A Rising Tide Lifts All Boats: MTQE Rewards for Idioms Improve General Translation Quality
Ishika Agarwal, Zhenlin He, Dhruva Patil, Dilek Hakkani-T\"ur

TL;DR
This paper explores fine-tuning neural machine translation models with MTQE rewards to improve idiom translation, resulting in significant gains in translating non-compositional expressions and enhancing overall translation quality.
Contribution
It introduces a novel MTQE-based reward training approach specifically targeting idiom translation in neural machine translation models.
Findings
Idiomatic translation improves by approximately 14 points.
General translation quality improves by around 8 points.
Cross-lingual translation abilities increase by about 6 points.
Abstract
Non-compositional expressions (e.g., idioms, proverbs, and metaphors) pose significant challenges for neural machine translation systems because their meanings cannot be derived from individual words alone. These expressions encode rich, cultural meaning, and have both figurative and literal meanings, making accurate translation difficult. Because models are fairly good at translating compositional text, we investigate GRPO-style fine-tuning using Machine Translation Quality Estimation (MTQE) models as reward functions to train models to better translate idioms. Using Chinese and Hindi idiom datasets, we find that idiom translation abilities improve by ~14 points, general, non-idiomatic translation implicitly improves by ~8 points, and cross-lingual translation abilities (trained on one language, evaluated on another) improves by ~6 points. Overall, our work quantifies the…
Peer Reviews
No public reviews on file for this paper yet. If you reviewed it on a platform where reviews are public (OpenReview, ICLR, NeurIPS, ICML), you can paste yours below so the community can read it here.
Code & Models
- 🤗ishikaa/Chinese_llama8b-damodel
- 🤗ishikaa/Chinese_llama8b-qe-consmodel· 1 dl1 dl
- 🤗ishikaa/Chinese_llama8b-qe-posmodel· 1 dl1 dl
- 🤗ishikaa/Chinese_llama8b-qe-negmodel
- 🤗ishikaa/Chinese_qwen3b-damodel· 1 dl1 dl
- 🤗ishikaa/Chinese_qwen3b-qe-consmodel
- 🤗ishikaa/Chinese_qwen3b-qe-posmodel· 1 dl1 dl
- 🤗ishikaa/Chinese_qwen3b-qe-negmodel
- 🤗ishikaa/Hindi_llama8b-damodel
- 🤗ishikaa/Hindi_llama8b-qe-consmodel
Videos
No videos yet. Explain this paper in a talk, walkthrough, or lecture? Add one.
Taxonomy
TopicsLanguage, Metaphor, and Cognition · Natural Language Processing Techniques · Topic Modeling
