Standard German Subtitling of Swiss German TV content: the PASSAGE Project

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 5063–5070 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 5063 Standard German Subtitling of Swiss German TV content: the PASSAGE Project Jonathan Mutal, Pierrette Bouillon, Johanna Gerlach, Veronika Haberkorn Faculty of translation and interpreting, University of Geneva 40, bd du Pont d’Arve, 1211 Geneva, Switzerland {jonathan.mutal, pierrette.bouillon, johanna.gerlach}@unige.ch, veronika.haberkorn@etu.unige.ch Abstract In Switzerland, two thirds of the population speak Swiss German, a primarily spoken language with no standardised written form. It is widely used on Swiss TV, for example in news reports, interviews or talk shows, and subtitles are required for people who cannot understand this spoken language. This paper focuses on the task of automatic Standard German subtitling of spoken Swiss German, and more specifically on the translation of a normalised Swiss German speech recognition result into Standard German suitable for subtitles. We compared different statistical and deep learning machine translation systems for this task. We also produced an aligned corpus of normalised Swiss German and Standard German subtitles. Results of two evaluations, automatic and human, show that the systems succeed in improving the content, but are currently not capable of producing entirely correct Standard German. Keywords: automatic subtitling, Swiss German, machine translation, automatic post-editing 1. Introduction In Switzerland, two thirds of the population speak Swiss German, thus this language is widely used on Swiss TV, for example in news reports, interviews or talk shows. Swiss German is primarily a spoken lan- guage, with many regional dialects and no standardised written form (Honnet et al., 2018). In order to make these Swiss German contents accessible to people who cannot understand spoken Swiss German, either due to hearing impairments, or because they only understand Standard German, these TV programs need to be subti- tled in Standard German. For daily TV content, where large amounts of subtitles need to be produced within a short time frame and in a cost-effective manner, being able to automate the subtitling process would be ad- vantageous. The PASSAGE project ”Sous-titrage au- tomatique du suisse allemand en allemand standard”1, a Swiss project financed by IMI (”Initiative for Media Innovation”), focuses on this task. One way to automate subtitling is to combine a speech recognition system with an intralingual machine trans- lation (MT) system. In this process, MT can be used to different ends, for example correcting speech recogni- tion issues or transforming content to achieve compli- ance with subtitling standards (Buet and Yvon, 2021). In this study, we specifically focus on the translation of normalised Swiss German speech recognition out- put into Standard German and explore different MT ar- chitectures to deal with the divergences between these two languages (Scherrer, 2011). Figure 1 illustrates the complete subtitling pipeline. The first step is automatic speech recognition (ASR) of Swiss German (GSW) to produce a normalised transcription (GSW REC), keep- ing the original syntax and expressions, but using Ger- 1https://www.media-initiative.ch/project/subtitling-of- swiss-german-into-standard-german-automatic-post-editing/ man words (Arabskyy et al., 2021) since there is no …

Content truncated to 500 words