<?xml version="1.0" standalone="yes"?>
<Paper uid="P06-2088">
  <Title>Simultaneous English-Japanese Spoken Language Translation Based on Incremental Dependency Parsing and Transfer</Title>
  <Section position="3" start_page="0" end_page="683" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> Recently, speech-to-speech translation has become one of the important research topics in machine translation. Projects concerning speech translation such as TC-STAR (Hoge, 2002) and DARPA Babylon have been executed, and conferences on spoken language translation such as IWSLT have been held. Though some speech translation systems have been developed so far (Frederking et al., 2002; Isotani et al., 2003; Liu et al., 2003; Takezawa et al., 1998), these systems, because of their sentence-by-sentence translation, cannot start to translate a sentence until it has been fully uttered. The following problems may arise in cross-language communication: * The conversation time become long since it takes much time to translate * The listener has to wait for the translation since such systems increase the difference between the beginning time of the speaker's utterance and the beginning time of its translation null These problems are likely to cause some awkwardness in conversations. One effective method of improving these problems is that a translation system begins to translate the words without waiting for the end of the speaker's utterance, much as a simultaneous interpreter does. This has been verified as possible by a study on comparing simultaneous interpretation with consecutive interpretation from the viewpoint of efficiency and smoothness of cross-language conversations (Ohara et al., 2003).</Paragraph>
    <Paragraph position="1"> There has also been some research on simultaneous machine interpretation with the aim of developing environments that support multilingual communication (Mima et al., 1998; Casacuberta et al., 2002; Matsubara and Inagaki, 1997).</Paragraph>
    <Paragraph position="2"> To realize simultaneous translation between languages with different word order, such as English and Japanese, our method utilizes the feature that the word order of a target language is flexible. To resolve the problem that translation systems generates grammatically dubious sentence,  our method utilizes dependency structures and Japanese dependency constraints to determine the word order of a translation. Moreover, by considering the fact that the inversion of predicate expressions occurs more frequently in Japanese spoken language, our method employs predicate inversion to resolve the problem that Japanese has the predicate at the end of the sentence. Furthermore, our method features the function of canceling an inversion by restating a predicate when the translation is incomprehensible due to the inversion. In the research described in this paper, we implement a prototype translation system, and to evaluate it, we conduct an experiment with all 578 sentences in the ATIS corpus.</Paragraph>
    <Paragraph position="3"> This paper is organized as follows: Section 2 discusses an important problem in English-Japanese simultaneous translation and explains the idea of utilizing flexible word order. Section 3 introduces our method for the generation in English-Japanese simultaneous translation, and Section 4 describes the configuration of our system. Section 5 reports the experimental results, and the paper concludes in Section 6.</Paragraph>
  </Section>
class="xml-element"></Paper>