<?xml version="1.0" standalone="yes"?>
<Paper uid="C04-1179">
  <Title>FrameNet-based Semantic Parsing using Maximum Entropy Models</Title>
  <Section position="3" start_page="0" end_page="1" type="relat">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> The first work using FrameNet for semantic parsing was done by Gildea and Jurafsky (G &amp; J, 2002) using conditional probabilistic models.</Paragraph>
    <Paragraph position="1"> They divide the problem into two sub-tasks: frame element identification and frame element classification. Frame element identification identifies the frame element boundaries in a sentence, and frame element classification classifies each frame element into its appropriate semantic role. The basic assumption is that the frame element (FE) boundaries match the parse constituents, and both identification and classification are then done for each constituent  .</Paragraph>
    <Paragraph position="2"> In addition to the separate two phase model of frame element identification and role classification, they provide an integrated model that exhibits improved performance. They define a frame element group (FEG) as a set of frame element roles present in a particular sentence. By integrating FE identification with role labeling, allowing FEG priors and role labeling decision to affect the determination of next FE identification, they accomplish F-score of 71.9% for FE identification and 62.8% for both of FE identification and role labeling. However, since this integrated approach has an exponential complexity in the number of constituents, they apply a pruning scheme of using only the top m  The final output performance measurement is limited to the number of parse constituents matching the frame element boundaries.</Paragraph>
    <Paragraph position="3"> hypotheses on the role for each constituent (m = 10).</Paragraph>
    <Paragraph position="4"> Fleischman et al.(FKH, 2003) extend G &amp; J's work and achieve better performance in role classification for correct frame element boundaries. Their work improves accuracy from 78.5% to 84.7%. The main reasons for improvement are first the use of Maximum Entropy and second the use of sentence-wide features such as Syntactic patterns and previously identified frame element roles. It is not surprising that there is a dependency between each constituent's role in a sentence and sentence level features reflecting this dependency improve the performance.</Paragraph>
    <Paragraph position="5"> In this paper, we extend our previous work (KFH) by adopting sentence level features even for frame element identification.</Paragraph>
  </Section>
class="xml-element"></Paper>