This project aims to develop, in collaboration with SRF and recapp IT, an automatic post-editing system to improve the quality of automatic German subtitling of Swiss German television programs. It has three main objectives. The first one is to compare different methods of automatic post-editing for this task, the second one is to collect the opinions of users on this type of transcription in order to generate subtitles that are as accessible as possible, and the third one is to share written and oral resources to promote research on Swiss German in the field of media.
Automatic post-editing will be done using modern Machine Translation (MT) techniques. On the scientific level, this project contributes to research on poorly endowed languages, on automatic post-editing, especially with new neural methods, and on the understandability of this type of transcription. In addition, it has a strong societal impact, as it aims to make television more accessible by subtitling programs in Swiss German and making them available in standard German for people who do not speak Swiss German or who suffer from a hearing impairment. This project therefore complies directly with the new legal requirements for accessibility in Europe and falls squarely within the multilingual context of Switzerland, offering solutions to promote linguistic diversity, multilingual cohesion and cultural exchange, which is also the primary mission of the media.
Now, the Department of Translation Technology (TIM) of the University of Geneva has set up a demo page to test the automatic post-editing API that will later become available to media partners.
Watch on YouTube