Semantic Similarity or Predictability? Quantifying Contextual Influences on Word Processing

Previous studies have shown that prior context facilitates the processing of thematically congruent words (Laurino & Kaczer, 2025), but the properties underlying that influence remain unclear. This study examines how the semantic content of a written context shapes the later processing of an auditorily presented word, and which properties best account for that influence. Twenty-three participants read short texts and then made an auditory semantic relatedness judgment on a word pair, congruent or incongruent with the preceding context. Eye movements, EEG and responses were recorded. We propose two quantifiable properties: semantic similarity (meaning overlap between context and target) and language-model predictability (how expected a word is given its context). Semantic similarity is quantified with WordNet and FastText; predictability is estimated with open-source language models. We will test whether these continuous measures account for more variability in neurophysiological (e.g., time-frequency representations) and behavioral (e.g., response times) responses than categorical congruence, characterizing how contextual information is represented and used. Additionally, we will test whether the individual measures of context reading (i.e., total fixation times) influence target-word processing. To do so, we will fit trial-level linear mixed-effects models including total fixation times and compare them with models without this predictor.