Linguistic Monoculture in LLM-Assisted Language Use

As LLMs become integral to the writing process, they may inadvertently homogenize human expression. This paper investigates the conditions under which LLM assistance leads to linguistic monoculture—a reduction in population-level variation—and whether personalization or strategic choices can mitigate this trend.

The authors develop a mathematical framework representing authors and LLMs as probability distributions over linguistic features. They model the co-evolution of these distributions through three interaction mechanisms: a fixed shared model, a recursively updated shared model, and personalized models that incorporate both individual and population-level feedback. The researchers then treat conformity as a strategic choice, where authors balance the private benefits of clear, fluent, and legible writing against the desire to maintain a distinct personal style.

The study reveals that shared models act as a powerful force for linguistic convergence, effectively pulling authors toward a common norm. While recursive feedback (updating the model based on user output) shifts the location of this norm, it does not inherently preserve diversity. In contrast, personalization allows for a family of distinct equilibria, enabling authors to maintain stylistic differences.

Critically, the authors identify a "price of monoculture": because individuals do not account for the social value of their unique linguistic contributions, they tend to conform more than is socially optimal. This creates a negative externality where the collective loss of diversity is greater than what would be chosen if authors internalized the value of their distinctiveness to the broader community.

This research provides a formal basis for understanding how AI-mediated communication might reshape human culture. It suggests that the design of LLM deployment—specifically the choice between shared and personalized systems—has profound implications for the long-term diversity of human language. The findings caution that without intentional design, the drive for individual clarity and efficiency may lead to a systemic, irreversible loss of stylistic variety in academic and professional discourse.