Terminology-Aware Retrieval-Augmented Knowledge Distillation for Biomedical Neural Machine Translation

Maria Zafar, Souhail Bakkali, Rejwanul Haque


Abstract
Knowledge distillation (KD) compresses large teacher models into smaller student models by transferring soft labels or intermediate activations. While effective in general domains, KD alone falls short in specialised machine translation (MT) settings, such as biomedical translation. The student inherits only the teacher’s compressed knowledge and lacks access to external domain information. Moreover, standard KD typically relies on abundant parallel data, which is often unavailable in domain-specific scenarios. To address these limitations, we combine KD with retrieval-augmented generation (RAG) in a few-shot setting. We propose a retrieval-augmented enhanced few-shot KD framework for French-to-English biomedical translation task. The student learns to retrieve relevant in-domain knowledge from an external database, complementing the teacher’s supervision. We design and compare several retrieval strategies to enhance student capacity. Experiments show that with our terminology-aware retrieval-based methods, the student achieves performance comparable to or better than the teacher, while preserving translation quality and efficiency.
Anthology ID:
2026.eamt-1.26
Volume:
Proceedings of the 26th Annual Conference of the European Association for Machine Translation (Volume 1)
Month:
June
Year:
2026
Address:
Tilburg, The Netherlands
Editors:
Dimitar Shterionov, Eva Vanmassenhove, Mirella De Sisto, Fred Blain, Javad Pourmostafa Roshan Sharami, Lisa Lepp, Chiara Manna, Argentina Anna Rescigno, Alina Karakanta, Ayla Rigouts Terryn, Manuel Lardelli, Natalia Resende, Elena Murgolo, Janiça Hackenbuchner, Anna Zaretskaya, Miquel Esplà-Gomis, Thierry Etchegoyhen, Dagmar Gromann, Rachel Bawden, Barry Haddow, Sara Szoc, Mikel Forcada, Helena Moniz
Venue:
EAMT
SIG:
Publisher:
European Association for Machine Translation
Note:
Pages:
399–411
Language:
URL:
https://aclanthology.org/2026.eamt-1.26/
DOI:
Bibkey:
Cite (ACL):
Maria Zafar, Souhail Bakkali, and Rejwanul Haque. 2026. Terminology-Aware Retrieval-Augmented Knowledge Distillation for Biomedical Neural Machine Translation. In Proceedings of the 26th Annual Conference of the European Association for Machine Translation (Volume 1), pages 399–411, Tilburg, The Netherlands. European Association for Machine Translation.
Cite (Informal):
Terminology-Aware Retrieval-Augmented Knowledge Distillation for Biomedical Neural Machine Translation (Zafar et al., EAMT 2026)
Copy Citation:
PDF:
https://aclanthology.org/2026.eamt-1.26.pdf