Reshaping The Narrative: Linguistic Evolution And Reception In Online Storytelling Communities

This study investigates the longitudinal evolution of linguistic patterns within an online storytelling community, utilizing a dataset spanning the public release of ChatGPT to empirically test the theoretical tension between the enhancement of linguistic fluency and the risk of semantic homogenization. Using an Interrupted Time Series design, we identify a distinct discontinuity. The post-intervention period is characterized by an immediate increase in text predictability and reduced perplexity variance. However, long-term dynamics diverge; while semantic content exhibits continuous convergence, burstiness and length display long-term trends that oppose the direction of the immediate shifts. Complementing these temporal dynamics, a Negative Binomial Regression analysis reveals a concave non-linear association between text predictability and post scores. Specifically, we observe diminishing marginal returns, indicating that the positive correlation between linguistic fluency and scores saturates as texts become highly predictable. These findings suggest directions for future work on platform design strategies that balance linguistic fluency with the preservation of diverse narrative voices in digital communities.