Research map / Label-efficient and robust learning / Recognition and long-tail
Generative 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 |
|---|---|---|---|
| Add a language-agnostic latent representation that generates both source and target sentences | GNMT · Missing-word translation across six language directions | BLEU: 26.99 -> 33.23; 27.39 -> 33.46; 23.79 -> 29.84; 23.51 -> 28.27; 22.46 -> 29.83; 25.75 -> 33.09 | Helped |
| Share parameters across languages using source- and target-language indicators | GNMT-Multi · Pure translation with 40K and 400K paired sentences | BLEU | Helped |
| Add monolingual reconstruction training | GNMT-Multi-SSL · Pure translation with limited paired data | BLEU | Helped |
| Add monolingual reconstruction training | GNMT-Multi-SSL · Unseen language pairs | BLEU: 36.72 -> 38.80; 35.05 -> 37.43; 32.81 -> 34.79; 32.62 -> 34.98; 32.94 -> 33.57; 36.77 -> 38.11 | Helped |
| Use GNMT rather than VNMT's conditional latent-variable model | GNMT · Pure translation, 40K and 400K paired sentences | BLEU | Helped |
| Use GNMT rather than VNMT | GNMT · Unseen language pairs | BLEU: 35.58 vs 35.35; 33.59 vs 33.76; 31.34 vs 31.55; 31.95 vs 31.38; 32.31 vs 32.39; 35.86 vs 35.85 | Mixed |
| Anneal the KL-divergence weight from 0 to 1 during the first 50,000 iterations | GNMT and VNMT · Latent-variable model training | BLEU | No clear effect |
| Apply 30% word dropout during training | GNMT · Translation | BLEU | Helped |
| Use the alternative factorization p(z)p(x|z)p(y|z), removing source x from the target-generation condition (ablation) | GNMT · Translation quality and grammatical coherence | Translation quality | Hurt |
Closest papers by meaning
- Mirror-Generative Neural Machine Translation (ICLR 2020 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.89)
- Variational Recurrent Neural Machine Translation (AAAI 2018 · similarity 0.89)
- Controlling Neural Machine Translation Formality with Synthetic Supervision (AAAI 2020 · similarity 0.88)
- Neural Machine Translation with Soft Prototype (NeurIPS 2019 · similarity 0.88)
- A Latent Morphology Model for Open-Vocabulary Neural Machine Translation (ICLR 2020 · similarity 0.88)
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.