Research map / Label-efficient and robust learning / Recognition and long-tail
Nearest Neighbor Machine Translation
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 a datastore of WMT'19 training examples | WMT'19 German-English model · newstest2019, German-English | BLEU: 37.59 -> 39.08; 1.5 BLEU | Helped |
| Add retrieval from same-language-pair datastores | 418M multilingual Transformer · newstest2019, 17 language pairs | BLEU: 26.00 -> 27.40; 1.4 BLEU | Helped |
| Retrieve from a same-language-pair datastore | 418M multilingual Transformer · newstest2019, English-German | BLEU: 36.47 -> 39.49 | Helped |
| Retrieve from a same-language-pair datastore | 418M multilingual Transformer · newstest2019, English-Chinese | BLEU: 30.22 -> 33.63 | Helped |
| Use an English-source datastore for cross-lingual retrieval | 418M multilingual Transformer · Ted Talks and newstest2019 cross-lingual pairs | BLEU: 16.98 -> 17.96; 1 BLEU on average | Helped |
| Use a domain-specific datastore | WMT'19 German-English model · Multi-domains, average | BLEU: 33.82 -> 42.63; 9.2 BLEU | Helped |
| Replace domain-specific datastore with WMT'19 out-of-domain data | kNN-MT · Multi-domains, average | BLEU: 42.63 -> 34.89 | Hurt |
| Combine domain-specific data into an all-domains datastore | kNN-MT · Multi-domains, average | BLEU: 42.63 -> 43.04 | Helped |
| Use a combined all-domains datastore | kNN-MT · Multi-domains, IT | BLEU: 37.98 -> 48.63 | Helped |
| Use an all-domains datastore rather than the best reported data-selection method | kNN-MT · Multi-domains, average | BLEU: 41.26 -> 43.04; 1.4 BLEU | Helped |
| Flatten the nearest-neighbor distribution with temperature greater than 1 rather than temperature 1 (ablation) | kNN-MT · Multilingual validation, English-Chinese | BLEU | Helped |
| Retrieve more than 64 neighbors (ablation) | kNN-MT · Multilingual validation | BLEU | Mixed |
Closest papers by meaning
- Simple and Scalable Nearest Neighbor Machine Translation (ICLR 2023 · similarity 0.94)
- Generalization through Memorization: Nearest Neighbor Language Models (ICLR 2020 · similarity 0.90)
- Building Earth Mover's Distance on Bilingual Word Embeddings for Machine Translation (AAAI 2016 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.89)
- Test-Time Training on Nearest Neighbors for Large Language Models (ICLR 2024 · similarity 0.88)
- MT-Ranker: Reference-free machine translation evaluation by inter-system ranking (ICLR 2024 · 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.