<?xml version="1.0" standalone="yes"?>
<Paper uid="A88-1011">
  <Title>TRIPHONE ANALYSIS: A COMBINED METHOD FOR THE CORRECTION OF ORTHOGRAPHICAL AND TYPOGRAPHICAL ERRORS.</Title>
  <Section position="6" start_page="81" end_page="81" type="concl">
    <SectionTitle>
4. CONCLUSION
</SectionTitle>
    <Paragraph position="0"> We have demonstrated that an integration of complementary correction methods performs better than single methods. With respect to orthographical errors, triphone analysis performs better than either grapheme-to-phoneme conversion or trigram analysis alone. Its capacity to correct typographical errors is still to be evaluated, but it is already clear that it will be better than that of SPELL THERAPIST although somewhat worse than trigram analysis in those cases where a typographical error drastically alters the pronunciation. In practice, however, one always finds both kinds of errors. Therefore, it would be interesting to compare the various methods in actual use. Future research will go into a number of variants on the basic ideas presented here. From a linguistic point of view, it is possible to make the phonological matching less stringent. One way to do this is to use a comparison at the level of phonological features rather than phonemes. However, greater emphasis on orthographical errors may deteriorate performance on the correction of typing errors.</Paragraph>
    <Paragraph position="1"> An area of current research is the extension of triphone analysis toward the correction of compounds.</Paragraph>
    <Paragraph position="2"> In languages like Dutch and German, new compounds such as taaltechnologie (language technology) are normally written as one word. Correction of errors in such compounds is difficult because the constituting words should be corrected separately but there is no 82, easy way to find the right segmentation. We have developed some heuristics to solve this problem.</Paragraph>
    <Paragraph position="3"> Of course, other combinations of methods are possible. One possibility which looks promising is to combine phonemic transcription with the PF-474 chip. Although triphone analysis is fairly fast, use of the PF-474 chip might further increase the speed. For the correction of large quantities of word material, speed is an essential factor. However, it should be kept in mind that there is a linear correlation between the size of the dictionary and the required processing time, and that the correlation curve is steeper for the PF-474 chip than for triphone analysis. This means that triphone analysis will still be faster for very large dictionaries.</Paragraph>
    <Paragraph position="4"> With an eye to commercial applications, TNO-ITI is extending the basic method with data compression techniques and an improved formalism for grapheme-to-phoneme conversion.</Paragraph>
  </Section>
class="xml-element"></Paper>