<?xml version="1.0" standalone="yes"?>
<Paper uid="C86-1023">
  <Title>Kagaku Kenky~-Hi IIojokin Ippan-Kenky-~-TB</Title>
  <Section position="1" start_page="0" end_page="0" type="abstr">
    <SectionTitle>
ABSTRACT
</SectionTitle>
    <Paragraph position="0"> Current practical machine translation systems (MT, in short), which are designed to deal with a huge amount of document, are general\]y structure-bound.</Paragraph>
    <Paragraph position="1"> That is, the translation process is done based on tile analysis and transformation of the structure of source sentence, not on the understanding and paraphrasing of the meaning of that. But each language has its own :~yntactic and semantic idiosyncrasy, and on this account, without understanding the total meaning of source sentences it is often difficult for MT to bridge properly the idiosyncratic gap between source~ and target- language. A somewhat new method called &amp;quot;Cross Translation Test (CTT, in short)&amp;quot; is presented that reveals the detail of idiosyncratic gap (IG, in short) together with the so-so satisfiable possibility of MT. It is also mentioned the usefulness olf sublanguage approach to reducing the IC between source- and target- language.</Paragraph>
    <Paragraph position="2"> i. Introduct:ion The majoJ:\[ty of the current practical machine translation system (MT, in short) (See \[Nagao 1985\] and \[Slocum 11.985\] for a good survey.) are structure-bound in tile sense that al\] the target sentences (i.e. translated ,~entences) are composed only from the syntactic st:ructure of the source sentences, not from the meaning understanding of those. Though almost all tile MT are utilizing some semantic devices such as semantic feature agreement checkers, semantic filters antl preference semantics (See \[Wilks 1975\] for example.) which are serving as syntactic structural disambiguation, they still remain Jn structurehound approaches far from tile total\[ meaning understanding approaches.</Paragraph>
    <Paragraph position="3"> \]?he structure-bound MT has a lot of advantageous features among which the easiness of formalizing translation process, that is, writing translation rules and the uniformity of lexicon description are vital from the practical standpoint that it must transact a huge vocabulary and \]numerable kinds of sentence patterns.</Paragraph>
    <Paragraph position="4"> On the other hand, the structure-bound MT has the inewttable limitation on the treatment of lingu:istic idiosyncrasy originated from the different way of&amp;quot; thinking.</Paragraph>
    <Paragraph position="5"> In this paper, first of all, we will sketch out the typical language modeling techniques on which the structure-bound MT(= current practical machine translation systems) are constructed. Secondly, we will examine the difference between the principal mechanism of machine translation and that of human translation irom the viewpoint of the language understanding abi\]ity, l'hirdly, we will illustrate the structural idiosyncratic gap (IG, in short) by comparing the sample sentences in English and that in ,lapanese. These sentences are sharing the same reCalling. This comparison will be made by a somewhat new method which we call &amp;quot;Cross Translation Test (CTT, in short)&amp;quot;, which will cventual\]y reveal the various IGs that have origins in the differences of culture, i.e., the way of thinking or the way of representing concepts. But at&amp;quot; the same Lime, CTT wiJl give some encouraging evidence that the principal technologies of today's not-yet-completed structure-bound HTs have the potentia\] for producing barely acceptable translation, if the source language sentences are taken from tile documents of less equivocations or are appropriately rewritten. Finally, we will briefly comment on the sub\]anguage to control or normalize source sentences as the promising and practical approaches to overcoming the IGs.</Paragraph>
  </Section>
class="xml-element"></Paper>