Research map / Graph learning / Graph algorithms and clustering
Deep Biaffine Attention for Neural Dependency Parsing
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 deep biaffine classifiers with dimension-reducing MLPs rather than shallow classifiers (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | LAS: 94.00* -> 94.22 | Helped |
| replace the biaffine classifiers with MLP-based attention and classification (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 95.53* | Hurt |
| reduce LSTM size from 400d to 300d (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 95.82 | No clear effect |
| reduce LSTM size from 400d to 200d (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 95.55* | Hurt |
| reduce LSTM depth from 3 to 2 (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 95.62* | Hurt |
| replace LSTM cells with GRU cells (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 93.18* | Hurt |
| replace LSTM cells with Cif-LSTM cells (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 95.67 | Hurt |
| increase shallow-classifier dropout to 50% (ablation) | Shallow biaffine parser · PTB-SD 3.5.0 test | UAS: 95.74 -> 95.73 | No clear effect |
| reduce shallow-classifier recurrent size to 300d (ablation) | Shallow biaffine parser · PTB-SD 3.5.0 test | UAS: 95.74 -> 95.63* | Hurt |
| remove word dropout (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | LAS: 94.22 -> 94.08* | Hurt |
| remove tag dropout (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | UAS: 95.75 -> 95.28* | Hurt |
| remove POS tags entirely (ablation) | Deep Biaffine parser · PTB-SD 3.5.0 test | LAS: 94.22 -> 93.91* | Hurt |
Closest papers by meaning
- Dependency Parsing is More Parameter-Efficient with Normalization (NeurIPS 2025 · similarity 0.92)
- Dynamic Syntactic Feature Filtering and Injecting Networks for Cross-lingual Dependency Parsing (AAAI 2025 · similarity 0.85)
- Scaling Stick-Breaking Attention: An Efficient Implementation and In-depth Study (ICLR 2025 · similarity 0.85)
- Tracing Hyperparameter Dependencies for Model Parsing via Learnable Graph Pooling Network (NeurIPS 2024 · similarity 0.84)
- Punctuation-aware Hybrid Trainable Sparse Attention for Large Language Models (NeurIPS 2026 · similarity 0.84)
- Dependency Structure-Enhanced Graph Attention Networks for Event Detection (AAAI 2024 · similarity 0.84)
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.