<?xml version="1.0" standalone="yes"?>
<Paper uid="W94-0333">
  <Title>Cross Modal Comprehension in ZARDOZ An English to Sign-Language Translation System</Title>
  <Section position="5" start_page="250" end_page="251" type="metho">
    <SectionTitle>
6. M0de-Interleaved Sign Generation
</SectionTitle>
    <Paragraph position="0"> We have already mentioned the distinction between native sign languages (e.g. ASL) and borrowed sign system (like SEE). Native sign language is the dominant means of communication among deaf signers, however, borrowed sign language is often used for educational purposes (where hearing signers are involved), and for such ends as signed news summaries on television. As a result, most native signers are comfortable with both types of sign, and encounter little difficulty in segueing between the two forms.</Paragraph>
    <Paragraph position="1"> This ability affords a system such as ZARDOZ with a base-level performance that can be guaranteed by</Paragraph>
    <Section position="1" start_page="251" end_page="251" type="sub_section">
      <SectionTitle>
7th International Generation Workshop * Kennebunkport, Maine * June 21-24, 1994
</SectionTitle>
      <Paragraph position="0"> the system. Should the source parser be unable to generate a syntactic structure that spans the entire input, the system can still produce a full output representation by interleaving native and coded sign. Native sign is used for those input fragments which produce case-frame interlingual representations, while the coded sign is used for troublesome link words which cause the parser to fail.</Paragraph>
    </Section>
  </Section>
  <Section position="6" start_page="251" end_page="251" type="metho">
    <SectionTitle>
6. Sign Articulation and Animation
</SectionTitle>
    <Paragraph position="0"> Having generated a syntactic form of a sign utterance it remains to create a fluid articulation of the utterance to display as an animated sequence. In the case of coded sign systems such as SEE this is a simple matter of mapping sign tokens to pre-stored animation sequences and smoothing over inter-sign gaps.</Paragraph>
    <Paragraph position="1"> However, native sign languages employ a much richer sign structure, which requires a correspondingly richer representation for the output lexicon.</Paragraph>
    <Paragraph position="2"> Native sign languages are heavily inflected, with a much of syntactic information encoded in parallel on a single lexeme. One example already mentioned is the use of verb movement to indicate agent and patient.</Paragraph>
    <Paragraph position="3"> Another example in ASL is the rich aspectual inflection system employed. For example the sign ASL-SICK with a circular motion added means 'sickly' or 'sickness prone'. A repeated, tense motion indicates the meaning 'extremely sick'. These and many more inflections apply in a regular manner to ASL signs, and may be compounded. For example, the verb ASL-LOOK-AT can be inflected to mean 'he watches it regularly' or 'I look at each of them in turn'.</Paragraph>
    <Paragraph position="4"> Because of the richness of the inflection system, it is impractical to store every inflected form directly as an animation sequence. We adopt the approach of storing signs in their citation or root form only, and storing inflection rules separately. Inflected signs are generated as needed by applying the appropriate rules to the root sign forms. Signs are stored using a phonological model of sign structure, based on Sandler's Hand-Tier model (Sandler, 1989). The phonological representations are not mapped to concrete animation values in a DCL program until after inflection rules have been applied.</Paragraph>
    <Paragraph position="5"> 7. Summary and Future Research To date, we have implementedthe infrastructure of the Zardoz system, including parsing, interlingua, generation and animation components, but have yet to implement a comprehensive sign grammar or lexicon.</Paragraph>
    <Paragraph position="6"> The phonological model of sign structure and inflection rules, mentioned in Section 6, is also in an early stage of development.</Paragraph>
    <Paragraph position="7"> Our current research efforts are concentrated on developing more a comprehensive computational grammar, morphology and lexicon for ISL, the native sign language of Ireland where our research is based.</Paragraph>
    <Paragraph position="8"> The examples in this paper are taken from ASL, as linguistic information on ASL is more readily available, but in future work will focus on ISL, as we feel that the evaluation and advice of native signers will be crucial to the success of our research.</Paragraph>
    <Paragraph position="9"> Though our work is still in an early stage, we are confident that the framework outlined here will provide a sound basis for tackling the challenges of cross-model translation. The issues of translation between different language media holds considerable theoretical interest, but we also believe that the A.I./linguistic technology is mature enough to build systems of value to sign users in the near future. We hope to contribute to the development of such systems.</Paragraph>
  </Section>
class="xml-element"></Paper>