Predictive Parsing

Abstract

Predictive parsing is an approach to natural language analysis based on the use of powerful programs associated primarily with individual lexical items which embody expectations about the form and content of the subsequent sentence/text input. The expectations are intended to determine the analysis of further inputs by providing semantic structures into which they must fit. (These structures may be those of preference semantics <190>, conceptual dependency <36>, or, by generalisation, scripts <223>.) The essentially word-driven basis, and emphasis on semantic rather than syntactic information and processing, of this style of parsing makes it predictive in a very different sense from that of classical top-down syntactic parsing; and allowing the exploitation of any type or piece of information in a word program offers great flexibility. The price includes a lack of generality, and overlaps or gaps between individual programs. The approach is therefore best suited to domainlimited language processing.