A chart parser to analyze large medical corpora

Abstract

We describe a natural language parser for large medical corpora. Sentence parsing is a necessary step to build a sentence representation and to support a wide-coverage semantic interpretation. When applied to limited domains, a good syntax coverage can be obtained from Phrase-Structure rules. Large medical corpora show a strong variability in word and phrase order that requires more specific parsing strategies. We describe a parser based on Chart techniques. It parses constituents from left to right as they appear in a sentence. It enables the incremental partial parsing of words and phrases coming from a speech recognition input. We report here first results obtained from a large corpus of cancer treatment reports.