Fine-Grained Spatio-Temporal Modeling of Reading Behavior

ACL ARR 2025 February Submission7605 Authors

Abstract: Reading is a process that unfolds across space and time. Standard modeling approaches, however, overlook much of the spatio-temporal dynamics involved in reading by relying on aggregated reading measurements - typically only focusing on fixation durations -and employing modeling techniques that impose strong assumptions. In this paper, we propose a model that captures not only how long fixations last, but also where they land in space and when they take place in time. This is achieved by considering reading as an alternating renewal process, in which the locations and durations of eye fixations are modeled separately yet cohesively. The location (and timing) of fixation shifts, so-called saccades, are modeled using a spatio-temporal Hawkes process, which captures how each fixation excites the probability of a new fixation occurring near it in time and space. Empirically, our Hawkes process model exhibits higher likelihood on held-out reading data than baselines. The duration time of fixation events is modeled as a function of fixation-specific features convolved across time, thus capturing non-stationary delayed effects. We evaluate goodness-of-fit across various time-to-event distributions and find evidence that previous convolution-based approaches (shain-schuler-2018-deconvolutional,shain-schuler-2020-continuous) are insufficiently expressive for modeling disaggregated durations. Finally, testing surprisal theory on disaggregated data, we find that it is weakly predictive of where fixations land but has virtually no predictive power for individual fixation durations.

Paper Type: Long

Research Area: Linguistic theories, Cognitive Modeling and Psycholinguistics

Research Area Keywords: linguistic theories; cognitive modeling; computational psycholinguistics;

Languages Studied: English

Submission Number: 7605