Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
Prompting Large Language Model for Machine Translation: A Case Study
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 |
|---|---|---|---|
| Use the simple English template naming the source and target languages (ablation) | GLM-130B translation prompting · Wiki Ablation sets | COMET: 38.78 | Helped |
| Add a line break to the simple English template (ablation) | GLM-130B zero-shot translation prompting · Wiki Ablation sets | COMET: 38.78 -> 31.17 | Hurt |
| Use a German-language template instead of the English template (ablation) | GLM-130B zero-shot translation prompting · Wiki Ablation sets | COMET: 38.78 -> -26.15 | Hurt |
| Use a Chinese-language template instead of the English template (ablation) | GLM-130B zero-shot translation prompting · Wiki Ablation sets | COMET: 38.78 -> 14.82 | Mixed |
| Increase the number of prompt demonstrations (ablation) | GLM-130B translation prompting · Wiki Ablation sets | translation quality | Helped |
| Use SemScore to select high-quality demonstrations instead of random selection | GLM-130B 1-shot translation · Wiki Full sets | COMET: 48.29 -> 49.34 | Helped |
| Use SemScore to select high-quality demonstrations instead of random selection | GLM-130B 5-shot translation · WMT Full sets | COMET: 33.87 -> 34.30 | Helped |
| Use the combined SemScore, LMScore, and TLength strategy instead of random selection from the low-quality pool | GLM-130B 1-shot translation · Wiki Full sets | COMET: 38.86 -> 39.88 | Helped |
| Use monolingual source-only, target-only, or randomly paired examples instead of parallel demonstrations (ablation) | GLM-130B translation prompting · Wiki Ablation sets | COMET | Hurt |
| Construct pseudo-parallel demonstrations by back-translation of target monolingual data (ablation) | GLM-130B translation prompting · Wiki Ablation sets | COMET | Helped |
| Use demonstrations from another language pair or domain (ablation) | GLM-130B 1-shot translation · Wiki and Multi-Domain transfer settings | COMET: +7.03 COMET; +9.67 COMET; +11.56 COMET; +19.52 COMET | Mixed |
| Transfer sentence-level SemScore-selected demonstrations to document-level translation (ablation) | GLM-130B · PDC Zh→En Full sets | d-BLEU: 30.2 -> 30.5 | Helped |
Closest papers by meaning
- Adaptive Few-shot Prompting for Machine Translation with Pre-trained Language Models (AAAI 2025 · similarity 0.90)
- Quantifying Language Models' Sensitivity to Spurious Features in Prompt Design or: How I learned to start worrying about prompt formatting (ICLR 2024 · similarity 0.88)
- Zero-Shot Continuous Prompt Transfer: Generalizing Task Semantics Across Language Models (ICLR 2024 · similarity 0.88)
- Prompt Sketching for Large Language Models (ICML 2024 · similarity 0.87)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.87)
- Scaling Prompt Synthesis for Large Language Model Reasoning (ICML 2026 · similarity 0.87)
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.