<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-2036">
  <Title>1- A Spelling Correction Program Based on a Noisy Channel Model</Title>
  <Section position="4" start_page="206" end_page="210" type="concl">
    <SectionTitle>
5. Conclusions
</SectionTitle>
    <Paragraph position="0"> There have been a number of spelling correction programs in the past such as Kucera (1988) that generated a list of candidates by looking for insertions, deletions, substitutions and reversals, rauch as we have been doing here. Our contribution is the emphasis on scoring.</Paragraph>
    <Paragraph position="1"> Mcllroy, the author of the Unix spell program (1982), intentionally focused on the spelling detection problem, and argued (private communication) that spelling correction was a bad idea so long as the corrector couldn't separate the plausible candidates from the implausible ones. He felt that it was probably more distracting than helpful to bury the user under a long list of mostly implausible candidates. In this work, we have attempted to show that it is possible to sort the candidates by a likelihood function that agrees well enough with human judges to be helpful.</Paragraph>
    <Paragraph position="2"> In future work, we would hope to extend the prior model to take advantage of context. We noticed that the human judges were extremely reluctant to cast a vote given only the information available to the program, and that they were much more comfortable when they could see a concordance line or two. Perhaps our program could take advantage of these contextual cues by adopting very simple language modeling techniques such as trigrams, that have proven effective for speech recognition applications (Jelinek, 1985). Hopefully more interesting language models would improve</Paragraph>
    <Paragraph position="4"/>
  </Section>
class="xml-element"></Paper>