A Hierarchical Bayesian Model for Unsupervised Induction of Script Knowledge Abstract: We present a hierarchical Bayesian model for unsupervised induction of knowledge from scripts. Scripts are abstract representations of common everyday scenarios which consist of a stereotypical sequence of events. In recent years there has been various work on inducing script knowledge automatically from text. We present a new approach to this task by tackling the three tasks of learning equivalence classes of events, equivalence classes of participants, and constraints on orderings of events in one unified framework. We develop a joint generative framework, arguing that the three tasks provide strong cues for each other. We incorporate a statistical model for permutations, the Generalized Mallows Model (GMM), for modeling ordering constraints. We further include prior knowledge of semantic similarity obtained from WordNet to guide the inference process and leverage the problem of relatively small data sets. We present an evaluation for all three tasks, comparing our results to a system which learns the same three objectives using a pipeline-based architecture. We show that the GMM is a robust model of event orderings in scripts. While we do not achieve state-of-the-art performance on participant class learning, we achieve results comparable to the related work on the task of event class learning.