<?xml version="1.0" standalone="yes"?>
<Paper uid="W99-0628">
  <Title>PP-Attachment: A Committee Machine Approach</Title>
  <Section position="8" start_page="236" end_page="237" type="concl">
    <SectionTitle>
4 Conclusions
</SectionTitle>
    <Paragraph position="0"> Neural networks have been shown to be very successful in tasks such as pattern recognition or prediction in many different applications of business, biomedicine, engineering, astronomy, high energy physics, etc. Their results are similar and often better than those of alternative models.</Paragraph>
    <Paragraph position="1"> The benefits of neural networks are well known as was explained above. Unfortunately neural networks have not been very successful in the domain of Natural Language Processing. However, our system has obtained better results than any that have been published to date using the complete Ratnaparkhi data set. We also obtained excellent results in word sense disambiguation \[Moliner, 1998\]. Our success can be attributed to two things: on one hand, the use of semantic classes is fundamental to keep from flooding the network's memory. In other hand, the use of canonic thematic structures. Finally, improvement on the generalization is an area in permanent development in the field of neural networks. We are developing new methods of generalization  which will allow us to improve our results even more. Provisional results place us in the environment of 88% with Ratnaparkhi's data set.</Paragraph>
  </Section>
class="xml-element"></Paper>