Are corpus-derived recurrent clusters psycholinguistically valid?