Research map / Label-efficient and robust learning / Recognition and long-tail
“Bilingual Expert” Can Find Translation Errors
Kai Fan, Jiayi Wang, Bo Li, Fengming Zhou, Boxing Chen, Luo Si
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Add model-derived and mis-matching features from a pretrained bilingual expert | Bi-LSTM sentence QE · WMT 2017 en-de and de-en | Pearson's: 0.6599 -> 0.6837; 0.6985 -> 0.7099 | Improved |
| Remove mis-matching features (ablation) | Ours Single sentence QE · WMT 2017 en-de test | Pearson's: 0.6837 -> 0.6763 | Hurt |
| Add mis-matching features (ablation) | Ours Single sentence QE · WMT 2017 de-en test | Pearson's: 0.7063 -> 0.7099 | Improved |
| Remove model-derived features (ablation) | Ours Single sentence QE · WMT 2017 en-de test | Pearson's: 0.6837 -> 0.6408 | Hurt |
| Add model-derived features (ablation) | Ours Single sentence QE · WMT 2017 de-en test | Pearson's: 0.6726 -> 0.7099 | Improved |
| Use the proposed ensemble QE system | Sentence-level QE · WMT 2018 en-de and de-en | Pearson's: 0.7000 -> 0.7308; 0.7667 -> 0.7631 | Mixed |
| Use the proposed system instead of the strongest listed baseline | Word-level QE · WMT 2017 en-de and de-en | F1-Multi: 0.568 -> 0.5826; 0.535 -> 0.5507 | Improved |
| Use BPE rather than word tokenization (ablation) | Quality estimation · Most evaluated sentence- and word-level tasks | QE performance | Improved |
| Replace Bi-LSTM with a self-attention module (ablation) | Quality estimation predictor · QE experiments | QE performance | Hurt |
| Add a CRF layer to the Bi-LSTM (ablation) | Word-level QE predictor · QE experiments | QE performance | No clear effect |
| Use WMT parallel data instead of CWMT-only parallel data (ablation) | Single-model bilingual expert for CWMT QE · CWMT 2018 Ch->En and En->Ch | Pearson's: 0.564 -> 0.612; 0.588 -> 0.620 | Improved |
Closest papers by meaning
- Calibrating Translation Decoding with Quality Estimation on LLMs (NeurIPS 2025 · similarity 0.88)
- MT-Ranker: Reference-free machine translation evaluation by inter-system ranking (ICLR 2024 · similarity 0.87)
- QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translation (NeurIPS 2024 · similarity 0.86)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.85)
- Learning from others' mistakes: Finetuning machine translation models with span-level error annotations (ICML 2025 · similarity 0.85)
- Are Large Reasoning Models Good Translation Evaluators? Analysis and Performance Boost (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
