Research map / Label-efficient and robust learning / Recognition and long-tail
Analyzing Uncertainty in Neural 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 |
|---|---|---|---|
| Increase beam width from 5 to 200 | NMT decoding · WMT'14 En-Fr 500-sentence test subset | Sentence BLEU, single reference: 41.4 -> 36.2 | Hurt |
| Use sampling instead of beam search | NMT decoding · WMT'14 En-Fr 500-sentence test subset | Distinct human references covered: 5.0 -> 7.4 | Helped |
| Use sampling instead of beam search | NMT decoding · WMT'14 En-Fr 500-sentence test subset | Sentence BLEU, average oracle: 56.4 -> 39.1 | Hurt |
| Increase training copy noise to 1% | WMT'17 En-De news-commentary NMT · WMT'17 En-De validation | BLEU with beam 20: -3.3 BLEU | Hurt |
| Increase training copy noise to 10% | WMT'17 En-De news-commentary NMT · WMT'17 En-De validation | BLEU | Hurt |
| Filter low-scoring training sentence pairs | WMT'17 En-De NMT · WMT'17 En-De newstest2017 | BLEU | Helped |
| Prune completed hypotheses overlapping at least 50% with the source | WMT'17 En-De beam search · WMT'17 En-De newstest2017 | BLEU | Helped |
| Combine training-data filtering and source-copy pruning | WMT'17 En-De NMT · WMT'17 En-De newstest2017 | BLEU: +0.5 BLEU | Helped |
| Increase the number of sampled hypotheses | NMT sampling · WMT'14 En-Fr | Sentence BLEU | Hurt |
| Increase beam width | WMT'14 En-Fr NMT · WMT'14 En-Fr test | BLEU | Hurt |
Closest papers by meaning
- Calibrating Translation Decoding with Quality Estimation on LLMs (NeurIPS 2025 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.88)
- Forking Paths in Neural Text Generation (ICLR 2025 · similarity 0.88)
- QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translation (NeurIPS 2024 · similarity 0.88)
- Language Model Cascades: Token-Level Uncertainty And Beyond (ICLR 2024 · similarity 0.87)
- Decomposing Uncertainty for Large Language Models through Input Clarification Ensembling (ICML 2024 · 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.