Preposition sense disambiguation for semantic information extraction based on maximum entropy

Prepositions are usually senseless but can have different senses depending on its context in a sentence. In extracting relevant information from texts, the exact sense of prepositions can often become crucial in extracting the exact relations among elements the preposition connects. We integrate the natural language parser MINIPAR syntactic structure parser and thesauri such as WORDNET, MeSH, and Gene Ontology to conduct the semantic analysis of complicated sentences involving prepositional phrases. Our system applies the maximum entropy method to build a model for preposition semantic role assignment. We collect the biological literature from a set of two hundred thousand test sentences retrieved from the NCBI PubMed database based on biological pathways in Apoptosis domain and evaluate the system performance using a statistical cross validation method.