<?xml version="1.0" standalone="yes"?>
<Paper uid="E06-2013">
  <Title>Automatic Annotation for All Semantic Layers in FrameNet</Title>
  <Section position="4" start_page="135" end_page="136" type="metho">
    <SectionTitle>
3 Semantic Entities in FrameNet
</SectionTitle>
    <Paragraph position="0"> The semantic annotation in FrameNet consists of a set of layers. One of the layers defines the target, and the other layers provide additional information with respect to the target. The following layers are used: * The FE layer, which defines the spans and semantic roles of the arguments of the predicate. null * A part-of-speech-specific layer, which contains aspectual information for verbs; and copulas, support expressions, and slot filling information for nouns and adjectives.</Paragraph>
    <Paragraph position="1"> * The &amp;quot;Other&amp;quot; layer, containing special cases such as null arguments.</Paragraph>
    <Paragraph position="2"> The semantic entities that we consider in this article are defined in the second and third of these layers.</Paragraph>
    <Section position="1" start_page="135" end_page="135" type="sub_section">
      <SectionTitle>
3.1 Support Expressions
</SectionTitle>
      <Paragraph position="0"> Some noun targets, typically denoting events, are oftenconstructed using support verbs. Inthiscase, the noun carries most of the semantics (that is, it evokes the frame), while the verb allows the slots of the frame to be filled. Thus, the dependents of a support verb are annotated as FEs, just like for a verb target. Support verbs are annotated using the SUPP label on the Noun or Adjective layer.</Paragraph>
      <Paragraph position="1"> In the following sentence, there is a support verb (underwent) for the noun target (operation).</Paragraph>
      <Paragraph position="2"> [Frances Patterson]PATIENT underwent an operation at RMH today and is expected to be hospitalized for a week or more.</Paragraph>
      <Paragraph position="3"> The support verbs do not change the core semantics of the noun target (that is, they bear no relation to the frame). However, they maydetermine therelation betweentheFEsandthetarget (&amp;quot;pointof-view supports&amp;quot;, such as &amp;quot;undergo an operation&amp;quot; or &amp;quot;perform an operation&amp;quot;) or provide aspectual information (such as &amp;quot;start an operation&amp;quot;).</Paragraph>
      <Paragraph position="4"> The following sentence shows an example where agoverning verb is not asupport verb ofthe noun target. An automatic system must be able to distinguish support verbs from other verbs.</Paragraph>
      <Paragraph position="5"> A senior nurse observed the operation.</Paragraph>
      <Paragraph position="6"> Although alarge majority ofthe support expressions are verbs, there are additionally some cases of support prepositions, such as the following example: null Secret agents of this ilk are at work all the time.</Paragraph>
    </Section>
    <Section position="2" start_page="135" end_page="135" type="sub_section">
      <SectionTitle>
3.2 Copulas
</SectionTitle>
      <Paragraph position="0"> Copular verbs, typically be, may be seen as a special kind of support verb. They are marked using the COP label on the Noun or Adjective layer.</Paragraph>
      <Paragraph position="1"> There are several uses of copulas:  * Class membership: John is a sailor.</Paragraph>
      <Paragraph position="2"> * Qualities: Your literary masterpiece was delicious. * Location: This was inside a desk drawer.</Paragraph>
      <Paragraph position="3"> * Identity: Smithers is the vice-president of the armchair division.</Paragraph>
      <Paragraph position="4"> In FrameNet annotation, these uses of the copular verb are not distinguished.</Paragraph>
    </Section>
    <Section position="3" start_page="135" end_page="135" type="sub_section">
      <SectionTitle>
3.3 Null Arguments
</SectionTitle>
      <Paragraph position="0"> There are constructions that require special arguments to be syntactically valid, but where these arguments have no relation to the semantics of the sentence. In the example below, it is an example of this phenomenon.</Paragraph>
      <Paragraph position="1"> I hate it when you do that.</Paragraph>
      <Paragraph position="2"> Other common cases include existential constuctions (&amp;quot;there are&amp;quot;) and subject requirement of zero-place predicates (&amp;quot;it rains&amp;quot;). These null arguments are tagged as NULL on the Other layer.</Paragraph>
    </Section>
    <Section position="4" start_page="135" end_page="136" type="sub_section">
      <SectionTitle>
3.4 Aspectual Particles
</SectionTitle>
      <Paragraph position="0"> Verb particles that indicate aspectual information are marked using the ASPECT label. These particles must be distinguished from particles that are parts of multiword units, such as carry out.</Paragraph>
      <Paragraph position="1"> They just moan on and on about Fergie this and Fergie that and I 've simply had enough.</Paragraph>
    </Section>
  </Section>
  <Section position="5" start_page="136" end_page="136" type="metho">
    <SectionTitle>
3.5 Slot Fillers: GOV and X
</SectionTitle>
    <Paragraph position="0"> FrameNet annotation contains some information about the relation of predicates in the same sentence when one predicate is a slot filler (that is, an argument) of the other. This is most common for noun target words, typically referring to natural kinds or artifacts.</Paragraph>
    <Paragraph position="1"> In the following example, the target word fingertips evokes the OBSERVABLE_BODYPARTS frame, involving two FEs: POSSESSOR (&amp;quot;his&amp;quot;) and BODY_PART (&amp;quot;fingertips&amp;quot;). This noun phrase is also a slot filler (that is, an argument) of another predicate in the sentence: cling on. In FrameNet, such predicates are annotated using the GOV label. The constituent that contains the slot filler in question is called (for lack of a better name) X.</Paragraph>
    <Paragraph position="2"> Shares will boom and John Major will [cling on]GOV [by [his]POSSESSOR [fingertips]BODY_PART ]X.</Paragraph>
    <Paragraph position="3"> If GOV and X are present, all FEs must be contained in the span of the X node, such as BODY_PART and POSSESSOR above. This may be of use for automatic FE identifiers.</Paragraph>
  </Section>
  <Section position="6" start_page="136" end_page="136" type="metho">
    <SectionTitle>
4 Identifying Semantic Entities
</SectionTitle>
    <Paragraph position="0"> To find the semantic entities in the text, we used the method that has previously been used for FE detection: classification of nodes in a parse tree. We divide the identification process into two stages:  The reason for this division is that we expect that the knowledge of the presence of SUPP, COP, and GOV, which are almost always verbs, is useful when detecting the other entities. The second stage makes use of the information found in the first stage. Above all, it is necessary to have information about GOV to be able to detect X.</Paragraph>
    <Paragraph position="1"> Totrain the classifiers, weselected the 150 most common frames and divided the annotated example sentences for those frames into a training set of 100,000 sentences and a test set of 8,000 sentences. null The classifiers used the Support Vector learning method using the LIBSVM package (Chang and Lin, 2001). The features used by the classifiers are listed in Table 1. Apart from the features used by  Stage 2, most of them are well-known from previous literature on FE identification and labeling (Gildea and Jurafsky, 2002; Litkowski, 2004). For all path features, we used both the traditional constituent parse tree path (as by Gildea and Jurafsky (2002)) and a dependency tree path (as by Ahn et al. (2004)). We produced the parse trees using the parser of Collins (1999).</Paragraph>
  </Section>
class="xml-element"></Paper>