Research map / Architectures and efficiency / RNNs and normalization
Prompting Neural Machine Translation with Translation Memories
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 sentence-level TM prompting | NMT translation · DGT-TM De-En, DGT-TM En-De, H-m En-Zh | BLEU: maximum BLEU score increases of 8.63, 5.74, and 7.74 points | Helped |
| Add sentence-level TM prompting | NiuTrans WMT20 En-Zh · UNPC En-Zh | BLEU: slightly | Mixed |
| Use fragment-level rather than sentence-level TM prompting (ablation) | NMT translation · TM specialized translation datasets | BLEU: about 1 to 2 points | Mixed |
| Add sentence-level TM prompting | Facebook WMT19 De-En · IT, Law, Medical, Koran, Subtitles | BLEU: 2.54, 3.05, 4.64; 0.42 | Mixed |
| Use sentence-level TM prompting instead of no TM | NiuTrans Enterprise · UNPC En-Zh, Fr-Zh, Ru-Zh | BLEU: 2.94, 3.55, 3.06 BLEU points | Helped |
| Use sentence-level TM prompting instead of no TM (ablation) | NiuTrans Enterprise · UNPC En-Zh, Fr-Zh, Ru-Zh | BLEU | Hurt |
| Increase NMT training-data scale (ablation) | Sentence-level TM prompting · UNPC En-Zh | BLEU: maximum BLEU score improvement ... b15M and b20M are higher than ... b5M and b10M | Helped |
| Use a stronger model architecture at the same training-data scale (ablation) | Sentence-level TM prompting · UNPC En-Zh | BLEU | Helped |
| Prompt with a retrieved TM whose FMS is below 0.6 (ablation) | NiuTrans WMT20 En-Zh · UNPC En-Zh, FMS below 0.6 | BLEU | Hurt |
| Prompt with a retrieved TM whose FMS is above 0.8 (ablation) | NiuTrans WMT20 En-Zh · UNPC En-Zh, FMS above 0.8 | BLEU: maximum improvement of 7.09 and 6.11 BLEU points | Helped |
| Concatenate the TM with long input sentences (ablation) | Sentence-level TM prompting · UNPC En-Zh, longer sentence groups | BLEU | Hurt |
| Compare sentence-level TM prompting with kNN-MT | Facebook WMT19 De-En · IT, Koran, Law, Medical, Subtitles | BLEU | Hurt |
Closest papers by meaning
- Graph Based Translation Memory for Neural Machine Translation (AAAI 2019 · similarity 0.90)
- Towards Making the Most of BERT in Neural Machine Translation (AAAI 2020 · similarity 0.89)
- Improved Neural Machine Translation with SMT Features (AAAI 2016 · similarity 0.89)
- Neural Machine Translation Advised by Statistical Machine Translation (AAAI 2017 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.87)
- Adaptive Few-shot Prompting for Machine Translation with Pre-trained Language Models (AAAI 2025 · similarity 0.86)
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.