Search Abstracts | Symposia | Slide Sessions | Poster Sessions
Goal-oriented modulation of reading - the neural mechanisms
Poster Session D, Thursday, October 1, 4:30 - 6:30 pm, Wangari Maathai
Pepita Alex1, Marta Brzeska1, Catherine Mulder1, Julia Schwarz2, Benjamin Tatler1, Agnieszka Konopka1, Anastasia Klimovich-Gray1; 1University of Aberdeen, 2Basque Center on Cognition, Brain and Language
Introduction: Reading behaviour is highly sensitive to contextual demands, with readers flexibly adapting their strategies according to overarching goals and situational constraints (Lorch et al., 1993). Previous research has shown goals modulate both visual sampling (eye-movements) and semantic engagement with text. For example, when reading to study or answer questions, readers strategically allocate more attention to goal-relevant sections, read more slowly, and regress more frequently to extract specific information based on task demands (Kaakinen & Hyönä, 2007, 2010). However, while the behavioural outcomes of goal-directed reading are well established, the neurocognitive mechanisms enabling this flexibility remain poorly understood. One possible mechanism is the dynamic re-weighting of goal-relevant versus goal-irrelevant linguistic features, consistent with precision weighting accounts of predictive processing (e.g., Friston, 2010). To test this directly, we combined eye-tracking and 64-channel EEG during naturalistic reading to examine how short-term reading goals selectively modulate the neural tracking of linguistic features. We hypothesised that goals would selectively enhance goal-relevant processing, such that proofreading goals would increase tracking of word-form information, whereas semantic/comprehension goals would increase tracking of contextual information, demonstrating functional plasticity of linguistic analysis under cognitive demands. Method: 40 native English speakers (mean Age=19.7±2.0; range:18-26) read continuous passages in three within-subject goal conditions: Baseline (no goal), Orthographic (proofreading goal), and Semantic (attend-to-content goal). Fixation-aligned word-level EEG activity was modelled using multivariate temporal response function (mTRF; Crosse et al., 2016) to estimate the strength of neural tracking of different features. We modelled orthographic neighbourhood size (ONS: number of words differing from the word by one letter), Zipf (log word frequency: context-independent base probability of encountering the word), and contextual Surprisal (how unexpected a word is given its preceding context) to capture orthographic, lexical, and semantic/contextual-level processing, respectively. Analyses focused on fronto-temporal and occipito-parietal regions of interest (ROI). Results: Mixed-effects modelling showed significant effects of Condition and ROI for ONS, Zipf, and Surprisal tracking. Across conditions, neural tracking of all features was consistently stronger in occipito-parietal than fronto-temporal regions. Importantly, goal condition selectively modulated feature tracking. Post-hocs revealed that ONS word-form tracking was enhanced selectively in the Orthographic condition compared to both Baseline and Semantic conditions. Zipf lexical tracking was enhanced selectively in Semantic condition contrasted to both Orthographic and Baseline conditions. Finally, contextual Surprisal tracking was inhibited in the Orthographic condition, but did not differ between Semantic and Baseline conditions. Discussion: We were the first to use a decoding approach to quantify online linguistic feature tracking during goal-directed naturalistic reading. Our findings suggest that naturalistic reading is adaptively goal-directed: short-term task demands flexibly enhance and/or down-weight specific feature-level processing. Word-form (ONS) and lexical (Zipf) feature analysis was selectively up-regulated when these features were relevant for the given goal (proofreading and content, respectively). However, neural tracking of contextual semantic feature (Surprisal) was selectively inhibited when reading goal did not require context processing (proofreading). These findings demonstrate that top-down goals can dynamically and differentially re-weight neural processing of features across the linguistic hierarchy, supporting a flexible predictive processing account of goal-directed comprehension during reading.
Topic Areas: Reading, Computational Approaches