<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3215">
  <Title>Object-Extraction and Question-Parsing using CCG</Title>
  <Section position="9" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have presented a detailed evaluation of a CCG parser on object extraction dependencies in WSJ text. Given the difficulty of the task, the accuracy of the parser is encouraging. The errors made by the parser suggest that general improvements in the coverage of the lexicon and parsing models derived from CCGbank will lead to improved recovery of long-range object dependencies.</Paragraph>
    <Paragraph position="1"> In contrast, we have suggested that general improvements in CCGbank parsing models will not lead to satisfactory performance on question parsing. The reason is that the Wh-question domain is syntactically distinct from WSJ text. We have presented a novel method for porting the parser to the question domain, which has led to good performance on question parsing. This has also demonstrated the close integration of the supertagger and the CCG parser on which our method depends.</Paragraph>
    <Paragraph position="2"> One of the major drawbacks of current NLP technology is that in general it performs very poorly outside of the training data domain. Our porting method only requires lexical category data, which is far easier to produce than full parse trees. This is an efficient method for porting the parser to other domains. The method may also be applicable to other lexicalised grammar formalisms.</Paragraph>
    <Paragraph position="3"> We will extend the question corpus to other question types. We are also continuing to develop the supertagger, which we have demonstrated is central to efficient portable wide-coverage CCG parsing.</Paragraph>
  </Section>
class="xml-element"></Paper>