Stroke affecting the basal ganglia and thalamus can lead to language deficits. In addition to the lesion’s direct impact on language processing, connectional diaschisis involving cortical-subcortical interactions also plays a critical role. This study investigated connectional diaschisis using the “dynamic meta-networking framework of language” in patients with basal ganglia and thalamus stroke, analyzing longitudinal resting-state fMRI data collected at 2 weeks (n = 32), 3 months (n = 19), and one year post-stroke (n = 23). As expected, we observed dynamic cortico-subcortical interactions between cortical language regions and subcortical regions in healthy controls (HC, n = 25). The cortical language network exhibited dynamic domain-segregation patterns in HCs, severely disrupted in the acute phase following stroke. The connectional diaschisis manifested as dual effects characterized by both hypo- and hyper-connectivity, which positively and negatively correlated with language deficits, respectively. State-specific changes in nodal and topological properties were also identified. Throughout language recovery, cortical language network dynamics gradually normalized toward sub-optimal domain-segregation patterns, accompanied by the normalization of nodal and topological properties. These findings underscore the crucial role of cortico-subcortical interactions in language processing.