Register Accommodation in ChatGPT-Generated Primary Science Explanations: A Conceptual Framework and Research Protocol

Generative artificial intelligence can produce science explanations addressed to children, yet fluent prose and favourable readability scores do not establish that an explanation is linguistically appropriate or scientifically comprehensible. This conceptual and methodological article develops a framework for examining register accommodation in ChatGPT-generated primary science explanations. Drawing on systemic functional linguistics, corpus-based register research, and recent studies of generated educational texts, it distinguishes lexical substitution from coordinated changes in grammatical packaging, causal explicitness, technical vocabulary, and audience positioning. Lexical density is treated as a descriptive property rather than a direct measure of comprehension. A proposed corpus study will compare explanations of thirty science topics across three audience conditions and repeated generations with a matched human-authored corpus. Complementary age-only and register-guided prompts, independent scientific review, manual validation of automated annotations, and multilevel analysis are specified. Constructed examples illustrate how apparently simple language can retain conceptual compression or introduce misleading agency. No corpus has been collected and no empirical findings are reported. The article offers a testable research programme and a basis for teacher evaluation, while reserving claims about learning benefits for subsequent investigations involving children.