Research map / Architectures and efficiency / Attention and transformers
Alignment-Enhanced Transformer for Constraining NMT with Pre-Specified Translations
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 dedicated alignment-supervised attention head for dictionary-guided decoding (ablation) | Transformer + dict. guid. · En-Ro test2016 BLEU | BLEU: 26.37 -> 28.17 | Helped |
| add dedicated alignment-supervised attention head for dictionary-guided decoding (ablation) | Transformer + dict. guid. · En-De test2014 BLEU | BLEU: 28.09 -> 31.19 | Helped |
| add dedicated alignment-supervised attention head for dictionary-guided decoding (ablation) | Transformer + dict. guid. · En-Ru test18 BLEU | BLEU: 34.17 -> 36.99 | Helped |
| alignment-enhanced dict. guid. vs lexical constraints | Align. Enhan. · 12 of 13 test sets | BLEU: avg +4.48 vs +3.51 BLEU | Helped |
| alignment-enhanced dict. guid. vs lexical constraints | Align. Enhan. · Ch-En test2018 | BLEU: 23.45 vs 23.43 | No clear effect |
| vanilla dictionary-guided decoding vs lexical constraints | Transformer + dict. guid. · 12 of 13 test sets | BLEU | Hurt |
| add alignment head (no constraints used) (ablation) | Transformer · translation quality without constraints | BLEU: -0.01 to +0.03 (En-Ro/De/Fr) | No clear effect |
| add alignment-supervised head (ablation) | Transformer · En-Ro alignment | AER (%): 60.7 -> 36.2 | Helped |
| add alignment-supervised head (ablation) | Transformer · En-De alignment | AER (%): 75.7 -> 25.3 | Helped |
| use GIZA instead of FastAlign alignment supervision (ablation) | Align. Enhan. · En-Ro test2016 with dictionary | BLEU: 29.46 -> 30.37 | Helped |
| add alignment-supervised head (ablation) | Transformer + dict. guid. · five language pairs | constraint success rate: 77.00 -> 85.17 (En-Ro) | Helped |
Closest papers by meaning
- Lexically Constrained Neural Machine Translation with Explicit Alignment Guidance (AAAI 2021 · similarity 0.92)
- Syntax-Directed Attention for Neural Machine Translation (AAAI 2018 · similarity 0.88)
- Deep Fusing Pre-trained Models into Neural Machine Translation (AAAI 2022 · similarity 0.87)
- L3Ms — Lagrange Large Language Models (ICLR 2025 · similarity 0.85)
- AEQA-NAT : Adaptive End-to-end Quantization Alignment Training Framework for Non-autoregressive Machine Translation (ICML 2025 · similarity 0.85)
- StableMask: Refining Causal Masking in Decoder-only Transformer (ICML 2024 · similarity 0.85)
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.