Design tools, methods, or datasets to enable AI to accurately “speak” Swiss German—both in text and potentially in voice—without losing general German comprehension. Possible approaches could include:

• Detect and quantify High German bias in AI-generated text (e.g., spelling, word frequency, syntax patterns).
• Develop fine-tuning or retraining methods to adapt LLM outputs to Swiss German orthography and vocabulary.
• Build Swiss German–specific datasets (text, transcripts, or parallel corpora) while respecting privacy and copyright.
• Create evaluation metrics for measuring Swiss German accuracy and cultural authenticity in AI.

Deliverables

• Working prototype (e.g., web tool, LLM fine-tuning script, Swiss German spell-checker).
• An evaluation of trade-offs (e.g., risk of overfitting to dialect vs. preserving general German fluency).
• Documentation explaining your approach and datasets.

See also:

• https://www.supertext.com/
• https://hack.opendata.ch/event/22