Prediction in language processing and learning

Abstract

Understanding how a child’s language system develops into an adult-like system is a central question in language development research. An increasingly influential account proposes that the brain constantly generates top-down predictions and matches them against incoming input, with higher-level cognitive models serving to minimize prediction errors at lower levels of the processing hierarchy. However, the role of prediction in facilitating language development remains unclear. This dissertation aims to understand the contribution of prediction to language development by addressing the following research questions: First, we examined developmental differences in the ability to detect auditory prediction errors in auditory inputs by assessing whether adults and children differ in perceiving and detecting temporal statistics in speech using an artificial language learning task (Chapter 2). To overcome limitations of traditional prediction measures, we developed a new metric that captures individuals’ real-time prediction during incremental sentence processing by combining simultaneous EEG and eye-tracking with the visual world paradigm (Chapter 3). Applying these real-time prediction metrics, we investigated the relationship between linguistic prediction and learning performance in adults (Chapter 4) and children (Chapter 5). In Chapter 4, we found that the efficacy of updating existing verb biases in skilled language users is associated with 1) real-time predictive processes during learning, 2) predictive ability in a different context, and 3) prediction errors experienced during learning. In Chapter 5, we examined the relationship between prediction and verb bias learning in school-age children, who have developed verb biases but show greater sensitivity to distributional information than adults. We found that children’s ability to learn novel combinatorial information about known verbs is closely tied to their prediction skills, potentially even more so than adults. Collectively, these studies further knowledge about the role of prediction in supporting real-time language processing and development, providing novel insights into whether and how prediction occurs over the course of language development and during real-time language processing. Ultimately, these findings shed light on prediction-based learning frameworks and suggest directions for future research in language development.