Text Watermarking via Segmented Language Model Embedding
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The protection of textual content copyright and the tracing of information generated or modified by language models (LMs) have become increasingly important, as existing methods struggle to effectively embed and detect imperceptible watermarks in text while maintaining the original semantic meaning.
Innovation Solution
A method for watermark processing that divides an original text into segments, determines target symbols from a symbol sequence in watermark information, and converts these segments into watermarked text segments using a set of language models for embedding, ensuring the watermark is imperceptible yet detectable.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If text watermarks are embedded to protect copyright and trace information, then the ability to identify and protect content generated by language models is improved, but the detectability and stealthiness of the watermark must be maintained against attackers who may try to predict or remove watermarks
Solution Approach 1:
The text is divided into multiple segments, and different language models are applied to different segments. This segmentation prevents attackers from easily predicting or removing watermarks across the entire text, as each segment is processed independently with different models, enhancing both detectability and stealthiness
Solution Approach 2:
A set of multiple language models is used instead of a single static model. The system dynamically selects and applies different language models to different text segments, making the watermark embedding process adaptive and harder to predict or attack, while maintaining reliability in watermark detection
2Object-affected harmful factors
If multiple language models are used to convert text segments into watermarked segments, then the security and stealthiness of the watermark are enhanced, but the complexity of the watermark embedding system increases
Solution Approach 1:
The text processing is segmented into multiple independent operations, each handled by a different language model. This allows the system to distribute complexity across multiple simpler model applications rather than requiring one highly complex model, making the overall system more manageable while enhancing security
Solution Approach 2:
Multiple language models serve the dual function of both text generation and watermark embedding. Each model in the set can process text segments independently, providing universal functionality that reduces the need for specialized complex components, thereby enhancing security without proportionally increasing overall system complexity
Data Source
AI summary
Embodiments of the present disclosure provide a solution for watermark processing. A method includes: dividing at least one portion of a n original text for watermark embedding into a plurality of original text segments; determining, for an original text segment of the plurality of original text segments, a target symbol from a symbol sequence in watermark information; converting, based on respective target symbols determined for the plurality of original text segments, the plurality of original text segments into a plurality of watermarked text segments by using a set of language models for watermark embedding, the set of language models corresponding to a set of symbol values respectively; and generating a watermarked text for the original text based on the plurality of watermarked text segments.