<?xml version="1.0" standalone="yes"?>
<Paper uid="W04-3214">
  <Title>The Influence of Argument Structure on Semantic Role Assignment</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 FrameNet
</SectionTitle>
    <Paragraph position="0"> FrameNet is a lexical resource based on Fillmore's Frame Semantics (Fillmore, 1985). It is designed as an ontology of frames, representations of prototypical situations. Each frame provides a set of predicates (nouns, verbs or adjectives) which can introduce the frame. The semantic roles are framespecific, since they are defined as categories of entities or concepts pertaining to the particular situation a predicate evokes.</Paragraph>
    <Paragraph position="1"> The following sentences are examples for the semantic annotation provided in the FrameNet corpus for verbs in the IMPACT frame, which describes a situation in which typically &amp;quot;an Impactor makes sudden, forcible contact with the Impactee, or two  Impactors both ... [make] forcible contact&amp;quot;3.</Paragraph>
    <Paragraph position="2"> (1) a. [Impactee His car] was struck [Impactor by a third vehicle].</Paragraph>
    <Paragraph position="3"> b. [Impactor The door] slammed [Result shut].</Paragraph>
    <Paragraph position="4"> c. [Impactors Their vehicles] collided  [Place at Pond Hill].</Paragraph>
    <Paragraph position="5"> Note that the frame-specificity of semantic roles in FrameNet has important consequences for semantic role assignment, since there is no direct way to generalise across frames. Therefore, the learning for automatic assignment of semantic roles has to proceed frame-wise. Thus, the data sparseness problem is especially acute, and automatic assignment for frames with no training data is very difficult (see Gildea and Jurafsky (2002)).</Paragraph>
  </Section>
class="xml-element"></Paper>