Revisiting the interaction between word frequency and predictability in natural reading: Large-scale evidence based...

Word predictability-that is, the predictability of a word in a sentence based on the preceding context-reliably influences processing of that word. Prior research has consistently reported additive effects of word predictability and frequency, suggesting that contextual influence may operate independently of bottom-up lexical access. We leveraged an ensemble of large language models (LLMs) to derive predictability estimates across six natural reading eye-tracking datasets in English and Chinese. Across both languages, we found that the frequency effect was attenuated with increasing word probability yet remained significant under the high predictability conditions. These findings support the interactive processing account and Chinese Reading Model, which assume that top-down preactivation reduces the demand for bottom-up lexical access, but does not fully replace it. The findings are inconsistent with the independent relationship implemented in some other computational models of eye-movement control. However, such models could readily be extended to accommodate this interaction.