Research map / Label-efficient and robust learning / Recognition and long-tail
Denoising Pre-training for Machine Translation Quality Estimation with Curriculum Learning
Xiang Geng, Yu Zhang, Jiahuan Li, Shujian Huang, Hao Yang, Shimin Tao, Yimeng Chen, Ning Xie, Jiajun Chen
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 |
|---|---|---|---|
| Use the CED noise metric with cleaner-to-noisier curriculum instead of DirectQE pre-training | CLQE on XLM-R large · WMT19 EN-DE | Pearson: +2.87 | Improved |
| Use the CED noise metric with cleaner-to-noisier curriculum instead of DirectQE pre-training | CLQE on XLM-R large · WMT19 EN-DE | F1-MULT: +1.94 | Improved |
| Use target-MT QE Prob noise scoring in CLQE instead of DirectQE pre-training | CLQE on XLM-R large · WMT20 EN-ZH | Pearson: +1.52 | Improved |
| Use target-MT QE Prob noise scoring in CLQE instead of DirectQE pre-training | CLQE on XLM-R large · WMT20 EN-ZH | MCC: +1.67 | Improved |
| Use target-MT QE Prob noise scoring in CLQE instead of DirectQE pre-training | CLQE on XLM-R large · WMT21 EN-ZH | Pearson: +1.79 | Improved |
| Use target-MT QE Prob noise scoring in CLQE instead of DirectQE pre-training | CLQE on XLM-R large · WMT21 EN-ZH | MCC: +1.53 | Improved |
| Fine-tune the surrogate MT model on QE training data for QE Prob scoring (ablation) | CLQE on XLM-R large · WMT20/WMT21 EN-ZH | Pearson | No clear effect |
| Present pseudo examples from cleaner to noisier rather than using no curriculum (ablation) | CLQE with CED on XLM-R large · WMT19 EN-DE | Pearson: 57.11 -> 59.82 | Improved |
| Present pseudo examples from cleaner to noisier rather than from noisier to cleaner (ablation) | CLQE with CED on XLM-R large · WMT19 EN-DE | Pearson: 58.08 -> 59.82 | Improved |
| Use a cleaner-to-noisier curriculum rather than filtering to the 75% least noisy examples (ablation) | CLQE with CED on XLM-R large · WMT19 EN-DE | Pearson: 58.52 -> 59.82 | Improved |
| Use only the 50% least noisy pseudo examples rather than all examples without curriculum (ablation) | CLQE with CED on XLM-R large · WMT19 EN-DE | Pearson: 57.11 -> 57.42 | Improved |
| Use CLQE with CED rather than DirectQE with 100K parallel pairs | QE on XLM-R large · WMT19 EN-DE low-resource | Pearson: 53.01 -> 54.58 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- DirectQE — DirectQE: Direct Pretraining for Machine Translation Quality Estimation (AAAI 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Calibrating Translation Decoding with Quality Estimation on LLMs (NeurIPS 2025 · similarity 0.86)
- QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translation (NeurIPS 2024 · similarity 0.86)
- Error Norm Truncation: Robust Training in the Presence of Data Noise for Text Generation Models (ICLR 2024 · similarity 0.85)
- MBR and QE Finetuning: Training-time Distillation of the Best and Most Expensive Decoding Methods (ICLR 2024 · similarity 0.85)
- Overestimation in LLM Evaluation: A Controlled Large-Scale Study on Data Contamination’s Impact on Machine Translation (ICML 2025 · similarity 0.85)
- QuRating: Selecting High-Quality Data for Training Language Models (ICML 2024 · 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.
