Automated Skimming System in Response to Questions for NonVisual Readers

This paper presents factors in designing a system for automatically skimming text documents in response to a question. The system will take a potentially complex question and a single document and return a Web page containing links to text related to the question. The goal is that these text areas be those that visual readers would spend the most time on when skimming for the answer to a question. To identify these areas, we had visual readers skim for an answer to a complex question while being tracked by an eye-tracking system. Analysis of these results indicates that text with semantic connections to the question are of interest, but these connections are much looser than can be identified with traditional Question-Answering or Information Retrieval techniques. Instead, we are expanding traditional semantic treatments by using a Web search. The goal of this system is to give nonvisual readers information similar to what visual readers get when skimming through a document in response to a question.