DE-COP: Detecting Copyrighted Content in Language Models Training Data
This addresses the issue of copyright infringement detection for model developers and legal stakeholders, offering a novel method for a known bottleneck.
The paper tackles the problem of detecting whether copyrighted content was used in training language models, proposing DE-COP, which uses multiple-choice questions with verbatim text and paraphrases to probe models; it improves detection performance by 9.6% in AUC on models with logits available and achieves 72% accuracy on black-box models.
How can we detect if copyrighted content was used in the training process of a language model, considering that the training data is typically undisclosed? We are motivated by the premise that a language model is likely to identify verbatim excerpts from its training text. We propose DE-COP, a method to determine whether a piece of copyrighted content was included in training. DE-COP's core approach is to probe an LLM with multiple-choice questions, whose options include both verbatim text and their paraphrases. We construct BookTection, a benchmark with excerpts from 165 books published prior and subsequent to a model's training cutoff, along with their paraphrases. Our experiments show that DE-COP surpasses the prior best method by 9.6% in detection performance (AUC) on models with logits available. Moreover, DE-COP also achieves an average accuracy of 72% for detecting suspect books on fully black-box models where prior methods give approximately 4% accuracy. The code and datasets are available at https://github.com/LeiLiLab/DE-COP.