<?xml version="1.0" standalone="yes"?>
<Paper uid="P92-1029">
  <Title>Association-based Natural Language Processing with Neural Networks</Title>
  <Section position="10" start_page="229" end_page="230" type="concl">
    <SectionTitle>
6 Discussion
</SectionTitle>
    <Paragraph position="0"> Although the result of conversion test is satisfactory, word associations by neural network are not human-like ones yet. Following is a list of improvements that many further enhance the system: * Improvements for generating a network The quality of the network depends on how to reduce noisy word occurrence in the network from the point of view of association. The existence of noisy words is inevitable in automatic generation but plays a role to make unwanted associations. One approach to reducing noisy words is to identify those words which are context independent and remove them from the network generation stage. The identification can be based on word categories and meanings. In most cases, words representing very abstract concepts are noisy because they force unwanted activations in unrelated contexts. Therefore they should be detected through experiments. Another problem arises because of the ambiguity of morphological analysis. Word extraction from real documents is not always correct because of the agglutinative nature of the Japanese language. Other possibility for network improvement is to consider a syntactic relationship or co-occurrence relationship while deciding link weights. In addition, there are keywords in a document in general which play a central role in association. They will be reflected in a network more in consideration of technical terms.</Paragraph>
    <Paragraph position="1"> Preference decision in kana-kanji conversion null The reinforcement of associative information complicates the decision of homonym preference in kana-kanji conversion. We already have several means of semantic disambiguation of homonyms: co-occurrence restrictions and selectional restrictions. As building a complete thesaurus is very difficult, our thesaurus is still not enough to select the correct meaning(kanfi-conversion) of kanawritten word. So selectional restrictions should be weak constraints in homonym selection. In the same vein, associative information should be considered a weak constraint because associations by neural networks are not always reliable. Possible conflict between selectional restrictions and associative information, added to tile grammatical ambiguities remaining in the stage of homonym selection, make kanji selection very complex. The problem of multiply and weakly constrained  homonyms is one to which we have not yet found the best solution.</Paragraph>
  </Section>
class="xml-element"></Paper>