<?xml version="1.0" standalone="yes"?>
<Paper uid="P98-1059">
  <Title>Spelling Correction Using Context*</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> This paper describes how context-dependent spelling correction is performed in a natural language dialogue system under control of the parser. Our spelling correction system is a functioning part of an intelligent tutoring system called Circsim-Tutor \[Elmi, 94\] designed to help medical students learn the language and the techniques for causal reasoning necessary to solve problems in cardiovascular physiology. The users type in answers to questions and requests for information.</Paragraph>
    <Paragraph position="1"> In this kind of man-machine dialogue, spelling correction is essential. The input is full of errors.</Paragraph>
    <Paragraph position="2"> Most medical students have little experience with keyboards and they constantly invent novel abbreviations. After typing a few characters of a long word, users often decide to quit. Apparently, the user types a few characters and decides that (s)he has given the reader enough of a hint, so we get 'spec' for 'specification.' The approach to spelling correction is necessarily different from that used in word processing or other authoring systems, which submit candidate corrections and ask the user to make a selection. Our system must make automatic corrections and make them rapidly since the system has only a few seconds to parse the student input, update the student model, plan the appropriate response, turn it into sentences, and display those sentences on the screen.</Paragraph>
    <Paragraph position="3"> Our medical sublanguage contains many long *This work was supported by the Cognitive Science Program, Office of Naval Research under Grant No. N00014-941-0338, to Illinois Institute of Technology. The content does not reflect the position or policy of the government and no official endorsement should be inferred.</Paragraph>
    <Paragraph position="4"> phrases that are used in the correction process. Our filtering system is adaptive; it begins with a wide acceptance interval and tightens the filter as better candidates appear. Error weights are position-sensitive. The parser accepts several replacement candidates for a misspelled string from the spelling corrector and selects the best by applying syntactic and semantic rules. The selection process is dynamic and context-dependent. We believe that our approach has significant potential applications to other types of man-machine dialogues, especially speech-understanding systems. There are about 4,500 words in our lexicon.</Paragraph>
  </Section>
class="xml-element"></Paper>