A Distributed Word-Based Approach to Parsing

Abstract

People have an incredible facility for organizing and selecting word senses in arriving at the intended meaning of sentences in context. The process takes place with such unnoticed and subconscious ease that it has been often overlooked in the study of language. To those building computer programs to understand language, however, this phenomenon represents a central problem. While many syntactic constructions and conceptual relations can be described through systems of rewrite rules, the word sense selection problem remains. The reason for this lies in the incompatibility of the sense discrimination problem and the rule-based problem-solving method.

The computational theory of Word Expert Parsing approaches natural language understanding as a non-uniform distributed process of interacting words. An expert process for each word actively pursues its intended meaning in the context of other word experts and real-world knowledge. The theory perceives understanding as a behavior of memory interactions, and the computer model emphasizes process rather than output structures.

The Lexical Interaction Language (LIL) formalizes the interactions among individual word experts, and the Sense Discrimination Language (SDL) specifies their actions to determine intended word senses in context. These languages constitute the formal theory of Word Expert Parsing, and permit the representation of all linguistic knowledge in terms of active word-based distributed agents. An existing computer program translates word experts represented in these languages into executable processes that interact to cooperatively analyze sentences.

This chapter makes a number of claims about the processes of natural language comprehension and its computational realization. A formal theory is developed and a computer model constructed to provide evidence to support those claims. Word Expert Parsing explains the understanding of sentences containing highly ambiguous words and complex structures. The distributed word-based approach is advanced as a framework for a full-scale theory of discourse comprehension.

The preparation of this manuscript has been supported by the National Science Foundation under Grants IST-8208571 and MCS-8209971. Their support of this basic research is gratefully acknowledged.