Jannis Vamvas, PhD
Language AI Researcher at University of Zurich
Recent posts
Just Released: Machine Translation Engine for Romansh
Covering the six varieties of the Romansh language.
Read post
OpenAI's Speculative Decoding,
Reverse-Engineered
Why LLMs are faster if we give them a draft to complete.
Read postRecent publications
Apertus Team. 2026. Apertus 1.5: From Open Text Models to Open Multimodal Reasoners. Pre-print. [cite] [model]
Vilém Zouhar, Niyati Bafna, Mukund Choudhary, Maike Züfle, Sara Rajaee, Pinzhen Chen, Jannis Vamvas, and 237 others. 2026. Last Translation Benchmark. Pre-print. [cite] [data] [code]
Jannis Vamvas, Ignacio Pérez Prat, Angela Heldstab, Dominic P. Fischer, Sina Ahmadi, and Rico Sennrich. 2026. Translation Asymmetry in LLMs as a Data Augmentation Factor: A Case Study for 6 Romansh Language Varieties. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 4837–4861, Budapest, Hungary. Association for Computational Linguistics. [cite] [data] [model] [code]
Michelle Wastl, Jannis Vamvas, and Rico Sennrich. 2026. Scaling Unsupervised Word Alignment to Documents via Structural Constraints. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 7419–7436, Budapest, Hungary. Association for Computational Linguistics. [cite] [data] [code, code]
Hanxu Hu, Zdenek Snajdr, Pinzhen Chen, Jannis Vamvas, and Rico Sennrich. 2026. Reinforcement Learning Elicits Contextual Learning of Unseen Language Translation. In Findings of the Association for Computational Linguistics: EMNLP 2026, pages 44643–44657, Budapest, Hungary. Association for Computational Linguistics. [cite] [code]