Naturally, a combination of criteria should be taken into account but I am trying to include all relevant literature, be it quantitative, pragmatic, or semantic approaches. The goal is to have objective criteria and avoid the usual clash between prescriptivists who reject everything, and anti-prescriptivists who say that recurrent errors are just idiolects. I know that many idioms and phrases (e.g. I could care less vs I couldn't care less) have been debated but I am trying to work out a generic solution. Thanks a lot for your help.