Research data supporting the publication "Error-correction mechanisms in language learning: modeling individuals"

This study investigates the effectiveness of a computational model called Rescorla-Wagner error-correction learning in explaining language learning. Rescorla-Wagner is a model of classical conditioning that captures how humans and animals learn the predictive relationship between cues and outcomes in an environment (think about Pavlov’s famous experiment where a dog learns to associate the sound of a bell with food). While traditionally the model has been applied to study animal behavior and some aspects of causal learning in humans, it has recently been successfully adapted to the language domain, with cues and outcomes ranging from form units to abstract linguistic categories. Here, we used the model to analyze the language learning behavior of individual participants in a controlled natural language learning task, which was inspired by the challenge of learning subject-verb agreement in the plural past tense in Polish. We showed that the model accurately predicted the participants’ choices, response times, and levels of response agreement. We also showed that gender and working memory capacity influenced how well the Rescorla-Wagner model explained language learning. Our findings help to further our understanding of how humans learn language and shed light on the importance of integrating cognitive and personal characteristics when modeling language learning.