LFG and psychological explanation

Abstract

The publication of The Mental Representation of Grammatical Relations (hereafter, MRGR)' is an important event for linguists and all concerned with the psychology of language. It provides the most definitive statement to date of Lexical-Functional Grammar (LFG), a theory of natural language grammar which is claimed to generalise across radically different language types, to be recursive, learnable, realistically parsable and producible, and to provide "a stronger basis than ever before for a psychologically realistic theory of grammar". The present paper begins by briefly reviewing the nature of the psychological problem, before outlining LFG as a theory of grammar in section 1.2 Section 2 then considers the broader psychological claim. Grammars of natural languages are psychologically problematic in at least two ways. The first is a major concern among linguists: natural language syntax is complicated by the inclusion of various startling discontinuous and fragmentary constituents, which occur for example in relative clauses and "reduced" coordinate sentences, Other less surprising constructions still tend to be characterised by a puzzling complexity in the relation between surface grammatical functions like subject and object and underlying semantic or thematic categories like agent and patient. Natural languages are also characterised by a surprising degree of syntactic ambiguity. In particular, points of "local" syntactic ambiguity, where it is locally unclear which of two or more analyses should be followed, are widespread, a fact which has led psychologists and com putational linguists to postulate processing regimes embodied in various algorithms and heuristic "strategies". The psychological implications of these two types of untoward complexity are serious, both for learning and processing. The two problems of grammar and local ambiguity resolution are logically quite separate. The former is a (partial) specification of WHAT is computed in natural language comprehension, whereas the latter relates to HOW it is computed. Since it is obvious that the space of algorithms for computing the relevant class of functions is very much larger than that class itself, there is an equally obvious methodological priority across the two problems: whether we count ourselves as psychologists or linguists, until the problem of grammar has been solved, we are unlikely to make much headway with the other problems of processing, such as local