Coping with Ambiguity in Knowledge-based Natural Language Analysis

This paper describes the strategies and techniques used by the English analysis component of the KANT Knowledgebased Machine Translation system to cope with ambiguity. The constraints for elimination of ambiguity are distributed across the various knowledge sources in the analyzer. As a result, efficiency in the analysis component is maintained, and output quality is improved. 1 INTRODUCTION The KANT system [Nyberg and Mitamura, 1992] is a Knowledge-based Machine Translation (KBMT) system designed to translate English source documents into multiple target languages. The current application of the KANT system is the translation of service information publications for all major products of Caterpillar, Inc. (heavy equipment) into the major export languages. One of the difficulties in implementing a KBMT system, as with many other natural language applications, is the ambiguity that arises during analysis. Previous knowledge-based natural language analysis systems have identified a numbe..