<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1103">
  <Title>investigations</Title>
  <Section position="9" start_page="21" end_page="21" type="concl">
    <SectionTitle>
6 Conclusion and future work
</SectionTitle>
    <Paragraph position="0"> This paper presented some ideas from two angles of study (human and metrics) into the intricate problem of similarity judgments. A larger study  is under way on both angles. First, we suggested, based on some psychological and philosophical model analysis, a two-axis Osgood-like benchmarking approach for &amp;quot;ordinary human&amp;quot; word-sense judgments. We intend to perform an empirical experiment to validate this idea by looking at inter-judge agreement.</Paragraph>
    <Paragraph position="1"> On the algorithm side, although the approaches based on the cardinality of sets are not central to WSD, we presented them first as we find it inspiring to see an effort of classification on those measures. We then attempted a somewhat more broad classification by emphasizing properties of different groups of similarity measures: cardinality of sets, distance, probabilistic measures and angular metrics. Although each group has a particular subset of properties, we noted that all of them share a property of transitivity. This is interestingly different from the psychological contrast model of Tversky where differences and similarities are measured differently on different criteria. We think investigations into similarity measures which reproduce such a non-transitive differentiation approach should be performed. We are on that path in our larger study. We also suggest that any proposal of a measure for a task should be preceded by a study of which properties seem adequate for such a task. We conclude by opening up the debate for the WSD task.</Paragraph>
  </Section>
class="xml-element"></Paper>