NSF Award
0742022
Interoperation between systems which use extensive formalized knowledge<br/>is hampered by the wide variety of formalisms currently in use in such<br/>systems. Attempts to create World-wide Web standards have already<br/>yielded a collection of such formalisms, not all mutually compatible. <br/>This situation results in part from a methodology which views deductive<br/>efficiency of inference engines, rather than semantic clarity, as the<br/>primary design criterion for new formalisms. Recent work (reviving an <br/>older idea) has focussed instead on the design of a single, <br/>highly expressive, logic into which a wide number of existing <br/>formalisms can be straightforwardly mapped, essentially treating <br/>them as subsets or simple ontologies (theories) within the <br/>single common formalism. The resulting logic, <br/>called the Interoperation Knowledge Language (IKL), <br/>can express many representational strategies and the relationships<br/>between them in a single formalism, in principle overcoming this <br/>interoperation problem in many common cases. To realize this potential <br/>in practice requires an inference engine to process IKL.<br/><br/>Although IKL is more expressive than first-order logic and so is<br/>not decidable, many of the formalisms translated into it have very <br/>tractable inference behavior. This project implements a new design <br/>of a reasoner which is theoretically complete for IKL, while also having<br/>the ability to recognize many of the known tractable subcases and <br/>use efficient inference strategies on inputs which fit into these <br/>known cases. The behavior of the engine is directed by scripts<br/>which control its search behavior in a highly flexible multi-directional <br/>search space combining hyper-resolution, rule-saturation and <br/>tableaux reasoning. The engine is designed as an experimental <br/>workbench rather than a production engine, with a focus on inventing<br/>useful techniques for controlling its behavior, running large<br/>experiment suites semi-automatically, and gathering information <br/>relevant to the search process. We plan to make use of this engine<br/>in a series of projects devoted to learning new inference strategies <br/>from empirical ontological data. <br/><br/>Broader Impacts. <br/><br/>The code created by this project will be publicly available <br/>under the GNU Lesser General Public Licence. <br/><br/>Details of the project can be found on the project web page at <br/>http://homam.ihmc.us/silkie/Home.html
