Research map / Label-efficient and robust learning / Knowledge distillation
Improving Simultaneous Machine Translation with Monolingual Data
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 randomly sampled monolingual data to distilled parallel data | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: +3.15 | Helped |
| Use LM chunk-length and monotonicity sampling instead of random sampling | SiMT student · En⇒Zh and En⇒Ja | BLEU: +0.84 and +0.60 BLEU points | Helped |
| Select monolingual data by alignment-based chunk length instead of randomly | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: +0.80 | Helped |
| Select monolingual data by LM-based chunk length instead of randomly | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: +0.73 | Helped |
| Select monolingual data by monotonicity instead of randomly | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: -0.17 | Hurt |
| Add monotonicity reranking to LM-based chunk-length sampling (ablation) | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: +0.11 | Helped |
| Add monotonicity reranking to alignment-based chunk-length sampling (ablation) | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: +0.04 | No clear effect |
| Increase monolingual data scale from 1:1 to 1:10 (ablation) | Ours sampling on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: 43.32 -> 43.66 | Helped |
| Replace Transformer Base teacher with Transformer Big (ablation) | Ours SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: 43.32 -> 43.88 (+0.56) | Helped |
| Replace distilled bilingual data with raw bilingual data (ablation) | Ours SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEU: 43.32 -> 43.73 (+0.41) | Helped |
| Use LM chunk-length and monotonicity sampling instead of random sampling | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | BLEURT: +0.0050 | Helped |
| Use LM chunk-length and monotonicity sampling instead of random sampling | SiMT student on CWMT En⇒Zh · CWMT En⇒Zh, average wait-k | chrF2: +0.50 | Helped |
Closest papers by meaning
- Joint Training for Neural Machine Translation Models with Monolingual Data (AAAI 2018 · similarity 0.91)
- SimulPL: Aligning Human Preferences in Simultaneous Machine Translation (ICLR 2025 · similarity 0.89)
- Divergence-Guided Simultaneous Speech Translation (AAAI 2024 · similarity 0.88)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.88)
- QUEST: Quality-Aware Metropolis-Hastings Sampling for Machine Translation (NeurIPS 2024 · similarity 0.88)
- Calibrating Translation Decoding with Quality Estimation on LLMs (NeurIPS 2025 · 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.