Question Asking and Answering for ADL

There is a pressing need for networked intelligent Question Asking and Answering (QAA) systems capable of interacting with soldiers and DoD civilians anywhere/anytime. These users may have training needs, on-the-job questions, or be deployed in the field. While SMEs could answer many questions, they are often unavailable, and if available, not able to respond instantly. Fortunately, QAA systems are maturing. Recent advances in automatic text understanding (Latent Semantic Analysis) coupled with deeper computational linguistic analyses, such as understanding how language is used to form requests or questions (speech acts) and relating speech acts to responses users are seeking, may allow us to semi automatically construct mature QAA systems whose performance is indistinguishable from knowledgeable human intermediaries. In Phase I we propose constructing a SCORM conformant QAA system that uses LSA as a cornerstone, augmented with paragraph retrieval, relevance feedback, and automated summaries. We add to this syntactic parsing and speech act classification to answer queries about material in distributed learning courseware and other DoD information repositories. While LSA tools give broad brush answers, for finding text containing the answer, the other computational linguistic tools will be used to construct more refined answers for the user.