Cognitive Grammar as a Predictor of Differential Cognitive Effort in Translation from Japanese into English and Spanish

We investigate source-text features that lead to high levels of cognitive effort during written translation. Understanding the nature of such features can assist with gaining insight into the translation process as well as with designing pedagogical materials for novice translators or second language learners. Naturally, these features and the effort they induce depend on the language pair. Cognitive grammar predicts differences in cognitive effort patterns for Japanese-to-Spanish and Japanese-to-English translation, for example in the areas of subject expression and indirect expression. To test these predictions,we prepared short, general Japanese texts containing features like the ones discussed above. Participants translated these texts into Spanish or English. We examined variability in cognitive effort during translation of areas of interest in the source texts. Variables we considered were target-language (Spanish vs. English) and translator's L1 (Japanese vs. Spanish or English). Cognitive effortis frequentlyassessed through pausesin eyemovementsmeasured byan eye-tracker and pauses in keystroke activity measured by a keystroke logger. In this study, we used a relatively new approach to measuring cognitive effort. Wide variability in the target-texts produced by different translators indicates that high cognitive effort is exerted as translators resolve uncertainty over their chosen translation solution. The degree of variability is quantified by translation entropy HTra (Carl et al. 2016), which has its origins in information theory. We probed deeper into the standard concept of translation entropy, distinguishing between variability due to difficulties in source-text comprehension (manifested by high translation entropy and errors in the two target-languages) and variability due to translation difficulty (manifested, for example, by differential entropy measures in the same areas of the two target-languages). This work extends a pilot study presented at Congreso CANELA, 2019.