Incorporating Stylistic Lexical Preferences in Generative Language Models

While recent advances in language modeling have resulted in powerful\ngeneration models, their generation style remains implicitly dependent on the\ntraining data and can not emulate a specific target style. Leveraging the\ngenerative capabilities of a transformer-based language models, we present an\napproach to induce certain target-author attributes by incorporating continuous\nmulti-dimensional lexical preferences of an author into generative language\nmodels. We introduce rewarding strategies in a reinforcement learning framework\nthat encourages the use of words across multiple categorical dimensions, to\nvarying extents. Our experiments demonstrate that the proposed approach can\ngenerate text that distinctively aligns with a given target author's lexical\nstyle. We conduct quantitative and qualitative comparisons with competitive and\nrelevant baselines to illustrate the benefits of the proposed approach.\n

Paper

Similar papers

© 2026 NYSGPT2525 LLC