Research map / Architectures and efficiency / Attention and transformers
Attending to Entities for Better Text Understanding
Pengxiang Cheng, Katrin Erk
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 CorefAll auxiliary supervision (ablation) | BiDAF-SA-Early on LAMBADA · LAMBADA test | accuracy (%): 60.54 -> 61.51 | Improved |
| Add ELMo embeddings without CorefAll supervision (ablation) | BiDAF-SA-Early on LAMBADA · LAMBADA test | accuracy (%): 60.54 -> 61.38 | Improved |
| Add CorefAll supervision to the ELMo-augmented model (ablation) | BiDAF-SA-Early + ELMo on LAMBADA · LAMBADA test | accuracy (%): 61.38 -> 63.71 | Improved |
| Replace Stanford CoreNLP coreference signals with neural coreference signals (ablation) | BiDAF-SA-Early + CorefAll on LAMBADA · LAMBADA test | accuracy (%): 61.51 -> 61.13 | Hurt |
| Add CorefAll supervision (ablation) | BiDAF-SA-Early + ELMo on LAMBADA · LAMBADA dev, instances requiring coreference | accuracy: 48.5 ± 4.1 -> 62.0 ± 3.5 | Improved |
| Add CorefAll supervision (ablation) | BiDAF-SA-Early + ELMo on LAMBADA · LAMBADA dev, pronoun targets | accuracy: 72.31 ± 0.15 -> 76.72 ± 0.53 | Improved |
| Add CorefAll supervision (ablation) | BiDAF-SA-Early on LAMBADA · LAMBADA test | accuracy (%): 59.48 -> 61.19 | Improved |
| Place the self-attention encoder early rather than late (ablation) | BiDAF on LAMBADA · LAMBADA test, without auxiliary supervision | accuracy (%): 59.48 -> 60.54 | Improved |
| Apply CorefAll supervision at the early rather than late encoder (ablation) | BiDAF-SA-Both on LAMBADA · LAMBADA test | accuracy (%): 61.54 -> 61.72 | Improved |
| Use dependency-parse supervision instead of CorefAll (ablation) | BiDAF-SA-Early on LAMBADA · LAMBADA test | accuracy (%): 61.51 -> 61.06 | Hurt |
| Use CorefPrev instead of CorefAll supervision (ablation) | BiDAF-SA-Early on LAMBADA · LAMBADA test | accuracy (%): 61.51 -> 60.94 | Hurt |
| Use CorefNext instead of CorefAll supervision (ablation) | BiDAF-SA-Early on LAMBADA · LAMBADA test | accuracy (%): 61.51 -> 61.27 | Hurt |
Closest papers by meaning
- Is attention required for ICL? Exploring the Relationship Between Model Architecture and In-Context Learning Ability (ICLR 2024 · similarity 0.87)
- SelfIE: Self-Interpretation of Large Language Model Embeddings (ICML 2024 · similarity 0.86)
- Self-Updatable Large Language Models by Integrating Context into Model Parameters (ICLR 2025 · similarity 0.86)
- Scaling Stick-Breaking Attention: An Efficient Implementation and In-depth Study (ICLR 2025 · similarity 0.86)
- From Self-Attention to Markov Models: Unveiling the Dynamics of Generative Transformers (ICML 2024 · similarity 0.85)
- Self-Play Fine-Tuning Converts Weak Language Models to Strong 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.
