<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2057">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics A FrameNet-based Semantic Role Labeler for Swedish</Title>
  <Section position="7" start_page="442" end_page="442" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> We have described the design and implementation of a Swedish FrameNet-based SRL system that was trained using a corpus that was annotated using cross-language transfer from English to Swedish. With no manual effort except for translating sentences for evaluation, we were able to reach promising results. To our knowledge, the system is the first SRL system for Swedish in literature. We believe that the methods described could be applied to any language, as long as there exists a parallel corpus where one of the languages is English. However, the relatively close relationship between English and Swedish probably made the task comparatively easy in our case.</Paragraph>
    <Paragraph position="1"> As we can see, the figures (especially the FE bracketing recall) leave room for improvement for the system to be useful in a fully automatic setting. Apart from the noisy training set, probable reasons for this include the lower robustness of the Swedish parsers compared to those available for English. In addition, we have noticed that the European Parliament corpus is somewhat biased. For instance, a very large proportion of the target words evoke the STATEMENT or DIS-CUSSION frames, but there are very few instances of the BEING_WET and MAKING_FACES frames.</Paragraph>
    <Paragraph position="2"> While training, we tried to balance the selection somewhat, but applying the projection methods on other types of parallel corpora (such as novels available in both languages) may produce a better training corpus.</Paragraph>
  </Section>
class="xml-element"></Paper>