Previous neuroimaging studies demonstrated that a network of left-hemispheric frontal and temporal brain regions contributes to the integration of contextual information into a sentence. However, it remains unclear how these cortical areas influence and drive each other during contextual integration. The present study used dynamic causal modeling (DCM) to investigate task-related changes in the effective connectivity within this network. We found increased neural activity in left anterior inferior frontal gyrus (aIFG), posterior superior temporal sulcus/middle temporal gyrus (pSTS/MTG) and anterior superior temporal sulcus/MTG (aSTS/MTG) that probably reflected increased integration demands and restructuring attempts during the processing of unexpected or semantically anomalous relative to expected endings. DCM analyses of this network revealed that unexpected endings increased the inhibitory influence of left aSTS/MTG on pSTS/MTG during contextual integration. In contrast, during the processing of semantically anomalous endings, left aIFG increased its inhibitory drive on pSTS/MTG. Probabilistic fiber tracking showed that effective connectivity between these areas is mediated by distinct ventral and dorsal white matter association tracts. Together, these results suggest that increasing integration demands require an inhibition of the left pSTS/MTG, which presumably reflects the inhibition of the dominant expected sentence ending. These results are important for a better understanding of the neural implementation of sentence comprehension on a large-scale network level and might influence future studies of language in post-stroke aphasia after focal lesions.