<?xml version="1.0" standalone="yes"?>
<Paper uid="W93-0223">
  <Title>HOW COULD RHETORICAL RELATIONS BE USED IN MACHINE TRANSLATION? (AND AT LEAST TWO OPEN QUESTIONS)</Title>
  <Section position="1" start_page="0" end_page="0" type="metho">
    <SectionTitle>
HOW COULD RHETORICAL RELATIONS BE
USED IN MACHINE TRANSLATION?
(AND AT LEAST TWO OPEN QUESTIONS)
</SectionTitle>
    <Paragraph position="0"> My position paper addresses more or less Workshop question No. 5: &amp;quot;How are rhetorical relations used in discourse understanding? How are linguistic clues and word knowledge brought to bear?&amp;quot; The paper shows how rhetorical relations could be used in Machine Translation (MT). It introduces in brief, a discourse-oriented approach for MT which uses schemata of rhetorical predicates for describing the structure of a paragraph. At the same time, it poses at least two questions (in my opinion practically unsolved  problems): 1) How can rhetorical predicates be computationally recognized? 2) Are the so far defined predicates sufficient and precise enough to describe the real world?</Paragraph>
  </Section>
  <Section position="2" start_page="0" end_page="86" type="metho">
    <SectionTitle>
INTRODUCTION: DISCOURSE-ORIENTED MACHINE TRANSLATION
</SectionTitle>
    <Paragraph position="0"> The discourse-oriented MT should be regarded as a very important research topic, since it is expected to make the translation more natural in MT systems.</Paragraph>
    <Paragraph position="1"> Unfortunately, not much attention has been given to this problem yet and the availability of a discourse component in a MT system has been reported very briefly in \[7\] only.</Paragraph>
    <Paragraph position="2"> Most of the MT systems perform sentence-by-sentence translation. Only a few try to translate paragraph-by-paragraph and in these cases, the discourse structure of the output language is identical with that of the input language. However, I have shown that the discourse structures across the different sublanguages are not always the same for any pair of natural languages \[5\].</Paragraph>
    <Paragraph position="3"> Paragraph-by-paragraph machine translation seems to be for now, an unjustifiably complicated task for practical needs. It involves the complete understanding of the paragraph, the determination of discourse topic(s), goals, intentions, so that the output can be produced in accordance with the respective discourse rules and purposes. However, recognizing topic, goal, intention by a computer program seems to be a very tough problem. Moreover, analyzing a paragraph is itself a complicated task which does not always yield satisfactory results.</Paragraph>
    <Paragraph position="4">  On the other hand, translating sentence-by-sentence with the sequence of the original sentences preserved is a general approach, which guarantees in most of the cases an understandable output. However, in order for a translated message to sound as natural as possible, it should be conveyed in accordance with the discourse organization rules of the target language. If we examine more closely the work of a professional translator, we shall inevitably note that he/she does not always follow the order of sentences in the source text.</Paragraph>
    <Paragraph position="5"> Taking into account the complexity of paragraph understanding and the necessity of observing the specific target sublanguage rules, I have been working on a practical discourse-oriented MT approach (within an English to Malay MT system) which analyzes a source paragraph as a schema of rhetorical predicates and generates the target text possibly as another schema of rhetorical predicates. Towards this end, I have developed a Text Organization Framework Grammar which maps source paragraph structures of rhetorical predicates into the specific target paragraph structures of rhetorical predicates \[6\].</Paragraph>
  </Section>
  <Section position="3" start_page="86" end_page="86" type="metho">
    <SectionTitle>
SELECTION OF TEXT ORGANIZATION APPROACH
</SectionTitle>
    <Paragraph position="0"> I have been studying different approaches which have been so far used to describe the organization of a given text (paragraph). From a practical point of view, I argue that the most appropriate approach would be the &amp;quot;schemata-based approach&amp;quot; introduced by K. McKeown \[3\] and used by other researchers.</Paragraph>
    <Paragraph position="1"> Though some researchers point out the relatively missing flexibility of this approach, I found this approach more suitable for the needs of MT. The plan-based approach \[4\] seems to be too complicated and unrealistic to be implemented in an MT system because its rhetorical relations are dependent on an expected effect on the hearer achieved by their combination. In a MT system, as already mentioned, it is very hard, if practically not possible, to recognize automatically in a paragraph the goals and intentions of the speaker.</Paragraph>
  </Section>
class="xml-element"></Paper>