A tool that uses natural language processing techniques to extract causal relations from text and output useful Bayesian network fragments is described. Previous research indicates that a primarily syntactic approach to causal relation detection can yield good results. We used such an approach to identify subject-verb-object triples and then applied various rules to determine which of the triples were causal relations. Overall, precision and recall were low; however, causal relations with a subject-verb-object structure accounted for a low percentage of the total causal relations in the texts we analyzed. Our research shows that additional methods are needed in order to reliably detect explicit causal relations in text.
