<?xml version="1.0" standalone="yes"?>
<Paper uid="W03-1002">
  <Title>Statistical Machine Translation Using Coercive Two-Level Syntactic Transduction</Title>
  <Section position="3" start_page="0" end_page="0" type="intro">
    <SectionTitle>
2 Prior Work
</SectionTitle>
    <Paragraph position="0"> Statistical machine translation, as pioneered by IBM (e.g. Brown et al., 1993), is grounded in the noisy channel model. And similar to the related channel problems of speech and handwriting recognition, the original SMT language pair French-English exhibits a relatively close linear correlation in source and target sequence. Much common local motion that is observed for French, such as adjective-noun swapping, is adequately modeled by the relative-position-based distortion models of the classic IBM approach. Unfortunately, these distortion models are less effective for languages such as Japanese or Arabic, which have substantially different top-level sentential word orders from English, and hence longer distance constituent motion.</Paragraph>
    <Paragraph position="1"> Wu (1997) and Jones and Havrilla (1998) have sought to more closely tie the allowed motion of constituents between languages to those syntactic transductions supported by the independent rotation of parse tree constituents. Yamada and Knight (2000, 2001) and Alshawi et al. (2000) have effectively extended such syntactic transduction models to fully functional SMT systems, based on channel model tree transducers and finite state head transducers respectively. While these models are well suited for the effective handling of highly divergent sentential word orders, the above frameworks have a limitation shared with probabilistic context free grammars that the preferred ordering of subtrees is insufficiently constrained by their embedding context, which is especially problematic for very deep syntactic parses.</Paragraph>
    <Paragraph position="2"> In contrast, Och et al. (1999) have avoided the constraints of tree-based syntactic models and allow the relatively flat motion of empirically derived phrasal chunks, which need not adhere to traditional constituent boundaries. null Our current paper takes a middle path, by grounding motion in syntactic transduction, but in a much flatter 2-level model of syntactic analysis, based on flat embedded noun-phrases in a flat sentential constituent-based chunk sequence that can be driven by syntactic bracketers and POS tag models rather than a full parser, facilitating its transfer to lower density languages. The flatter 2-level structures also better support transductions conditioned to full sentential context than do deeply embedded tree models, while retaining the empirically observed advantages of translation ordering independence of nounphrases. null Another improvement over Och et al. and Yamada and Knight is the use of the finite state machine (FSM) modelling framework (e.g. Bangalore and Riccardi, 2000), which offers the considerable advantage of a flexible framework for decoding, as well as a representation which is suitable for the fixed two-level phrasal modelling employed here.</Paragraph>
    <Paragraph position="3"> Finally, the original cross-part-of-speech lexical coercion models presented in Section 4.3.3 have related work in the primarily-syntactic coercion models utilized by Dorr and Habash (2002) and Habash and Door (2003), although their induction and modelling are quite different from the approach here.</Paragraph>
  </Section>
class="xml-element"></Paper>