Researchers have proposed a new framework for understanding how language evolves in the presence of large language models (LLMs). They argue that LLMs act as distributional mediators, aggregating human-generated language, transforming its distribution through training and post-training, and redistributing model-specific outputs. This process can alter the relative frequencies of competing linguistic variants, potentially leading to changes in how humans select and use language. The framework extends Mufwene's ecological model of language evolution, which posits that language change arises from competition among individual idiolects and speakers' selection from available linguistic material. Emerging evidence suggests that LLMs can influence linguistic profiles and lexical uptake, but human social evaluation remains the decisive factor in determining whether model-associated forms become conventionalized or not. --- Why it matters: This matters to researchers in AI because it provides a new framework for understanding how language evolves in the presence of large language models. It has implications for the development of more effective and socially acceptable LLMs, as well as for our understanding of the long-term impact of these models on human communication. Source: https://arxiv.org/abs/2608.21088