Learning Textual Graph Patterns to Detect Causal Event Relations

This paper presents a novel method for discovering causal relations between events encoded in text. In order to determine if two events from the same sentence are in a causal relation or not, we first build a graph representation of the sentence that encodes lexical, syntactic, and semantic information. In a second step, we automatically extract multiple graph patterns (or subgraphs) from such graph representations and sort them according to their relevance in determining the causality between two events from the same sentence. Finally, in order to decide if these events are causal or not, we train a binary classifier based on what graph patterns can be mapped to the graph representation associated with the two events. Our experimental results show that capturing the feature dependencies of causal event relations using a graph representation significantly outperforms an existing method that uses a flat representation of features.