Verb-patient co-occurrences in Internet texts and verb meaning in memory

Verb meaning in semantic memory contains both core and peripheral knowledge. This consists of specific representations of the verb's possible thematic roles, in particular possible agents and possible patients. Each is linked to the verb core with a particular degree of typicality. The primary source of thematic knowledge and its typicality lies in the real world. Another source may be discourse. Is the specific information about possible patients for verbs available in discourse? And if so, is it accompanied by the same detailed statistical properties as thematic knowledge in semantic memories? Possible patients for verbs were collected in an experiment and their production frequencies were recorded. Co-occurrences of these verbs with their possible patients were searched in Internet texts using an automatic mechanism. They were found to be present with a linguistic frequency paralleling the production frequency in the experiment. Relevant distributional information concerning verbs and their patients is thus available in texts. It can contribute, through incidental learning and repetition, to the integration of verb/possible patient links and degrees of typicality into verb meanings.