Research map / Architectures and efficiency / Attention and transformers
Deep Semantic Role Labeling With Self-Attention
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 FFN nonlinear sub-layers rather than recurrent or convolutional sub-layers (ablation) | DeepAtt · CoNLL-2005 | F1: 82.3 -> 83.4 | Helped |
| Ensemble 5 FFN models | DeepAtt · CoNLL-2005 Combined | F1: 83.4 -> 84.6 | Helped |
| Ensemble 5 FFN models | DeepAtt · CoNLL-2012 Test | F1: 82.7 -> 83.9 | Helped |
| Reduce depth from 10 to 4 (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 83.1 -> 79.9 | Hurt |
| Increase depth from 10 to 12 (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 83.1 -> 83.0 | Hurt |
| Increase width from 200 to 600 (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 83.1 -> 83.4 | Helped |
| Replace random word embeddings with pretrained GloVe (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 79.6 -> 83.1 | Helped |
| Remove position encoding (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 83.1 -> 20.0 | Hurt |
| Use timing signals instead of position embeddings (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 79.4 -> 83.1 | Helped |
| Add nonlinear sub-layers (ablation) | DeepAtt · CoNLL-2005 development | F1: 79.9 -> 83.1 | Helped |
| Use constrained decoding instead of argmax decoding (ablation) | DeepAtt (FFN) · CoNLL-2005 development | F1: 83.1 -> 83.0 | Hurt |
Closest papers by meaning
- Syntax-Aware Neural Semantic Role Labeling (AAAI 2019 · similarity 0.93)
- Encoder-Decoder Based Unified Semantic Role Labeling with Label-Aware Syntax (AAAI 2021 · similarity 0.90)
- End-to-end Semantic Role Labeling with Neural Transition-based Model (AAAI 2021 · similarity 0.89)
- Frame Semantic Role Labeling Using Arbitrary-Order Conditional Random Fields (AAAI 2024 · similarity 0.88)
- Dependency or Span, End-to-End Uniform Semantic Role Labeling (AAAI 2019 · similarity 0.87)
- RAT: Bridging RNN Efficiency and Attention Accuracy via Chunk-based Sequence Modeling (NeurIPS 2025 · similarity 0.83)
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.