<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-1013">
  <Title>Generation for Dialogue Translation Using Typed Feature Structure Unification</Title>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3. Introducing Disjunctive Feature
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Structure Unification
</SectionTitle>
      <Paragraph position="0"> Introduction of disjunctive feature structures solves the inefficiency caused by making copies of whole trees when a node can be applied to multiple candidates of rules.</Paragraph>
      <Paragraph position="1"> For example, multiple copying is caused by the mutual restriction between the subject and the verb (subcategorizing by verb and subject-verb agreement). The verb cannot be determined until the subject is determined and the derivation tree must be copied for each verb candidate.</Paragraph>
      <Paragraph position="2"> Instead of copying the derivation tree, the surface entries of a verb are packed into a 2 65 disjunctive feature structure in a lexical entry as follows.</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
(DEFLEX-UNIT Ibe-Unitl DYADIC
</SectionTitle>
    <Paragraph position="0"/>
    <Paragraph position="2"> ......</Paragraph>
    <Paragraph position="3"> When the derivation proceeds and the subject is determined, one surface string is selected from these three candidates (see Fig. 2). The unification of disjunctive feature structures is implemented according to Kasper's algorithm (Kasper, 1987).</Paragraph>
  </Section>
  <Section position="6" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4. Grammar and Examples
</SectionTitle>
    <Paragraph position="0"> The grammar developed for this generation system is based on HPSG (Pollard and Sag, 1987) and its modification by Borsley (1987).</Paragraph>
    <Paragraph position="1"> Relating illocutionary forces to utterances is achieved in this grammar.</Paragraph>
    <Paragraph position="2"> For example, consider the following feature structure including the REQUEST illocutionary force.</Paragraph>
  </Section>
  <Section position="7" start_page="0" end_page="0" type="metho">
    <SectionTitle>
\[CIRC\[RELN \[*REQUEST*\]\]
\[AGEN ?X03\[IND-OBJ\[LABEL *SPEAKER*\]\]\]
\[RECP ?X02\[IND-OBJ\[LABEL *HEARER*\]\]\]
\[OBOE \[CIRC\[RELN \[*SEND-I*\]\]
</SectionTitle>
    <Paragraph position="0"/>
    <Paragraph position="2"> ;abbreviated here From this feature structure, the following generation results can be obtained. &gt; (gen3 fs-l) (&amp;quot;would you send me a registration form&amp;quot; &amp;quot;could you send me a registration form&amp;quot; &amp;quot;send me a registration form&amp;quot;) ;; &amp;quot;send NP to NP&amp;quot; form is suppressed here. Specifying one of these results can be done by enriching the input feature structure.</Paragraph>
    <Paragraph position="3"> 5. Current Status and Further Tasks This article described how the generation process is effectively controlled by typed feature structures and disjunctive feature structures. The generation mechanism described here is implemented in Common Lisp on Symbolics Lisp Machines and Sun Workstations. A screen hardcopy of the environment is shown in Fig. 2. The grammar for this generation system is now under enrichment. The relationships between surface utterances and intentions need to be further explored.</Paragraph>
    <Paragraph position="4"> Kume et al. (1989) and Kogure et al. (1989) introduced illocutionary force type planning from deep illocutionary force type. Combining this method with the generator is the next task. ~( UED.'I NP . CH - &gt; NPSPECP Uq //~u n~D ov~p ~ nm o,m ~ -- * i ~ ow, i n~ e,,~ ~</Paragraph>
  </Section>
class="xml-element"></Paper>