Header

Search

Extending the Bouquet Machine Translation Evaluation to Swiss German

Supervisor: Jannis Vamvas

Summary

BOUQuET: dataset, Benchmark and Open initiative for Universal Quality Evaluation in Translation (Andrews et al., EMNLP 2025) is a benchmark for machine translation in more than 250 languages.

One interesting feature of BOUQuET is that human translators can choose from several source languages, such as English or German, when adding a new language. It also covers extremely low-resource languages and different registers within a language.

The goal of this thesis is to add a Swiss German version of the benchmark. You could focus on your own dialect or, with the help of friends or volunteers, include several dialects.

You can then use the extended benchmark to evaluate machine translation systems. For example, you could compare whether ChatGPT or Claude performs better when translating:

  • from Swiss German into English;
  • from Standard German into Swiss German.