Validity of a one-stop automatic algorithm for counting clusters and shifts in the Semantic Fluency Task

Abstract

Abstract We introduce a computational algorithm for the Semantic Fluency Task (SFT), which automatically counts clusters and shifts. We compared its output relative to human coders, and its performance in predicting executive functions (EF), intelligence, processing speed, and semantic retrieval, also against human coders. Correlations with EF subdomains and other cognitive factors closely resembled those of human coders, evidencing convergent validity. We also used Naïve Bayes and Decision Tree for age classification, with algorithm outputs successfully discriminating age groups, evidence of discriminant validity. Clusters and shifts were found to be more important than word counts. The algorithm's consistency extended across semantic categories (animals, clothing, foods), suggesting its robustness and generalizability. We believe that our algorithm is applicable beyond the specifics of the SFT, and to many tasks in which people list items from semantic memory (e.g., tasks like free associates, top-of-mind, feature listing). Practical implications of the algorithm’s ease of implementation and relevance for studying the relation of the SFT to EFs and other research problems are discussed.