Text Watermarking in the Era of Large Language Models

Format-based Watermarking

• Electronic marking and identification techniques to discourage document copying.
• UniSpaCh: A text-based data hiding method using Unicode space characters.
• Content-preserving text watermarking through unicode homoglyph substitution.
• Embarrassingly simple text watermarks.

Lexical-based Watermarking

• Deeptextmark: Deep learning based text watermarking for detection of large language model generated text.
• The hiding virtues of ambiguity: quantifiably resilient watermarking of natural language text through synonym substitutions.
• Watermarking text generated by black-box language models.
• Robust multi-bit natural language watermarking through invariant features.

Syntactic-based Watermarking

• Natural language watermarking: Design, analysis, and a proof-of-concept implementation.
• Natural language watermarking via morphosyntactic alterations.
• Words are not enough: sentence level natural language watermarking.

Generation-based Watermarking

• Adversarial watermarking transformer: Towards tracing text provenance with data hiding.
• Waterfall: Framework for Robust and Scalable Text Watermarking.
• REMARK-LLM: A Robust and Efficient Watermarking Framework for Generative Large Language Models.
• PostMark: A Robust Blackbox Watermark for Large Language Models. (Future)

Watermarking during Logits Generation

• A watermark for large language models.
• Who Wrote this Code? Watermarking for Code Generation.
• Unbiased Watermark for Large Language Models.
• DiPmark: A Stealthy, Efficient and Resilient Watermark for Large Language Models.
• Advancing Beyond Identification: Multi-bit Watermark for Language Models.
• Provable Robust Watermarking for AI-Generated Text.
• Towards Codable Watermarking for Injecting Multi-Bits Information to LLMs.
• On the Reliability of Watermarks for Large Language Models.
• A Semantic Invariant Robust Watermark for Large Language Models.
• Can watermarks survive translation? on the cross-lingual consistency of text watermark for large language models.
• An Unforgeable Publicly Verifiable Watermark for Large Language Models.
• Three bricks to consolidate watermarks for large language models.
• Publicly Detectable Watermarking for Language Models.
• A Robust Semantics-based Watermark for Large Language Model against Paraphrasing.
• An Entropy-based Text Watermarking Detection Method.
• Watermarking conditional text generation for ai detection: Unveiling challenges and a semantic-aware watermark remedy.
• Codeip: A grammar-guided multi-bit watermark for large language models of code.
• Adaptive Text Watermark for Large Language Models.
• Optimizing watermarks for large language models.
• WatME: Towards Lossless Watermarking Through Lexical Redundancy.
• Stylometric Watermarks for Large Language Models. (Future)
• Topic-Based Watermarks for LLM-Generated Text. (Future)
• Distortion-free Watermarks are not Truly Distortion-free under Watermark Key Collisions. (Future)
• A Nested Watermark for Large Language Models. (Future)
• Balanced Watermark: A Simple High-Imperceptibility Watermark for Large Language Models. (Future)
• Necessary and Sufficient Watermark for Large Language Models. (Future)
• MCGMark: An Encodable and Robust Online Watermark for LLM-Generated Malicious Code. (Future)
• Less is More: Sparse Watermarking in LLMs with Enhanced Text Quality. (Future)
• Watermarking Language Models for Many Adaptive Users. (Future)
• GumbelSoft: Diversified Language Model Watermarking via the GumbelMax-trick (Future)

Watermarking during Sampling

• Undetectable watermarks for language models.
• Watermarking GPT outputs.
• Robust Distortion-free Watermark for Language Models.
• SemStamp: A Semantic Watermark with Paraphrastic Robustness for Text Generation.
• k-SemStamp: A Clustering-Based Semantic Watermark for Detection of Machine-Generated Text.
• A Watermark for Low-entropy and Unbiased Generation in Large Language Models. (Future)
• Bileve: Securing Text Provenance in Large Language Models Against Spoofing with Bi-level Signature. (Future)
• Edit Distance Robust Watermarks for Language Models. (Future)
• A Statistical Framework of Watermarks for Large Language Models: Pivot, Detection Efficiency and Optimal Rules. (Future)
• A Watermark for Black-Box Language Models. (Future)

Watermarking during LLM Training

• CodeMark: Imperceptible Watermarking for Code Datasets against Neural Code Completion Models.
• Coprotector: Protect open-source code against unauthorized training usage with data poisoning.
• On the Learnability of Watermarks for Language Models.
• Hufu: A Modality-Agnositc Watermarking System for Pre-Trained Transformers via Permutation Equivariance.
• Learning to watermark llm-generated text via reinforcement learning.
• Proving membership in LLM pretraining data via data watermarks. (Future)
• Watermarking Makes Language Models Radioactive. (Future)
• Can Watermarking Large Language Models Prevent Copyrighted Text Generation and Hide Training Data. (Future)

• Can watermarks survive translation? on the cross-lingual consistency of text watermark for large language models.
• Is Watermarking LLM-Generated Code Robust?
• Lost in Overlap: Exploring Watermark Collision in LLMs.
• Can AI-Generated Text be Reliably Detected?
• Bypassing LLM Watermarks with Color-Aware Substitutions.
• Large Language Model Watermark Stealing With Mixed Integer Programming.
• Watermark stealing in large language models.
• Watermarks in the Sand: Impossibility of Strong Watermarking for Generative Models.
• On the Learnability of Watermarks for Language Models.
• No Free Lunch in LLM Watermarking: Trade-offs in Watermarking Design Choices. (Future)
• Watermark Smoothing Attacks against Language Models. (Future)
• Discovering Clues of Spoofed LM Watermarks (Future)
• Optimizing Adaptive Attacks against Content Watermarks for Language Models (Future)
• Can Watermarked LLMs be Identified by Users via Crafted Prompts? (Future)
• Black-Box Detection of Language Model Watermarks (Future)

• WaterBench: Towards Holistic Evaluation of Watermarks for Large Language Models.
• WaterJudge: Quality-Detection Trade-off when Watermarking Large Language Models.
• Mark My Words: Analyzing and Evaluating Language Model Watermarks.
• MarkLLM: An Open-Source Toolkit for LLM Watermarking.
• Evaluating Durability: Benchmark Insights into Multimodal Watermarking. (Future)
• New Evaluation Metrics Capture Quality Degradation due to LLM Watermarking. (Future)
• On Evaluating The Performance of Watermarked Machine-Generated Texts Under Adversarial Attacks. (Future)