<?xml version="1.0" standalone="yes"?>
<Paper uid="W99-0208">
  <Title>Coreference resolution in dialogues in English and Portuguese</Title>
  <Section position="5" start_page="59" end_page="59" type="concl">
    <SectionTitle>
5. Conclusion and future developments
</SectionTitle>
    <Paragraph position="0"> The process of building solutions for natural language processing on the basis of corpus information may rely simply on a classification model of any kind that would enable decision trees to be created inductively. However, the direct observation of corpus tokens allows the sort of refinement that may prove crucial for the actual operational success of the model in real-life processing situations. The approach described in this paper is an attempt to find an appropriate balance between the practicality of automatically inducing decision trees out of a training set and the thoroughness that the contrastive analysis of the various cases in the corpus is likely to accomplish.</Paragraph>
    <Paragraph position="1"> The systematisation of observed regularities in combination with statistical evidence proved very successful in dealing with the testing set of cases previously analysed for the purpose. It is also true, nevertheless, that the complexity introduced by the inclusion of a large amount of information to be taken into account during the processing makes actual implementation extremely hard. Therefore, the high score of the manual test must be seen cautiously. Future developments of the. approach described in the present paper aim at testing the actual gain of dealing with a thorough account of anaphoric relations in dialogues as compared to the increased difficulty of implementation, of which the inclusion of topicality and segmentation in the model are obvious examples. It is expected that the above mentioned balance will be eventually reached, preserving the satisfactory results to an extent that offsets the undesirable processing complexity.</Paragraph>
  </Section>
class="xml-element"></Paper>