<?xml version="1.0" standalone="yes"?>
<Paper uid="P02-1032">
  <Title>The Descent of Hierarchy, and Selection in Relational Semanticsa0</Title>
  <Section position="8" start_page="0" end_page="0" type="relat">
    <SectionTitle>
6 Related Work
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
6.1 Noun Compound Relation Assignment
</SectionTitle>
      <Paragraph position="0"> Several approaches have been proposed for empirical noun compound interpretation. Lauer &amp; Dras (1994) point out that there are three components to the problem: identification of the compound from within the text, syntactic analysis of the compound (left versus right association), and the interpretation of the underlying semantics. Several researchers have tackled the syntactic analysis (Lauer, 1995), (Pustejovsky et al., 1993), (Liberman and Church, 1992), usually using a variation of the idea of finding the subconstituents elsewhere in the corpus and using those to predict how the larger compounds are structured.</Paragraph>
      <Paragraph position="1"> We are interested in the third task, interpretation of the underlying semantics. Most related work relies on hand-written rules of one kind or another. Finin (1980) examines the problem of noun compound interpretation in detail, and constructs a complex set of rules. Vanderwende (1994) uses a sophisticated system to extract semantic information automatically from an on-line dictionary, and then manipulates a set of hand-written rules with handassigned weights to create an interpretation. Rindflesch et al. (2000) use hand-coded rule-based systems to extract the factual assertions from biomedical text. Lapata (2000) classifies nominalizations according to whether the modifier is the subject or the object of the underlying verb expressed by the head noun.</Paragraph>
      <Paragraph position="2"> Barker &amp; Szpakowicz (1998) describe noun compounds as triplets of information: the first constituent, the second constituent, and a marker that can indicate a number of syntactic clues. Relations are initially assigned by hand, and then new ones are classified based on their similarity to previously classified NCs. However, similarity at the lexical level means only that the same word occurs; no generalization over lexical items is made. The algorithm is assessed in terms of how much it speeds up the hand-labeling of relations. Barrett et al. (2001) have a somewhat similar approach, using WordNet and creating heuristics about how to classify a new NC given its similarity to one that has already been seen.</Paragraph>
      <Paragraph position="3"> In previous work (Rosario and Hearst, 2001), we demonstrated the utility of using a lexical hierarchy for assigning relations to two-word noun compounds. We use machine learning algorithms and MeSH to successfully generalize from training instances, achieving about 60% accuracy on an 18-way classification problem using a very small training set. That approach is bottom up and requires good coverage in the training set; the approach described in this paper is top-down, characterizing the lexical hierarchies explicitly rather than implicitly through machine learning algorithms.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
6.2 Using Lexical Hierarchies
</SectionTitle>
      <Paragraph position="0"> Many approaches attempt to automatically assign semantic roles (such as case roles) by computing semantic similarity measures across a large lexical hierarchy; primarily using WordNet (Fellbaum, 1998). Budanitsky &amp; Hirst (2001) provide a comparative analysis of such algorithms. null However, it is uncommon to simply use the hierarchy directly for generalization purposes. Many researchers have noted that WordNet's words are classified into senses that are too fine-grained for standard NLP tasks. For example, Buitelaar (1997) notes that the noun book is assigned to seven different senses, including fact and section, subdivision. Thus most users of WordNet must contend with the sense disambiguation issue in order to use the lexicon.</Paragraph>
      <Paragraph position="1"> The most closely related use of a lexical hierarchy that we know of is that of Li &amp; Abe (1998), which uses an information-theoretic measure to make a cut through the top levels of the noun portion of WordNet. This is then used to determine acceptable classes for verb argument structure, and for the prepositional phrase attachment problem and is found to perform as well as or better than existing algorithms.</Paragraph>
      <Paragraph position="2"> Additionally, Boggess et al. (1991) &amp;quot;tag&amp;quot; veterinary text using a small set of semantic labels, assigned in much the same way a parser works, and describe this in the context of prepositional phrase attachment.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>