<?xml version="1.0" standalone="yes"?>
<Paper uid="H93-1008">
  <Title>Gemini: A Natural Language System for Spoken-Language Understanding*</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
1. INTRODUCTION
</SectionTitle>
    <Paragraph position="0"> Gemini is a natural language 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 sub-components in detail.</Paragraph>
    <Paragraph position="1"> The demands on a natural language understanding system used for spoken language differ somewhat from the demands of text processing. For processing spoken language, there is a tension between the system being as robust as necessary, and as constrained as possible. The robust system will attempt to find as sensible an interpretation as possible, even in the presence of performance errors by the speaker, or recognition errors by the speech recognizer. In contrast, in order to provide language constraints to a speech recognizer, a system should be able to detect that a recognized string is not a sentence of English, and disprefer that recognition hypothesis from the speech recognizer. If the coupling is to be tight, with parsing and recognition interleaved, then the parser should be able to enforce as many constraints as possible for partial utterances. The approach taken in Gemini is to tightly constrain language recognition to limit overgeneration, but to extend the language analysis to recognize certain characteristic patterns of spoken utterances (but not generally thought of as part of grammar) and to recognize specific types of performance errors by the speaker.</Paragraph>
    <Paragraph position="2"> 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 semanticallyacceptable utterance-spanning edges are found during this phase, a component to recognize and correct certain grammatical disfluencies is applied. When an acceptable interpretation is found, a set of parse preferences are 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 will describe each of these components in detail, with the exception of the query answering subsystem, which will not be described in this paper.</Paragraph>
    <Paragraph position="3"> Since this paper describes a component by component view of Gemini, we will provide detailed statistics on the size, speed, coverage, and accuracy of the various components. These numbers detail our performance on the subdomain of air-travel planning that is currently being used by the DARPA spoken language understanding community\[13\]. 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 multiple times) to monitor our performance on a dataset that we didn't train on. We will 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 the both the training and fair test sets. More detailed explanations of these numbers are given in the relevant sections.</Paragraph>
    <Paragraph position="4">  this document are those of the authors and should not he interpreted as necessarily representing the official policies, either expressed or implied, of the Advanced Research Projects Agency of the U.S. Government.</Paragraph>
  </Section>
class="xml-element"></Paper>