<?xml version="1.0" standalone="yes"?>
<Paper uid="E95-1014">
  <Title>Corpus-based Method for Automatic Identification of Support Verbs for Nominalizations</Title>
  <Section position="4" start_page="99" end_page="99" type="intro">
    <SectionTitle>
3 Method
</SectionTitle>
    <Paragraph position="0"> In this section we explain our method for extracting support verbs for nominalizations.</Paragraph>
    <Paragraph position="1"> We suppose that we are given a pair of words: ayerb and its nominalized form. As explained in the previous section, we are interested in extracting only nominalized forms which have not become concrete nouns, and that this will be done by comparing syntactic structures attached to the verb and noun forms. In order to extract corpus evidence related to these phenomena, we proceed as follows: 1. We generate all the morphologically related forms of the word pair using a lexical transducer for English (Karttunen et al., 1992). This list of words will be used as corpus filter.</Paragraph>
    <Paragraph position="2">  (Grefenstette, 1994), we derive the local syntactic patterns involving the verbal form and the nominalized form.</Paragraph>
    <Paragraph position="3"> 5. Considering that nominalized forms retain some of the verbal characteristics of the underlying predicate, we want to extract the most common argument/adjunct structures found around verbal uses of the predicate. As an approximation, we extract here all the prepositional phrases found after the verb.</Paragraph>
    <Paragraph position="4"> 6. For nominal forms, we select only those uses which involve argument/adjunct structures similar to phrases extracted in the previous step. For these selected nominalized forms, we extract the verbs of which these forms are the direct  object. We sort these verbs by frequency.</Paragraph>
    <Paragraph position="5"> 7. This sorted list is the list of candidate support verbs for the nominalization.</Paragraph>
    <Paragraph position="6"> This method assumes that the verb and the nominalized form of the verb are given. We have experimented with automatically extracting the nominalized form by using the prepositional patterns extracted for the verb in step 5. We extracted 6 megabytes of newspaper articles containing a form of the verb propose: propose, proposes, proposed, proposing. Since one use of nominalization is to avoid repetition of the verb form, we suppose that the nominalization of propose is likely to appear in the same articles. We extracted the three most common prepositions following a form of propose (step 5). We then extracted the nouns appearing in these same artigles and which preceded these prepositions.~oThe results 4 appear in figure 1. Since a nomilmlized form is normally morphologically related to the verb form, almost any morphological comparison method will pick proposal from this list.</Paragraph>
    <Paragraph position="7"> 4Further experimentation has confirmed these results, but indicate that it may sufficient to simply tag a text, and perform morphological comparison with the most commonly cooccurring nouns in order to extract the nominalized forms of verbs.</Paragraph>
  </Section>
class="xml-element"></Paper>