International Research Fellowship Program: Probabilistic Models for Reasoning in Natural Language Dialog

点击翻译按钮获取中文摘要

0853021ZettlemoyerThis award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).The International Research Fellowship Program enables U.S. scientists and engineers to conduct nine to twenty-four months of research abroad. The program's awards provide opportunities for joint research, and the use of unique or complementary facilities, expertise and experimental conditions abroad.This award will support a twenty-four month research fellowship by Dr. Luke Zettlemoyer to work with Dr. Mark Steedman at the University of Edinburgh in the UK.The PI is developing probabilistic models for pragmatic reasoning, the type of context-dependent reasoning that is required for automated systems to participate in natural language conversations. They are investigating whether these new methods will improve performance when used in dialog systems. The focus is on two specific questions. (1) Can we develop methods for learning to recover the context-dependent meanings of a sequence of natural language statements? For each sentence, we want to be able to automatically construct a rich, logical representation of its underlying meaning. In general, later statements can elaborate on, correct, or refer to parts of previous statements, leading to a challenging context-dependent reasoning problem. (2) Can we use probabilistic, game-theoretic models of multi-agent interaction to build effective dialog systems? Such an approach will explicitly model dialog participants jointly interacting in an uncertain world where each conversant has independent beliefs and desires that influence the conversational flow. Building automated systems that participate effectively in natural language conversations is one of the classic goals of research in artificial intelligence. Such dialog systems have the potential to revolutionize the way we interact with computers. Although probabilistic techniques have been used successfully in a wide range of natural language processing problems, researchers have only recently started to develop them for modeling conversation. The methods we are developing should enable deployed systems to participate in significantly more complex conversations.