<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-2926">
  <Title>A Pipeline Model for Bottom-Up Dependency Parsing</Title>
  <Section position="7" start_page="189" end_page="189" type="concl">
    <SectionTitle>
5 Further Work and Conclusion
</SectionTitle>
    <Paragraph position="0"> In the shared task of CoNLL-X, we have shown that our dependency parsing system can do well on multiple languages without requiring special knowledge for each of the languages.</Paragraph>
    <Paragraph position="1"> From a technical perspective, we have addressed the problem of using learned classi ers in a pipeline fashion, where a task is decomposed into several stages and classi ers are used sequentially to solve each stage. This is a common computational strategy in natural language processing and is known to suffer from error accumulation and an inability to correct mistakes in previous stages. We abstracted two natural principles, one which calls for making the local classi ers used in the computation more reliable and a second, which suggests to devise the pipeline algorithm in such a way that it minimizes the number of actions taken.</Paragraph>
    <Paragraph position="2"> However, since we tried to build a single approach for all languages, we have not fully utilized the capabilities of our algorithms. In future work we will try to specify both features and local search parameters to the target language.</Paragraph>
    <Paragraph position="3"> Acknowledgement This research is supported by NSF ITR IIS-0428472, a DOI grant under the Re ex program and ARDA's Advanced Question Answering for Intelligence (AQUAINT) program.</Paragraph>
  </Section>
class="xml-element"></Paper>