Individual corpora predict fast memory retrieval during reading

TL;DR

This study shows that individual reading experiences, captured through personal corpora, better predict early eye movement patterns during reading than generic norm-based corpora, highlighting the importance of personalized language models.

Contribution

It demonstrates that individual-specific corpora improve predictions of reading behavior over traditional norm-based models, emphasizing personalized language representations.

Findings

Individual corpora better predict eye movement durations.

Recently acquired information is accessed rapidly during reading.

Personalized models outperform norm-based models in predicting reading behavior.

Abstract

The corpus, from which a predictive language model is trained, can be considered the experience of a semantic system. We recorded everyday reading of two participants for two months on a tablet, generating individual corpus samples of 300/500K tokens. Then we trained word2vec models from individual corpora and a 70 million-sentence newspaper corpus to obtain individual and norm-based long-term memory structure. To test whether individual corpora can make better predictions for a cognitive task of long-term memory retrieval, we generated stimulus materials consisting of 134 sentences with uncorrelated individual and norm-based word probabilities. For the subsequent eye tracking study 1-2 months later, our regression analyses revealed that individual, but not norm-corpus-based word probabilities can account for first-fixation duration and first-pass gaze duration. Word length additionally affected gaze duration and total viewing duration. The results suggest that corpora representative for an individual's longterm memory structure can better explain reading performance than a norm corpus, and that recently acquired information is lexically accessed rapidly.