Ancestral Visuo-Motor Computations in the Midbrain Underly Readers’ Oculomotor Behavior Across Spaced and Unspaced Languages

Modeling eye movements during reading is challenging because it requires identifying principles that generalize across many different font types and print sizes, as well as the world’s countless languages and writing systems. Existing, top-down, models circumvented this challenge by taking letters as input, essentially agreeing that saccades are programmed towards the center of target words regardless of print properties, and that inter-word spacing, which enables fast text segmentation into word(-object)s, is all that matters. However, these assumptions, specific to spaced Western-alphabetic languages, imply that Eastern (ideographic) scripts that lack inter-word spacing are read using less efficient word segmentation and/or different saccade-targeting strategies. Yet, although differences in eye-movement behavior have been reported for Chinese/Japanese readers, most word-based oculomotor patterns generalize across spaced and unspaced languages. Previous work using a Model of Attention in the Superior Colliculus (MASC) showed that illiterate visuo-motor computations generated prototypical word-based eye-movement behavior over French sentences (Vitu et al., 2021). Here we show that the same ancestral visuo-motor computations generalize to Chinese reading. MASC predicts reading scanpaths by first computing the bottom-up saliency over the retina-transformed sentence image and then averaging the visual-saliency map over retinotopically arranged and size-invariant visual and motor neuronal populations in SC space. Its behavior over 120 sentences from the Beijing Sentence Corpus was compared to that observed in 60 Chinese reading the sentences for comprehension (Pan et al., 2021). MASC nearly perfectly reproduced Chinese readers’ forward eye-movement patterns, including their tendency to fixate preferentially the words’ first character rather than towards the words’ center (as in Westerners). MASC additionally revealed that such differences, rather than being attributable to inter-word spacing, simply result from angular print size being two-to-four times larger in Chinese/Japanese studies. Thus, rudimentary midbrain visuo-motor computations broadly explain readers’ oculomotor behavior across languages and print properties.