<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0907">
  <Title>Making Sense of Japanese Relative Clause Constructions</Title>
  <Section position="7" start_page="85" end_page="85" type="concl">
    <SectionTitle>
6 Discussion
</SectionTitle>
    <Paragraph position="0"> Perhaps the most directly comparable research to that outlined in this paper is that of Abekawa et al.</Paragraph>
    <Paragraph position="1"> (2001), who disambiguate RCCs according to simplex dependency data and KL divergence. That is, they extract out a0a2a1a4a3a6a5a7a1a9a8a11a10a13a12a15a14a17a16 a18a19a12a6a20a22a21a23a16a24a20a7a8a17a25a6a16a24a20a11a26a28a27 triples from corpus data, and disambiguate RCCs according to which case slot the head noun occurs in most commonly in simplex data. The accuracy for their method over a task where they distinguished between attributive and 6 types of case-slot gapping RCCs (defined according to case marker) was a relatively modest 65.3%. For a binary attributive vs.</Paragraph>
    <Paragraph position="2"> case-slot gapping task, the accuracy was a more respectable 88.8%, but still considerably lower than that achieved in this research.</Paragraph>
    <Paragraph position="3"> An alternate point of reference is found in the work of Li et al. (1998) on Korean RCCs, which display the same structural ambiguities as Japanese RCCs. Li et al. (1998) attain an accuracy of 90.4% through statistical analysis of the distribution of verb-case filler collocates, except that they classify relative clauses according to only 5 categories and consider only case-slot gapping RCCs. With our method, restricting analysis to only gapping RCCs (still retaining a total of nineteen RCC types) produces an accuracy of 94.1% for the ANDCI system with C4.5.</Paragraph>
    <Paragraph position="4"> In conclusion, we have proposed a method for interpreting Japanese relative clause constructions according to surface evidence and a generalised semantic representation. The method is designed to cope with analytical ambiguity in the head verb and head noun, and also interpretational parallelism in cosubordinated RCCs. In evaluation using C4.5, we showed our system to have a classification accuracy of 89.3%, marginally below the 90% upper bound for the described task.</Paragraph>
    <Paragraph position="5"> We have totally ignored the effects of pragmatics and context in this research, and in doing so, shown that it is possible to reliably derive a default RCC interpretation using only shallow syntactic and semantic features. In future research, we are interested in exploring methods of incorporating pragmatic and contextual features into our method, and the impact of these factors on both human and machine RCC interpretation.</Paragraph>
  </Section>
class="xml-element"></Paper>