Achievement

IGERT Trainee Emily Pitler gained expertise in computational linguistics to develop procedures to automatically extract "local discourse relations" between sentences of written English. Local discourse relations are sometimes conveyed by connecting words such as "so", "because" "and then", etc., but are often implicit. Although it is now commonplace to have computers automatically extract syntactic relations in written text, discourse relations are seldom extracted despite their importance. Using the Penn Discourse Tree Bank, a large human annotated corpus of explicitly and implicitly realized discourse relations, Pitler, with her advisor Professor Ani Nenkova (computer science) have developed a computer learning algorithm that can successfully extract discourse relations from English text that is not already annotated by humans. Performance by the algorithm is of course not perfect but extremely promising and is likely to lead to important advances in natural language processing.