8 permit the parser to operate with minimal back;up, and to make optimal use of the cascaded interaction between parsing and interpretation. The result is that the RUS system has many of the "deterministic" properties of Marcus' PARSIFAL system, while still providing for non-deterministic search to handle cases that are not within the scope of PARSIFAL. We are working on extensions to these strategies to handle ill-formed input, and to allow various types of feedback from discourse processes as well as semantics. In addition to the research-oriented work described above, we are also trying to facilitate the use of RUS as a front-end to other systems. As a first step in this direction we have provided a lexical entry and maintenance package that allows one to define and edit words without having to know the internal dictionary representation. This makes it easy to develop dictionaries for new domains and to use the parser on sentences beyond the scope of any particular semantic component. References: Bates, Madeleine "The Dictionary Package," in C. Sidner, et. al., "Research in Knowledge Representation for Natural Language Understanding; Annual Report: 1 September 1980 - 31 August 1981," Report 4785, Bolt Beranek and Newman Inc., forthcoming. Bobrow, Robert J. and Webber, Bonnie L. "RUS and PSI-KLONE: Parsing as an Incremental Recognition Process," in C. Sidner, et. al. (see above) Bobrow, Robert J. and Webber, Bonnie U "PSI-KLONE: Parsing and Semantic Interpretation in the BBN Natural Language Understanding System," in Proceedings of Third Biennial Conference of the Canadian Society for Computational Studies of Intelligence, May 14-16, 1980 [1]Bobrow, Robert J and Webber, Bonnie L."Knowledge Representation for Syntactic/Semantic Processing," in Proceedings of the First Annual National Conference on Artificial Intelligence, Aug. 19-21, 1980. Bobrow, Robert J. and Webber, Bonnie L. "Systems Considerations for Search by Cooperating Processes: Providing Continually Available Output," to appear in IEEE Transactions on Pattern Analysis and Machine Intelligence. [2]W.A. Woods, "Cascaded ATN Grammars." in American Journal of Computational Linguistics, Vol.6, No.l, January-March 1980. The RUS Parsing System and PSI-KLONE Semantic Interface Bolt, Beranek, and Newman Rusty Bobrow, Madeleine Bates, Boanie Webber (consultant) Bolt Beranek and Newman Inc. 50 Moulton St. Cambridge, MA 02238 Summary: The RUS parsing system provides a practical framework for natural language processing that is as flexible and extensible as a modular syntactic/semantic processor (such as LUNAR) and as efficient as a semantic grammar (such as the SOPHIE parser). RUS combines a large ATN grammar for English (with coverage greater than that of the LUNAR parser) with an interface that permits semantic interpretation to be performed efficiently during the parsing process. The interaction between the parser and the interpreter takes the form of a cascade [2] in which the interpreter receives information about syntactic constituents incrementally during the parsing process, and provides immediate feedback to the parser about the semantic acceptability of proposed syntactic structures. This results in an extremely efficient parser that has the advantages of both a syntax-first and a semantics-first approach. The grammar and semantic interpreter are written as two separate modules, whose only interaction is mediated by the interface. Because of this, we have been able to develop a number of natural language systems which share the same basic RUS grammar, but differ in the form of their semantic component. These systems are in use at BBN, ISI, Sperry-Univac, General Motors, the National Library of Medicine, and the University of Delaware. The RUS system was developed in Interlisp, but a Franz Lisp version is expected to be available sometime in 1982. For simple data-base front ends and work in instructional systems, the interpreter is based on a case frame dictionary with inheritance of interpretation rules. BBN's research group uses the KLONE knowledge representation system to represent both semantic interpretation rules and the resulting interpretations. This latter system is called PSI-KLONE (for Parsing and Semantic Interpretation in KLONE). PSI-KLONE uses an inference algorithm called incremental description refinement (IDR) [1] to simultaneously check semantic plausibility and discover (and apply) semantic interpretation rules as the parsing progresses. Work is going on to extend IDR to provide feedback on semantic preferences and to allow semantically anomalous parsings (this will provide the basis for dealing with metonymy and metaphor). We have also developed control strategies for ATN's which

/lp/association-for-computing-machinery/the-rus-parsing-system-and-psi-klone-semantic-interface-bolt-beranek-1MbKq6cMmL