Abstract
The smart help we propose based on application software manuals features four phases: (1) understanding the user's input text, which is a question about software operation; (2) matching the result of understanding with software manual text; (3) planning a dialog with the user; and (4) paraphrasing matching manual text to output for the user. Manual text found during matching is paraphrased using the analysis result found during understanding. In planning, the hetorical structure of the selected manual is used to determine which clauses in matching manual text are specifically relevant to the user's question. Four paraphrases for a novice user are identified in a subject experiment, and rules for these are applied based on linguistic features extracted during the understanding. We evaluated an implemented prototype, and discuss its applications and problems.