Combining Logic-Based and Corpus-Based Methods for Resolving Translation Mismatches
Megumi Kameyama, Stanley Peters, Hinrich Schfitze · 1993
Very often in good quality translations, a target language expression (word, sentence or text) differs meaning from the corresponding expression in the source language. Fortunately quality translation consists in approximating the meaning of a source document, for the same meaning often cannot be rendered exactly in the target language. This impossibility is called translation mismatch by Kameyama, Ochitani, and Peters [4], where two of us discussed some factors involved in choosing an appropriate meaning approximation as required. The present paper continues that discussion, and draws conclusions regarding bi-lingual lexicons. Languages differ in the concepts and real-world entities that they have words and grammatical constructs to describe. This is why translation must frequently be a matter of approximating the meaning of a source language text rather than finding an exact counterpart in the target language. A major cause of the inadequate quality presently achieved by machine translation (MT) is the inability to generate contextually appropriate approximations when no exact translation exists between the source and target languages. An MT system needs to recognize mismatches when it encounters them, and to resolve mismatches either by recovering implicit information from their context, as required by the target language, or by leaving out some information the source text gives. English has lexical items like picture, photograph, painting, water color, oil painting, drawing, and line drawing, whose meanings are partially ordered by the relation of hyponymy (inclusion of meaning). In the same semantic field, Japanese has e, syasin, snisaiga, aburae, and senbyou. Only four words in each language have synonyms in the other language (i.e., words with the same meaning): water color (when it refers to the painting) and suisaiga, oil painting and aburae, line drawing and senbyou, photograph and syasin. The Japanese word e is more specific than English picture, as it does not apply to photographs; however, it is more general than painting and drawing, both of whose meanings it includes. To translate an English stateHinrich Schfitze Stanford University