<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-0305">
  <Title>Reducing Parameter Space for Word Alignment</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> We participated the workshop shared task for English-French and Romanian-English word alignment. We use IBM Model 4 as a baseline. The number of parameters in this model roughly scales as the product of the vocabulary sizes (ie number of types) in the source and target languages. In order to obtain better alignment performance, we wish to investigate techniques that may reduce the number of parameters, therefore increasing the datato-parameter ratio. For that purpose, we preprocessed the training corpus using a word lemmatizer and a bilingual lexicon extraction algorithm. Section 2 briefly describes the base alignment algorithm, Section 3 describes our additional components, and Section 4 shows our experimental results, followed by Discussion and Conclusion in Section 5 and 6, respectively.</Paragraph>
  </Section>
class="xml-element"></Paper>