<?xml version="1.0" standalone="yes"?>
<Paper uid="W06-1628">
  <Title>Sydney, July 2006. c(c)2006 Association for Computational Linguistics A Discriminative Model for Tree-to-Tree Translation</Title>
  <Section position="4" start_page="233" end_page="233" type="intro">
    <SectionTitle>
2 Related Work
</SectionTitle>
    <Paragraph position="0"> There has been a substantial amount of previous work on approaches that make use of syntactic information in statistical machine translation. Wu (1997) and Alshawi (1996) describe early work on formalisms that make use of transductive grammars; Graehl and Knight (2004) describe methods for training tree transducers. Melamed (2004) establishes a theoretical framework for generalized synchronous parsing and translation. Eisner (2003) discusses methods for learning synchronized elementary tree pairs from a parallel corpus of parsed sentences. Chiang (2005) has recently shown significant improvements in translation accuracy, using synchronous grammars. Riezler and Maxwell (2006) describe a method for learning a probabilistic model that maps LFG parse structures in German into LFG parse structures in English. null Yamada and Knight (2001) and Galley et al.</Paragraph>
    <Paragraph position="1"> (2004) describe methods that make use of syntactic information in the target language alone; Quirk et al. (2005) describe similar methods that make use of dependency representations. Syntactic parsers in the target language have been used as language models in translation, giving some improvement in accuracy (Charniak et al., 2001).</Paragraph>
    <Paragraph position="2"> The work of Gildea (2003) involves methods that make use of syntactic information in both the source and target languages.</Paragraph>
    <Paragraph position="3"> Other work has attempted to incorporate syntac- null and for the noun obstacle. The EPs were taken from the parse tree for the sentence We know that the main obstacle has been the predictable resistance of manufacturers.</Paragraph>
    <Paragraph position="4"> tic information through reranking approaches applied to n-best output from phrase-based systems (Och et al., 2004). Another class of approaches has shown improvements in translation through reordering, where source language strings are parsed and then reordered, in an attempt to recover a word order that is closer to the target language (Collins et al., 2005; Xia and McCord, 2004).</Paragraph>
    <Paragraph position="5"> Our approach is closely related to previous work on synchronous tree adjoining grammars (Shieber and Schabes, 1990; Shieber, 2004), and the work on TAG approaches to syntax described by Frank (2002). A major departure from previous work on synchronous TAGs is in our use of a discriminative model that incrementally predicts the information in the AEP. Note also that our model may include features that take into account any part of the German clause.</Paragraph>
  </Section>
class="xml-element"></Paper>