WHEN FLUENCY BECOMES FREE: A PRODUCTIVE FRICTION FRAMEWORK FOR LANGUAGE EDUCATION AND LINGUISTIC RESEARCH IN THE AGE OF ARTIFICIAL INTELLIGENCE

Abstract

Over the centuries, the ability to write fluently became a costly indicator of underlying competence: it was assumed that a fluent well-organized text demonstrated a sustained period of exposure, effort, and skill. Large language models (LLMs) have broken that connection as they have made fluent and register appropriate text available in seconds. The central issue this article addresses is that the most significant issue that AI poses to language education is not plagiarism or factual error but the breakdown of fluency as an indicator of linguistic knowledge, cognitive effort, and communicative interest, as summarized in the Fluency Decoupling Thesis (FDT). The article is based on second language acquisition, cognitive psychology, language assessment, sociolinguistics and natural language processing, and builds the Productive Friction Framework (PFF) consisting of four constructs (linguistic offloading, fluency inflation, register flattening, and productive friction), a five-part typology of the design of the friction (in terms of source, target, type, and magnitude), and three pillars (provenance over product, friction by design, and linguistic justice). A three-study, mixed-methods research programme is outlined, with a Fluency Inflation Index, a qualitative taxonomy of offloading actions, and an analysis of the performance of LLM across linguistic varieties in Pakistan specified, as well as five falsifiable hypotheses and a power analysis. There are five classroom protocols provided for immediate use. The article ends with the idea that language education should "tolerate costless fluency" and not police machines, but rather re-think tasks to make human judgment, voice and intent visible and assessable. No empirical data are provided, the article is a conceptual contribution with a research design provided and pre-registered.