Research map / Image generation / Autoregressive and latent generation
Modeling Coherence for Discourse Neural Machine Translation
Hao Xiong, Zhongjun He, Hua Wu, Haifeng Wang
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 |
|---|---|---|---|
| Add second-pass document deliberation | first-pass Transformer · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 20.41 -> 20.92 | Improved |
| Add reward-teacher policy learning | two-pass Transformer · IWSLT 2015 Zh-En TED talks (test average) | BLEU_doc: 26.89 -> 27.50 | Improved |
| Add reward-teacher policy learning | first-pass Transformer · IWSLT 2015 Zh-En TED talks (test average) | BLEU_doc: 25.99 -> 26.82 | Improved |
| Add discourse BLEU as an additional reward | two-pass-rl Transformer · IWSLT 2015 Zh-En TED talks (test average) | BLEU_doc: +2.2 BLEU | Improved |
| Train with talks in original sentence order rather than randomly shuffled sentences | Transformer training · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 20.10 -> 20.41 | Improved |
| Pretrain on WMT2018 bilingual data before TED fine-tuning | two-pass-bleu-rl · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 21.33 -> 27.55 | Improved |
| Replace TED reward-teacher training corpus with AFP (ablation) | two-pass-rl · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 21.11 -> 20.54 | Hurt |
| Replace TED reward-teacher training corpus with XIN (ablation) | two-pass-rl · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 21.11 -> 20.92 | Hurt |
| Replace TED reward-teacher training corpus with Gigaword (ablation) | two-pass-rl · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 21.11 -> 20.49 | Hurt |
| Set balance factors to λ1=0.85 and λ2=0.80 (ablation) | first-pass-rl and two-pass-rl · IWSLT 2015 Zh-En TED talks | BLEU | Improved |
| Give the second pass context from up to 16 surrounding sentences rather than one preceding sentence | two-pass versus context-encoder · IWSLT 2015 Zh-En TED talks (test average) | BLEU: 20.31 -> 20.92 | Improved |
Closest papers by meaning
- BBScore: A Brownian Bridge Based Metric for Assessing Text Coherence (AAAI 2024 · similarity 0.87)
- Quantifying the Plausibility of Context Reliance in Neural Machine Translation (ICLR 2024 · similarity 0.85)
- Exploring the Translation Mechanism of Large Language Models (NeurIPS 2025 · similarity 0.84)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.84)
- Open-Domain Text Evaluation via Contrastive Distribution Methods (ICML 2024 · similarity 0.83)
- Top-H Decoding: Adapting the Creativity and Coherence with Bounded Entropy in Text Generation (NeurIPS 2025 · similarity 0.83)
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.
