The third problem researchers face is in deciding which levels of corpus data are most relevant for modeling the psycholinguistic phenomenon in question – typically referred to as the issue of grain size. For example, in deciding whether her goals is a direct object or the subject of a sentential complement in the athlete realized her goals…, one could consider the probabilities of transitive vs. intransitive sentences, the probabilities of a direct object or sentential complement across all verbs where both are possible, the probabilities associated just with the verb realize, the probabilities associated with realize when the subject is animate, or the probabilities when the subject of realize is athlete.