Research map / Image generation / Text-to-image diffusion
Automatic Fact-Guided Sentence Modification
Darsh Shah, Tal Schuster, Regina Barzilay
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 |
|---|---|---|---|
| Use the two-step masker and two-encoder pointer generator instead of the strongest reported baseline | Automatic fact-guided sentence modification · Wikipedia fact updates | SARI: 26.5 -> 31.5 | Improved |
| Remove the copy mechanism from the two-encoder generator (ablation) | Automatic fact-guided sentence modification · Wikipedia fact updates | SARI: 31.5 -> 15.1 | Hurt |
| Augment FEVER training with generated supporting evidence | BERT fact-checking classifier · FEVER unbiased symmetric Test | accuracy: 66.1 -> 67.8 | Improved |
| Augment FEVER training with generated supporting evidence | BERT fact-checking classifier · FEVER unbiased symmetric +Turk | accuracy: 77.0 -> 80.0 | Improved |
| Use paraphrase-generated augmentation | BERT fact-checking classifier · FEVER unbiased symmetric Test | accuracy: 66.1 -> 64.6 | Hurt |
| Use masked concatenation instead of the proposed generator for augmentation | BERT fact-checking classifier · FEVER unbiased symmetric Test | accuracy: 66.1 -> 63.7 | Hurt |
| Add syntactic regularization to the masker at λ=.4 (ablation) | Neutrality masker · FEVER dev masker | Acc: 87.8 -> 80.0 | Hurt |
| Add syntactic regularization to the masker at λ=.4 (ablation) | Neutrality masker · FEVER dev masker | F1: 37.6 -> 39.9 | Improved |
| Set the masking penalty λ to .5 with syntactic regularization (ablation) | Neutrality masker · FEVER dev masker | Acc: 80.0 -> 5.1 | Hurt |
| Set λ to 0 and remove syntactic regularization (ablation) | Neutrality masker · FEVER dev masker | Acc: 80.0 -> 91.6 | Mixed |
Closest papers by meaning
- WikiSQE: A Large-Scale Dataset for Sentence Quality Estimation in Wikipedia (AAAI 2024 · similarity 0.87)
- Precise Information Control in Long-Form Text Generation (NeurIPS 2025 · similarity 0.82)
- Neighboring Perturbations of Knowledge Editing on Large Language Models (ICML 2024 · similarity 0.82)
- Synthetic continued pretraining (ICLR 2025 · similarity 0.82)
- How to Synthesize Text Data without Model Collapse? (ICML 2025 · similarity 0.82)
- WikiBigEdit: Understanding the Limits of Lifelong Knowledge Editing in LLMs (ICML 2025 · similarity 0.82)
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.
