Incremental interpretation using a tree description language

Abstract

This paper briefly reports on the implementation of a variant of the LDSNL framework described by Meyer Viol (this volume) and Kempson et al (this volume) and proceeds to a programmatic discussion of proposed future developments, bringing out some consequences of the methodological goal of incrementality combined with the assumption that the single level of representational structure required is a (decorated) binary function-argument tree with a type-logical 'backbone'. The current system has been implemented in Sicstus Prolog with the following results: (i) translation of natural language input into formulas of the epsilon calculus; (ii) correct predictions for a broad range of 'crossover' data in English questions and relative clauses; (iii) runtime options for interaction of 'gap' construal and pronoun resolution producing consistent results for clear cases and varying results for marginal cases, (such as those involving resumptive pronouns) offering a possible way of modelling between-speaker differences; and (iv) 'incremental' determination of quantifier scope in simple sentences via online choice of term dependencies. The parser is implemented as a state transition system with parse states represented as sequences of unit clauses in the Prolog database². A fuller description appears in (Kibble et al. forthcoming).