<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2004">
  <Title>The Effect of Corpus Size in Combining Supervised and Unsupervised Training for Disambiguation</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The best performing systems for many tasks in natural language processing are based on supervised training on annotated corpora such as the Penn Treebank (Marcus et al., 1993) and the prepositional phrase data set first described in (Ratnaparkhi et al., 1994). However, the production of training sets is expensive. They are not available for many domains and languages. This motivates research on combining supervised with unsupervised learning since unannotated text is in ample supply for most domains in the major languages of the world. The question arises how much annotated and unannotated data is necessary in combination learning strategies. We investigate this question for two attachment ambiguity problems: relative clause (RC) attachment and prepositional phrase (PP) attachment. The supervised component is Collins' parser (Collins, 1997), trained on the Wall Street Journal. The unsupervised component gathers lexical statistics from an unannotated corpus of newswire text.</Paragraph>
    <Paragraph position="1"> The sizes of both types of corpora, annotated and unannotated, are of interest. We would expect that large annotated corpora (training sets) tend to make the additional information from unannotated corpora redundant. This expectation is confirmed in our experiments. For example, when using the maximum training set available for PP attachment, performance decreases when &amp;quot;unannotated&amp;quot; lexical statistics are added. For unannotated corpora, we would expect the opposite effect. The larger the unannotated corpus, the better the combined system should perform. While there is a general tendency to this effect, the improvements in our experiments reach a plateau quickly as the unlabeled corpus grows, especially for PP attachment. We attribute this result to the noisiness of the statistics collected from unlabeled corpora. null The paper is organized as follows. Sections 2, 3 and 4 describe data sets, methods and experiments. Section 5 evaluates and discusses experimental results. Section 6 compares our approach to prior work. Section 7 states our conclusions.</Paragraph>
  </Section>
class="xml-element"></Paper>