Evidence of Parallel Processing During Translation

Abstract

Three eye tracking experiments test the hypothesis that translation involves parallel rather than sequential processing of the source and target texts. In Experiment 1, a group of professional translators translated texts from their native language Danish into English. The texts included both segments where the order of verb and subject was congruent between source and target text and segments that were non-congruent. Translators gazed significantly longer at the non-congruent segments of the source text, indicating that the structure of the target text is anticipated during source text reading. Two follow-up experiments on first and second language reading demonstrate that this congruence effect in translation is not the result of the non-congruent Danish segments being inherently more difficult than the congruent ones and that the effect is not a general effect in bilingual reading. We conclude that translation is a parallel process and that literal translation is likely to be a universal initial default strategy in translation. This conclusion is strengthened by the fact that all three experiments were relatively naturalistic, due to the combination of remote eye tracking and mixed-effects regression modeling.