Kim Setayert's PhD project studies de effectiveness and efficiency of machine translation for audio description from English into Dutch.

A key concern within Translation Studies is the profound impact of technological developments on the dynamic human-machine interactions. In this respect, the introduction of neural machine translation systems has had a profound influence on the study and practice of translation. The role of technology is particularly outspoken in the area of focus of the present project, namely Audio Description (AD). AD is an access service that translates images into words, which are inserted in between the music, sound and dialogue of the original audiovisual source text so that audiences who (cannot) see, still have access to the text's content. Despite technology being key in AD research and practice, machine translation for audio description has not been studied yet. Due to recent EU legislation, however, Flemish audiovisual content providers will have to drastically increase the amount of AD that they provide. The translation of existing English descriptions of foreign films and series into Dutch with the use of machine translation systems is an obvious avenue to be explored to meet these new legislative demands. However, limited preliminary research suggests that current machine translation systems do not generate an acceptable quality level for AD, because these systems have not been developed to meet the specific exigencies of this text type. ADs pose domain-specific translation challenges. It is a multimodal and intersemiotic type of translation and constitutes a unique transfer of information between semiotically distinct modes of communication; a fact that has not been taken into account in current research and a fact that poses methodological challenges given the lack of translation studies frameworks to study technology for multimodal text types such as AD. Against this background, the current project aims to explore machine-assisted translation for AD and the exigencies of audio description versus the possibilities of technology and human input, following three research objectives: • Applied objective: to explore the effectiveness and efficiency of machine translation for audio description into Dutch. • Strategic objective: to explore what innovative optimizations could improve the quality level of machine translation for audio description. • Fundamental objective: contribute to the discussion about the interdisciplinary and methodological challenges related to the study of technology and its interaction with humans in Translation Studies more generally, and for multimodal texts specifically. The project is a mixed-methods study, combining human-centered approaches and automatic evaluation methods with product as well as process-oriented research. It includes the human and machine evaluation of a corpus of translated audio descriptions, as well as an experiment with professional audio describers. This will allow us to gather data about the types of errors in the machine translation output, the number of errors made, the number and types of corrections made by professional describers and the time spent on correcting machine translation output. The text analysis and experiment will be supported by a thorough, interdisciplinary literature study, setting our findings off against current insights in literature and against the newest developments in machine translation research. The project constitutes a first step to gather fundamental knowledge regarding the study of technology for multimodal text types and strategic knowledge to start developing machine translation for audio description more systematically.