Research map / Architectures and efficiency / Attention and transformers
A Dynamic Window Neural Network for CCG Supertagging
Huijia Wu, Jiajun Zhang, Chengqing Zong
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 dynamic filter gates to the context window (ablation) | Stacked bi-LSTM · CCGBank section 23 (test) | 1-best supertagging accuracy: 94.09 -> 94.69 | Improved |
| add dynamic filter gates to the context window (ablation) | Stacked bi-LSTM · CCGBank section 00 (dev) | 1-best supertagging accuracy: 94.02 -> 94.4 | Improved |
| add character-level information (ablation) | Stacked bi-LSTM · CCGBank section 23 (test) | 1-best supertagging accuracy: 94.21 -> 94.69 | Improved |
| apply dropout to dynamic filters rather than word embeddings (ablation) | Stacked bi-LSTM · CCGBank section 23 (test) | 1-best supertagging accuracy: 94.33 -> 94.69 | Improved |
| add dropout (ablation) | Stacked bi-LSTM · CCGBank section 23 (test) | 1-best supertagging accuracy: 94.25 -> 94.69 | Improved |
| apply filters elementwise to word embeddings instead of using word-level gated concatenation (ablation) | Stacked bi-LSTM dynamic filters · CCGBank section 00 (dev) | 1-best supertagging accuracy: 94.4 -> 94.05 | Hurt |
| use a two-layer MLP to learn dynamic filters (ablation) | Stacked bi-LSTM dynamic filters · CCGBank section 00 (dev) | 1-best supertagging accuracy: 94.4 -> 94.23 | Hurt |
| replace gated concatenation with weighted averaging (ablation) | Stacked bi-LSTM dynamic filters · CCGBank section 00 (dev) | 1-best supertagging accuracy: 94.4 -> 93.95 | Hurt |
| use orthogonal recurrent-weight initialization instead of other initialization methods | Recurrent neural networks · CCGBank | relative performance gain: 2% | Improved |
| use 400 or 600 LSTM cells instead of 512 (ablation) | Stacked bi-LSTM · CCGBank | 1-best supertagging accuracy | No clear effect |
| use momentum, Adadelta, or Adam instead of SGD (ablation) | Neural supertaggers · CCGBank | model performance | Hurt |
Closest papers by meaning
- Dynamic Bundling with Large Language Models for Zero-Shot Inference on Text-Attributed Graphs (NeurIPS 2025 · similarity 0.81)
- MAPLE: Many-Shot Adaptive Pseudo-Labeling for In-Context Learning (ICML 2025 · similarity 0.80)
- #InsTag: Instruction Tagging for Analyzing Supervised Fine-tuning of Large Language Models (ICLR 2024 · similarity 0.80)
- Exploiting Vocabulary Frequency Imbalance in Language Model Pre-training (NeurIPS 2025 · similarity 0.80)
- Stay on Topic with Classifier-Free Guidance (ICML 2024 · similarity 0.80)
- Adaptive Classifier-Free Guidance via Dynamic Low-Confidence Masking (NeurIPS 2025 · similarity 0.80)
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.
