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

TL;DR

This study investigates why larger Transformer language models, despite lower perplexity, better fit to human reading times, actually show less predictive power due to systematic deviations in their surprisal estimates.

Contribution

The paper provides a detailed linguistic analysis revealing how larger models' tendency to memorize causes their surprisal estimates to diverge from human reading behavior.

Findings

Larger models have a positive log-linear relationship between perplexity and fit to reading times.

Residual error analysis shows larger models underpredict reading times for named entities.

Larger models overpredict reading times for function words, indicating divergence from human expectations.

Abstract

This work presents a detailed linguistic analysis into why larger Transformer-based pre-trained language models with more parameters and lower perplexity nonetheless yield surprisal estimates that are less predictive of human reading times. 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, replicating earlier results limited to just GPT-2 (Oh et al., 2022). 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 pre-trained language models to study human language processing.