A Minimal Cognitive Model for Translating and Post-editing

A Minimal Cognitive Model for Translating and Post-editing Moritz Schaeffer moritzschaeffer@gmail.com Gutenberg University, Mainz, Germany Michael Carl m.gummiball@gmail.com Renmin University of China, and Copenhagen Business School, Denmark Abstract This study investigates the coordination of reading (input) and writing (output) activities in from-scratch translation and post-editing. We segment logged eye movements and keylogging data into minimal units of reading and writing activity and model the process of post-editing and from-scratch translation as a Markov model. We show that the time translators and post- editors spend on source or target text reading predicts with a high degree of accuracy how likely it is that they engage in successive typing. We further show that the typing probability is also conditioned by the degree to which source and target text share semantic and syntactic properties. The minimal cognitive Markov model describes very basic factors which play a role in the processes occurring between input (reading) and output (writing) during translation. 1 Introduction We build a cognitive model of the translation process (from-scratch translation and post-editing) which aims at predicting where translation problems occur. We ground the model in transla- tion activity data that consists of keystrokes and gaze data that was captured during translation sessions. We decompose the translation process into minimal cycles of iterative reading and writing. We assume that the typing activities represent the solution to a translation problem that emerged during the preceding reading event. We show that the complexity (i.e. non-literality) of the produced translation as well as the duration and distribution of gaze activities on the source and target texts has an effect on the probability of a successive typing event. Schaeffer et al. (2016); Hvelplund (2016); Carl et al. (2016); L¨aubli and Germann (2016) describe methods to decompose the stream of eye movements and keystrokes into sequences of minimal activity units. In this paper we relate the duration of activity units with properties of the translation product — the degree of translation literality — to predict the probability when post-editors and translators will type after reading either the source (henceforth ST) or the target text (henceforth TT). Carl et al. (2016) show that a measure of translation literality has a great predictive power for behavioral observations in the translation process. According to this definition, a translation is literal if: 1. Word order is identical in the ST and TT 2. ST and TT items correspond one-to-one Proceedings of MT Summit XVI, vol.1: Research Track Nagoya, Sep. 18-22, 2017 | p. 144 3. Each ST word has only one possible translated form in a given context A translations which completely fulfills all three criteria is an absolutely literal transla- tions. A literal translation consists of the same number of ST and TT tokens where each TT token corresponds to exactly one ST token, and tokens in both texts are ordered in the same way. A change in word order or a situation in which one ST word is aligned to more than one TT word or vice versa weakens …

Content truncated to 500 words