Neural dynamics of sentiment processing during naturalistic sentence reading

Area of Science:

• Neuroscience
• Cognitive Science
• Computational Linguistics

Background:

• Reading involves eye movements (fixations, saccades) to extract information and meaning, including sentiment.
• Understanding how the brain processes word sentiment during natural reading is challenging due to limitations of traditional controlled experimental setups.
• Previous research often lacked ecological validity, using artificial word presentations instead of naturalistic reading.

Purpose of the Study:

• To investigate the electrical neural correlates of word sentiment processing during naturalistic reading.
• To analyze electroencephalography (EEG) and eye-tracking data to understand brain responses to word sentiment.
• To determine if sentiment processing occurs at the word level or is influenced by phrase/sentence context.

Main Methods:

• Utilized a dataset with simultaneous EEG, eye-tracking, and word-level semantic annotations from natural English sentence reading.
• Computed fixation-related potentials (FRPs) time-locked to eye fixations.
• Employed general linear mixed models and decoding analyses to assess neural responses to word sentiment, controlling for linguistic and eye-movement variables.

Main Results:

• Identified topographical differences in FRPs between positive and negative word sentiment in the 224-304 ms interval after fixation onset.
• Found that FRP differences were associated with word-level sentiment, but not phrase- or sentence-level sentiment predictors.
• Achieved an average decoding accuracy of 0.60 for classifying word sentiment from FRPs, with control analyses ruling out confounds.

Conclusions:

• The emotional valence of words evokes a rapid electrical neural response during naturalistic reading, specifically upon word fixation.
• Word sentiment processing during reading is primarily a word-level phenomenon, reflected in early neural activity.
• These findings advance our understanding of lexico-semantic processing in ecologically valid conditions and inform natural language processing algorithms.