<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1082">
  <Title>Tagging with Hidden Markov Models Using Ambiguous Tags</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
4 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> We have presented a method for computing the probability distributions associated to ambiguous tags, denoting subsets of the tagset, in an HMM based part of speech tagger. An iterative method for discovering ambiguous tags, based on the mistakes made by the tagger allowed to reach a recall of 0:982 for an ambiguity rate of 1:232. These flgures can be compared to the baseline model which achieves a recall of 0:979 and an ambiguity rate of 1:418 using the same ambiguous tags. An analysis of ambiguous tags showed that they do not always behave in the way expected; some of them introduce a lot of ambiguity without correcting many mistakes.</Paragraph>
    <Paragraph position="1"> This work will be developed in two directions.</Paragraph>
    <Paragraph position="2"> The flrst one concerns the study of the difierent behaviour of ambiguous tags which could be in uenced by computing difierently the flctitious counts of each ambiguous tag, based on its behaviour on a development corpus in order to force or prevent its introduction during tagging.</Paragraph>
    <Paragraph position="3"> The second direction concerns experiments on supertagging (Bangalore and Joshi, 1999) followed by a parsing stage the tagging stage associates to each word a supertag. The supertags are then combined by the parser to yield a parse of the sentence. Errors of the supertagger (almost one out of 5 words is attributed the wrong supertag) often impede the parsing stage. The idea is therefore to allow some ambiguity during the supertagging stage, leaving to the parser the task of selecting the right supertag using syntactic constraints that are not available to the tagger. Such experiments will constitute one way of testing the viability of our approach.</Paragraph>
  </Section>
class="xml-element"></Paper>