A Template Matcher for Robust NL Interpretation

A Template Matcher for Robust NL Interpretation Eric Jackson, Douglas Appelt, John Bear, Robert Moore, and Ann Podlozny SRI International 333 Ravenswood Avenue Menlo Park, CA 94025 Abstract In this paper, we describe the Template Matcher, a sys- tem built at SRI to provide robust natural-language interpretation in the Air Travel Information System (ATIS) domain. The system appears to be robust to both speech recognition errors and unanticipated or dif- ficult locutions used by speakers. We explain the mo- tivation for the Template Matcher, describe in general terms how it works in comparison with similar systems, and examine its performance. We discuss some limita- tions of this approach, and sketch a plan for integrating the Template Matcher with an analytic parser, which we believe will combine the advantages of both. Introduction One of the conclusions SRI has drawn from working with the ATIS common task data is that, even with a very constrained user task, there will always be unanticipated expressions and difficult constructions in the spoken lan- guage elicted by the task that will cause problems for a conventional, analytical approach to natural-language processing. However, it also seems that requests for only a few types of information account for a very large pro- portion of the utterances produced by users performing a task like air travel planning. This point is illustrated by some of the more difficult queries in the June 1990 test set: discontinuity. The third example would be straightfor- ward, except for the fact that the verb "servicing" has been substituted for the more conventional "serving." Despite the difficult linguistic problems posed by these queries, the information they request is very simple--- just fares, flights, and airlines for travel between a pair of specified cities. Consideration of examples such as these has led us to modify our approach to natural-language processing in spoken language systems. The key modification to our system is the addition of a Template Matcher to pro- vide robust interpretation for the most common types of requests in the task domain. The Template Matcher achieves robustness in two ways: (1) it provides an inter- pretation when not all the words or constructions in an utterance have been accounted for, and (2) it provides a mechanism for trading-off the risk of wrong answers with the degree of coverage. These properties arise from a mechanism that assigns scores to interpretations, pe- nalizing interpretations that do not account for words in the utterance. The bulk of this paper is devoted to describing the Template Matcher and discussing its per- formance as a stand-alone system for interpretation of naturM-language queries for the ATIS task. Later in the paper we consider how such a module might best fit into a complete system for spoken-language understanding. Give me a list of all airfares for round-trip tick- ets from Dallas to Boston flying on American Airlines. Show me all the flights and their fares from San Francisco to Boston on June second. I need information on airlines servicing Boston flying from …

Content truncated to 500 words