Assessing eye-tracking accuracy in translation studies
Christian B. Jensen · Copenhagen studies in language · 2008
This paper describes an experiment which was designed to test the ability of an eye-tracker and the associated software to correctly identify which words in a text are fixated by the reader. In particular, the experiment attempted to establish (1) whether the reading pattern which we often observe in a translation task gives rise to more detection problems than a more linear reading task, and (2) what kind of overall accuracy can be expected from the investigated tracker and software. The results indicate that in a normal linear reading style, fixations are detected on an average of (at best) 65-75 % of all longer words (= 4 characters or more). The accuracy is slightly better for very long words but worse for short and medium-length words. Two tasks using a masked text design in order to force specific reading patterns yielded somewhat lower scores, but suggested that detection accuracy was the same for non-linear reading (as in a translation task) as for linear reading, or indeed better. However, the overall lower scores may be an indication of technical problems with the experimental design, and certain modifications are suggested for future applications of the design for testing accuracy.