Delegation boundaries: A multiple-case study of teacher pedagogical content knowledge for generative AI in Chinese-as-a-foreign-language classrooms

Abstract

Generative AI has entered Chinese-language classrooms more as a new participant than a syllabus component, shifting teachers' core task from delivering content to orchestrating a content-generating system. This study asks what knowledge Chinese as a foreign language (CFL) teachers retain rather than delegate when working with generative AI, and whether that knowledge is specific to Chinese as a target language. In an instrumental multiple-case design, we tracked twelve CFL teachers across five institutions in four national contexts for one academic semester, drawing on 24 interviews, 31 observed class sessions, and 148 teaching artefacts. Analysis combined inductive thematic coding with a deductive TPACK-based framework, yielding four cross-case patterns: teachers moved from prompt crafting to task architecture, sequencing activities so generative output appeared only where they kept interpretive decision-making power; they set boundaries for Chinese-distinctive features, refusing to delegate such judgments even when AI output was plausible; assessment demanded the most improvisation, as teachers built verification routines beyond existing institutional guidance; and incompatible infrastructure -workload models, class sizes, absent shared practices-constrained pedagogical possibilities.We propose that AI-specific pedagogical content knowledge for CFL is best defined as knowledge of delegation boundaries, and argue teacher development should center on this concept rather than tool familiarity.