Linguistic Monoculture in LLM-Assisted Language Use
Source: https://arxiv.org/abs/2607.27134
Authors: Suhas Thejasvm, Juhi Kulshreshta, Lutz Oettershagen
Published: 2026-07-29
Categories: cs.AI, cs.CL, cs.GT
PDF: https://arxiv.org/pdf/2607.27134v1
Abstract
Writing and communication are increasingly mediated by large language models (LLMs) used to draft, revise, and polish text. Although such assistance can improve clarity and help authors meet institutional expectations, widespread reliance on shared models may reduce population-level variation in linguistic form — a phenomenon the authors term linguistic monoculture.
The paper develops a mathematical framework in which authors and LLMs are represented as distributions over linguistic features and coevolve through repeated interaction. Three interaction mechanisms are analyzed:
1. Shared model with fixed linguistic distribution — can drive authors toward a common norm 2. Shared model recursively updated from author outputs — relocates the shared norm without altering pairwise spread under common conformity 3. Personalized models — can preserve a family of distinct author-model equilibria with nonzero linguistic diversity
Key Findings
The authors characterize resulting equilibria and convergence rates. Crucially, when conformity is endogenized as a strategic choice (trading off private benefits from clarity, legibility, and perceived fluency against distinctive style):
- Individually rational authors may conform more than is socially optimal
- This creates a negative externality — authors do not internalize the value their distinctiveness provides to others
- The resulting price of monoculture is finite for each fixed instance but can grow without bound when distinctiveness dominates authenticity
Synthetic simulations illustrate how fixed shared assistance, recursive feedback, and personalization produce different long-run diversity outcomes.
中文概要
本文提出“语言单一文化” (linguistic monoculture) 概念：广泛使用共享 LLM 辅助写作可能压缩人群层面的语言多样性。作者建立数学框架，将作者和 LLM 表示为语言特征分布，分析三种交互机制：固定共享模型、递归反馈模型、个性化模型。核心发现：共享模型会驱动作者趋同；个人理性的顺从选择会导致超出社会最优的一致性，产生负外部性，“单一文化代价”在个性化主导时可能无界增长。