<?xml version="1.0" standalone="yes"?>
<Paper uid="W98-0604">
  <Title>Using NOMLEX to Produce Nominalization Patterns for Information Extraction</Title>
  <Section position="2" start_page="0" end_page="25" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Although, nominalizationQ are very common in written text, the computational linguistics literature provides few systematic accounts of how to deal with phrases containing these words. This paper focuses on this problem in the context of Information Extraction (IE). 2 Many extraction systems use either parsing combined with some form of syntactic regularization, or a meta-rule mechanism to automatically match variants of clausal syntactic structures (active main clause, passive, relative clause etc.), e.g., FASTUS (Appelt et al., 1995) and the Proteus Extraction System (Grishman, 1995). However, this mechanism does not extend to nominalization patterns, which must be coded separately from the clausal patterns. NOMLEX, a dictionary of nominalizations currently under development at NYU, (Macleod et al., 1997) provides a way to handle nominalizations more automatically, and with INominalizations are nouns which are related to words of another part of speech, most commonly verbs. In this paper, only verbal nominalizatious will be discussed.</Paragraph>
    <Section position="1" start_page="0" end_page="25" type="sub_section">
      <SectionTitle>
2The Message Understanding Colfference Scenario
</SectionTitle>
      <Paragraph position="0"> Template Task (MUC, 1995), (MUC, 1998) is ore&amp;quot; model for the kind of information that we are attempting to extract (who does what to whom, and when and where).</Paragraph>
      <Paragraph position="1">  greater coverage. NOMLEX includes information about mappings between verbs and nominalizations that will help generalize information from verbal patterns to create nominalization patterns. This paper describes the structure of the dictionary and a procedure for creating nominalization patterns. This procedure takes into account lexical information about nominalizations which is encoded in NOMLEX.</Paragraph>
      <Paragraph position="2"> The Proteus Extraction System starts with a semantic pattern for an active clause: np(C-company) vg(appoint) np(C-person) &amp;quot;as&amp;quot; np(C-position) which matches a clause beginning with a noun phrase headed by a noun of type COMPANY, followed by a verb group (verb plus auxilliaries) headed by appoint, a noun phrase headed by a noun of type PERSON, the literal as, and a noun phrase headed by a noun of type position, e.g., IBM appointed Alice Smith as vice president. Proteus applies meta-rules to this pattern to produce new patterns for other clausal types, e.g., a passive clause: np(C-person) vg-pass(appoint) &amp;quot;as&amp;quot; np(C-position) &amp;quot;by&amp;quot; np(C-company) (vg-pass is a passive verb group). This new pattern would match Alice Smith was appointed as vice president by IBM. When a pattern matches input text, the pieces of the text corresponding to the constituents of the pattern are used to build a semantic representation of the text.</Paragraph>
      <Paragraph position="3"> To avoid the need for having users code such patterns, we have developed the Proteus Extraction Tool (PET) (Yangarber and Grishman, 1997). PET allows the user to input an example sentence and specify the mappings from syntactic to semantic form. The system then produces generalized patterns to perform these mappings.</Paragraph>
      <Paragraph position="4"> This paper shows how PET can use NOMLEX to create nominalization patterns as well. For example, given the sentence IBM appointed Alice Smith as vice president, human input and dictionary entries identify IBM as the employer, Alice Smith as the employee, and vice president as the position. The meta-rules add a slot for temporal PPs which state the date (e.g., on June 1, 1998). PET creates patterns to fill the semantic slots (employer, employee, position) from the gramatical roles (subject, object, NP following as, etc.) in the sentence. PET generates patterns to cover passive sentences, active sentences and relative clauses. Enhanced with NOMLEX, PET can also cover examples like Alice Smith's appointment as vice president; IBM's June 1, 1998 appointment of Alice Smith; and the June 1, 1998 appointment of Alice Smith by IBM. The correspondence between nominal and verbal positions is determined by explicit information in the NOMLEX dictionary entry and by general linguistic constraints.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>