Multi-modal estimation of cognitive load in post-editing of machine translation

Nico Herbig, Santanu Pal, Antonio Krüger, Josef van Genabith

In: Tra&Co Group (editor). Translation, Interpreting, Cognition: The Way Out of the Box. Chapter 1 Pages 1-32 Language Science Press 2021.


In this paper, we analyze a wide range of physiological, behavioral, performance, and subjective measures to estimate cognitive load (CL) during post-editing (PE) of machine translated (MT) text. To the best of our knowledge, the analyzed feature set comprises the most diverse set of features from a variety of modalities that has to date been investigated in the translation domain. Our focus lies on predicting the subjectively reported perceived CL based on the other measures, which could for example be used to better capture the usefulness of MT proposals for PE, including the mental effort required, or to develop cognition-aware translation environments that support human translators according to their current level of CL. Based on the data gathered from 10 professional translators, we show that feature sets from all different modalities outperform our baseline measures in terms of predicting the subjectively perceived level of CL, and that especially eye-, heart-, or skin-based features yield good results in a simple ``top-down'' regression analysis using feature selection. When passing the participant and segment to the regression models, other modalities like keyboard, text, body posture, or time, also perform well. An additional correlation analysis provides insights into redundancies among the features which may be used to further improve the currently achieved best regression score of 0.7 mean squared error (MSE) on a 9-point scale.


Weitere Links

paper_herbig_multi_modal_CL_LangSci.pdf (pdf, 664 KB )

German Research Center for Artificial Intelligence
Deutsches Forschungszentrum für Künstliche Intelligenz