Computational tools for quantifying schemas in autobiographical narratives

According to schema theory, human knowledge is functionally organized in network-like structures called "schemas". Schemas are known to influence memory encoding and recall, and have been proposed to similarly structure our imagination of the future. An automated, general-purpose, and scalable means of quantifying reliance on schemas in autobiographical narratives would therefore be an invaluable tool in research on memory and imagination. Here, we introduce several novel computational measures of narrative schematicity and present a systematic comparison of these with previously published measures. We compare these measures to human judgments of schematicity (180 human raters, 300 narratives) and examine their ability to differentiate between schematic versus idiosyncratic narratives produced by experimental manipulation (over 7000 narrratives across two datasets). We show that the most promising measure is based on identifying the largest network of related concepts that can be extracted from a narrative. The measures of schematicity considered here are available at pypi.org/project/narsche/.