<?xml version="1.0" standalone="yes"?>
<Paper uid="C02-1050">
  <Title>Bidirectional Decoding for Statistical Machine Translation</Title>
  <Section position="2" start_page="0" end_page="0" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> The statistical approach to machine translation regards the machine translation problem as the maximum likelihood solution of a translation target text given a translation source text. According to the Bayes Rule, the problem is transformed into the noisy channel model paradigm, where the translation is the maximum a posteriori solution of a distribution for a channel target text given a channel source text and a prior distribution for the channel source text (Brown et al., 1993).</Paragraph>
    <Paragraph position="1"> Although there exists e cient algorithms to estimate the parameters for the statistical machine translation (SMT), one of the problems of SMT is the search algorithms for the translation given a sequence of words. There exists stack decoding algorithm (Berger et al., 1996), A* search algorithm (Och et al., 2001; Wang and Waibel, 1997) and dynamic-programming algorithms (Tillmann and Ney, 2000; Garcia-Varea and Casacuberta, 2001), and all translate a given input string word-by-word and render the translation in left-to-right, with pruning technologies assuming almost linearly aligned translation source and target texts. The algorithms proposed above cannot deal with drastically di erent word correspondence, such as Japanese and English translation, where Japanese is SOV while SVO in English. Germann et al. (2001) suggested greedy method and integer programming decoding, though the rst method su er from the similar problem as described above and the second is impractical for the real-world application.</Paragraph>
    <Paragraph position="2"> This paper presents two decoding methods, one is the right-to-left decoding based on the left-to-right beam search algorithm, which generates outputs from the end of a sentence. The second one is the bidirectional decoding method which decodes in both of the left-to-right and right-to-left directions and merges the two hypothesized partial sentences into one. The experimental results of Japanese and English translation indicated that the right-to-left decoding was better for English-to-Japanese translation, while the left-to-right decoding was better for Japanese-to-English decoding. The above results could be justi ed by the structural di erence of Japanese and English, where English takes the pre x structure that places emphasis at the beginning of a sentence, hence prefers left-to-right decoding. On the other hand, Japanese takes post x structure, setting attention around the end of a sentence, therefore favors right-to-left decoding. The bidirectional decoding, which can take both of the bene ts of decoding method, was superior to mono-directional decoding methods.</Paragraph>
    <Paragraph position="3"> The next section brie y describes the SMT focusing on the IBM Model 4. Then, the Section 3 presents decoding algorithms in three direction, leftto-right, right-to-left and bi-direction. The Section 4 presents the results of Japanese and English translation followed by discussions.</Paragraph>
  </Section>
class="xml-element"></Paper>