Research map / Large language models / Language, safety and interpretability
IACW: Intent-Aware Controllable Watermarking for Scalable Authorial Intent Attribution
Hao Huang, Ruihua Zhou, JiaTang Luo, Yunpeng Li, Yuling Liu
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 entropy gating (ablation) | IACW · IACW-Instruct editing | PPL: 89.1 -> 42.1 | Improved |
| add entropy gating (ablation) | IACW · IACW-Instruct | intent accuracy: 99.0% -> 99.5% | No clear effect |
| add ECC (ablation) | IACW · IACW-Instruct | intent accuracy: 90.0% -> 99.5% | Improved |
| add confidence-based erasure decoding (ablation) | IACW · IACW-Instruct | intent accuracy: 85.5% -> 99.5% | Improved |
| use entropy-guided skipping instead of random skipping (ablation) | IACW · IACW-Instruct clean text | TPR: 0.000 -> 1.000 | Improved |
| IACW vs Qu et al. under 20% token deletion | IACW · IACW-Instruct | intent accuracy: 90.5% -> 95.0% | Improved |
| IACW vs Qu et al. under paraphrasing | IACW · IACW-Instruct | intent accuracy: 58.0% -> 69.0% | Improved |
| IACW vs Qu et al. under back-translation | IACW · IACW-Instruct | intent accuracy: 36.5% -> 54.5% | Improved |
| IACW entropy gating vs constant-bias Qu et al. on perplexity | IACW · IACW-Instruct editing | PPL: 90.92 -> 42.10 | Improved |
| IACW vs Qu et al. human preference against no watermark | IACW · IACW-Instruct editing (human study) | preference rate: 38.7% -> 48.2% | Improved |
| IACW vs Qu et al. annotator watermark detection rate | IACW · IACW-Instruct editing (human study) | watermark detected by humans: 67.3% -> 52.2% | Improved |
| increase AI-polished content ratio from 10% to 100% | IACW · Human-AI mixed documents | intent accuracy: 64.0 -> 99.5 | Improved |
Closest papers by meaning
- Break the Trade-off Between Watermark Strength and Speculative Sampling Efficiency for Language Models (ICLR 2026 · similarity 0.87)
- Token-Specific Watermarking with Enhanced Detectability and Semantic Coherence for Large Language Models (ICML 2024 · similarity 0.85)
- Unbiased Watermark for Large Language Models (ICLR 2024 · similarity 0.85)
- TextSeal: A Localized LLM Watermark for Provenance & Distillation Protection (NeurIPS 2026 · similarity 0.85)
- StealthInk: A Multi-bit and Stealthy Watermark for Large Language Models (ICML 2025 · similarity 0.85)
- Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy (AAAI 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.
