<?xml version="1.0" standalone="yes"?>
<Paper uid="P84-1111">
  <Title>HANDLING SYNTACTICAL AMBIGUITY IN MACHINE TRANSLATION</Title>
  <Section position="7" start_page="523" end_page="523" type="concl">
    <SectionTitle>
5. CONCLUDING REMARKS
</SectionTitle>
    <Paragraph position="0"> To conclude, some syntactically ambiguous strings in English can have literal, others non-llteral, and still others do not have any correspondences in Bulgarian. In summary, from a total number of approximately 200 simple strings treated in Engllsh more than 3/4 can, and only 1/4 cannot, be literally translated; about half of the latter strings can be put into correspondence with syntactically ambiguous strings in Bulgarian preserving their ambiguity. This gives quite a strong support to the usefulness of our approach in an English into Bulgarian translation system.</Paragraph>
    <Paragraph position="1"> Several advantages of this way of handling of syntactical ambiguity can be mentioned.</Paragraph>
    <Paragraph position="2"> First, in the processing of the majority of syntactically ambiguous sentences within an English into Bulgarian translation system it dispenses with semantical and world knowledge information at the very low cost of studying the ambiguity correspondences in both languages. It could be expected that investigations along this line will prove to be frultful for other pairs of languages as well.</Paragraph>
    <Paragraph position="3"> Secondly, whenever this way of handling syntactical ambiguity is applicable, the impossibility of previous approaches to translate sentences with unresolvable ambiguity, or such with verbal Jokes and the like, turns out to be an easily attainable task.</Paragraph>
    <Paragraph position="4"> Thirdly, the approach seems to have a very natural extension to another principal difficulty in MT, viz. coreference (cf. the three-ways ambiguity of Jim hit John and then he (Jim, John or neither?) went away and the same ambiguity of tQ~ (=he) in its literal translation into Bulgarian: D$im udari DJon i togava toj(?) si otide).</Paragraph>
    <Paragraph position="5"> And, finally, there is yet another reason for adopting the approach discussed here. Even if we choose to go another way and (somehow) dlsamblguate sentences in the SL, almost certainly their translational equivalents will be again syntactlcally ambiguous, and quite probably preserve the very ambiguity we tried to resolve. In this sense, for the purposes of MT (or other man-oriented applications of CL) we need not waste our efforts to disambiguate e.g. sentences like John hit the dog with th___ee lon~ hat or John hit th____ee do~ with the long woo1, since, even if we have done that, the correct Bulgarian translations of both these sentences are syntactically ambiguous in exactly the same way, the resolution of ambiguity thus proving to be an entirely superfluous operation (cf. D~on udari kucheto s dal~ata palka and Djon udari kucheto s dal~ata valna).</Paragraph>
  </Section>
class="xml-element"></Paper>