<?xml version="1.0" standalone="yes"?>
<Paper uid="P02-1014">
  <Title>Improving Machine Learning Approaches to Coreference Resolution</Title>
  <Section position="5" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> We investigate two methods to improve existing machine learning approaches to the problem of 8Soon et al. (2001) present only the tree learned for the MUC-6 data set.</Paragraph>
    <Paragraph position="1"> noun phrase coreference resolution. First, we propose three extra-linguistic modifications to the machine learning framework, which together consistently produce statistically significant gains in precision and corresponding increases in F-measure.</Paragraph>
    <Paragraph position="2"> Our results indicate that coreference resolution systems can improve by effectively exploiting the interaction between the classification algorithm, training instance selection, and the clustering algorithm. We plan to continue investigations along these lines, developing, for example, a true best-first clustering coreference framework and exploring a &amp;quot;supervised clustering&amp;quot; approach to the problem. In addition, we provide the learning algorithms with many additional linguistic knowledge sources for coreference resolution. Unfortunately, we find that performance drops significantly when using the full feature set; we attribute this, at least in part, to the system's poor performance on common noun resolution and to data fragmentation problems that arise with the larger feature set. Manual feature selection, with an eye toward eliminating low-precision rules for common noun resolution, is shown to reliably improve performance over the full feature set and produces the best results to date on the MUC-6 and MUC-7 coreference data sets -- F-measures of 70.4 and 63.4, respectively. Nevertheless, there is substantial room for improvement. As noted above, for example, it is important to automate the precision-oriented feature selection procedure as well as to investigate other methods for feature selection. We also plan to investigate previous work on common noun phrase interpretation (e.g. Sidner (1979), Harabagiu et al.</Paragraph>
    <Paragraph position="3"> (2001)) as a means of improving common noun phrase resolution, which remains a challenge for state-of-the-art coreference resolution systems.</Paragraph>
  </Section>
class="xml-element"></Paper>