Summarization design for interactive cross-language question answering

This paper describes an experimental investigation of interactive techniques for cross-language information access. The task was to answer factual questions from a large collection of documents written in a language in which the user has little proficiency. An interactive cross-language retrieval system that included optional user-assisted query translation, display of translated summaries for individual document ranked in order of decreasing degree of match to the user's query, and optional full-text examination of individual documents was provided. Two alternative types of extractive summaries were tried using a systematically varied presentation order, one drawn from a single segment of the translated document and the other drawn from three (usually) shorter segments of the translated document. On average, users were able to correctly answer just 62% of the sixteen assigned questions in an average of 176 seconds per question. Little difference was found between the two summary types for this task in an experiment using eight human subjects. Time on task and the number of query iterations were found to exhibit a positive correlation with question difficulty.