<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-2249">
  <Title>A Cognitive Model of Coherence-Driven Story Comprehension</Title>
  <Section position="4" start_page="1499" end_page="1500" type="metho">
    <SectionTitle>
3 Coherence-Driven Comprehension
</SectionTitle>
    <Paragraph position="0"> In this section, I outline some general principles which may attenuate the performance of a comprehension system. I begin with the general definition of a schema: Cl, ..., C.n --)' I.</Paragraph>
    <Paragraph position="1"> where cl, ..., c~ are the elements connected by I. The left-hand side of a schema is its condition set, and the right-hand side represents the interpretation of those conditions in terms of other concepts (e.g. a temporal relation, or a cornpound event sequence). During each processing cycle, condition sets are matched against the set of observations.</Paragraph>
    <Paragraph position="2"> At present, I am developing a metric which measures coherence contribution with respect to a schema and a set of observations:</Paragraph>
    <Paragraph position="4"> where C = coherence contribution; V = Coverage; U-- Utility; P -- Completion; and S = Skepticism. This metric is based on work in categorisation and diagnosis, and measures the similarity between the observations and a condition set (Tversky, 1977).</Paragraph>
    <Section position="1" start_page="1499" end_page="1499" type="sub_section">
      <SectionTitle>
3.1 Coverage and Completion
</SectionTitle>
      <Paragraph position="0"> Coverage captures the principle of conflict resolution in production systems. The more elements matched by a schema, the more coherence that schema imparts on the representation, and the higher the Coverage. By contrast, Completion represents the percentage of the schema that is matched by the input (i.e. the completeness of the match). Coverage and Completion thus measure different aspects of the applicability of a schema. A schema with high Coverage may match all of the observations; however, there may be schema conditions that are unmatched. In this case, a schema with lower Coverage but higher Completion may generate more coherence.</Paragraph>
    </Section>
    <Section position="2" start_page="1499" end_page="1500" type="sub_section">
      <SectionTitle>
3.2 Utility
</SectionTitle>
      <Paragraph position="0"> The more observations a schema can explain, the greater its coherence contribution. Utility measures this inherent usefulness: schemas with many conditions are considered to contribute more coherence than schemas with few. Utility is independent of the number of observations matched, and reflects the structure of the knowledge base (KB). In previous comprehension models, the importance of schema size is often ignored: for example, an explanation requiring a long chain of small steps may be less costly than a proof requiring a single large step.</Paragraph>
      <Paragraph position="1"> To alleviate this problem, I have made a commitment to schema 'size', in line with the notion of 'chunking' (Laird et al., 1987). Chunked schemas are more efficient as they require fewer processing cycles to arrive at explanations.</Paragraph>
    </Section>
    <Section position="3" start_page="1500" end_page="1500" type="sub_section">
      <SectionTitle>
3.3 Skepticism
</SectionTitle>
      <Paragraph position="0"> This parameter represents the unwillingness of the comprehender to 'jump to conclusions'. For example, a credulous comprehender (with low Skepticism) may make a thematic inference that a trip to a restaurant is being described, when the observations lend only scant support to this inference. By raising the Skepticism parameter, the system may be forced to prove that such an inference is valid, as missing evidence now decreases coherence more drasticallyJ</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="1500" end_page="1500" type="metho">
    <SectionTitle>
4 Example
</SectionTitle>
    <Paragraph position="0"> Skepticism can have a significant impact on the coherence contribution of a schema. Let the set of observations consist of two statements: enter(john, restaurant), order(john, burger) Let the KB consist of the schema (with Utility of 1, as it is the longest schema in the KB):</Paragraph>
    <Paragraph position="2"> rant) being the assumption. If S is raised to 1, C now equals 2 5, with the same assumption.</Paragraph>
    <Paragraph position="3"> Raising S makes the system more skeptical, and may prevent hasty thematic inferences.</Paragraph>
  </Section>
  <Section position="6" start_page="1500" end_page="1500" type="metho">
    <SectionTitle>
5 Future Work
</SectionTitle>
    <Paragraph position="0"> Previous models of comprehension have relied on an 'all-or-nothing' approach which denies partial representations. I believe that changing the goal of comprehension from top-levelpattern instantiation to coherence-need satisfaction may produce models capable of producing partial representations.</Paragraph>
    <Paragraph position="1"> One issue to be addressed is how coherence is incrementally derived. The current metric, and many previous ones, derive coherence from a static set of observations. This seems implausible, as interpretations are available at any point during comprehension. A second issue is  schema applications. Local weights could also be attached to individual conditions (see section 5). the cost of assuming various conditions. Some models use weighted conditions, which differentially impact on the quality of the representation (Hobbs et al., 1993). A problem with these schemes is the sometimes ad hoc character of weight assignment: as an antidote to this, I am currently constructing a method for deriving weights from condition distributions over the KB. This moves the onus from subjective decisions to structural criteria.</Paragraph>
  </Section>
class="xml-element"></Paper>