<?xml version="1.0" standalone="yes"?>
<Paper uid="W99-0313">
  <Title>A Two-level Approach to Coding Dialogue for Discourse Structure: Activities of the 1998 DRI Working Group on Higher-level Structures*</Title>
  <Section position="5" start_page="106" end_page="106" type="concl">
    <SectionTitle>
4 Summary and Future Work
</SectionTitle>
    <Paragraph position="0"> In addition to the quantitative analysis of eodings, the subgroup at the 1998 DRI meeting reiterated some goals for the scheme in general and made progress on several open theoretical issues3 First and foremost, it was agreed upon that CGU analysis at the meso-level allowed coders to abstract the &amp;quot;messy&amp;quot; bits of dialogue (e.g., local repair, turntaking, grounding) into common ground units, making the structures at both the meso- and macro-levels cleaner. The consensus was that many NLP applications would benefit from this abstraction, which can help separate to a large degree the processing of dialogic phenomena from the processing of intentions and informational units at the dialogue planning level.</Paragraph>
    <Paragraph position="1"> As for theoretical issues, the subgroup laid out initial proposals for exploring the interface between Damsl tagging at the dialogue act micro-level, and CGU analysis at the meso-level. One important open issue was whether to modify the coding scheme to identify different types of acknowledgments separately, especially when the acknowledgment function was parasitic on a more direct relation, such as an answer to a question. It was found that alternative proposals for placing CGU boundaries patterned a Full details of the subgroup proceedings can be found in the DRI report of the 1998 meeting, also available from the first author.</Paragraph>
    <Paragraph position="2"> with differences in backward- and forward-looking properties of the ambiguous tokens. The general principle that was agreed upon was that we should investigate further the situations in which dialogue act coding can serve as the basis for CGU coding decisions, just as CGU codings serve as the primitive units for constraining IU analysis in a substantial way. A more general principle was to identify when independent decisions at one level could influence the coding decisions at a second level, e.g. when an IU boundary resolved a difficult CGU boundary decision. Defining non-circular coding guidelines appears feasible, if difficult.</Paragraph>
    <Paragraph position="3"> While the reliability results presented here are already close to acceptable, directions for future work are clear. In particular, extensions to include additional dimensions of dialogue content would be desirable; the current scheme considers only grounding at the meso-range, and information/intention content at the macro-range. Secondly, we expect refinement and revision of the initial coding manual, (Nakatani and'Traum, 1999), will facilitate both greater reliability and utility of the two levels we do cover. We hope other researchers will explore whether a more productive synergy can be found between the two levels, both in theory and in practice. The relation we hypothesize between the two levels, and our supposition that important relations may be found between micro-level schemes and the two-level scheme posited here, lay the groundwork for more focused investigations of coding schemes for discourse structure in dialogue than have previously existed within the DR/initiative.</Paragraph>
  </Section>
class="xml-element"></Paper>