<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-1020">
  <Title>User-Friendly Text Prediction for Translators</Title>
  <Section position="8" start_page="0" end_page="0" type="concl">
    <SectionTitle>
7 Conclusion and Future Work
</SectionTitle>
    <Paragraph position="0"> We have described an approach to text prediction for translators that is based on maximizing the benefit to the translator according to an explicit user model whose parameters were set from data collected in user evaluations of an existing text prediction prototype. Using this approach, we demonstrate in simulated results that our current predictor can reduce the time required for an average user to type a text in the domain of our training corpus by over 10%.</Paragraph>
    <Paragraph position="1"> We look forward to corroborating this result in tests with real translators.</Paragraph>
    <Paragraph position="2"> There are many ways to build on the work described here. The statistical models which are the backbone of the predictor could be improved by making them adaptive--taking advantage of the user's input--and by adding features to capture the alignment relation between h and s in such a way as to preserve the efficient search properties. The user model could also be made adaptive, and it could be enriched in many other ways, for instance so as to capture the propensity of translators to accept at the beginnings of words.</Paragraph>
    <Paragraph position="3"> We feel that the idea of creating explicit user models to guide the behaviour of interactive systems is likely to have applications in areas of NLP apart from translators' tools. For one thing, most of the approach described here carries over more or less directly to monolingual text prediction, which is an important tool for the handicapped (Carlberger et al., 1997). Other possibilities include virtually any application where a human and a machine communicate through a language-rich interface.</Paragraph>
  </Section>
class="xml-element"></Paper>