Language Models Outperform Cloze Predictability in a Cognitive Model of Reading

Abstract

AbstractAlthough word predictability is commonly considered an important factor in reading, sophisticated accounts of predictability in theories of reading are yet lacking. Computational models of reading traditionally use cloze norming as a proxy of word predictability, but what cloze norms precisely capture remains unclear. This study investigates whether large language models (LLMs) can fill this gap. Contextual predictions are implemented via a novel parallel-graded mechanism, where all predicted words at a given position are pre-activated as a function of contextual certainty, which varies dynamically as text processing unfolds. Through reading simulations with OB1-reader, a cognitive model of word recognition and eye-movement control in reading, we compare the model’s fit to eye-movement data when using predictability values derived from a cloze task against those derived from LLMs (GPT2 and LLaMA). Root Mean Square Error between simulated and human eye movements indicates that LLM predictability provides a better fit than Cloze. This is the first study to use LLMs to augment a cognitive model of reading with higher-order language processing while proposing a mechanism on the interplay between word predictability and eye movements.Author SummaryReading comprehension is a crucial skill that is highly predictive of later success in education. One aspect of efficient reading is our ability to predict what is coming next in the text based on the current context. Although we know predictions take place during reading, the mechanism through which contextual facilitation affects ocolarmotor behaviour in reading is not yet well-understood. Here, we model this mechanism and test different measures of predictability (computational vs. empirical) by simulating eye movements with a cognitive model of reading. Our results suggest that, when implemented with our novel mechanism, a computational measure of predictability provide better fits to eye movements in reading than a traditional empirical measure. With this model, we scrutinize how predictions about upcoming input affects eye movements in reading, and how computational approches to measuring predictability may support theory testing. In the short term, modelling aspects of reading comprehension helps reconnect theory building and experimentation in reading research. In the longer term, more understanding of reading comprehension may help improve reading pedagogies, diagnoses and treatments.