Research map / Label-efficient and robust learning / Recognition and long-tail
Unsupervised Machine Translation Using Monolingual Corpora Only
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 |
|---|---|---|---|
| Iteratively refine the model with additional training iterations | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 27.48 -> 31.72 -> 32.76 | Helped |
| Iteratively refine the model with additional training iterations | Unsupervised MT · WMT en-fr | BLEU: 12.10 -> 14.42 -> 15.05 | Helped |
| Add cross-domain training loss (ablation) | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 25.44 -> 27.48 | Helped |
| Add pretrained embeddings (ablation) | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 25.29 -> 27.48 | Helped |
| Add both pretrained embeddings and cross-domain loss (ablation) | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 8.78 -> 27.48 | Helped |
| Add input corruption to the reconstruction process (ablation) | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 16.76 -> 27.48 | Helped |
| Add auto-encoding loss (ablation) | Unsupervised MT · Multi30k-Task1 fr-en | BLEU: 20.02 -> 28.07 | Helped |
| Add adversarial loss (ablation) | Unsupervised MT · Multi30k-Task1 fr-en | BLEU: 22.74 -> 28.07 | Helped |
| Use unsupervised MT rather than oracle word reordering | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 11.62 -> 32.76 | Helped |
| Use unsupervised MT rather than supervised NMT | Unsupervised MT · Multi30k-Task1 en-fr | BLEU: 56.83 -> 32.76 | Hurt |
| Use the unsupervised model-selection criterion rather than a small parallel validation set | Model selection · WMT test sets | BLEU: 1 or 2 BLEU points | Hurt |
Closest papers by meaning
- Unsupervised Neural Machine Translation (ICLR 2018 · similarity 0.94)
- Word translation without parallel data (ICLR 2018 · similarity 0.92)
- Contrastive Clustering to Mine Pseudo Parallel Data for Unsupervised Translation (ICLR 2022 · similarity 0.91)
- Improving Simultaneous Machine Translation with Monolingual Data (AAAI 2023 · similarity 0.90)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.88)
- A Benchmark for Learning to Translate a New Language from One Grammar Book (ICLR 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.