Research map / Probabilistic and scientific modeling / Variational and generative sampling
Dependency Grammar Induction with a Neural Variational Transition-Based Parser
Bowen Li, Jianpeng Cheng, Yang Liu, Frank Keller
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 universal-rule posterior regularization (ablation) | Neural variational transition parser · WSJ-10 | DDA: 33.3 (39.0) -> 54.7 (58.2) | Improved |
| add pretraining with direct posterior-regularization reward (ablation) | Posterior-regularized model · WSJ-10 | DDA: 47.5 (59.8) -> 64.8 (67.1) | Improved |
| add pretraining with direct posterior-regularization reward (ablation) | Posterior-regularized model · WSJ | DDA: 36.7 (46.3) -> 42.0 (43.7) | Improved |
| replace conventional learned baseline with within-sample score normalization (ablation) | Posterior-regularized model with pretraining · WSJ-10 | DDA: 58.7 -> 60.8 | Improved |
| add polarity correction using the number of universal rules (ablation) | RL-SN · WSJ-10 | DDA: 60.8 -> 66.7 | Improved |
| replace polarity-corrected normalized score with normalized rule-count score (ablation) | Posterior-regularized model with pretraining · WSJ-10 | DDA: 66.7 -> 64.4 | Hurt |
| initialize rule expectations from annotated training rules instead of uniformly (ablation) | Posterior-regularized parser · WSJ-10 | DDA: 54.7 -> 66.7 | Improved |
| add lexicalized word-token features (ablation) | WeaklySupervised parser · WSJ-10 | DDA: 66.7 (67.6) -> 68.2 (71.1) | Improved |
| use the proposed transition-based parser instead of RF+H1+H2 | Dependency grammar induction · WSJ-10 | DDA: 51.0 (52.7) -> 54.7 (58.2) | Improved |
| use the proposed transition-based parser instead of RF+H1+H2 | Dependency grammar induction · UD average, length ≤ 40 | DDA: 31.9 (36.2) -> 50.8 (52.5) | Improved |
| use the proposed unlexicalized parser instead of LC-DMV | Dependency grammar induction · French UD, all lengths | tokens per second: 74 -> 1191 | Improved |
Closest papers by meaning
- Global Greedy Dependency Parsing (AAAI 2020 · similarity 0.90)
- Bidirectional Transition-Based Dependency Parsing (AAAI 2019 · similarity 0.89)
- Dependency Parsing is More Parameter-Efficient with Normalization (NeurIPS 2025 · similarity 0.86)
- Syntax vs. Semantics: How Transformers Learn Deep Dependencies (ICML 2026 · similarity 0.86)
- Dynamic Syntactic Feature Filtering and Injecting Networks for Cross-lingual Dependency Parsing (AAAI 2025 · similarity 0.85)
- Implicit Word Reordering with Knowledge Distillation for Cross-Lingual Dependency Parsing (AAAI 2025 · 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.
