<?xml version="1.0" standalone="yes"?>
<Paper uid="W02-2033">
  <Title>Learning to Distinguish PP Arguments from Adjuncts</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
6 Conclusions and Future Work
</SectionTitle>
    <Paragraph position="0"> In this paper we described one possible approach to deal with the problem of disambiguating between arguments or adjuncts. This approach is tested by a learning system used to investigate the automatic acquisition of language from data. The learning system is equipped with a plausible model of the Universal Grammarandithastosetitsparameterstothetarget null language based on the input data. The ambiguity between arguments and adjuncts is one of several difficulties encountered by the learning system during the acquisition process and the approach proposed to overcome this problem, proved to be effective and helped the learner decide the appropriate case for the ambiguities found in the data available. The implemented learning system can successfully learn from a corpus of real child-directed data, containing noise and ambiguity, in a more realistic account of parameter setting (Villavicencio 2002).</Paragraph>
    <Paragraph position="1"> Disambiguation based on frequency information and semantically motivated selection provides a plausible strategy, compatible with some research on language acquisition. Although this is primarily a cognitive computational model, it is potentially relevant to the development of more adaptive NLP technology, by indicating possible paths for future developments in the area. However, larger scale tests still need to be conducted to see how the approach would generalise, and for that we need more annotated data. These two tasks of annotating more data and undertaking this larger scale investigation are included in the future directions of this work.</Paragraph>
  </Section>
class="xml-element"></Paper>