Information properties of morphologically complex words modulate brain activity during word reading

Area of Science:

• Neuroscience
• Computational Linguistics
• Cognitive Science

Background:

• Neuroimaging reveals distinct stages in reading and word recognition.
• Current understanding of these stages is largely descriptive.
• Computational linguistics offers quantitative methods to study brain dynamics.

Purpose of the Study:

• To assess if a statistical morphology model (Morfessor) can quantitatively capture neural dynamics during reading.
• To use information theory's surprisal as a common measure.
• To correlate brain responses with model-derived surprisal values and other psycholinguistic variables.

Main Methods:

• Utilized the Morfessor model for unsupervised discovery of morphemes based on the minimum description length principle.
• Correlated magnetoencephalography (MEG) data with word surprisal values from Morfessor and other psycholinguistic predictors.
• Analyzed spatially, temporally, and functionally distinct cortical activation components during a word recognition task.

Main Results:

• Early occipital and occipito-temporal brain responses correlated with visual complexity and orthographic properties.
• Later bilateral superior temporal activation correlated with whole-word and morphological models.
• Morfessor model's estimated word processing costs significantly related to late-stage reading brain dynamics.

Conclusions:

• Statistical models of morphology, like Morfessor, provide valuable quantitative insights into the neural processes of reading.
• Morphological complexity, as captured by computational models, plays a role in later stages of word recognition.
• This approach bridges computational linguistics and neuroscience for a deeper understanding of reading.