The lexico-syntactic patterns for causality acquisition define point-wise causal assertions. If pattern instances are found in texts, the extracted event pairs suggest but not confirm a causal relation. It may happen that these detected relations are accidental or are only valid in the given contexts. That is, they cannot be considered commonly agreed. In this phase, we perform a statistical analysis over event frequencies to assess and prune candidate causal relations.
To score the pairs for causal association, we propose a measure called Causal Strength which gauges how likely these two events are to be in a causal relationship without prior knowledge of any context. Causal Strength is calculated via the following formula