The first suggestion of a link between subcortical structures and language was made by Broadbent (1872), who proposed that words were generated as motor acts in the basal ganglia. Despite this suggestion, according to the classical anatomo-functional models of language organization proposed by Wernicke (1874) and Lichtheim (1885), subcortical brain lesions could only produce language deficits if they disrupted the white matter fibers that connect the various cortical language centres. Consequently, aphasia has traditionally been regarded as a language disorder resulting from damage to the language areas of the dominant cerebral cortex. Since the late 1970s, however, this traditional view has been challenged by the findings of an increasing number of cliniconeuroradiological correlation studies that have documented the occurrence of adult language disorders in association with apparently subcortical vascular lesions. In particular, the introduction in recent decades of new neuroradiological methods for lesion localization in vivo, including computed tomography in the 1970s and more recently magnetic resonance imaging, has led to an increasing number of reports in the literature of aphasia following apparently purely subcortical lesions. (For reviews of in vivo correlation studies, see Alexander, 1989; Cappa and Vallar, 1992, and Murdoch, 1996.) Therefore, although the concept of subcortical aphasia remains controversial, recent years have seen a growing acceptance of a role for subcortical structures in language. Despite an abundance of theoretical models, however, the precise nature of that role remains elusive.

Subcortical structures most commonly purported to have a linguistic role include the basal ganglia, the thalamus, and the subcortical white matter pathways. Some evidence for a role for the cerebellum in language has also been reported (Leiner, Leiner, and Dow, 1993). The basal ganglia comprise the corpus striatum (including the caudate nucleus and the putamen and internal capsule), the globus pallidus, the subthalamic nucleus, and the substantia nigra. Although these nuclei are primarily involved in motor functions, the corpus striatum and globus pallidus have frequently been included in models of subcortical participation in language. In addition, several thalamic nuclei have also been implicated in language, in particular the ventral anterior nucleus, which has direct connections to the premotor cortex and indirect connections to the temporoparietal cortex via the pulvinar. The basal ganglia and the thalamus are linked to the cerebral cortex by way of a series of circuits referred to as the cortico-striato-pallido-thalamo-cortical loops. The majority of contemporary theories specify these loops as the neuroanatomical basis of subcortical participation in language.

Although there is general agreement that critical white matter pathways and the thalamus are involved in language, controversy and uncertainty surround the possible linguistic role of striatocapsular structures. Although in vivo correlation studies have documented beyond reasonable doubt that language impairments can occur in association with lesions confined to the striatocapsular region of the dominant hemisphere, considerable variability has been reported in the nature and degree of these language impairments, with no unitary striatocapuslar aphasia being identified (Kennedy and Murdoch, 1993; Nadeau and Crosson, 1997). Varied impairments have been noted in spontaneous speech, confrontation naming, repetition, auditory comprehension, and reading comprehension. A number of authors have suggested that a difference exists between the type of aphasia associated with anterior striatocapsular lesions compared to posterior striatocapsular lesions. For example, Naeser et al. (1982) noted that patients with capsular-putaminal lesions extending into the anterior-superior white matter typically had good comprehension and slow but grammatical speech. In contrast, those with capsular-putaminal lesions including posterior white matter extension showed poor comprehension and fluent Wernicke's-type speech, while those with anterior-superior and posterior white matter involvement were globally aphasic. Further support for this anterior-posterior distinction was provided by Cappa et al. (1983) and Murdoch et al. (1986). Despite this apparent consensus, several other studies have questionned the accuracy and utility of the anterior-posterior dichotomy by describing a number of cases in which the patterns of language impairment could not be accounted for in terms of this anatomical distinction (Kennedy and Murdoch, 1993; Wallesch, 1985).

In contrast to the striatocapsular lesions, language disturbances following thalamic lesions present a more uniform clinical picture, and it is generally accepted that a typical thalamic aphasia can be characterized by the clinical presentation. Most commonly the aphasia resulting from thalamic injury is of a mixed transcortical presentation, sharing some features with both transcortical motor and transcortical sensory aphasia (Cappa and Vignolo, 1979; Murdoch, 1996). The features of thalamic aphasia most commonly reported include preserved repetition, variable but often relatively good auditory comprehension, a reduction in spontaneous speech output, a predominance of semantic paraphasic errors, and anomia. Lesions of the dominant anterolateral thalamus (including the ventral anterior, ventral lateral, and anterior nuclei) have been highlighted as the loci of aphasic deficits, given that infarctions in this region more consistently lead to aphasic disturbances than lesions involving the posterior parts of the thalamus (Cappa et al., 1986).

Attempts to explain the clinical manifestations of subcortical aphasia have culminated in the formulation of several theories of subcortical participation in language. These theories, largely developed on the basis of speech and language data collected from subjects who have sustained cerebrovascular accidents involving the thalamus or striatocapsular region, have been expressed as neuroanatomically based models. Two models of subcortical participation in language have been quite influential. The first of these, the response/release/semantic feedback model (Crosson, 1985), proposes a role for subcortical structures in regulating the release of preformulated language segments from the cerebral cortex. According to this model, the conceptual, word-finding, and syntactic processes that fall under the rubric of language formulation occur in the anterior cerebral cortex. The monitoring of anteriorly formulated language segments, as well as the semantic and phonological decoding of incoming language, occurs in the posterior temporoparietal cortex. Language segments are conveyed from the anterior language formulation center to the posterior language center via the thalamus prior to release for motor programming. This operation allows the posterior semantic decoding centers to monitor the language segment for semantic accuracy. If an inaccuracy is detected, then the information required for correction is conveyed via the thalamus back to the anterior cortex. If the language segment is found to be accurate during monitoring, then it is released from a buffer in the anterior cortex for subsequent motor programming. In addition to subcortical structures participating in the preverbal semantic monitoring process, the model also specifies that the striatocapsular structures are involved in the release of the formulated language segment for motor programming. Specifically, it is suggested that this release occurs through the cortico-striato-pallido-thalamo-cortical loop in the following way. Once the language segments have been verified for semantic accuracy, the temporoparietal cortex releases the caudate nucleus from inhibition. The caudate nucleus then serves to weaken inhibitory pallidal regulation of thalamic excitatory outputs in the anterior language center, which in turn arouses the cortex to enable the generation of motor programs for semantically verified language segments. According to this model, Crosson (1985) hypothesized that subcortical lesions within the cortico-striato-pallido-thalamo-cortical loop would produce language deficits confined to the lexical-semantic level.

Crosson's (1985) original conception of the response-release mechanism has since been revised and elaborated in terms of the neural substrates involved (Crosson, 1992a, 1992b). Although the actual response-release mechanism in the modified version resembles that in the original conception, the route for this release is altered. The formulation of a language segment causes frontal excitation of the caudate, which increases inhibition of specific fields within the globus pallidus; however, this level of inhibition alone is not sufficient to alter pallidal output to the thalamus. An increase in posterior language cortex excitation to the caudate, which occurs once a language segment has been semantically verified posteriorly, provides a boost to the inhibition of the pallidum. The pallidal summation of this anterior and posterior inhibitory input allows the release of the ventral anterior thalamus from inhibition by the globus pallidus, causing the thalamic excitation of the frontal language cortex required to trigger the release of the language segment for motor programming. Overall, the revised model provides an integrated account of how subcortical structures might influence language output through a neuroregulatory mechanism that is consistent with knowledge of cortical-subcortical neurotransmitter systems and structural features.

A second model of subcortical participation in language was proposed by Wallesch and Papagno (1988). This model, referred to as the lexical selection model, also proposes that subcortical structures participate in language processes via a cortico-striato-pallido-thalamo-cortical loop. Wallesch and Papagno (1988) postulated that the subcortical components of the loop constitute a frontal lobe system comprised of parallel modules with integrative and decision-making capabilities rather than the simple neuroregulatory function proposed in Crosson's (1985) model. Specifically, the basal ganglia system and thalamus were hypothesized to process situational as well as goal-directed constraints and lexical information from the frontal cortex and posterior language area, and to subsequently participate in the process of determining the appropriate lexical item, from a range of cortically generated lexical alternatives, for verbal production. The most appropriate lexical alternative is then released by the thalamus for processing by the frontal cortex and programming for speech. Cortical processing of selected lexical alternatives is made possible by inhibitory influences of the globus pallidus on a thalamic gating mechanism. This most appropriate lexical alternative has an inhibitory effect on the thalamus, promoting closure of the thalamic gate, resulting in activation of the cerebral cortex and production of the desired response. Cortical processing of subordinate alternatives is suppressed as a consequence of pallidal disinhibition of the thalamus, and the inhibition of cortical activity.

Despite the apparent differences in the two models, they both ascribe an important role to the subcortical nuclei in language processing, especially at the lexical level of language organization. It is equally apparent that each of these models has a number of limitations and that no one model has achieved uniform acceptance. A major limitation of these models is that neither explains the considerable variability in clinical presentation of subcortical aphasia. According to Cappa (1997), a further problem is that the models suggest such extensive and widely distributed systems subserving lexical processing that specific predictions appear to be difficult to disprove on the basis of pathological evidence. Put more simply, these models do not lend themselves readily to empirical testing. Yet another limitation arises from the nature of the research on which these models are based. The available models of subcortical participation in language are largely based on the observation that certain contrasting deficits of language production arise in subjects with particular subcortical vascular lesions when tested on traditional tests of language function. These language measures were typically designed for taxonomic purposes regarding traditional cortical-based aphasia syndromes and may be inadequate for developing models of brain functioning (Caramazza, 1984). It has also been argued that language deficits associated with subcortical vascular lesions may actually be related to concomitant cortical dysfunction via various pathophysiological mechanisms. For instance, cortical infarction may not have been detected by neuroimaging. Also, subcortical lesions may result in diaschisis or the functional deactivation of distant related cortical structures (Metter et al., 1983). Further, language dysfunction following subcortical lesions may be related to decreases in cortical perfusion, causing widespread cortical damage that may or may not be detected by neuroimaging (Nadeau and Crosson, 1997). As yet, however, the relationship between the structural site and etiology of subcortical lesions, the extent of cortical hypometabolism and hypoperfusion, and associated language function remains to be fully elucidated.

Further clarification of the role of subcortical structures in language is likely to come through the use of functional imaging techniques and neurophysiological methods such as electrical and magnetic evoked responses, as well as from the study of the language abilities of patients with circumscribed neurosurgical lesions involving subcortical structures (e.g., thalamotomy and pallidotomy). Functional imaging techniques such as positron emission tomography (PET) and functional magnetic resonance imaging (fMRI) enable brain images to be collected while the subject is performing various language production tasks (e.g., picture naming, generating nouns) or during language comprehension (e.g., listening to stories). These techniques therefore enable visualization of the brain regions involved in a language task, with a spatial resolution as low as a few millimeters. The use of fMRI in the future is therefore likely to further inform the debate as to the role of subcortical structures in language. A review of the extensive literature on PET studies indicates that some studies published since 1994 have demonstrated activation of the thalamus and basal ganglia during completion of language tasks such as picture naming (Price, Moore, et al., 1996) and word repetition (Price, Wise, et al., 1996).

In summary, although a role for the thalamus in language is generally accepted, some controversy still exists as to whether the structures of the striatocapsular region participate directly in language processing or play a role as supporting structures for language. Contemporary theories suggest that the role of subcortical structures in language is essentially neuroregulatory, relying on quantitative neuronal activity. Although these theories have a number of limitations, for the present they do serve as frameworks for generating experimental hypotheses which can then be tested in order to advance our understanding of subcortical brain mechanisms in language.