Research map / Architectures and efficiency / Attention and transformers
Lexically Constrained Neural Machine Translation with Explicit Alignment Guidance
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 |
|---|---|---|---|
| Use EAM-OUTPUT with an explicit alignment module trained on ATT-INPUT alignment labels | Transformer constrained decoding · WMT news translation testset, average over directions | BLEU: 36.1 -> 37.6 | Helped |
| Use EAM-OUTPUT instead of ATT-INPUT | Transformer constrained decoding · WMT news translation testset, average over directions | BLEU: 37.0 -> 37.6 | Helped |
| Use EAM-OUTPUT instead of ATT-INPUT | Transformer constrained decoding · WMT news translation testset, De-En decoding | decoding time (seconds): 1442 -> 968 | Helped |
| Use ATT-INPUT alignment extraction rather than ATT-OUTPUT | Transformer alignment extraction · Alignment testset, De-En | AER: 29.1 -> 20.2 | Helped |
| Use ATT-INPUT alignment extraction rather than ATT-OUTPUT | Transformer alignment extraction · Alignment testset, Ro-En | AER: 43.8 -> 35.2 | Helped |
| Use ATT-INPUT instead of ATT-OUTPUT | Transformer constrained decoding · WMT news translation testset, average over directions | BLEU: 34.1 -> 37.0 | Helped |
| Use ATT-INPUT instead of ATT-OUTPUT | Transformer constrained decoding · WMT news translation testset, De-En decoding | decoding time (seconds): 997 -> 1442 | Hurt |
| Use EAM-OUTPUT rather than Add-AlignHead or Code-Switch | Constrained NMT · Alignment testset | copying success rate (CSR) | Helped |
| Use one-to-many constraints with EAM-OUTPUT | De-En WSD testset · De-En WSD testset | BLEU: 30.5 -> 30.9 | Helped |
| Extract ATT-OUTPUT alignments from decoder layer 5 rather than layer 1 (ablation) | Transformer alignment extraction · Development set, De-En | AER: 75.7 -> 24.9 | Helped |
| Extract ATT-INPUT alignments from decoder layer 3 rather than layer 1 (ablation) | Transformer alignment extraction · Development set, De-En | AER: 17.0 -> 16.3 | Helped |
| Extract ATT-INPUT alignments from decoder layer 3 rather than layer 1 (ablation) | Transformer alignment extraction · Development set, Ro-En | AER: 90.5 -> 14.3 | Helped |
Closest papers by meaning
- Alignment-Enhanced Transformer for Constraining NMT with Pre-Specified Translations (AAAI 2020 · similarity 0.92)
- Regularizing Neural Machine Translation by Target-Bidirectional Agreement (AAAI 2019 · similarity 0.89)
- Deep Fusing Pre-trained Models into Neural Machine Translation (AAAI 2022 · similarity 0.88)
- Towards Making the Most of BERT in Neural Machine Translation (AAAI 2020 · similarity 0.87)
- Syntax-Directed Attention for Neural Machine Translation (AAAI 2018 · similarity 0.87)
- AEQA-NAT : Adaptive End-to-end Quantization Alignment Training Framework for Non-autoregressive Machine Translation (ICML 2025 · similarity 0.86)
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.