A computational language approach to modeling prose recall in schizophrenia

Area of Science:

• Cognitive Neuroscience
• Computational Linguistics
• Psychiatry

Background:

• Verbal memory deficits are a key feature of schizophrenia, but their precise nature is not fully understood.
• Computational language modeling offers a novel approach to quantify verbal memory processes.

Purpose of the Study:

• To apply computational language approaches to analyze semantic and sequential properties of prose recall in schizophrenia patients, unaffected siblings, and healthy controls.
• To assess the ability of these computational models to mimic human scoring and predict group membership.

Main Methods:

• Prose recall was assessed at immediate, 30-minute, and 24-hour intervals.
• N-gram sequence and Latent Semantic Analysis (LSA) were used to model recall performance.
• Models were evaluated for their ability to predict group membership (schizophrenia, sibling, control).

Main Results:

• Automated analyses using n-grams and LSA accurately mimicked human scoring of recall performance.
• Semantic features derived from LSA were the most predictive of group membership individually.
• A combined model of semantic and sequential features slightly improved prediction accuracy over semantic features alone and human ratings.

Conclusions:

• Computational language analysis provides a valid and sensitive method for assessing verbal memory deficits in schizophrenia.
• Semantic features are particularly informative for differentiating diagnostic groups.
• This approach has significant implications for understanding the mechanisms of prose recall in cognitive neuroscience.