Ongoing Research Projects Knowledge Representation and Reasoning Deduction with Constraints Constraint Solving

This document provides brief descriptions of research projects that are representative of those conducted within the Intelligent Systems Group. For convenience the document is divided into three sections|knowledge representation and reasoning, machine learning, and natural language processing|although there is signiicant overlap among these. One of the most widely-used and successful approaches to increasing the ef-ciency of general-purpose automated reasoning systems has been that of integrating special-purpose reasoning systems into them, resulting in what are often called hybrid reasoning systems. Though the resulting hybrid reasoning systems are appealing, their construction and analysis can be diicult 17]. Our research helps to remedy this problem for a particular class of hybrid reasoners that we have identiied and dubbed \substitutional reasoners". Substitutional reasoners share certain architectural features; most notably they (1) operate on a language that contains a distinguished set of symbols for representing constraints on the values over which quantiied variables range, and (2) employ a special purpose reasoning system to test the satissability of these constraints. One of the distinguishing features of substitutional reasoners is that the constraints are manipulated exclusively by the special-purpose reasoner. Though the substitutional architecture has been one of the most common and successful architectures for hybrid reasoning, our research is the rst to identify these reasoners as a single class and to investigate their common properties and the general principles that underly them. Our results support a framework that enables the systematic production of substitutional reasoners and their completeness proofs from certain kinds of non-hybrid reasoners and their completeness proofs 15, 20]. Within the substitutional framework we have studied reasoning systems for knowledge retrieval, constraint logic programming, modal logic deduction 21], parsing feature-based grammars 16], inductive learning with background information 25], and planning in temporally rich domains. In contrast to our results on deduction with constraints, which have been ob