Hidden Semi-Markov Models to Segment Reading Phases from Eye Movements.

Area of Science:

• Cognitive Science
• Computational Linguistics
• Human-Computer Interaction

Background:

• Understanding reading processes is crucial for educational and technological applications.
• Eye-tracking provides valuable insights into cognitive strategies during reading.
• Existing models may not fully capture individual variability in reading behavior.

Purpose of the Study:

• To analyze scanpaths during a reading task to answer a binary topic relevance question.
• To develop a data-driven method for segmenting scanpaths into distinct cognitive phases.
• To investigate individual differences in reading strategies.

Main Methods:

• Utilized hidden semi-Markov chains (HSMCs) for scanpath segmentation.
• Identified cognitive strategies including normal reading, fast reading, information search, and slow confirmation.
• Validated model states using external covariates, including semantic text information.

Main Results:

• The HSMC model successfully segmented scanpaths into interpretable cognitive phases.
• Semantic information from texts served as a significant covariate for phase identification.
• Significant individual preferences for specific reading strategies were observed.
• Substantial inter-individual variability in eye-movement characteristics was quantified.

Conclusions:

• Hidden semi-Markov models offer a robust framework for analyzing reading scanpaths and cognitive strategies.
• Individual heterogeneity in reading behavior is a key factor to consider in reading models.
• This approach can enhance our understanding of complex reading tasks and inform adaptive systems.