Deriving Narrative Morphologies via Analogical Story Merging

It has long been suspected that stories drawn from the same cultural setting share important narrative structure. One example of such structure, first identified by Vladimir Propp in 1928, is the morphology of a set of stories, which describes the set of plot elements and their allowed sequences. Until now, the extraction of morphologies has remained a manual task, the purview of anthropological virtuosos (e.g., Propp, Lévi-Strauss, Dundes, Campbell); reproduction or validation of their analyses is a time-consuming, prohibitively difficult endeavor. I demonstrate a technique called Analogical Story Merging that derives a morphology given a set of stories. It incorporates Forbus & Gentner's Structure Mapping Engine as well as a method called Bayesian Model Merging for inducing a grammar from a corpus of positive examples. I present the output of the basic implementation when applied to a small example story corpus, a set of summaries of Shakespearean plays.