<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-1097">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics Semi-Supervised Training for Statistical Word Alignment</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The most widely applied training procedure for statistical machine translation IBM model 4 (Brown et al., 1993) unsupervised training followed by post-processing with symmetrization heuristics (Och and Ney, 2003) yields low quality word alignments. When compared with gold standard parallel data which was manually aligned using a high-recall/precision methodology (Melamed, 1998), the word-level alignments produced automatically have an F-measure accuracy of 64.6 and 76.4% (see Section 2 for details).</Paragraph>
    <Paragraph position="1"> In this paper, we improve word alignment and, subsequently, MT accuracy by developing a range of increasingly sophisticated methods:  1. We rst recast the problem of estimating the IBM models (Brown et al., 1993) in a discriminative framework, which leads to an initial increase in word-alignment accuracy.</Paragraph>
    <Paragraph position="2"> 2. We extend the IBM models with new  (sub)models, which leads to additional increases in word-alignment accuracy. In the process, we also show that these improvements are explained not only by the power of the new models, but also by a novel search procedure for the alignment of highest probability. null 3. Finally, we propose a training procedure that interleaves discriminative training with maximum likelihood training.</Paragraph>
    <Paragraph position="3"> These steps lead to word alignments of higher accuracy which, in our case, correlate with higher MT accuracy.</Paragraph>
    <Paragraph position="4"> The rest of the paper is organized as follows. In Section 2, we review the data sets we use to validate experimentally our algorithms and the associated baselines. In Section 3, we present iteratively our contributions that eventually lead to absolute increases in alignment quality of 4.8% for French/English and 4.8% for Arabic/English, as measured using F-measure for large word alignment tasks. These contributions pertain to the casting of the training procedure in the discriminative framework (Section 3.1); the IBM model extensions and modi ed search procedure for the Viterbi alignments (Section 3.2); and the interleaved, minimum error/maximum likelihood, training algorithm (Section 4). In Section 5, we assess the impact that our improved alignments have on MT quality. We conclude with a comparison of our work with previous research on discriminative training for word alignment and a short discussion of semi-supervised learning.</Paragraph>
  </Section>
class="xml-element"></Paper>