roles in descending order: agent, beneficiary and maleficiary, recipient and experiencer, instrumental, patient and theme, locative, motive:
Syntactic functions are decomposed according to the features [+/- r] (thematically restricted or unrestricted) and [+ o] (objective or not) as follows:
Individually, each value for the two features [+/- r] and [+/- o] defines a partial grammatical function as indicated below.
Semantic roles are associated with partially specified grammatical functions according to the following Lexical Mapping Principles:
After mapping principles have applied, any remaining underspecified grammatical function is fully instantiated. This instantiation is free as long as Biuniqueness (17) and the Subject Condition (18) are observed.
|(17)||Biuniqueness of Function-Argument Assignments|
|Within the predicate-argument structure of a lexical form,|
|there is a one-to-one relation between grammatical functions and arguments.|
|(18)||The Subject Condition|
|Every lexical form must have a subject.|
As an illustrative example, consider the treatment of passive proposed in Bresnan & Kanerva (1988). Prior to passivisation, the theme and agent roles of a verb such as find are intrinsically associated with partially specified grammatical functions, as shown in (19).
As indicated above, the passive rule introduces the functional specification [+r] (i.e. thematically restricted) for the highest role of a lexical form, , as seen in 20:
When passive applies to the predicate argument structure in (19), the agent role acquires the specification [+r] which in conjunction with [-o] defines an oblique function. The agent argument of a passive verb is thus realised as an oblique complement, while the theme can be either subject or object. Well-formedness constraints induced by the subject condition require that the subject option be chosen in this case. Example (21) below provides a schematic representation of the process described as a whole.
Finally, control relations with raising and control verbs are handled lexically with reference to grammatical functions. For example, subject control with both raising and equi verbs would be stated in the lexicon as indicated by the functional control equations in the relevant parts of of the lexical entries for seem (22) and try (23)
Since control is dealt with lexically and no empty categories are used to bind the complement subject, it follows that both raising and equi verbs subcategorise for VPs rather than sentences as in GB.