Understanding messages in a diagnostic domain

摘要

The problem of coping with subject-matter sublanguages in text processing is well known in the natural language processing field. The main problem is to balance the use of generic knowledge sources and the specific needs of the sublanguages. This paper introduces the characteristics of the sublanguage found in diagnostic messages about automotive equipment failures, and discusses an architecture to analyse those messages. The model is based on a two-level partial parsing approach that uses a syntax-driven strategy to parse fragments of a sentence. A set of semantics-driven strategies is used to collapse the fragments. General knowledge sources are proposed for use as an independent syntax, a knowledge-based semantics, a pragmatic module and a two-level lexicon. The problem of balancing accuracy, robustness and efficiency in the message analysis is addressed. Finally some applied results are shown.