Stylistic Choice in Machine Translation

The study of comparative stylistics attempts to catalogue and explain the differences in style between languages. We describe a formal model of comparative syntactic stylistics that could be implemented in a machine translation system, enabling the system to make a more informed decision about possible translation choices and their potential stylistic effects. 1 Stylistic adequacy of translation We have all had the experience of reading a translated text, and feeling that it was not quite right, but not being able to articulate exactly why this was so. The text may be divided into the correct number of paragraph units, the semantic content may all be there, but something in the way the material is presented strikes a false note: while the meaning of the source text may be intact, the style of the target text is faulty. We have argued (DiMarco and Hirst 1993) that style is not mere surface appearance, but conveys a good deal of meaning, so that if the style of a translation is not appropriate, then some significant aspects of meaning will have been lost. In earlier work (DiMarco and Hirst 1990), we showed the need to account for aspects of style in machine translation, and noted that style influences translations on all linguistic levels: lexical, syntactic, semantic, and textual. The particular choices of words, the specific arrangement of sentences, the selection of which details to carry over in the translation, all convey a particular stylistic effect. The linguistic choices the translator makes in selecting various ways of realizing the source text in the target language carry pragmatic import, and therefore affect the meaning communicated to the reader. As we observed in (DiMarco and Hirst 1990), consideration of style in translation involves two complementary, but sometimes conflicting, aims: • The translation must preserve, as much as possible, the author's stylistic intent—the information conveyed through the manner of presentation. • But it must have a style that is appropriate and natural to the target language. From comparative stylistics, we recognize that languages differ in their stylistic approaches, that each has its own characteristic stylistic preferences. Thus, what is needed for truly effective machine translation is a way of incorporating knowledge of comparative stylistics to make the most appropriate choices during the process of generating target text. In (DiMarco and Hirst 1990), we first proposed the outline of a computational model of comparative stylistics for machine translation. The model focused on syntactic aspects of style; theoretical