Discourse Information for Disambiguation: the Phoenix Approach in Janus M.s. Project Report Committee: 1 Problem Statement

For any given utterance out of what we can loosely call context, there is usually more than one possible interpretation. A speaker's utterance of an elliptical expression, like the gure \twelve fteen", might have a diierent meaning depending on the discourse context, the way the conversation has evolved until that point, and the previous speaker's utterance. If this is a problem for any human listener, the problem grows considerably when it is a parser doing the disambiguation. In this project, I intend to help a parser disambiguate among diierent possible parses for an input sentence, with the nal goal of improving the translation in an end-to-end speech translation system. One of the problems machine translation within a speech system faces is the presence of dissuencies and recognizer errors. A grammar designed to accept only perfectly formed sentences will fail on this type of input. The Phoenix parser Ward 91, Ward 94] was designed to capture the content of spoken dialogue through semantic grammars, which assign an input string to a concept, putting together a sequence of concepts that might form a complete thought or sequence, usually assigning a speech act to the sequence. If we are able to understand illocutionary force in the input language, we can produce the same illocutionary force in the output language, irrespective of the actual 1