The present work introduces a synthesis of neocerebellar state estimation and feedforward control with multi-level language processing. The approach combines insights from clinical, imaging, and modelling work on the cerebellum with psycholinguistic and historical linguistic research. It finally provides the first experimental attempts towards the empirical validation of this synthesis, employing transcranial magnetic stimulation. A neuroanatomical locus traditionally seen as limited to lower sensorimotor functions, the cerebellum has, over the last decades, emerged as a widely accepted foundation of feedforward control and state estimation. Its cytoarchitectural homogeneity and diverse connectivity with virtually all parts of the central nervous system strongly support the idea of a uniform, domain-general cerebellar computation. Its reciprocal connectivity with language-related cortical areas suggests that this uniform cerebellar computation is also applied in language processing. Insight into the latter, however, remains an elusive desideratum; instead, research on cerebellar language functions is predominantly involved in the frontal cortical-like deficits (e.g. aphasias) seldom induced by cerebellar impairment. At the same time, reflections on cerebellar computations in language processing remain at most speculative, given the lack of discourse between cerebellar neuroscientists and psycholinguists. On the other hand, the fortunate contingency of the recent accommodation of these computations in psycholinguistic models provides the foundations for satisfying the desideratum above. The thesis thus formulates a neurolinguistic model whereby multi-level, predictive, associative linguistic operations are acquired and performed in neocerebello-cortical circuits, and are adaptively combined with cortico-cortical categorical processes. A broad range of psycholinguistic phenomena, involving, among others, "pragmatic normalization", "verbal/semantic illusions", associative priming, and phoneme restoration, are discussed in the light of recent findings on neocerebellar cognitive functions, and provide a rich research agenda for the experimental validation of the proposal. The hypothesis is then taken further, examining grammaticalization changes in the light of neocerebellar linguistic contributions. Despite a) the broad acceptance of routinization and automatization processes as the domain-general core of grammaticalization, b) the growing psycholinguistic research on routinized processing, and c) the evidence on neural circuits involved in automatization processes (crucially involving the cerebellum), interdisciplinary discourse remains strikingly poor. Based on the above, a synthesis is developed, whereby grammaticalization changes are introduced in routinized dialogical interaction as the result of maximized involvement of associative neocerebello-cortical processes. The thesis then turns to the first steps taken towards the verification of the hypothesis at hand. In view of the large methodological limitations of clinical research on cerebellar cognitive functions, the transcranial magnetic stimulation apparatus is employed instead, producing the very first linguistic experiments involving cerebellar stimulation. Despite the considerable technical difficulties met, neocerebellar loci are shown to be selectively involved in formal- and semantic-associative computations, with far-reaching consequences for neurolinguistic models of sentence processing. In particular, stimulation of the neocerebellar vermis is found to selectively enhance formal-associative priming in native speakers of English, and to disrupt, rather selectively, semantic-categorical priming in native speakers of Modern Greek, as well as to disrupt the practice-induced facilitation in processing repeatedly associated letter strings. Finally, stimulation of the right neocerebellar Crus I is found to enhance, quite selectively, semantic-associative priming in native speakers of English, while stimulation of the right neocerebellar vermis is shown to disrupt semantic priming altogether. The results are finally discussed in the light of a future research agenda overcoming the technical limitations met here.