<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1037">
  <Title>Optimizing disambiguation in Swahili</Title>
  <Section position="6" start_page="14" end_page="14" type="evalu">
    <SectionTitle>
9 Discussion
</SectionTitle>
    <Paragraph position="0"> The disambiguation of a language is a process where the cooperation of linguistic rules and probability should be optimised. It was shown above briefly that different disambiguation operations should be cascaded so that the most reliable disambiguation is carried out first and the least reliable cases last. Multi-word concepts can be handled so that such constructions that do not have inflecting constituent parts are treated as part of morphology, and those with inflecting parts, especially idioms, are handled with disambiguation rules. We have also seen that linguistic rules should precede rules based on probability. It is also possible to simplify the writing of semantic rules by constructing the morphological parser so that semantic readings come in order of frequency, whereby the most frequent interpretation is considered a default case, and only other interpretations need rules.</Paragraph>
    <Paragraph position="1"> The experiments with the SOM algorithm indicate that it is possible to find significant relationships between adjacent words on the one hand and between words and tags on the other.</Paragraph>
    <Paragraph position="2"> Such information can then be encoded in the morphological dictionary and used in generalising disambiguation rules. Ambiguity resolution can be enhanced further by constructing explicit dependencies between constituent parts of a sentence (Jarvinen and Tapanainen 1997; Tapanainen and Jarvinen 1997; Tapanainen 1999) or by making use of a parse tree bank of the type of WordNet (Hirst and Onge 1998).</Paragraph>
  </Section>
class="xml-element"></Paper>