Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling

Craig Messner, Tom Lippincott


Abstract
We present an ngram model-based logit scaling technique that effectively transfers extreme subword stylistic variation to large language models at inference time. We demonstrate its efficacy by tracking the perplexity of generated text with respect to the ngram interpolated and original versions of an evaluation model. Minimizing the former measure while the latter approaches the perplexity of a text produced by a target author or character lets us select a sufficient degree of adaptation while retaining fluency.
Anthology ID:
2025.nlp4dh-1.24
Volume:
Proceedings of the 5th International Conference on Natural Language Processing for Digital Humanities
Month:
May
Year:
2025
Address:
Albuquerque, USA
Editors:
Mika Hämäläinen, Emily Öhman, Yuri Bizzoni, So Miyagawa, Khalid Alnajjar
Venues:
NLP4DH | WS
SIG:
Publisher:
Association for Computational Linguistics
Note:
Pages:
272–280
Language:
URL:
https://preview.aclanthology.org/fix-sig-urls/2025.nlp4dh-1.24/
DOI:
Bibkey:
Cite (ACL):
Craig Messner and Tom Lippincott. 2025. Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling. In Proceedings of the 5th International Conference on Natural Language Processing for Digital Humanities, pages 272–280, Albuquerque, USA. Association for Computational Linguistics.
Cite (Informal):
Transferring Extreme Subword Style Using Ngram Model-Based Logit Scaling (Messner & Lippincott, NLP4DH 2025)
Copy Citation:
PDF:
https://preview.aclanthology.org/fix-sig-urls/2025.nlp4dh-1.24.pdf