Research map / Architectures and efficiency / RNNs and normalization
Data Diversification: A Simple Strategy For 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 |
|---|---|---|---|
| Augment parallel training data with multiple forward and backward model translations | Transformer NMT · WMT'14 En-De newstest2014 | BLEU: 29.3 -> 30.7 | Helped |
| Augment training data using bidirectional model translations | Transformer NMT · Low-resource En-Ne, Ne-En, En-Si, Si-En | BLEU: 1.4, 1.3, 2.2 and 1.5 BLEU | Helped |
| Add data diversification to back-translation | Big Transformer with BT · IWSLT'14 En-De and De-En; WMT'14 En-De | BLEU: additional +1.0 BLEU | Helped |
| Use identical initial parameters for constituent models and final model (ablation) | Data Diversification · IWSLT'14 En-De and De-En | BLEU | Hurt |
| Use forward-only or backward-only diversification instead of both directions (ablation) | Data Diversification · Translation tasks | BLEU | Hurt |
| Set dropout to 0.0 (ablation) | Data Diversification · IWSLT'14 En-De | BLEU: 30.1 -> 27.5 | Hurt |
| Set dropout to 0.0 (ablation) | Data Diversification · IWSLT'14 De-En | BLEU: 36.5 -> 32.5 | Hurt |
| Use greedy decoding rather than beam size 5 to generate synthetic data (ablation) | Data Diversification · IWSLT'14 En-De | BLEU: 30.4 -> 30.3 | No clear effect |
| Use greedy decoding rather than beam size 5 to generate synthetic data (ablation) | Data Diversification · IWSLT'14 De-En | BLEU: 36.8 -> 36.6 | No clear effect |
| Train the diversified final model until overfitting, increasing confidence on synthetic predictions (ablation) | Data Diversification · IWSLT'14 En-De | BLEU: 30.6 -> 28.6 | Hurt |
| Train the diversified final model until overfitting, increasing confidence on synthetic predictions (ablation) | Data Diversification · IWSLT'14 En-De, De-En, IWSLT'13 En-Fr, Fr-En | BLEU: 30.6 vs 28.6; 37.0 vs 34.8; 45.5 vs 43.8; 45.0 vs 43.3 | Hurt |
Closest papers by meaning
- Joint Training for Neural Machine Translation Models with Monolingual Data (AAAI 2018 · similarity 0.90)
- Unsupervised Neural Machine Translation (ICLR 2018 · similarity 0.89)
- Meta Back-Translation (ICLR 2021 · similarity 0.88)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (ICLR 2024 · similarity 0.88)
- Transductive Ensemble Learning for Neural Machine Translation (AAAI 2020 · similarity 0.88)
- From Fully Trained to Fully Random Embeddings: Improving Neural Machine Translation with Compact Word Embedding Tables (AAAI 2022 · 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.