Learning to Translate Ambiguous Terminology by Preference Optimization on Post-Edits

2025-07-04

AI Summary

In real-world translation, terminology often exhibits polysemy, and its correct rendering depends on context and stylistic guidelines—nuances that conventional neural machine translation (NMT) struggles to capture due to its inability to model fine-grained, preference-driven decisions. Method: We propose the first end-to-end terminology disambiguation framework grounded in post-editing preferences, requiring neither hard dictionary constraints nor manual intervention during decoding. Our approach integrates term-level preference signals with sequence-level generation objectives, augmenting supervised fine-tuning with preference optimization to enable context-aware terminology selection. Results: On English–German translation, our method significantly improves terminology accuracy while maintaining stable COMET scores. We publicly release a human-annotated test set and a domain-specific terminology lexicon, establishing a new paradigm for controllable NMT.

Technology Category

• Natural Language Processing: Machine Translation, Multilinguality, Cross-Lingual NLP
• Machine Learning: Learning Preferences or Rankings
• Knowledge Representation and Reasoning: Preferences

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

In real world translation scenarios, terminology is rarely one-to-one. Instead, multiple valid translations may appear in a terminology dictionary, but correctness of a translation depends on corporate style guides and context. This can be challenging for neural machine translation (NMT) systems. Luckily, in a corporate context, many examples of human post-edits of valid but incorrect terminology exist. The goal of this work is to learn how to disambiguate our terminology based on these corrections. Our approach is based on preference optimization, using the term post-edit as the knowledge to be preferred. While previous work had to rely on unambiguous translation dictionaries to set hard constraints during decoding, or to add soft constraints in the input, our framework requires neither one-to-one dictionaries nor human intervention at decoding time. We report results on English-German post-edited data and find that the optimal combination of supervised fine-tuning and preference optimization, with both term-specific and full sequence objectives, yields statistically significant improvements in term accuracy over a strong NMT baseline without significant losses in COMET score. Additionally, we release test sets from our post-edited data and terminology dictionary.

Problem

• Disambiguate ambiguous terminology in neural machine translation
• Leverage human post-edits for preference optimization
• Improve term accuracy without compromising translation quality

Innovation

• Preference optimization on post-edits
• No need for one-to-one dictionaries
• Combines supervised fine-tuning and preference optimization