ChatGPT as a Grammar Tutor for Indonesian as a Foreign Language (BIPA): Evaluating Pedagogical Quality

Abstract

Learners of Bahasa Indonesia untuk Penutur Asing (BIPA), like language learners generally, consult ChatGPT for grammatical explanations outside class hours, yet the pedagogical quality of those explanations remains unexamined. This study evaluates ChatGPT (GPT-4o) explanations of four grammatical concepts that BIPA error-analysis research identifies as persistent sources of difficulty: the multifunctional di-, the meN- prefixation system, the suffixes -kan and -i, and the circumfixes me-kan and me-i. Using qualitative content analysis, outputs were coded against a seven-dimension rubric—accuracy, clarity, completeness, organisation, example quality, readability, and pedagogical usefulness—each operationalised through explicit four-level descriptors. Ratings function as ordinal descriptive summaries of qualitative judgements, not as inferential evidence, and no generalisation beyond the analysed corpus is claimed. Of 28 ratings, 22 fell in the highest band and 6 in the second: accuracy, clarity, organisation, and example quality were uniformly strong, whereas readability was weakest, followed by pedagogical usefulness. The recurring shortcoming was not linguistic error but pedagogical omission: explanations did not anticipate the affixation misconceptions documented among BIPA learners, and treated related affixes as unconnected topics. On the evidence of these four explanations, ChatGPT is better positioned to supplement structured BIPA instruction than to substitute for it.