Research map / Architectures and efficiency / Attention and transformers
SG-Net: Syntax Guided Transformer for Language Representation
Zhuosheng Zhang, Yuwei Wu, Junru Zhou, Sufeng Duan, Hai Yan Zhao, Rui Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 syntax-guided attention with dual aggregation | BERT · SQuAD 2.0 dev | EM: 84.1 -> 85.1 | Improved |
| add syntax-guided attention with dual aggregation | ALBERT · SQuAD 2.0 dev | EM: 87.0 -> 87.4 | Improved |
| add syntax-guided attention | BERT · RACE | accuracy: 72.6 -> 74.2 | Improved |
| add syntax-guided attention | BERT · SNLI test | accuracy: 91.5 -> 91.8 | Improved |
| add syntax-guided attention | Transformer · EN-DE NMT (word-level) | BLEU: 22.06 -> 23.02 | Improved |
| add syntax-guided attention | Transformer · EN-DE NMT (subword) | BLEU: 27.31 -> 27.68 | Improved |
| add dual contextual attention (vanilla plus syntax-guided) (ablation) | BERT · SQuAD 2.0 dev | EM: 84.2 -> 85.1 | Improved |
| add extra vanilla attention layer only (ablation) | BERT · SQuAD 2.0 dev | EM: 84.1 -> 84.2 | No clear effect |
| use syntax-guided attention only (ablation) | BERT · SQuAD 2.0 dev | EM: 84.1 -> 84.4 | No clear effect |
| use weighted dual aggregation instead of concatenation (ablation) | SG-Net · SQuAD 2.0 dev | EM: 84.5 -> 85.1 | Improved |
| use weighted dual aggregation instead of bi-attention (ablation) | SG-Net · SQuAD 2.0 dev | EM: 84.9 -> 85.1 | Improved |
| add answer verifier | SG-Net on BERT · SQuAD 2.0 dev | EM: 85.1 -> 85.6 | Improved |
Closest papers by meaning
- SG-Net: Syntax-Guided Machine Reading Comprehension (AAAI 2020 · similarity 0.95)
- MiniLM: Deep Self-Attention Distillation for Task-Agnostic Compression of Pre-Trained Transformers (NeurIPS 2020 · similarity 0.87)
- Neural Attention Search (NeurIPS 2025 · similarity 0.85)
- AttnLRP: Attention-Aware Layer-Wise Relevance Propagation for Transformers (ICML 2024 · similarity 0.85)
- SAS: Simple Attention Sparsification via End-to-End Optimization of Context Ranking (NeurIPS 2026 · similarity 0.85)
- Transformers Get Stable: An End-to-End Signal Propagation Theory for Language Models (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.
