Lexico-Semantic Pattern Matching as a Companion to Parsing in Text Understanding

LEXICO-SEMANTIC PATTERN MATCHING AS A COMPANION TO PARSING IN TEXT UNDERSTANDING Paul S. Jacobs, George R. Krupka and Lisa F. Rau Artificial Intelligence Laboratory GE Research and Development Schenectady, NY 12301 ABSTRACT Ordinarily, one thinks of the problem of natural language un- derstanding as one of making a single, left-to-right pass through an input, producing a progressively refined and detailed interpre- tation. In text interpretation, however, the constraints of strict left-to-right processing are an encumbrance. Multi-pass meth- ods, especially by interpreting words using corpus data and as- sociating units of text with possible interpretations, can be more accurate and faster than single-pass methods of data extraction. Quality improves because corpus-based data and global context help to control false interpretations; speed improves because pro- cessing focuses on relevant sections. The most useful forms of pre-processing for text interpre- tation use fairly superficial analysis that complements the style of ordinary parsing but uses much of the same knowledge base. Lexico-semantic pattern matching, with rules that combine lex- local analysis with ordering and semantic categories, is a good method for this form of analysis. This type of pre-processing is efficient, takes advantage of corpus data, prevents many garden paths and fruitless parses, and helps the parser cope with the complexity and flexibility of real text. INTRODUCTION The interpretation of large volumes of text poses many con- trol problems, including limiting the complexity of analysis and ensuring the production of valid interpretations without consid- ering too many possibifities. These problems are especially severe in processing news stories, where long sentences, information-rich news-style constructions, and the complex structure of events make normal syntax-first analysis especially impractical. Normal left-to-right syntactic parsing, in virtually all its forms, is a disaster for interpreting broad classes of extended texts. Multiple-path methods are haunted by attachment problems that can lead to a combinatoric explosion of paths, while simple deter- ministic methods bring on parser failures and problems in com- bining preferences. In previous work aimed at word sense coding of news stories [1], we have found that even heavy pruning of a multiple-path chart parsing strategy often leaves hundreds of parses to consider for a single sentence. Even worse, minor ir- regularities in linguistic structure or word usage bring on parser failures and inadequate interpretations. Better parsing strategies, including control using statistical data, flexible partial parsing, and recovery, can certainly help with some of these problems, but some of the easiest improve- ments in the control of parsing come from the creative use of pre-processing. Our system incorporates a lexico-semantic pat- tern mateher, which uses much of the same knowledge base as the parser and semantic interpreter but performs a global, superficial analysis of text prior to parsing. The design and implementation of the pattern matcher is simple; instead of concentrating on its details, this paper focuses on the functionality of pre-processing and its impact on parser control. Three aspects of pre-processing have particular promise for the quality and efficiency of later processing--tagging, template aciiva~ion (including topic analysis), and segmentation …

Content truncated to 500 words