The authors formulate translation as a sequential policy whose available actions include reasoning, issuing a retrieval query, incorporating returned evidence and producing a final translation.
Translation agents learn to retrieve evidence only when it improves results
SRAT jointly learns whether to retrieve, what to search for and how to use evidence by rewarding final translation quality.
Big Tech
Shijun Li · Yang Liu · Najmeh Sadoughi · Dongmei Wang · Xinjie Li · Zhu Liu · +2 more
The University of Texas at Austin · Amazon
Research Digest··2 min read
Li and colleagues present Selective Retrieval-Augmented Translation, an agentic reinforcement-learning framework that treats retrieval as an optional translation action rather than fixed preprocessing.
Why this paper
From Amazon and The University of Texas at Austin
In one line
LLM translation improves by learning when and what to retrieve instead of always retrieving.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§