Abduction is inference to the best explanation. If a logically implies b then given an observation b, an abductive reasoner will assume a as a possible explanation of b. When abduction is used for discourse processing, the process of interpreting text fragments is viewed as the process of providing the best explanation of why these text fragments would be true given the knowledge base.

Approaches were proposed for estimating the likelihood of alternative abductive explanations: cost- based abduction (Charniak and Shimony, 1990), weighted abduction (Hobbs et al., 1993), abduction based on Bayesian Networks (Pearl, 1988; Charniak and Goldman, 1989; Raghavan and Mooney, 2010), abduction based on Markov Logic Networks (Kate and Mooney, 2009).