<?xml version="1.0" standalone="yes"?>
<Paper uid="C86-1089">
  <Title>Learning the Space of Word Meanings for Information Retrieval Systems</Title>
  <Section position="6" start_page="378" end_page="378" type="relat">
    <SectionTitle>
5. Related Works
</SectionTitle>
    <Paragraph position="0"> 'PShe most related work is Rieger's\[7,8\]. He also made some kind of semantic space from the infer ~ marion of relations among words. However, he made thc space based only oil tile frequency of co-occurrences of words ill sentences. Our semantic space carl have more realil,y about meo, nings than his, because symbols that are actually connected to entities stored in tile database exist in our st)ace.</Paragraph>
    <Paragraph position="1"> In the sense of treating entities which do not have symbolic represeul, ation, conneetionist models\[4,9\] have some relation with our approach.</Paragraph>
    <Paragraph position="2"> llowever, it seems that the eonneetionists have not yet realized the process of yielding symbols on con ~ nection networks that can represent such abstract notions as dealt with by our system.</Paragraph>
    <Paragraph position="3"> From the methodological point of view, methods developed in the studies of multi-dimensional synthesis in statistics are related to ours. The reason we have developed methods to calculate tile space independently from those methods is similar to one menUoned by Lebowitz\[6\]. That ix, we are noL interested in mathematical rigidness of the process of analyzing data, but in modeling what occurs m tile human mind.</Paragraph>
    <Paragraph position="4"> 1,'rein the viewpoint of information retrievalsysteens, there are many more requests than those answered by our approach. They include treating structures of key words more explicitly, understanding tile user's intention from his query\[l\], or making co-operative responses\[5\]. However, the previous studies on those problems started ignoring tile most basic problem of understanding and learning the meanings of one word. Our work can give an assured starting point for' those further studies.</Paragraph>
  </Section>
class="xml-element"></Paper>