<?xml version="1.0" standalone="yes"?>
<Paper uid="C90-2003">
  <Title>FINDING TRANSLATION EQUIVALENTS: AN APPLICATION OF GRAMMATICAL METAPHOR</Title>
  <Section position="10" start_page="0" end_page="0" type="concl">
    <SectionTitle>
4 Note that I am not claiming that these sentences provide
</SectionTitle>
    <Paragraph position="0"> the only translations of each other. The argument rests upon these being possible desired translation equivalents in some particular circumstances. 3 Although a similar range of possibilities (V-C1 - V-C4) are theoretically available for the realization of the SPL in Germma, not all possible variations are equally acceptable in all contexts. For example, the realization of the cause-effect relation at clause rank calls for a process to realize that relation: a plausible selection on the basis of similarity with the English sentence might appear to be the verb 'resultieren' -- however, this is very marked. The realization at circumstantial adjunct group rank (in German with, e.g., a wegenadjunct) is often preferable.</Paragraph>
    <Section position="1" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
5.3 Realizations of modality
</SectionTitle>
      <Paragraph position="0"> Consider the following translation pair (Schlitz, 1989), which is again problematic in a transfer-based fl'amework that relies on representations less abstract than that of the upper model and SPL because the structures are very different:</Paragraph>
      <Paragraph position="2"> As before, several differential realizations of this SPL can be generated by allowing the grammatical metaphor mechanisms to order the consumption of terms as heads at differing ranks in the grammar.</Paragraph>
      <Paragraph position="3"> For example, the orderings \[D A B C\], \[A D B C\] and \[A B C D\] support clause structures with the following dependency organizations respectively, showing realizations of the statement of probability at clause complex rank, verbal group rank, and circumstantial  What would previously have required rather complex transfer mechanisms, therefore, is here an automatic consequence of tile realizational mechanisms.</Paragraph>
    </Section>
    <Section position="2" start_page="0" end_page="0" type="sub_section">
      <SectionTitle>
5.4 Across the text/grammar
</SectionTitle>
      <Paragraph position="0"> boundary As a final example of the generality of the mechanisms described here (for further examples, see Bateman, 1989), it is also possible to consider dependency relations defined above the clause complex as possible targets. The 'highest' rank assignment for the cause-effect relationship suggested so far has been that of clause complex; this is the highest rank available in the grammar. However, in current work in the PENMAN group on Rhetorical Structure Theory (RST: Mann and Thompson, 1987), there is the possibility of extending this view to consider the dependency relationships defined across sentences in texts. If the following translation pair were desirable, for example, then we see precisely this kind of variation multilingually: null  (7) After you had explained your views, I could see them much better.</Paragraph>
      <Paragraph position="1"> (8) Du erkl~irtest deine Position, Danach verstand ich sie viel besser.</Paragraph>
      <Paragraph position="2">  Focusing on the temporal/causal relationship expressed between the two situations described, both sentences can receive the combined SPL and RST specification:  Sentence (8) preserves the realization of the RST relation at the text level and so produces two independent sentences. The next available strategy for realization is within the grammar, at clause rank with complexing. This realization is adopted in the English variant (7). Again, therefore, the metaphorset provides a powerful link between translationally equivalent texts, even beyond the boundaries of the grammar. Work on this will continue as our development of RST and text organization continues. o o 6 Conclusmn. Future directions null Following from earlier experiments in the integration of Machine Translation (MT) and text generation (Bateman, Kasper, Schlitz, and Steiner, 1989), we have found that the combination of two components of the Systemic-Functional Linguistic (SFL) model of language (Halliday, 1985; Matthiessen, 1987), such as are being developed for computational use within the PENMAN text generation system, significantly reduces tlhe need for structural transfer in machine translation (MT) without requiring deep modelling of specific domains. These two components may be described thus: (r) a linguistically motivated organization of general semantic distinctions that are not highly language-specific and that hold across both differing domains and differing languages -- this level of organization is realized computationally within the PENMAN system as the upper model; (r) a particular class of mapping relations between this abstract semantic organization and linguistic form- these mappings are motivated by the notion of 'grammatical metaphor'.</Paragraph>
      <Paragraph position="3"> The examples of the previous section made the point that the level of grammatical metaphorical realization is not preserved during translation, but the set of possibilities defined by granmmtical metaphor provides strong candidates for high quality translation equivalents, ttowever, since SPL forms decommit from so much of the surface syntax, it is necessary to control the range of linguistic realizations that are compatible with the constraints SPL specifications represent. The types of control that are necessary already constitute major areas of active research within the PENMAN teXt generation project, where the principal type of constraint under development is textual. It is reasonable to assume that textual organization, including the particular patterns of thematic development adopted by a text, is one type of meaning which needs to be preserved in translation. Metaphorically related clauses often have rather different textual organizations and so particular choices of clause may be motivated by the need to achieve particular thematic developments.</Paragraph>
      <Paragraph position="4"> A rather more global kind of textual constraint is that created by the 'style' of a text as a whole. Thus, for example, the translation process should not select as translation equivalents sentences whose stylistic import were radically divergent. This, however, can only be established on the basis of analyses of what constitutes an appropriate text style for a particular communicative situation. This is being addressed in text generation in terms of the SFL notion of register (Bateman and Paris, 1989), and its application to MT is pursued in Paris (forthcoming).</Paragraph>
      <Paragraph position="5"> The relevance of textual constraints is also not limited to the simple types of grammatical metaphor illustrated in this paper; for example, there are also 'metaphors of transitivity' which chang(: the process-type involved, e.g.:  (9) electricity was discharged : directed- action (10) there was a discharge of electricity: re- lationM</Paragraph>
      <Paragraph position="7"> Variations such as these provide significantly different thematization options. In certain cases, therefore, in order to achieve a particular thematic development for a text in a given language, this kind of metaphor may be the only means available for achieving a syntactically thematizeable constituent.</Paragraph>
      <Paragraph position="8"> This type of grammaticM metaphor can also be motivated by the need to preserve Aklionsarl (cf.</Paragraph>
      <Paragraph position="9"> Vendler, 1!)67). In the following translation pair, for example, we need to change the transitivity type of the predicate in order to preserve the accomplishment Aktionsart. 6 (11) Wit erarbeiteten uns die Theorie (12) We worked ourselves into the theory Finally, the lexical entries of certain languages may also restrict the types of alternatives that are available. If one language simply does not have a lexical item for realizing a meaning at a given rank, then another rank will be pursued.</Paragraph>
      <Paragraph position="10"> Many of the mechanisms that need to be supported for these capabilities to be achieved are equally relevant in the generation of texts from single languages. We expect, therefore, that the results we are obtaining in a monolingual setting --- particularly those concerned with the creation of texts, rather than sentences -- will generalize well to application in multilingual environments such as MT.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>