Abstract
Event schemas encode knowledge of stereotypical structures of events andtheir connections. As events unfold, schemas are crucial to act as ascaffolding. Previous work on event schema induction either focuses on atomicevents or linear temporal event sequences, ignoring the interplay betweenevents via arguments and argument relations. We introduce the concept ofTemporal Complex Event Schema: a graph-based schema representation thatencompasses events, arguments, temporal connections and argument relations.Additionally, we propose a Temporal Event Graph Model that models the emergenceof event instances following the temporal complex event schema. To build andevaluate such schemas, we release a new schema learning corpus containing 6,399documents accompanied with event graphs, and manually constructed gold schemas.Intrinsic evaluation by schema matching and instance graph perplexity, provethe superior quality of our probabilistic graph schema library compared tolinear representations. Extrinsic evaluation on schema-guided event predictionfurther demonstrates the predictive power of our event graph model,significantly surpassing human schemas and baselines by more than 17.8% onHITS@1.