Research map / Architectures and efficiency / RNNs and normalization
Regularizing Neural Machine Translation by Target-Bidirectional Agreement
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 L2R-R2L agreement regularization and joint training | Transformer NMT · NIST Chinese-English datasets average | BLEU: 40.28 -> 43.01 | Helped |
| Add L2R-R2L agreement regularization and joint training | Transformer NMT · WMT17 English-German newstest2017 | BLEU: 25.48 -> 27.18 | Helped |
| Add L2R-R2L agreement regularization and joint training | Transformer NMT · WMT17 Chinese-English newstest2017 | BLEU: 23.01 -> 25.38 | Helped |
| Add agreement regularization to the Transformer-big back-translation system | Transformer-big+BT · WMT17 English-German newstest2017 | BLEU: 28.34 -> 29.46 | Helped |
| Add agreement regularization to the Transformer-big back-translation system | Transformer-big+BT · WMT17 Chinese-English newstest2017 | BLEU: 25.53 -> 27.21 | Helped |
| Use bidirectional agreement regularization during training rather than bidirectional reranking at inference | Transformer NMT · NIST datasets average | BLEU: 41.51 -> 43.01 | Helped |
| Run one joint-training iteration (ablation) | Transformer NMT · WMT17 English-German validation | BLEU: 32.58 -> 33.86 | Helped |
| Run two joint-training iterations (ablation) | Transformer NMT · WMT17 English-German validation | BLEU: 32.58 -> 34.56 | Helped |
| Continue joint training for a third iteration (ablation) | Transformer NMT · WMT17 English-German validation | BLEU: 34.56 -> 34.58 | No clear effect |
| Increase the number of pseudo-sentence samples from m=1 to m=2, 3, or 4 (ablation) | Transformer+RT · Development sets | BLEU | No clear effect |
| Tune regularization weight lambda; lambda=1 performed best among tested values (ablation) | Transformer+RT · Validation sets | BLEU | Helped |
Closest papers by meaning
- Asynchronous Bidirectional Decoding for Neural Machine Translation (AAAI 2018 · similarity 0.90)
- Joint Training for Neural Machine Translation Models with Monolingual Data (AAAI 2018 · similarity 0.90)
- Non-Autoregressive Machine Translation with Auxiliary Regularization (AAAI 2019 · similarity 0.89)
- Neural Machine Translation with Joint Representation (AAAI 2020 · similarity 0.89)
- Deep Fusing Pre-trained Models into Neural Machine Translation (AAAI 2022 · similarity 0.89)
- A Paradigm Shift in Machine Translation: Boosting Translation Performance of Large Language Models (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.