<?xml version="1.0" standalone="yes"?>
<Paper uid="H91-1099">
  <Title>SRI'S REAL-TIME SPOKEN LANGUAGE SYSTEM</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
SRI'S REAL-TIME SPOKEN LANGUAGE SYSTEM
</SectionTitle>
    <Paragraph position="0"/>
  </Section>
  <Section position="2" start_page="0" end_page="0" type="metho">
    <SectionTitle>
PROJECT GOALS
</SectionTitle>
    <Paragraph position="0"> This project involves the integration of speech and natural-language processing for spoken language systems (SLS). The goal of this project, to develop a multi-modal interface to the Official Airline Guide database, is being developed along two overlapping research and development lines: one focussed on an SLS kernel for database query, and the other on the full interactive system.</Paragraph>
  </Section>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
RECENT RESULTS
</SectionTitle>
    <Paragraph position="0"> Speaker-dependent and speaker-independent demos to illustrate the combined recognition/natural language system and accompanying graphical user interfaces.</Paragraph>
    <Paragraph position="1"> Improved robustness through the development of a template matcher for generating database queries; the template matcher has a tunable parameter to control how much constraint is ignored, so that wrong answers can be traded with no answers.</Paragraph>
    <Paragraph position="2"> Implementation of a bottom-up parser for CLE-formalism grammars; the new parser is about twice as fast as our original left-corner parser, and about 17 times faster than an initial bottom-up parser.</Paragraph>
    <Paragraph position="3"> Exploration of two schemes to integrate the recognizer and current NL schemes: N-best recognition with a statistical grammar, and recognition guided by a probabilistic finite-state representation of the templates.</Paragraph>
    <Paragraph position="4"> Evaluation of SRI's NL, SLS, and speech recognition technologies. SRI's February-91 weighted sentence error rate for ATIS Class A sentences was 33.8% (NL) and 44.1% (SLS); word error rate for the Resource Management speaker independent speech recognition evaluation was 17.6% with no grammar and 4.8% with the standard word-pair grammar.</Paragraph>
    <Paragraph position="5"> Improvements in the CLE grammar: extended coverage of numerical expressions, ATIS domain sortal restrictions, and conjoined noun phrases.</Paragraph>
    <Paragraph position="6"> Implementation of tied-mixture hidden Markov models, which resulted in a 20% reduction in the word-error rate compared to the discrete-density version.</Paragraph>
    <Paragraph position="7"> Training of a discrete-density DECIPHER using 20,000 sentences of read and spontaneous speech from Resource Management, TIMIT, and ATIS corpora. We achieved 10% word error on the June-90 ATIS test set.</Paragraph>
    <Paragraph position="8"> New techniques for statistical language models: a back-off estimation algorithm, Good-Turing estimates, and interpolation of word-based grammars with class-based grammars. Current test-set perplexity ranges from 15 to 30.</Paragraph>
    <Paragraph position="9"> Initial implementation of fast-search recognition algorithms for near real-time recognition.</Paragraph>
    <Paragraph position="10"> Initial implementation of speaker-adaptation using tied-mixture codebook adaptation.</Paragraph>
    <Paragraph position="11"> Implementation of an HMM reject-word model for dealing with noises and out-of-vocabulary items in digit recognition tasks; we plan to incorporate this in our ATIS SLS.</Paragraph>
  </Section>
class="xml-element"></Paper>