We present the psycholinguistically motivated task of predicting human plausibility judgements for verb-role-argument triples and introduce a probabilistic model that solves it. We also evaluate our model on the related role-labelling task, and compare it with a standard role labeller. For both tasks, our model benefits from classbased smoothing, which allows it to make correct argument-specific predictions despite a severe sparse data problem. The standard labeller suffers from sparse data and a strong reliance on syntactic cues, especially in the prediction task. .