<?xml version="1.0" standalone="yes"?>
<Paper uid="P93-1008">
  <Title>GEMINI: A NATURAL LANGUAGE SYSTEM FOR SPOKEN-LANGUAGE UNDERSTANDING*</Title>
  <Section position="2" start_page="0" end_page="54" type="intro">
    <SectionTitle>
1. INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Gemini is a natural language (NL) understanding system developed for spoken language applications. This paper describes the details of the system, and includes relevant measurements of size, efficiency, and performance of each of its components.</Paragraph>
    <Paragraph position="1"> In designing any NL understanding system, there is a tension between robustness and correctness. Forgiving an error risks throwing away crucial information; furthermore, devices added to a system to enhance robustness can sometimes enrich the ways of finding an analysis, multiplying the number of analyses for a given input, and making it more difficult to find the correct analysis. In processing spoken language this tension is heightened because the task of speech recognition introduces a new source of error. The robust system will attempt to find a sensible interpretation, even in the presence of performance errors by the speaker, or recognition errors by the speech recognizer. On the other hand, a system should be able to detect that a recognized string is not a sentence of English, to help filter recognition errors by the speech recognizer. Furthermore, if parsing and recognition are interleaved, then the parser should enforce constraints on partial utterances.</Paragraph>
    <Paragraph position="2"> The approach taken in Gemini is to constrain language recognition with fairly conventional grammar, but to augment that grammar with two orthogonal rule-based recognition modules, one for glueing together the fragments found during the conventional grammar parsing phase, and another for recognizing and eliminating disfluencies known as &amp;quot;repairs.&amp;quot; At the same time, *This research was supported by the Advanced Research Projects Agency under Contract ONR N0001490-C-0085 with the Office of Naval Research. The views and conclusions contained in this document are those of the authors and should not be interpreted as necessarily representing the official policies, either expressed or implied, of the Advanced Research Projects Agency of the U.S. Government.</Paragraph>
    <Paragraph position="3"> the multiple analyses arising before and after all this added robustness are managed in two ways: first, by highly constraining the additional rule-based modules by partitioning the rules into preference classes, and second, through the addition of a postprocessing parse preference component.</Paragraph>
    <Paragraph position="4"> Processing starts in Gemini when syntactic, semantic, and lexical rules are applied by a bottom-up all-paths constituent parser to populate a chart with edges containing syntactic, semantic, and logical form information. Then, a second utterance parser is used to apply a second set of syntactic and semantic rules that are required to span the entire utterance. If no semantically acceptable utterance-spanning edges are found during this phase, a component to recognize and correct certain grammatical disfluencies is applied.</Paragraph>
    <Paragraph position="5"> When an acceptable interpretation is found, a set of parse preferences is used to choose a single best interpretation from the chart to be used for subsequent processing. Quantifier scoping rules are applied to this best interpretation to produce the final logical form, which is then used as input to a query-answering system. The following sections describe each of these components in detail, with the exception of the query-answering subsystem, which is not described in this paper.</Paragraph>
    <Paragraph position="6"> In our component-by-component view of Gemini, we provide detailed statistics on each component's size, speed, coverage, and accuracy.</Paragraph>
    <Paragraph position="7"> These numbers detail our performance on the sub-domain of air-travel planning that is currently being used by the ARPA spoken language understanding community (MADCOW, 1992). Gemini was trained on a 5875-utterance dataset from this domain, with another 688 utterances used as a blind test (not explicitly trained on, but run nmltiple times) to monitor our performance on a dataset on which we did not train. We also report here our results on another 756-utterance fair test set that we ran only once. Table 1 contains a summary of the coverage of the various components on both the training and fair test sets. More detailed  explanations of these numbers are given in the relevant sections.</Paragraph>
  </Section>
class="xml-element"></Paper>