<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1621">
  <Title>Lexical Reference: a Semantic Matching Subtask</Title>
  <Section position="12" start_page="177" end_page="178" type="ackno">
    <SectionTitle>
6 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper proposed an explicit task definition for lexical reference. This task captures directly the goal of common lexical matching models, which typically operate within more complex systems  fied example types at an overall system's recall of 25%. Disagreement refers to examples for which  theannotatorsdidnotagreeonthesubcategoryannotation (word vs. phrase/context). that address more complex tasks. This definition enabled us to create an annotated dataset for the lexical reference task, which provided insights into interesting sub-classes that require different types of modeling. The dataset enabled us to makeadirectevaluationandcomparisonoflexical matching models, reveal insightful differences between them, and create a simple improved model combination. In the long run, we believe that the availability of such datasets will facilitate improved models that consider the various sub-cases oflexicalreference, aswellasapplyingsupervised learning to optimize model combination and performance. null</Paragraph>
  </Section>
class="xml-element"></Paper>