<?xml version="1.0" standalone="yes"?>
<Paper uid="C96-2191">
  <Title>Spoken-Language Translation Method Using Examples</Title>
  <Section position="3" start_page="0" end_page="0" type="metho">
    <SectionTitle>
2 Seven requirements for
</SectionTitle>
    <Paragraph position="0"> spoken-language translation The following new design features are critical for success in spoken-language translation:</Paragraph>
  </Section>
  <Section position="4" start_page="0" end_page="0" type="metho">
    <SectionTitle>
1. Incremental processing
</SectionTitle>
    <Paragraph position="0"> Incremental processing is required so as to handle fragmental phrases or incomplete utterances and to realize a real-time response.</Paragraph>
    <Paragraph position="1"> This has a very close relation with item 5 below. null 2. Handling spoken language Fragmental phrases, isolated phrases, a gradient of case role changing, complex topicalization, metonymical phrases, idiomatic expressions for etiquette, and inconsistent expressions in one utterance are main characteristics of spoken language. They strongly depend on dialogue situations.</Paragraph>
    <Paragraph position="2"> 3. Handling euphemistic expressions Under the influence of social position or situation, euphemistic expressions appear in various scenes in various forms.</Paragraph>
  </Section>
  <Section position="5" start_page="0" end_page="0" type="metho">
    <SectionTitle>
4. Deterministic processing
</SectionTitle>
    <Paragraph position="0"> Neither pre-editing nor post-editing can be relied on in a speech translation system. Interactive disambiguation by speakers does not necessarily converge a correct interpretation. null 5. Sufficient speed to avoid to break communication null As an interpreter intervenes between speakers, real-time response is required to keep smooth turn taking.</Paragraph>
    <Paragraph position="1"> .</Paragraph>
    <Paragraph position="2"> .</Paragraph>
    <Paragraph position="3"> High-quality translation This is necessary in order to ensure correct information exchange between speakers. Recovering from speech recognition errors null There are various aspects to recovering from speech recognition errors, for example in correcting phoneme sequences, syllable sequences, word sequences (including compound words and collocations).</Paragraph>
  </Section>
  <Section position="6" start_page="0" end_page="1075" type="metho">
    <SectionTitle>
3 Meeting the seven requirements
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.1 Incremental processing
</SectionTitle>
      <Paragraph position="0"> This is an essential technology if one is to build an incremental translation system like a simultaneous interpreter, and the proper way to grasp a chunk of a translation unit corresponding to some chunk in a target language is to extend 'constituent boundary parsing' to bottom-up-type parsing \[Furuse96\].</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
3.2 Recovering from errors
</SectionTitle>
      <Paragraph position="0"> A certain recovery method is now under consideration: a re-entrizing model for phoneme candidates by means of searching the correct phonemes using modification depending on recognition error characteristics in an example-based framewbrk \[Wakita95\]. This approach provides a recovery effect in handling phoneme or syllable sequences, and the effect depends on the particular speakers because of individual error characteristics.</Paragraph>
    </Section>
    <Section position="3" start_page="0" end_page="1075" type="sub_section">
      <SectionTitle>
3.3 Requirements covered by
</SectionTitle>
      <Paragraph position="0"> EBMT/TDMT The remaining requirements are handled effectively by an example-based approach as explained here.</Paragraph>
      <Paragraph position="1"> In NLP systems, especially for spoken language, many possibile syntactic structures are produced. It is an important and difficult process to choose the most plausibile structure. Conventional approachs, such as knowledge-based one, cannot easily handle continuous phenomena: gradation of case role changing; derivation of a metonymical  relation; and relationship between a topicalized word and the main predicate.</Paragraph>
      <Paragraph position="2"> We have proposed Example-Based Machine 3Y=anslation (EBMT) to deal with these difliculties\[Sumita92-a\]. The EBMT method prepares a large number of translation examples; the translation example that most closely matches the input expression is retrieved; and tile example is nfimicked.</Paragraph>
      <Paragraph position="3"> When applying F, BM'F to sentence translation, the sentence must be analyzed by matching transaltion patterns of phrases \[Furuse94\]. This model is in a sense &amp;quot;driven by transfer&amp;quot;, and we call it Transfer-Driven Machine %anslation (TDMT).</Paragraph>
      <Paragraph position="4"> 3.3.1 Handling spoken language Spoken language includes many phenomena; here, howew'.r, we concentrate on the following ones: (1) &amp;quot;wa&amp;quot; is a Japanese topic marker and, in general, this marker can t)e replaced by other case particles. But some usages cannot be identified as to case role because of gradation of case role changing. Moreover, if there are double topic markers in a sentence, they cannot I)e replaced by other particles 1. The first sentence in our Japanese-to-English (JE) translation &amp;quot;snapshot&amp;quot; (Figure 1), for exam-. ple, is properly translated in our TI)MT pro- null totype system.</Paragraph>
      <Paragraph position="5"> (i) &amp;quot;Chikatetsu-wa ichiban-chikai eki-wa doko desu-ka.&amp;quot; ('subway-topiealized,' 'the nearest,' 'station-topicalized,' 'where,' 'bequestion') null (2) Two sentences are mixed in one utterance. The tirst is pended, then inunedaitely the second sentence starts without conjunction. (ii) &amp;quot;Shiharai-wa ginkou-fllrikomi-o o-machishite-oriInasu.&amp;quot; null ('payment-topicalized,' 'bank-transferobjective,' 'wait-for-polite-modest') a.a.= Handling euphemistic expressions (1) There are various types of expressions for politeness, modesty, and euphemism. Such expressions are used depending on social roles. The fourth sentence in our Japaneseto-Korean (JK) translation snapshot (Figure 2) is a sample of this type, which is properly dealt with by TI)MT.</Paragraph>
      <Paragraph position="6"> (iii) &amp;quot;Yoyaku-wo  ten alphabetically and surrounded by double quotes, and the corresponding English words with usage modifiers follow in parenthesis.</Paragraph>
      <Paragraph position="8"> ConventionM MT methods provide multiple translation candidates but no information to use in selecting among them, or else just the first possible sentence that is generated.</Paragraph>
      <Paragraph position="9"> On the contrary, EBMT generates all the possible candidates combining suitable phrases. It also provides proper scores to each candidate using a similarity calculation. The scores realize &amp;quot;deterministic&amp;quot; translation.</Paragraph>
      <Paragraph position="10">  \[Furuse96\] has improved a matching mechanism over translation patterns. By accepting input in left-to-right order and dealing with best-only substructures, the explosion of structural ambiguity is restrained and an efficient translation of a lengthy input sentence can be achieved, l)re liminary experimentation has shown that average translation times are reduced from 1.15 seconds to 0.55 seconds for input of 10 words in length and from 10.87 seconds to 2.04 seconds for input of 20 words in length. The incorporation of incremental morphological analysis and generation \[Akamine95\] into the new-version TDMT, is promising for achieving incremental (simultaneous) translation for a practical spoken-language translation system.</Paragraph>
      <Paragraph position="11"> If instantaneous response is required, the rest dominant process is retrieval of the closest translation patterns from bulk collection. It is effectively solved by using a massively parallel algorithms and machines \[Sumita95-a, Snmita95-b, Oi93\].</Paragraph>
      <Paragraph position="12">  First, a well-known difficult problem in Japanese to English translation was selected as a test. The Japanese noun phrase of the form &amp;quot;noun + NO + noun&amp;quot; using the Japanese adnominM particle &amp;quot;NO&amp;quot; is an expression whose meaning is continuous. A translation success rate of about 80% has been demonstrated in a Jacknife test \[Sumita92-a\]. Also, for other Japanese and English phrases, similar effectiveness in target word selection and structural dsiambiguation has been demonstrated\[Sumita92-b\].</Paragraph>
      <Paragraph position="13"> We have evaluated a experimental TDMT system, with 825 model sentences about conference registration. These sentences cover basic expressions in an inquiry dialogue. The success rate is 71% for a test data set consisting of 1,050 unseen sentences in the same domain.</Paragraph>
      <Paragraph position="14">  target : &amp;quot;T would like to arrive at Las Vegas by nine o' clock at night&amp;quot; target : &amp;quot;If you get on the bus at nine fifteen, you ~ll arrive by e~ht o' clock at night&amp;quot; source : &amp;quot;AB~-CT~ ~&amp;quot; target : &amp;quot;At eight 0' clock 2&amp;quot;</Paragraph>
      <Paragraph position="16"> source : &amp;quot;C-~%T'~Pe~,gtCZSf~'8~9-~ ~'' (Hi is it possible to make hotel reservation from here?) target : &amp;quot;~1~ ~ ~ ~ + ~@~1~?&amp;quot; source : &amp;quot;C~fr(~-C'~ ~/~-c~'C-~,,~a)ZIat~*~Z~X-c'I~ bT~ ~ ~:~i~ ~&amp;quot; (OK, what we do is to give you all the ~nformation you need and then ~e ask you to go ahead and make the call yourself. )</Paragraph>
      <Paragraph position="18"/>
    </Section>
  </Section>
  <Section position="7" start_page="1075" end_page="1075" type="metho">
    <SectionTitle>
4 JE 8C/ JK prototype systems
</SectionTitle>
    <Paragraph position="0"> The TDMT system is being expanded so as to handle travel arrangement dialogues including the topics of hotel reservation, room services, troubleshooting during hotel stays, various information queries, and various travel arrangements. At present the JE system has about a 5,000-word vocabulary and a transfer knowledge from 2,000 training sentences. The JK system is half this size. While some modules, such as morphological analysis and generation, are language-specific, the transfer module is a common part of every language pair. Through JE and JK implementation, we believe that the translation of every language pair can be achieved in the same framework using TDMT. On the other hand, it has turned out that the linguistic distance between source and target languages reflects the variety of target expression patterns in the transfer knowledge. Table 1 shows the number of target expression patterns corresponding a Japanese particles in JE and JK. These numbers are counted from the current TDMT system's transfer knowledge, and the numbers of examples are token numbers (i.e., not including duplications). null</Paragraph>
  </Section>
class="xml-element"></Paper>