Prediction in language comprehension has become a key mechanism in recent psycholinguistic theory, with evidence from lexical prediction as a primary source. Less work has focused on whether comprehenders also make structural predictions above the lexical level. Previous research shows that processing is facilitated for syntactic structures which are predictable based on context; however, there is so far no direct evidence that speakers formulate structural predictions ahead of encountering input. We investigated whether subject noun animacy cues comprehenders to predict different verb phrase (VP) structures, with the incompatibility between a low animacy subject and an Agent interpretation of transitives/unergative VPs predicting a derived (passive/unaccusative) VP structure, using Italian auxiliaries. Native Italian speakers read sentences with subject nouns varying from high to low animacy followed by the auxiliary avereHAVE, which is compatible with underived VPs, or essereBE, which is compatible with derived VPs. The auxiliary avereHAVE elicited greater frontal negativity when preceded by a subject noun with lower animacy. The auxiliary essereBE elicited no differential ERP given subject animacy. We propose that this frontal negativity reflects violation of a structural prediction, with amplitude reflecting the strength of initial commitment or difficulty in revising a predicted structure. Differences between auxiliaries are proposed to follow from the more specific distribution of avereHAVE. We argue that this evidence unambiguously supports a predictive mechanism for phrase-level structure.