AI Research at Bolt, Beranek & Newman, Inc

BBN’s project in knowledge representation for natural language understanding is developing techniques for computer assistance to a decision maker who is collecting information about and making choices in a complex situation. In particular, we are designing a system for natural language control of an intelligent graphics display. This system is intended for use in situation assessment and information management. Our work is concentrated in the development of fundamental techniques for knowledge representation and language understanding. Specifically, we are working on advanced parsing techniques, syntactic/semantic interaction, recognition of speaker intent, anaphora and deixis, fundamental knowledge representation techniques, and parallel algorithms and techniques for knowledge-based inference. (Contact person: Bill Woods) One major accomplishment so far is the development of the knowledge representation system KL-ONE. Another is a prototype display system that understands requests, assertions, and questions about display manipulation. This prototype is based upon several general tools, including the RUS parser, a lexical acquisition system, the PSI-KLONE interface, and a model of a speaker-meaning recognizer. We have also pursued research on an abstract parallel machine for marker passing, and implementations of knowledge-based inference algorithms. The RUS parsing system provides a practical framework for natural language processing that is as flexible and extensible as a modular syntactic/semantic processor and as efficient as a semantic grammar. RUS combines a large ATN grammar for English 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 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 the proposed syntactic structures. Because the grammar and semantic interpreter are two separate modules, we have been able to use the RUS grammar in a number of natural language systems which are in use at BBN, ISI, Sperry-Univac, General Motors, the National Library of Medicine, and the University of Delaware. (Contact person: Rusty Bobrow) We are also working on natural language generation as exemplified by the ILIAD system, a computer-assisted instruction system designed to help deaf people improve their understanding of the English language. ILIAD is based on a transformational grammar that is capable of generating such structures as transitive and intransitive sentences, indirect objects, passives, predicate adjectives, predicate nominals, predicate adverbs, comparative and superlative structures, complex verb auxiliaries, modals, yes-no ques-