<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2018">
  <Title>Using Machine-Learning to Assign Function Labels to Parser Output for Spanish</Title>
  <Section position="10" start_page="141" end_page="142" type="concl">
    <SectionTitle>
8 Conclusions and Future Research
</SectionTitle>
    <Paragraph position="0"> Our research has shown that machine-learning-based Cast3LB tag assignment as a post-processing step to raw tree parser output statistically significantly outperforms a baseline where the parser itself is trained to learn category / Cast3LB-function pairs. In contrast to the parser-based method, the machine-learning-based method avoids some sparse data problems and allows for more control over Cast3LB tag assignment. We have found that the SVM algorithm out-performs the other two machine learning methods used.</Paragraph>
    <Paragraph position="1">  In addition, we evaluated Cast3LB tag assignment in a task-based setting in the context of automatically acquiring LFG resources for Spanish from Cast3LB. Machine-learning-based Cast3LB tag assignment yields statistically-significantly improved LFG f-structures compared to parser-based assignment.</Paragraph>
    <Paragraph position="2"> One limitation of our method is the fact that it treats the classification task separately for each target node. It thus fails to observe constraints on the possible sequences of grammatical function tags in the same local context. Some functions are unique, such as the Subject, whereas others (Direct and Indirect Object) can only be realized by a full NP once, although they can be doubled by a clitic pronoun. Capturing such global constraints will need further work.</Paragraph>
  </Section>
class="xml-element"></Paper>