<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-0713">
  <Title>Sharing Problems and Solutions for Machine Translation of Spoken and Written Interaction</Title>
  <Section position="6" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusions
</SectionTitle>
    <Paragraph position="0"> This paper demonstrates that many of the problems which complicate translation of spoken interaction are shared by written interaction such as chat and instant messaging. It is proposed that solutions developed to solve similar problems in the two communication environments can be profitably shared, and several examples are presented where mutually beneficial approaches might be developed.</Paragraph>
    <Paragraph position="1"> Specifically, we noted that the problem of discriminating unrecognized OOV words from in-vocabularly words in spoken interaction is analogous to the problem of discriminating unrecognized OOV words from misspelled words in written interaction. We suggested that some of the same methods used in spelling correction might be adapted to speech recognition, especially language models that incorporate probabilities of errors in specific positions in the word. We also observed the potential of preprocessing operations that structure input for translation systems to allow special treatment of problematic language, including the possibility that much complexity can be avoided by processing and translating smaller units separately. We look forward to exploring these possibilities in future work.</Paragraph>
  </Section>
class="xml-element"></Paper>