The COUNSELOR Project: Understanding Answers to Questions.
MASSACHUSETTS UNIV AMHERST DEPT OF COMPUTER AND INFORMATION SCIENCE
Pagination or Media Count:
Typically, a computational model of question answering addresses the problem of understanding questions and producing answers. Once a question has been understood, the system accesses a knowledge base to retrieve the appropriate answer. Research motivated by the COUNSELOR system, a legal reasoning program, has turned the tables on this view of question answering. In COUNSELOR, questions are posed by the system and the natural language interface must understand the answer. Pursuant to these needs, we present a system model that utilizes the notion of predicitive processing in natural language understanding to interpret user answers to system questions. When we ask questions, we are seeking certain types of information. If we ask a yesno question, we are hoping to confirm or deny a hypothesis that we have. In the case of Wh-questions, we are looking for new information that we do not know. Regardless of the type of question, we create expections for the answer we want to receive. A precise and careful presentation of the question can limit possible responses. With a QA computational model of the type indicated earlier, the answer must be interpreted in whatever context and form it has been presented. Consequently, a mechanism that understands answers to questions must be robust enough to handle a diverse set of potential constructions.
- Computer Programming and Software