Comparing Post-editing based on Four Editing Actions against Translating with an Auto-Complete Feature

Comparing Post-editing based on Four Editing Actions against Translating with an Auto-Complete Feature Félix do Carmo Centre for Translation Studies University of Surrey f.docarmo@surrey.ac.uk Abstract This article describes the results of a work- shop in which 50 translators tested two ex- perimental translation interfaces, as part of a project which aimed at studying the de- tails of editing work. In this work, editing is defined as a selection of four actions: deleting, inserting, moving and replacing words. Four texts, machine-translated from English into European Portuguese, were post-edited in four different sessions in which each translator swapped between texts and two work modes. One of the work modes involved a typical auto-com- plete feature, and the other was based on the four actions. The participants an- swered surveys before, during and after the workshop. A descriptive analysis of the answers to the surveys and of the logs recorded during the experiments was per- formed. The four editing actions mode is shown to be more intrusive, but to allow for more planned decisions: although they take more time in this mode, translators hesitate less and make fewer edits. The ar- ticle shows the usefulness of the approach for research on the editing task. 1 Introduction1 1.1 Purpose This article describes an experiment that is based on a theoretical framework in which editing is de- fined as being composed of four actions (delete, insert, move and replace). This framework also in- cludes the definition of an editing threshold, which is a rate above which one may consider that the translator is no longer editing but translating the segment. The editing threshold was 1 © 2020 The author. This article is licensed under a Crea- tive Commons 3.0 licence, no derivative works, attribution, CC-BY-ND. experimentally set at 25% for the project, as an in- version of the 75% fuzzy match initial band used in the translation industry. (do Carmo 2017) The motivation for the experiment was to inves- tigate how translators edited machine translation (MT) output, with and without consideration for the four editing actions. At the beginning of the project, there was the expectation that this could contribute to the devel- opment of smart editing tools, which could learn patterns of editing based on these four actions, and then use this learned knowledge to support trans- lators' editing work. If such systems employed fea- tures like Online Learning (Ortiz-Martínez et al. 2016) to record and reuse, for example, the word substitutions that are required, each edit could be made more efficiently. Such a system would show the translator good candidates for deletion, sug- gest words that might be missing from the MT out- put and indicate possible new positions for words being moved. These features are particularly use- ful in texts with high internal repetition, and when the output only requires minor editing. After an analysis of the scope of the project, it was decided to focus on testing forms of interface for supporting editing work. The practical part was outlined as the …

Content truncated to 500 words