Representing Sentence Structure in Hidden Markov Models for Information Extraction

We study the application of Hidden Markov Models (HMMs) to learning information extractors for -ary relations from free text. We propose an approach to representing the grammatical structure of sentences in the states of the model. We also investigate using an objective function during HMM training which maximizes the ability of the learned models to identify the phrases of interest. We evaluate our methods by deriving extractors for two binary relations in biomedical domains. Our experiments indicate that our approach learns more accurate models than several baseline approaches.