# Improving the Generation Quality of Watermarked Large Language Models via Word Importance Scoring

arXiv: https://arxiv.org/abs/2311.09668 · HTML: https://mlsanity.com/paper/2311.09668/

## Key facts

- Cited by 6 papers in the map (data through 14 Oct 2026).
- Placed in Watermarking generated content and models, one of 12 directions in LLM safety and security.
- Builds on 17 of its 23 references that are in the map.
- First posted on arXiv 16 Nov 2023.

## Placed in

- LLM safety and security / [Watermarking generated content and models](https://mlsanity.com/safety/watermarking-generated-content-and-models/)

## Builds on (most cited first, 17 of 17)

- [BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding](https://mlsanity.com/paper/1810.04805/) (2018)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (2023)
- [BERTScore: Evaluating Text Generation with BERT](https://mlsanity.com/paper/1904.09675/) (2019)
- [PaLM: Scaling Language Modeling with Pathways](https://mlsanity.com/paper/2204.02311/) (2022)
- [Don't Give Me the Details, Just the Summary! Topic-Aware Convolutional Neural Networks for Extreme Summarization](https://mlsanity.com/paper/1808.08745/) (2018)
- [Abstractive Text Summarization Using Sequence-to-Sequence RNNs and Beyond](https://mlsanity.com/paper/1602.06023/) (2016)
- [DetectGPT: Zero-Shot Machine-Generated Text Detection using Probability Curvature](https://mlsanity.com/paper/2301.11305/) (2023)
- [ELI5: Long Form Question Answering](https://mlsanity.com/paper/1907.09190/) (2019)
- [Release Strategies and the Social Impacts of Language Models](https://mlsanity.com/paper/1908.09203/) (2019)
- [A Survey of Hallucination in Large Foundation Models](https://mlsanity.com/paper/2309.05922/) (2023)
- [Turning Your Weakness Into a Strength: Watermarking Deep Neural Networks by Backdooring](https://mlsanity.com/paper/1802.04633/) (2018)
- [Analysis Methods in Neural Language Processing: A Survey](https://mlsanity.com/paper/1812.08951/) (2018)
- [LLM Lies: Hallucinations are not Bugs, but Features as Adversarial Examples](https://mlsanity.com/paper/2310.01469/) (2023)
- [Evaluating the Factual Consistency of Large Language Models Through News Summarization](https://mlsanity.com/paper/2211.08412/) (2022)
- [Robust Multi-bit Natural Language Watermarking through Invariant Features](https://mlsanity.com/paper/2305.01904/) (2023)
- [Watermarking Conditional Text Generation for AI Detection: Unveiling Challenges and a Semantic-Aware Watermark Remedy](https://mlsanity.com/paper/2307.13808/) (2023)
- [Evaluating the Faithfulness of Importance Measures in NLP by Recursively Masking Allegedly Important Tokens and Retraining](https://mlsanity.com/paper/2110.08412/) (2021)

## Built on by (most cited first, 6 of 6)

- [Optimizing watermarks for large language models](https://mlsanity.com/paper/2312.17295/) (2023)
- [WaterJudge: Quality-Detection Trade-off when Watermarking Large Language Models](https://mlsanity.com/paper/2403.19548/) (2024)
- [B4: A Black-Box Scrubbing Attack on LLM Watermarks](https://mlsanity.com/paper/2411.01222/) (2024)
- [WaterPool: A Watermark Mitigating Trade-offs among Imperceptibility, Efficacy and Robustness](https://mlsanity.com/paper/2405.13517/) (2024)
- [Learning to Watermark: A Selective Watermarking Framework for Large Language Models via Multi-Objective Optimization](https://mlsanity.com/paper/2510.15976/) (2025)
- [CEFW: A Comprehensive Evaluation Framework for Watermark in Large Language Models](https://mlsanity.com/paper/2503.20802/) (2025)

## Abstract (arXiv)

The strong general capabilities of Large Language Models (LLMs) bring potential ethical risks if they are unrestrictedly accessible to malicious users. Token-level watermarking inserts watermarks in the generated texts by altering the token probability distributions with a private random number generator seeded by its prefix tokens. However, this watermarking algorithm alters the logits during generation, which can lead to a downgraded text quality if it chooses to promote tokens that are less relevant given the input. In this work, we propose to improve the quality of texts generated by a watermarked language model by Watermarking with Importance Scoring (WIS). At each generation step, we estimate the importance of the token to generate, and prevent it from being impacted by watermarking if it is important for the semantic correctness of the output. We further propose three methods to predict importance scoring, including a perturbation-based method and two model-based methods. Empirical experiments show that our method can generate texts with better quality with comparable level of detection rate.