![[2406.10281] Watermarking Language Models with Error Correcting Codes](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2406.10281] Watermarking Language Models with Error Correcting Codes
Summary
The paper presents a novel watermarking framework for language models using error correcting codes, ensuring robust detection of machine-generated text without compromising quality.
Why It Matters
As AI-generated content becomes more prevalent, distinguishing between human and machine-generated text is crucial for authenticity and trust. This research offers a reliable method to watermark language models, enhancing content integrity and addressing concerns about misinformation.
Key Takeaways
- Introduces a robust binary code (RBC) watermarking method for language models.
- Watermarking is designed to be undetectable to humans while maintaining text quality.
- Demonstrates resilience against edits, deletions, and translations.
- Provides theoretical guarantees and statistical tests for watermark detection.
- Compares favorably to existing state-of-the-art watermarking techniques.
Computer Science > Cryptography and Security arXiv:2406.10281 (cs) [Submitted on 12 Jun 2024 (v1), last revised 23 Feb 2026 (this version, v5)] Title:Watermarking Language Models with Error Correcting Codes Authors:Patrick Chao, Yan Sun, Edgar Dobriban, Hamed Hassani View a PDF of the paper titled Watermarking Language Models with Error Correcting Codes, by Patrick Chao and 3 other authors View PDF HTML (experimental) Abstract:Recent progress in large language models enables the creation of realistic machine-generated content. Watermarking is a promising approach to distinguish machine-generated text from human text, embedding statistical signals in the output that are ideally undetectable to humans. We propose a watermarking framework that encodes such signals through an error correcting code. Our method, termed robust binary code (RBC) watermark, introduces no noticeable degradation in quality. We evaluate our watermark on base and instruction fine-tuned models and find that our watermark is robust to edits, deletions, and translations. We provide an information-theoretic perspective on watermarking, a powerful statistical test for detection and for generating $p$-values, and theoretical guarantees. Our empirical findings suggest our watermark is fast, powerful, and robust, comparing favorably to the state-of-the-art. Subjects: Cryptography and Security (cs.CR); Computation and Language (cs.CL); Machine Learning (cs.LG) Cite as: arXiv:2406.10281 [cs.CR] (or arXiv:2406....