<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-0821">
  <Title>Meeting of the Association for Computational</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
Abstract
</SectionTitle>
    <Paragraph position="0"> For SENSEVAL-3, the University of Maryland (UMD) team focused on two primary issues: the portability of sense disambigation across languages, and the exploitation of real-world bilingual text as a resource for unsupervised sense tagging. We validated the portability of our supervised disambiguation approach by applying it in seven tasks (English, Basque, Catalan, Chinese, Romanian, Spanish, and &amp;quot;multilingual&amp;quot; lexical samples), and we experimented with a new unsupervised algorithm for sense modeling using parallel corpora.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
Samples
1.1 Tagging Framework
</SectionTitle>
      <Paragraph position="0"> For the English, Basque, Catalan, Chinese, Romanian, Spanish, and &amp;quot;multilingual&amp;quot; lexical samples, we employed the UMD-SST system developed for SENSEVAL-2 (Cabezas et al., 2001); we refer the reader to that paper for a detailed system description. Briefly, UMD-SST takes a supervised learning approach, treating each word in a task's vocabulary as an independent problem of classification into that word's sense inventory. Each training and test item is represented as a weighted feature vector, with dimensions corresponding to properties of the context. As in SENSEVAL-2, our system supported the following kinds of features:</Paragraph>
      <Paragraph position="2"> each word a2 in the vocabulary, there is a feature a3a5a4a7a6a8a2 representing the presence of word a2 at a distance of a1 words to the left of the word being disambigated; there is a corresponding set of features a9a10a4a10a6a11a2 for the local context to the right of the word.</Paragraph>
      <Paragraph position="3"> a0 Wide context. Each word a12 in the training set vocabulary has a corresponding feature indicating its presence. For SENSEVAL-3, wide context features were taken from the entire training or test instance. In other settings, one might make further distinctions, e.g. between words in the same paragraph and words in the document.</Paragraph>
      <Paragraph position="4"> We also experimented with the following additional kinds of features for English: a0 Grammatical context. We use a syntactic dependency parser (Lin, 1998) to produce, for each word to be disambiguated, features identifying relevant syntactic relationships in the sentence where it occurs. For example, in the sentence The U.S. government announced a new visa waiver policy, the word government would have syntactic features like DET:THE, MOD:U.S., and SUBJ-OF:ANNOUNCED.</Paragraph>
      <Paragraph position="5"> a0 Expanded context. In information retrieval, we and other researchers have found that it can be useful to expand the representation of a document to include informative words from similar documents (Levow et al., 2001). In a similar spirit, we create a set of expandedcontext features a13a15a14a8a16a17a6a18a13 by (a) treating the WSD context as a bag of words, (b) issuing it as a query to a standard information retrieval system that has indexed a large collection of documents, and (c) including the nonstopword vocabulary of the top a19 documents returned. So, for example, in a context containing the sentence The U.S. government announced a new visa waiver policy, the query might retrieve news articles like &amp;quot;US to Extend Fingerprinting to Europeans, Japanese&amp;quot; (Bloomberg.com, April 2, 2004), leading to the addition of features like EXT:EUROPEAN,  English lexical sample task As described by Cabezas et al. (2001), we have adopted the framework of support vector machines (SVMs) in order to perform supervised classification. Because we used a version of SVM learning designed for binary classification tasks, rather than the multi-way classification needed for disambiguating among a0 senses, we constructed a family of SVM classifiers for each word a2 -- one for each of the word's a0a2a1 senses. All positive training examples for a sense a3 a4 of a2 were treated as negative training examples for all the other senses a3a5a4 , a6a8a7a9 a1 . Table 1 shows the performance of our approach on the English lexical sample task from the previous SENSEVAL exercise (SENSEVAL-2), including the basic system (UMD-SST), the basic system with grammatical features added (UMD-SSTgram), and the basic system with document expansion features added (UMD-SST-docexp). (We have not done a run with both sets of features added.) The results showed a possible potential benefit for using grammatical features, in the fine-grained scoring. However, we deemed the benefit too small to rely upon, and submitted our official SENSEVAL-3 runs using UMD-SST without the grammatical or document-expansion features.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
1.2 SENSEVAL-3 Lexical Sample Tasks
</SectionTitle>
      <Paragraph position="0"> For SENSEVAL-3, the modularity of our system made it very easy to participate in the many lexical sample tasks, including the multilingual lexical sample, where the &amp;quot;sense inventory&amp;quot; consisted of vocabulary items from a second language.1 Indeed, we participated in several tasks without having anyone on the team who could read the language. (Whether or not this was a good idea remains to be seen.) For Basque, Catalan, and Spanish, we used the lemmatized word forms provided by the task organizers; for the other languages, including English, we used only simple tokenization.</Paragraph>
      <Paragraph position="1"> Table 2 shows the UMD-SST system's official 1Data format problems prevented us from participating in the Italian lexical sample task.</Paragraph>
      <Paragraph position="2">  English lexical sample task SENSEVAL-3 performance on the lexical sample runs in which we participated, using fine-grained scores.</Paragraph>
      <Paragraph position="3"> In unofficial runs, we also experimented with the grammatical and document-expansion features. Table 3 shows the results, which indicate that on this task the additional features did not help and may have hurt performance slightly. Although we have not yet reached any firm conclusions, we conjecture that value potentially added by these features may have been offset by the expansion in the size of the feature space; in future work we plan to explore feature selection and alternative learning frameworks.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>