The pioneering work in (Torisawa, 2003) constructed a statistical model for extracting commonsense inference rules from coordinate verb phrases based on Co-Occurrence feature and Object-Sharing feature. Each rule was selected from the highest combined probability of two verbs and their shared participant. Torisawa further extended his work in (Torisawa, 2006) by emphasizing the occurrence frequency of a single verb, which indicated how generic the meaning of the verb is.
The precision of the improved model achieved 60%.