Proceedings of the Annual Meeting of the Cognitive Science Society

This study explores a GPT-2 architecture's capacity to capture monolingual children's comprehension behaviour in Korean, a language underexplored in this context. We examine its performance in processing a suffixal passive construction involving verbal morphology and the interpretive procedures driven by that morphology. Through model fine-tuning via patching and hyperparameter variations, we assess their classification accuracy on test items used in Shin (2022a). Results show discrepancies in simulating children's response patterns, highlighting the limitations of neural networks in capturing child language features. This prompts further investigation into computational models' capacity to elucidate developmental trajectories of child language that have been unveiled through corpus-based or experimental research.