<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0612">
  <Title>Population Testing: Extracting Semantic Information On Near-Synonymy From Native Speakers</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
3 Visualizing Near-Synonymy in a Quasi
3-D Space
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.1 The Need for an Intuitive Way to View
and Review Semantic Information
</SectionTitle>
      <Paragraph position="0"> While we are in general satisfied with the results from the first round of our pilot study, we have come to realize that, in order to pass the results back to the informants for the next iteration of our test process, we need to present our findings in a more intuitive format.</Paragraph>
      <Paragraph position="1"> Furthermore, as researchers engaging in the design of modern electronic dictionaries and thesauri for human users, we are interested in creating a user friendly interface for a thesaurus like application. Indeed, we have in mind that our informants would also be the users of such a thesaurus/dictionary, and thus have an incentive to make contribution to its continuing update and improvement. The general configuration of such a setup is illustrated in Fig. 4 below and described in more detail in a forthcoming paper (Vanhatalo 2003).</Paragraph>
      <Paragraph position="2"> Fig. 4 A New Kind of Dictionary/Thesaurus The convergence of these interests and requirements resulted in the prototype visualization tool, currently implemented as a Java applet, described in the following sections.</Paragraph>
      <Paragraph position="3"> Fig. 5 Single-Axis Layout Based on gender</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Population
Tests
Analysis
Dictionary
/Thesaurus
</SectionTitle>
      <Paragraph position="0"> The screen-shot of the visualization tool in Fig. 5 above contains essentially the same information as in the bar chart of Fig. 3, except here the words themselves are the main objects being displayed. The words are distributed along a diagonal axis based on gender (of the agent), with lower left being more &amp;quot;male&amp;quot;-like, and upper right being more &amp;quot;female&amp;quot;-like.</Paragraph>
      <Paragraph position="1"> The view shown in Fig. 6 is similar, but in this case we use x-axis for gender and the y-axis for volume. In other words, Fig. 6 contains the same information as those in Fig. 2 and Fig. 3 combined.</Paragraph>
      <Paragraph position="2"> Actually, there is more. In both Fig. 5 and Fig 6 there is a third (z-) dimension shown via type size and color. This dimension is currently used to represent the semantic distance of each word from the focus, i.e. the currently selected word of interest highlighted in a box.</Paragraph>
      <Paragraph position="3"> The basic idea is that the word of interest would be closest to the viewer, and thus largest in type and darkest in color; while the other words (its near-synonyms) will be nearer or further from the viewer depending on how close they are semantically to this focus word. In other words, we hypothesize that the viewer would have an intuitive feel for this notion of 'semantic distance', and that he or she would instinctively translate this mental distance into a perceived visual distance, and vice versa.</Paragraph>
      <Paragraph position="4"> Fig. 6 Dual-Axis Layout (gender x volume) For example, in both Fig. 5 and Fig. 6 above, where the semantic distance is calculated as a weighted average of six semantic dimensions, one could see at a glance which word(s) are the closest near-synonyms to the selected focus word. Thus, in Fig. 5 for vaikeroida it is voivotella; while in Fig. 6 for nurista, they are purnata and mukista.</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.2 Towards a Web-Based, Visually Enriched
Extension to PTM
</SectionTitle>
      <Paragraph position="0"> We envision this Web-based, visual extension to PTM to work mostly the same as the paper-and-pencil version, except that (1) it would be conducted over the Internet; and (2) some, though not all, of the tests would be more visually oriented.</Paragraph>
      <Paragraph position="1"> More specifically, the visual tests would still consist of both multiple-choice tasks and open-ended tasks as before.</Paragraph>
      <Paragraph position="2"> For an open-ended task, one option is to present one of these displays, say Fig. 5 or Fig. 6 above, and ask the informants if the picture makes sense, and if not, to explain in their own words what in the picture appears odd to them, and why.</Paragraph>
      <Paragraph position="3"> As to multiple-choice tasks, consider the four views in Fig. 7a - 7d below: Fig. 7a Fig. 7b Fig. 7c Fig. 7d These four views are essentially variations on the same theme, i.e. they have the same x- and y-axis layout and the same focus word. The only difference among them is in the weight assignment for calculating semantic distance from the focus word jakattaa. The viewer will be asked to rank the four displays in terms of their &amp;quot;naturalness&amp;quot;.</Paragraph>
      <Paragraph position="4"> Another way to do this will allow the weights to be assigned by the viewer directly, e.g. via a set of sliders similar to those for photo manipulation programs. While this would require more work for the informant, it could actually be more fun and thus perhaps would have a greater potential as a successful method for an Internet based approach.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>