WaterSearch: A Quality-Aware Search-based Watermarking Framework for Large Language Models

2025-11-30

AI Summary

Existing LLM watermarking methods suffer from a significant trade-off between detection robustness and text quality. This paper proposes WaterSearch, a sentence-level search-based watermarking framework that enables parallel generation via seed pool control and introduces a distribution-aware joint optimization search mechanism. By formulating watermark embedding as a constrained search problem, WaterSearch simultaneously optimizes distribution fidelity and signal detectability, supporting seamless integration of multiple watermarking strategies. Experiments across three mainstream LLMs and ten diverse tasks demonstrate that WaterSearch achieves an average 51.01% performance gain over baselines; improvements reach 47.78% on short texts and 36.47% in low-entropy scenarios. Moreover, it maintains strong robustness against common adversarial attacks—including insertion and synonym substitution—while preserving high watermark detection accuracy (95%).

Technology Category

• Search and Optimization: Learning to Search
• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Generation

Application Category

• Search and Retrieval-Augmented AI: Search Tool Learning with LLM: Teaching LLMs to invoke search and make use of retrieved information
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work

Abstract

Watermarking acts as a critical safeguard in text generated by Large Language Models (LLMs). By embedding identifiable signals into model outputs, watermarking enables reliable attribution and enhances the security of machine-generated content. Existing approaches typically embed signals by manipulating token generation probabilities. Despite their effectiveness, these methods inherently face a trade-off between detectability and text quality: the signal strength and randomness required for robust watermarking tend to degrade the performance of downstream tasks. In this paper, we design a novel embedding scheme that controls seed pools to facilitate diverse parallel generation of watermarked text. Based on that scheme, we propose WaterSearch, a sentence-level, search-based watermarking framework adaptable to a wide range of existing methods. WaterSearch enhances text quality by jointly optimizing two key aspects: 1) distribution fidelity and 2) watermark signal characteristics. Furthermore, WaterSearch is complemented by a sentence-level detection method with strong attack robustness. We evaluate our method on three popular LLMs across ten diverse tasks. Extensive experiments demonstrate that our method achieves an average performance improvement of 51.01% over state-of-the-art baselines at a watermark detectability strength of 95%. In challenging scenarios such as short text generation and low-entropy output generation, our method yields performance gains of 47.78% and 36.47%, respectively. Moreover, under different attack senarios including insertion, synonym substitution and paraphrase attasks, WaterSearch maintains high detectability, further validating its robust anti-attack capabilities. Our code is available at href{https://github.com/Yukang-Lin/WaterSearch}{https://github.com/Yukang-Lin/WaterSearch}.

Problem

• Addresses the trade-off between watermark detectability and text quality in LLMs
• Proposes a search-based framework to enhance watermarking without degrading performance
• Improves robustness against attacks while maintaining high detection accuracy

Innovation

• Controls seed pools for diverse parallel watermarked text generation
• Jointly optimizes distribution fidelity and watermark signal characteristics
• Uses a sentence-level search-based framework adaptable to existing methods