Expert-to-Lay Communication: The Use of Automatic Speech Recognition and Machine Translation Post-Editing (PEMT) in Translations for the Medical Domain

Doctoral project - Raluca Chereji, BA MSc

In recent years, translation has become an increasingly technologised activity. Practices such as Machine Translation Post-Editing (PEMT) and dictated translation using Automatic Speech Recognition (ASR) tools are garnering attention from industry and academia alike for their potential to improve translators' productivity and output quality. However, there is currently little empirical research on the impact of said practices within low-resource language contexts and in high-risk, zero-error domains such as medical translation, and much less work on comparative investigations of different translation technologies in these environments.

This investigation is especially necessary in the medical domain. Here the work is particularly challenging: not only are translators required to perform interlingual adaptation from one language into another, but in the case of patient-facing medical texts such as Patient Information Sheets (PIS) and Informed Consent Forms (ICFs), they also need to translate texts intralingually, i.e., transferring the content from expert language to plain, non-expert language.

In practice however, research suggests translators may fail to suitably adapt to the conventions and requirements of a lay, non-expert audience, reverting instead to specialised language and jargon which are difficult to understand for lay persons. This in turn highlights the need for research into whether, and which, technologies and workflows could assist translators in mitigating this expert-to-lay bias in patient-facing medical texts.

Given these challenges, my doctoral thesis will investigate the impact of PEMT and ASR on professional medical translators' output quality and work processes, compared to standard, typed translation from scratch, in the context of patient-facing medical translations.

The project relies on a mixed-methods approach combining questionnaires, a corpus linguistics analysis, and an eye-tracking study, in order to:

  • identify the challenges reported by professional medical translators working on patient-facing texts, as well as what technologies and workflows they use for these translations;
  • evaluate the impact on translators' cognitive, technical and temporal effort of translating from scratch, post-editing and dictating translations using ASR tools for patient-facing texts;
  • propose best practice guidelines on integrating these tools into patient-facing medical translation workflows.