Jeremy Yeaton

The lesion correlates of syntactic deficits in aphasia remain poorly understood. Previous studies have suggested that distinct error types in expressive syntax may be associated with damage to different brain regions, but the specific lesion correlates of these errors have not been fully delineated. To address this gap, we analyzed spontaneous speech samples from individuals with chronic aphasia, categorizing errors into hierarchical (paragrammatic-like) and linearization (agrammatic-like) errors. Lesion-symptom mapping was conducted to identify brain regions associated with these error types. Lesion clusters in the medial superior temporal sulcus and inferior parietal lobe were associated with hierarchical errors, whereas frontal lesions, including those in the inferior and middle frontal lobe, were associated with linearization errors. These results provide support for a two-stage model of syntactic encoding, with distinct neural correlates for hierarchical and linearization processes. Our findings suggest that distinct brain regions contribute to different stages of syntactic encoding. Hierarchical processing seems to be supported by posterior temporal–parietal regions, whereas linearization seems to be supported by frontal regions. These results suggest that the diagnosis and treatment of aphasia could take into account the existence of distinct syntactic production deficits resulting from different patterns of brain damage.

Two-stage models of grammatical encoding posit that sentence production unfolds in two sequential steps: the construction of a hierarchical syntactic structure, followed by its linearization into a sequence suitable for articulation. While widely accepted in psycho- and neurolinguistics, such models lack formal computational implementations. Here, we introduce a novel multinomial processing tree (MPT) model that operationalizes this two-stage framework to explain syntactic error patterns in individuals with aphasia. Drawing on discourse samples annotated for distinct error types, we fit an MPT model to estimate individual abilities at each processing stage. These ability estimates correlated with observed error rates and localized to distinct neural substrates: hierarchical encoding ability was linked to the posterior superior temporal sulcus and parietal cortex, linearization to posterior inferior frontal regions, and omission-related processes to more dorsal frontal regions and the underlying white matter. Our findings support a neurocomputational dissociation between hierarchical and linear stages of grammatical encoding, aligning with prior lesion-symptom mapping and theoretical accounts of expressive agrammatism and paragrammatism. This work represents the first computational instantiation of a two-stage model of syntactic production, bridging formal modeling and lesion analysis to advance our understanding of the architecture and breakdown of grammatical encoding.

Background: The lesion correlates of syntactic deficits in aphasia remain poorly understood. Previous studies have suggested that distinct error types in expressive syntax, such as paragrammatic and agrammatic speech, may be associated with damage to different brain regions, but the specific lesion correlates of these errors have not been fully delineated.Objective: To identify the lesion correlates associated with distinct expressive syntactic error types in individuals with chronic post-stroke aphasia, using a novel utterance-level analysis of spontaneous speech errors.Methods: We analyzed spontaneous speech samples from individuals with chronic aphasia, categorizing errors into hierarchical (paragrammatic-like) and linearization (agrammatic-like) errors. Lesion-symptom mapping was conducted to identify brain regions associated with these error types. The analysis was based on a two-stage model of sentence production, with hierarchical processing and linearization analyzed as distinct stages.Results: Lesion clusters in the medial superior temporal sulcus and inferior parietal lobe were associated with hierarchical errors, while large frontal lesions, including those in the inferior and middle frontal lobe, were associated with linearization errors. These results provide support for a two-stage model of syntactic encoding, with distinct neural correlates for hierarchical and linearization processes. Conclusion: Our findings suggest a dichotomy of the lesion basis for syntactic deficits in aphasia, suggesting that distinct brain regions contribute to different stages of syntactic encoding. Hierarchical processing seems to be supported by posterior temporal-parietal regions, while linearization seems to be supported by frontal regions. These results suggest that the diagnosis and treatment of aphasia should take into account the existence of distinct syntactic production deficits resulting from different patterns of brain damage.

Two-stage models of grammatical encoding posit that sentence production unfolds in two sequential steps: the construction of a hierarchical syntactic structure, followed by its linearization into a sequence suitable for articulation. While widely accepted in psycho- and neurolinguistics, such models lack formal computational implementations. Here, we introduce a novel Multinomial Processing Tree (MPT) model that operationalizes this two-stage framework to explain syntactic error patterns in individuals with aphasia. Drawing on discourse samples annotated for distinct error types, we fit an MPT model to estimate individual abilities at each processing stage. These ability estimates correlated with observed error rates and localized to distinct neural substrates: hierarchical encoding ability was linked to the posterior superior temporal sulcus and parietal cortex, linearization to posterior inferior frontal regions, and omission-related processes to more dorsal frontal regions and the underlying white matter. Our findings support a neurocomputational dissociation between hierarchical and linear stages of grammatical encoding, aligning with prior lesion-symptom mapping and theoretical accounts of expressive agrammatism and paragrammatism. This work represents the first computational instantiation of a two-stage model of syntactic production, bridging formal modeling and lesion analysis to advance our understanding of the architecture and breakdown of grammatical encoding.

Background: Recent research suggests that damage to right hemisphere regions homotopic to the left hemisphere language network affects language abilities to a greater extent than previously thought. However, few studies have investigated acute disruption of language after lesion to the right hemisphere. Here, we examined lesion correlates of acute speech deficits following left and right hemisphere ischemic stroke to clarify the neural architecture underlying early language dysfunction. Methods: We retrospectively analyzed 410 patients (225 left, 185 right hemisphere lesions) from the Stroke Outcome Optimization Project dataset. Presence and severity of speech deficits was measured using the National Institute of Health Stroke Scale Best Language subscore within 48 h of onset. Manual lesion masks were derived from clinical MRI scans and normalized to MNI space. Lesion-symptom mapping was conducted using voxelwise and region-of-interest analyses with permutation correction (5000 iterations; p < 0.05), controlling for total lesion volume. Results: Speech deficits were observed in 53.7% of the cohort (58.2% left, 48.1% right hemisphere lesions). In the full sample, the presence of speech deficits was associated with bilateral subcortical and perisylvian damage, including the external and internal capsules, insula, putamen, and superior fronto-occipital fasciculus. Severity of speech deficits localized predominantly to left hemisphere structures, with peak associations in the external capsule (Z = 6.39), posterior insula (Z = 5.64), and inferior fronto-occipital fasciculus (Z = 5.43). In the right hemisphere cohort, the presence and severity of speech deficits were linked to homologous regions, including the posterior insula (Z = 3.70) and external capsule (Z = 3.63), although with smaller effect sizes relative to the left hemisphere cohort. Right hemisphere lesions resulted in milder deficits despite larger lesion volumes compared with left hemisphere lesions. Conclusions: Acute speech impairment following right hemisphere stroke is associated with damage to a homotopic network encompassing perisylvian cortical and subcortical regions analogous to the dominant left hemisphere language network. These findings demonstrate that damage to the right hemisphere consistently results in acute speech deficits, challenging the traditional left-centric view of post-stroke speech impairment. These results have important implications for models of bilateral language representation and the neuroplastic mechanisms supporting language recovery.

Syntactic impairments in aphasia can provide a powerful window into the neurobiology of language. Considerable research has focused on agrammatism in nonfluent aphasia, driving a strong association between frontal brain systems and syntax. However, the syndrome of paragrammatism, typically characterized by grammatical errors in fluent aphasia, has recieved far less attention. Recent work has suggested that paragrammatism is primarily associated with posterior temporal-parietal lesions, converging with functional neuroimaging evidence that these regions support hierarchical syntax. However, the holistic perceptual approaches to paragrammatism used in this work suffer from limited inter-rater reliability as well as conflating factors such as speech rate. To remedy these issues, this study reports a system to classify agrammatic and paragrammatic symptoms at the level of the utterance using an objective coding scheme, building on previous analyses by Matchin et al. (2020). Trained speech-pathology students rated transcriptions of 88 retellings of the story of Cinderella from persons with aphasia alongside transcriptions from 53 age-matched healthy controls. Each utterance was classified by the presence of errors corresponding to functional processing (i.e., hierarchical processing) or positional processing (i.e., linearization), along with non-grammatical errors. We found that patients defined holistically using the perceptual approach as agrammatic or paragrammatic doubly dissociated in rates of functional and positional processing, and that qualitative behavioral variables distinguish healthy speech from agrammatic and paragrammatic samples. We suggest that agrammatism and paragrammatism result from distinct breakdowns in syntactic structure building during speech, resulting from damage to distinct syntactic subsystems of the brain.

The influential model of sentence production developed by Bock and Levelt (1994) posits two stages of grammatical encoding: functional processing and positional processing. This model is sometimes referred to as the "consensus model" of grammatical encoding, given its wide support within psycholinguistic research. However, most research in the cognitive neuroscience of syntax does not align well with this two-stage consensus model. We will review recent research on the neurobiology of syntax, focusing on lesion-symptom mapping in people with aphasia, and in particular on the agrammatism-paragrammatism distinction. We will illustrate how a two-stage model of grammatical encoding aligns well with this distinction but does not precisely match the processing levels of functional and positional processing proposed by Bock and Levelt. We will illustrate how this consensus model can be revised in order to better accommodate the data. Finally, we will evaluate extant models of syntax and the brain, illustrating their respective adequacy and inadequacy with respect to accounting for these phenomena as well as their poor alignment with the consensus psycholinguistic model. We argue that models which involve two stages of grammatical encoding are more fruitful for pursuing relationships between psycholinguistics, aphasiology, and neuroscience.

Although there is a sizeable body of literature on sentence comprehension and processing both in healthy and disordered language users, the literature on sentence production remains much more sparse. Linguistic and computational descriptions of expressive syntactic deficits in aphasia are especially rare. In addition, the neuroimaging and (psycho)linguistic literatures operate largely separately. In this paper, I will first lay out the theoretical lay of the land with regard to psycholinguistic models of sentence production. I will then provide a brief narrative overview and large-scale meta-analysis of the neuroimaging literature as it pertains to syntactic computation, followed by an attempt to integrate the psycholinguistic models with the findings from functional and clinical neuroimaging. Finally, I provide a brief overview of the literature surrounding expressive syntactic deficits and propose a path forward to close some of the existing gaps.