Will 'translationese' ruin a contrastive study?

Translated texts have been known as source material for contrastive analyses for a long time. Their value as suitable data has tended to be controversial throughout, and a new controversy is springing up now that corpus linguistics offers new perspectives for contrastive studies as well. Now that we can access large databases in any language, is there much point in drawing on the traditional kind of data that translations offer? This paper argues that corpora of translated texts constitute a valuable source of evidence for contrastive research, since they fulfil many of the criteria that have generally been seen as strengths in corpus study — for example language that has been used in its normal communicative contexts by a large number of users. Translations should be recognised as the normal part of a natural language that they are. The paper goes on to investigate a lexical item — English think — and its translation equivalents in Finnish, showing how sense categories of even frequent items can be drawn up with the help of another language. Unsuspected semantic prosodies may also come up in this process, and connections of particular senses with specific meanings. It demonstrates that a parallel corpus can capture relations of sense as well as form, which would be very hard to capture without such data.