Generating cooperative answers in deductive databases
Theresa Gaasterland · 1992
This dissertation investigates the problem of presentation of useful responses to users' queries. Not only should a database interface collaborate with a user to find the information that he or she is seeking, but it should also present answers in a manner that is easily understood by the user. To better meet each user's needs, we show how to use constraints that describe users' restrictions over the database to tailor a set of answers values to meet the user's interests. User constraints serve to restrict queries that are too general. We formalize the use of user constraints through semantic query optimization techniques and give heuristic rules for determining when to include a user constraint in an answer to a query. We also describe a technique called relaxation that enables an interface to work interactively with the user to find alternative answers that are related to the answers of the original query. The relationships are defined through taxonomy clauses. With the relaxation technique, a user can ask a very specific query and get related answers as well as direct answers. The user constraint theory and the relaxation have been implemented along with Gal and Minker's work with integrity constraints within the context of a prototype cooperative answering interface called CARMIN. A cooperative response is a potentially large logical formula whose content could be confusing to the user. We give algorithms for selecting search tree information to ensure that a response is coherent. We have also developed formalisms for making a response more cohesive. We present linguistically motivated theories for lexical substitution, coordination and selection of tense, aspect, and temporal connecting words. The theory for lexical substitution uses the same taxonomy clauses and rewrite algorithms that are used for relaxation. The theory of coordination is general, applies to any relational or deductive database and operates independently of a generation grammar. The theory of tense and aspect uses a temporal database model as the source of time information and incorporates linguistic theories of tense and aspect to generate cooperative answers. We also present a method for managing grammatical information about database relations that enables the generation of text with subordinate clauses and pronouns as well as coordination. (Abstract shortened by UMI.)