Low load feedback improves machine translation quality consistently
Feedback as an Interface: Controlling Form and Detail in LLM Translation Refinement
Computation and Language
Summary
Improving translations with feedback from language models depends on how the feedback is given. The authors found that simple, compact feedback works better than detailed, complicated feedback for fixing translation errors. They tested this across many language pairs and models and showed that less is more when guiding improvements. This approach can make translation refinement more effective and efficient.
What this means in practice
- •For machine translation engineers: Implement simple and focused feedback to improve automatic translation refinement in high-resource language pairs.
- •For natural language processing teams: Design evaluator interfaces that minimize feedback detail to enhance model-guided correction of text generation tasks.
Authors
Yiqi Liu, Flavjo Tomori, Nafise Sadat Moosavi, Steffen Eger, Chenghua Lin
Abstract
Feedback presentation shapes how LLMs use evaluator judgments during translation refinement. We introduce a machine translation protocol that keeps the evaluator score, main issue, and recommendation fixed while varying feedback form and supporting detail. Across seven models and six translation directions, low-load feedback outperforms high-load feedback in all 42 translator-direction settings under xCOMET. On the 29,400-record rewriting subset, mean xCOMET scores multiplied by 100 are 83.30 for low-load feedback and 80.67 for high-load feedback; the full-matrix original-translation reference mean is 83.46. MetricX-24 supports the low-load advantage on a stratified subset. A no-feedback second pass yields a small improvement over the original translations, and direct evaluator prompting reproduces the low-load advantage without a separate rewriting stage. These results show that compact feedback limits the quality loss from evaluator-guided refinement in the evaluated high-resource MT settings.