<?xml version="1.0" standalone="yes"?>
<Paper uid="W97-1015">
  <Title>A Comparative Study of the Application of Different Learning Techniques to Natural Language Interfaces</Title>
  <Section position="7" start_page="0" end_page="0" type="concl">
    <SectionTitle>
5 Conclusion
</SectionTitle>
    <Paragraph position="0"> In this paper we have presented first results from a comparative study of applying different inductive learning techniques to natural language interfaces.</Paragraph>
    <Paragraph position="1"> We have implemented a representative selection of instance-based and model-based algorithms by making use of deductive object-oriented database functionality. The extensive case study for an interface to a production planning and control system shows the feasibility of the approach in that linguistic knowledge is learned the acquisition of which normally takes a large effort of human experts.</Paragraph>
    <Paragraph position="2"> Future work will concentrate on the important point of increasing: the reliability of test results in that we apply cross-validation trials and statistical tests for the significance of performance differences between two algorithms. Furthermore, we also want to generate learning functions that plot success rates as function of the size of the training collection. Besides this, we plan to test our learning algorithms on standard benchmark machine learning datasets and other typical natural language learning datasets.</Paragraph>
    <Paragraph position="3"> Finally, we intend to extend the implemented algorithms to include also unsupervised methods as well as connectionist and evolutionary techniques.</Paragraph>
    <Paragraph position="4"> In addition, we will implement incremental learning techniques, which continue the learning process during the test phase, and adaptive boosting methods, which apply several classifiers instead of just one.</Paragraph>
    <Paragraph position="5"> We believe that our study is a first promising step towards the challenging task of carrying out comparative evaluations of the performance of different machine learning algorithms for specific linguistic problems. null</Paragraph>
  </Section>
class="xml-element"></Paper>