This research examines how to enhance large language model (LLM) translation performance in specialized domains like medicine and law by incorporating domain knowledge at inference time.

• Investigates whether retrieving external domain knowledge or generating domain knowledge from the LLM itself is more effective
• Provides targeted recommendations for domain adaptation in technical translation
• Addresses key challenges in specialist translation without requiring full model retraining

Why it matters for Medical: Healthcare organizations can leverage these techniques to improve medical translation accuracy for international patients, research collaboration, and regulatory documentation—critical for patient safety and compliance.

Leveraging Domain Knowledge at Inference Time for LLM Translation: Retrieval versus Generation