<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0903">
  <Title>Constructing Text Sense Representations</Title>
  <Section position="3" start_page="2" end_page="2" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> Our notion of semantics is closely related to the notion of &amp;quot;naive semantics&amp;quot; discussed in (K. Dahlgren  In this paper, we would like to extend the notion of &amp;quot;Word Sense&amp;quot; onto &amp;quot;Text Sense&amp;quot;, i.e. texts of arbitrary length  Allen instead uses the term &amp;quot;logical form&amp;quot; for this kind of context-independent meaning representation, c.f. (Allen, 1995), p.14 et al., 1989). This article describes &amp;quot;naive semantics&amp;quot; as &amp;quot;a level of world knowledge that is general and common to many speakers of a language&amp;quot;, i.e. commonsense knowledge associated with words. Naive semantics identifies words with concepts which vary in type.</Paragraph>
    <Paragraph position="1"> A discussion of fundamental corpus-related aspects of word senses is provided by Kilgariff (Adam Kilgariff, 1997). Kilgariff herein questions the use of word sense disambiguation and concludes that word senses can only be defined &amp;quot;relative to a set of interests&amp;quot; and the &amp;quot;basic units of word meanings&amp;quot; are occurrences of words in contexts. Our notion of TSR trees aims at aggregating text meaning in it's topical context in order to construct a context independent representation.</Paragraph>
    <Paragraph position="2"> In Literature, there are several strong directions of representing text meaning or text sense: one prominent approach uses frame-based representation languages in combination with first order logic semantics. The analyzed text is matched against the frame database in order to construct text meaning representations. An example of this approach is presented by Clark et. al. (P. Clark et al., 2003).</Paragraph>
    <Paragraph position="3"> Dahlgren et. al. present &amp;quot;KT&amp;quot;, a complex text understanding system based on naive semantics. KT also uses frames to represent semantic content.</Paragraph>
    <Paragraph position="4"> A project that is based on a roughly similar notion of text meaning representation (TMR) concepts is the ukosmos project (Mahesh, 1996; Kavi Mahesh and Sergei Nirenburg, 1996). It is aimed at the creation of a machine translation system that uses a broad-coverage ontology and various input sources in order to translate english to spanish texts and vice versa. TMR concepts within ukosmos are hand-written frame-based data structures. Text meaning is represented by instances thereof that are derived by semantic rules from a linguistic rule database.</Paragraph>
    <Paragraph position="5"> Frame-based meaning representations are also the basis of AutoSlog-TS, an information extraction system that automatically acquires conceptual patterns from untagged texts, using only a preprocessed training corpus (Ellen Riloff and Jay Shoen, 1995). The thusly constructed concepts can be seen as text meaning representations.</Paragraph>
    <Paragraph position="6"> Approaches of computing text meaning similarities include using web directories for generating path-shaped data structures for text categorization (Fabrizio Sebastiani, 2003; Giuseppe Attardi et al., 1998). Sebastiani herein purports his efforts in by mining the structure of both web &amp;quot;catalogues&amp;quot; (web directories) for extracting category labels and mining web page structure for the actual classification task. This is an example for using path- and graph based methods rather than frame based structures.</Paragraph>
    <Paragraph position="7"> Another example would be the methodology described in this article.</Paragraph>
  </Section>
class="xml-element"></Paper>