Paper Overview
- Field: NLP
- Authors: Arnav Mazumder, Dengjia Zhang, Shuyue Stella Li, Yulia Tsvetkov, Niyati Bafna
- Published: 2026-06-25
- arXiv: 2606.27306
- Evaluated across nine multilingual benchmarks covering various task types, three backbone models, and 285 high-, mid-, and low-resource languages.
- Context-aware cascades deliver significant gains across models and resource regimes, particularly on open-ended generation tasks.
- The original-language question carries most of the beneficial context supplied to the final translation step.
Summary
Translation cascades for reasoning translate the query from another language into English, reason in English, and translate the answer back to the original language. This is a competitive approach to multilingual reasoning, but it is structurally lossy: each stage discards information that later stages may need, including cues for cultural grounding, register, and disambiguation.
The paper examines a simple, training-free intervention: a context-aware translation cascade, which additionally provides the original question, the English-translated question, and the reasoning trace to the context of the final translation module.
Key Findings
Takeaway
The study highlights the need to better design information flow in machine translation cascades to mitigate error propagation, and offers a simple, actionable default policy: retain the original user question at the end of the pipeline.
--- *Auto-collected on 2026-06-28*