Using Grammatical Inference to Improve Precision in Information Extraction

The eld of information extraction (IE) is concerned with applying natural language processing (NLP) and information retrieval (IR) techniques to the automatic extraction of essential details from text documents. We are exploring the use of machine learning methods for IE. While the most promising methods we have developed perform well for problems deened over a collection of electronic seminar announcements, they are imprecise in their identiication of the boundaries of relevant text fragments (elds). Here, we entertain the idea of using grammatical inference (GI) methods to learn the appropriate form of a eld. We describe one method for translating raw text into an abstract alphabet suitable for GI, and show that, by combining one IE learning method with the resulting inferred grammars, large improvements in precision can be realized for some elds.