Capturing Language-Specific Semantic Distinctions in Interlingua-based MT

[MT Summit III, Washington, DC, USA, 1-4 July 1991] Capturing Language-Specific Semantic Distinctions in Interlingua-Based MT James Barnett, Inderjeet Mani, Elaine Rich, Chinatsu Aone, Kevin Knight, Juan Carlos Martinez Microelectronics and Computer Technology Corporation (MCC) 3500 West Balcones Center Drive Austin, Texas 78759 Abstract We describe an interlingua-based approach to machine translation, in which a DRS represen- tation of the source text is used as the inter- lingua representation. A target DRS is then created and used to construct the target text. We describe several advantages of this level of representation. We also argue that problems of translation mismatch and divergence should properly bo viewed not as translation problems per se but rather as generation problems, al- though the source text can be used to guide the target generator. The system we have built relics exclusively on monolingual linguistic de- scriptions that are also, for the most part, bi- directional. 1 System Goals We describe an approach to machine translation that is motivated by the following goals. First, we want to maximize the ability of the system to produce correct and natural sounding translations even when translation mismatches and divergences occur. Second, we seek to minimize the cost of adding new languages to the system. Although the second of our two goals is clear, we should say a bit more about what we mean by the first. There are many cases in which the natural translation of one language into another has a very different form than the form in the original. These cases can be divided into two categories: • Translation divergences • Translation mismatches Translation divergences [Dorr, 90] arise when the same information is conveyed in the source and target texts, but the structures of the sentences are different. An ex- ample of this is the English/French pair, “I swam across the river” and “J’ai traversé la rivière à la nage” [Mounin 63]. Translation mismatches [Kameyama, et. al, 91] occur when there are actually differences in the information that is conveyed. If the source text is vague or ambigu- ous in ways that are not allowed in the target language, then the translation process must add information by first making the best possible guess about the intent of the source text and then rendering that intent into the target text. This problem comes up, for example, in translating from English into Japanese, where it is nec- essary to add politeness information. If, on the other hand, the target language allows vagueness or ambigu- ity that is not. allowed in the source, it may be neces- sary, in creating a natural translation, to throw away some information that is present in the source text and not render it completely into the target text. An ex- ample of this occurs in translating from English into Japanese, since it is often necessary to throw away num- ber information that is mandatory in English and not in Japanese. Although syntactic differences between lan- guages cause many cases of translation mismatch, they are …

Content truncated to 500 words