Conclusion
We aim to protect text generation APIs in this work. Previous LW methods compromised text quality and made watermarks easy to detect through error analysis due to not considering polysemy. To fit this, we propose meaning-preserving lexical substitution method that considers the target word’s correct meaning in context x. This enables high-confidence identification while making watermarks more invisible.
References
Chowdhery A, Narang S, Devlin J, Bosma M, Mishra G, et al. PaLM: scaling language modeling with pathways. Journal of Machine Learning Research, 2023, 24(240): 1–113
He X, Xu Q, Lyu L, Wu F, Wang C. Protecting intellectual property of language generation APIs with lexical watermark. In: Proceedings of the 36th AAAI Conference on Artificial Intelligence, 34th Conference on Innovative Applications of Artificial Intelligence, 12th Symposium on Educational Advances in Artificial Intelligence. 2022, 10758–10766
He X, Xu Q, Zeng Y, Lyu L, Wu F, Li J, Jia R. CATER: intellectual property protection on text generation APIs via conditional watermarks. In: Proceedings of the 36th Conference on Neural Information Processing Systems. 2022
Qiang J, Zhu S, Li Y, Zhu Y, Yuan Y, Wu X. Natural language watermarking via paraphraser-based lexical substitution. Artificial Intelligence, 2023, 317: 103859
Qiang J, Liu K, Li Y, Yuan Y, Zhu Y. ParaLS: lexical substitution via pretrained paraphraser. In: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics. 2023, 3731–3746
Yuan W, Neubig G, Liu P. BARTScore: evaluating generated text as text generation. In: Proceedings of the 35th Conference on Neural Information Processing Systems. 2021, 34
Sellam T, Das D, Parikh A. BLEURT: learning robust metrics for text generation. In: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics. 2020, 7881–7892
Miller G A. WordNet: An Electronic Lexical Database. Cambridge: MIT Press, 1998
Venugopal A, Uszkoreit J, Talbot D, Och F J, Ganitkevitch J. Watermarking the outputs of structured prediction with an application in statistical machine translation. In: Proceedings of the Conference on Empirical Methods in Natural Language Processing. 2011, 1363–1372
Acknowledgements
This research was partially supported by the National Natural Science Foundation of China (Grant Nos. 62076217 and U22B2037), and the Blue Project of Yangzhou University.
Author information
Authors and Affiliations
Corresponding author
Ethics declarations
Competing interests The authors declare that they have no competing interests or financial conflicts to disclose.
Additional information
Supporting information The supporting information is available online at https://journal.hep.com.cn and https://link.springer.com.
Rights and permissions
About this article
Cite this article
Zhu, S., Li, Y., Ouyang, X. et al. Safeguarding text generation API’s intellectual property through meaning-preserving lexical watermarks. Front. Comput. Sci. 17, 176352 (2023). https://doi.org/10.1007/s11704-023-3252-0
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s11704-023-3252-0