Is Human Annotation Necessary? Iterative MBR Distillation for Error Span Detection in Machine Translation
arXiv:2603.12983v3 Announce Type: replace Abstract: Error Span Detection (ESD) is a crucial subtask in Machine Translation (MT) evaluation, aiming to identify the location and severity of translation errors. While fine-tuning models on human-annotated data improves ESD performance, acquiring such data is expensive and prone to inconsistencies among annotators. To address this, we propose a novel self-evolution framework based on Minimum Bayes Risk (MBR) decoding, named Iterative MBR Distillation for ESD, which eliminates the reliance on human annotations by leveraging an off-the-shelf LLM to g
The increasing availability and capability of large language models (LLMs) are enabling novel approaches to automate tasks previously requiring expensive human annotation.
This development could significantly reduce the cost and improve the consistency of machine translation evaluation, accelerating MT progress and deployment across various applications.
Machine translation evaluation, specifically error span detection, can now be performed with reduced reliance on human-annotated data, lowering barriers to entry and increasing scalability.
- · Machine Translators
- · NLP Researchers
- · AI/ML Developers
- · Companies with MT needs
- · Human annotators for MT evaluation
Automated and more consistent evaluation of machine translation quality becomes more accessible.
Faster iteration and improvement cycles for machine translation models across various languages and domains.
Enhanced global communication and information exchange due to more accurate and reliable machine translation services.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL