Why Does Surprisal From Larger Transformer-Based Language Models Provide a Poorer Fit to Human Reading Times?

Abstract

This work presents results using multiple large pretrained language models showing that models with more parameters and lower perplexity nonetheless yield surprisal estimates that are less predictive of human reading times, replicating and expanding upon earlier results limited to just GPT-2 (Oh et al., 2022). First, regression analyses show a strictly monotonic, positive log-linear relationship between perplexity and fit to reading times for the more recently released five GPT-Neo variants and eight OPT variants on two separate datasets, providing strong empirical support for this trend. Subsequently, analysis of residual errors reveals a systematic deviation of the larger variants, such as underpredicting reading times of named entities and making compensatory overpredictions for reading times of function words such as modals and conjunctions. These results suggest that the propensity of larger Transformer-based models to 'memorize' sequences during training makes their surprisal estimates diverge from humanlike expectations, which warrants caution in using pretrained language models to study human language processing.