Audio description translation and expectancy norms: a comparison of human-translated and post-edited machine-translated output

Against the backdrop of the European Accessibility Act (EAA) and a growing interest in audio description (AD) translation workflows, this article reports on a small-scale exploratory study comparing human AD translation with post-edited machine translation (PEMT). Eight AD trainees translated an English AD script for a movie clip into Polish, having been explicitly instructed to observe the Polish AD style. Their output was compared with a PEMT version generated by the neural machine translation (NMT) engine DeepL. The results revealed substantial differences between the two translation types, both quantitatively (readability, word and sentence counts, and parts-of-speech analysis) and qualitatively (information sequencing, verbless sentences, theme-rheme structure, nominalisation, and pronouns). Drawing on Chesterman’s concept of expectancy norms, this study proposes a new AD creation method called ‘transdescription’, in which the interventions of the translator-cum-describer (‘transdescriber’) exceed those expected in regular interlingual translation in terms of both AD style and content. Moreover, the article further discusses implications for transdescriber training and for AD translation workflows.