This study investigated the neural basis of performance monitoring in patients with first-ever ischemic stroke. Performance monitoring involves the continuous evaluation of intended actions with actual performance, ensuring that behavior is adapted to the demands of the task (Ullsperger et al., 2014). Within this system, error monitoring detects errors and initiates post-error processing adjustments, including increased cognitive control, to prevent errors from occurring in the future. Thus, deficiencies in these processes can compromise the accuracy of performance. We examined error rates in modified spatial (five-point design) fluency (Regard et al., 1982; Lee et al., 1997) and verbal (word-fragment) completion (Tulving et al., 1982; Lustig and Hasher, 2001a, 2001b) tasks that do not provide external feedback. Using voxel-based lesion-behavior mapping (Bates et al., 2003; Rorden and Karnath, 2004; Vaidya et al., 2019; Pustina and Mirman, 2022), we analyzed lesion correlates of productivity and accuracy, using the former as a proxy for processing efficiency and the latter as a proxy for monitoring processes. Our general goal was to determine whether monitoring relies on domain-general or domain-specific mechanisms, as described below.

Working memory provides the task-relevant information needed to compare actions against intended outcomes (Logie et al., 2020). The dominant model (Baddeley, 2007) proposes two domain-specific storage subsystems - a left-lateralized phonological loop for verbal material and a right-lateralized visuospatial sketchpad for spatial content - that operate under the control of a domain-general central executive that allocates attention and cognitive resources (Fodor, 1983). In addition, Petrides (2000) distinguishes between storage and monitoring processes: whereas posterior brain regions support the maintenance of information in working memory, the mid-dorsolateral prefrontal cortex (PFC) is critical for monitoring the contents of working memory, particularly during self-structured tasks that require continuous evaluation and updating. This distinction is critical to understanding why stroke patients with prefrontal lesions may exhibit impairments in performance monitoring, as disruptions in these monitoring processes can lead to difficulties in accurately tracking one’s own behavior.

Compared to functional neuroimaging, lesion-behavior studies in stroke patients provide unique insights into how focal brain damage relates to deficits in executive function (e.g., Tsuchida and Fellows, 2013; Krämer et al., 2013). In this study, we investigated the neural correlates of productivity and accuracy in stroke patients performing both spatial and verbal production tasks. While productivity may be affected by domain-specific lesions - left lateralized for verbal tasks and right lateralized for spatial tasks—competing models of neural mechanisms for performance monitoring propose different sources of error. One model (which is consistent with Petrides’ view; Petrides, 2000) predicts that damage to a domain-general central monitoring process will facilitate errors in both types of tasks. A second model, consistent with the phonological and visuospatial subsystems of Baddeley and colleagues’ model (Baddeley, 2007), proposes that errors arise from lesions within the two specialized domain-specific systems, left lateralized for verbal tasks and right lateralized for spatial tasks. A third model, referred to as the hemispheric monitoring (Zaidel, 1987) or error monitoring (Hochman and Eviatar, 2004, 2006) hypothesis, suggests that errors are related to damage in a specific error-monitoring system that exhibits the opposite, i.e., domain-incongruent, lateralization, left lateralized for spatial tasks and right lateralized for verbal tasks, as described below. The primary aim of this study was to distinguish between these three models by analyzing lesion-error correlations in both spatial and verbal tasks.

The neural organization of error monitoring remains controversial. It may be centrally organized, relying on a unified executive system; it may follow hemispheric independence (Zaidel et al., 1990), with each hemisphere monitoring its own errors; or it may adhere to the hemispheric monitoring hypothesis (Zaidel, 1987), where error monitoring emerges from interhemispheric comparison. The error-monitoring hypothesis argues that because error monitoring is cognitively demanding, it follows a domain-incongruent lateralization pattern (Hochman and Meiran, 2005). Indeed, in tasks dominated by one hemisphere, such as lexical decision (left hemisphere) and bar-graph judgment (right hemisphere), a reverse lateralization effect has been observed, with the non-dominant hemisphere detecting errors more efficiently (Hochman and Eviatar, 2004, 2006). This effect increases under high cognitive load, suggesting that the non-dominant hemisphere compensates for processing limitations of the dominant hemisphere (Banich, 1998). Furthermore, Hochman et al. (2011) demonstrated that intact interhemispheric communication is essential for effective error monitoring, as patients with partial callosal disconnection showed impairments.

Thus, there are three models of how performance monitoring occurs across the cerebral hemispheres. One model attributes monitoring to a central, domain-general executive system that monitors errors equally in both verbal and spatial tasks, predicting either no lateralization if organized bilaterally or possible right hemisphere dominance. A second model, based on hemispheric independence, suggests that each hemisphere monitors errors within its specialized domain, with the left hemisphere primarily monitoring verbal errors and the right hemisphere focusing on spatial errors. In contrast, hemispheric interaction hypotheses, including the error-monitoring hypothesis, propose a reverse lateralization effect under high cognitive load, in which the non-dominant hemisphere monitors the dominant one; specifically, the right hemisphere monitors verbal errors while the left monitors spatial errors.

The error-monitoring hypothesis has remained unexplored in stroke patients, who often exhibit cognitive impairments - with left hemisphere lesions affecting verbal performance and right hemisphere lesions affecting spatial function (Bradshaw and Nettleton, 1981; Miceli et al., 1981; Hellige, 2001; Moore and Demeyere, 2022; Vallesi, 2012; Rost et al., 2022). Existing research suggests that key regions such as the inferior PFC are critical for successful performance monitoring (Gehring and Knight, 2000; Ullsperger et al., 2002; Ullsperger and von Cramon, 2006). Although Ullsperger et al. (2002) focused on left-sided lesions and Ullsperger and von Cramon (2006) included patients with left- and right-sided lesions, the systematic investigation of lesion laterality remains an often neglected aspect in the neuropsychology of executive functions. In addition, Mandal et al. (2020) linked increased speech errors in post-stroke aphasia to lesions in left frontal white matter and dorsolateral PFC, while McCall et al. (2022) showed that disrupted connections between left posterior medial PFC and speech areas enhance speech errors. Finally, some studies suggest that right hemisphere lesions are also associated with higher error rates on verbal tasks (Iacoboni et al., 1997; Kaplan and Zaidel, 2001; Robinson et al., 2021).

This study investigates in stroke patients whether performance monitoring is associated with damage to domain-general central processes, domain-specific working memory subsystems, or a system dedicated to error monitoring that exhibits reverse (domain-incongruent) lateralization. To address this question, we analyzed lesion-error correlations in both spatial and verbal tasks, building on existing evidence that regions in the PFC and intact interhemispheric connectivity are key for effective performance monitoring. To test these hypotheses, we conducted a lesion-behavior study in stroke patients who had experienced their first ischemic event. Participants completed two production tasks: a spatial five-point design (5PD) fluency task and a verbal word-fragment completion (WFC) task. Using voxel-based lesion-behavior mapping, we examined how lesion location correlated with two performance measures - productivity (fluency) and accuracy (error rates) - across both task domains. This approach allowed us to investigate whether performance monitoring relies on domain-general, domain-specific, or domain-incongruent neural mechanisms.

Fluency tasks, which require individuals to generate as many unique responses as possible under time constraints, are popular in neuropsychological assessment of executive function. Verbal fluency (Thurstone and Thurstone, 1962) typically engages left hemisphere regions, including prefrontal and temporal cortices, which support language production and comprehension (Baldo et al., 2001; Henry and Crawford, 2004; Robinson et al., 2012; Cipolotti et al., 2020; Biesbroek et al., 2021; Godefroy et al., 2023). In contrast, design fluency (Jones-Gotman and Milner, 1977) has traditionally been associated with the right hemisphere due to its spatial nature, although evidence for this lateralization is less consistent (Baldo et al., 2001; Marin et al., 2017; Cipolotti et al., 2020). Most neuropsychological research on fluency emphasizes productivity, while accuracy (error rates) is less frequently examined due to its relative rarity. For example, Stuss et al. (1998) attempted to classify verbal fluency errors but found low overall error rates. Cipolotti et al. (2020) reported that damage to the left posterior medial PFC was associated with phonemic rule violations. Recent work by Robinson et al. (2021) showed that fluency errors - including repetitions and rule violations - are primarily associated with right lateral PFC lesions across a variety of fluency tasks (verbal, design, gesture, and ideational fluency).

To increase the sensitivity of design fluency to performance monitoring in stroke patients, we introduced a modified five-point design (m5PD) fluency task. This adaptation increased error susceptibility by lengthening task duration, introducing a movable window to increase working memory demands, strictly limiting the number of valid designs to enforce strategic problem solving, and withholding external error feedback. These modifications were designed to enhance the role of working memory and performance monitoring in design fluency performance in stroke patients.

Verbal completion tasks assess language processing. For example, word-fragment completion tasks require participants to add missing letters to word fragments (e.g., in our study, complete “s_t” with “set”). These tasks primarily engage a left-lateralized network: the visual word form area in the left ventral occipitotemporal cortex processes letter patterns, the left inferior parietal lobule supports phonological working memory, and the left inferior prefrontal and temporal cortices facilitate semantic retrieval. Sentence-stem completion tasks that require completing sentences with semantically incongruent words, such as Part B of the Hayling task (Burgess and Shallice, 1996), recruit additional right-hemisphere regions (Volle et al., 2012; Cipolotti et al., 2016). However, evidence for right-hemisphere involvement in nonverbal tasks such as picture completion remains inconclusive (Bilder, 2011; Chelune, 2017).

To increase the sensitivity of verbal completion to performance monitoring in stroke patients, we implemented a modified WFC (mWFC) paradigm with four key adjustments: extended task duration, a moving display window to make previous choices irreversible, ambiguous trigrams to increase the difficulty of finding unique final letters that transform the two consonants into valid nouns, and withholding of external error feedback. These modifications were designed to increase cognitive demands and performance monitoring in verbal completion performance in stroke patients.

This study investigates the neural correlates of cognitive productivity and accuracy in stroke patients using modified spatial and verbal production tasks. Although previous research has examined hemispheric specialization for spatial fluency and verbal completion, the relationship between productivity and accuracy in lesion studies remains poorly understood, in part due to the typically low error rates in standard tasks. Based on existing evidence, productivity in verbal completion may be associated with left hemisphere lesions, whereas productivity in spatial fluency may be associated with right hemisphere lesions, following traditional lateralization patterns. However, the neural basis of accuracy may go beyond conventional domain-congruent hemispheric specialization. Error rates in both tasks may be associated with damage to a central domain-general monitoring process, which may be either bilateral or right dominant. Alternatively, error rates may be associated with damage to domain-congruent working memory systems that follow typical lateralization patterns, i.e., verbal errors-left lesions, spatial errors-right lesions. Finally, error rates may be associated with damage to an error-monitoring system that shows a reverse lateralization effect, i.e., verbal errors-right lesions, spatial errors-left lesions. It is important to note that these mechanisms are not mutually exclusive and can coexist.

This study investigated the neural correlates of cognitive performance in spatial fluency and verbal completion tasks in stroke patients, with a focus on accuracy. Although no significant correlations between productivity and lesion locations were found, our results reveal distinct neural correlates for error measures that support both domain-general central monitoring and a reverse lateralized error-monitoring system. In spatial fluency, lesion clusters in the right frontal operculum and left white matter suggest that these regions contribute to the maintenance of spatial accuracy. In verbal completion, accuracy was associated with lesions in the right frontal operculum and right white matter, implicating these areas in the maintenance of verbal accuracy. Thus, our results suggest a complex pattern of lateralization: a common lesion cluster in the right frontal operculum supports central monitoring across domains, whereas domain-incongruent lateralization of white matter - left for spatial errors and right for verbal errors - is consistent with the error-monitoring hypothesis (Hochman and Eviatar, 2004, 2006). This hypothesis posits an inversely lateralized system, with left hemisphere dominance for spatial error monitoring and right hemisphere dominance for verbal error monitoring, which is consistent with our observed white matter-error correlations.

The common lesion-error correlation observed in the right inferior frontal region is consistent with previous studies (summarized in the Introduction) highlighting its critical role in maintaining accuracy across task domains. Functional neuroimaging studies have implicated the inferior frontal opercular region in domain-general task control, suggesting its involvement in central executive processes, including performance monitoring (Assem et al., 2020; Cai et al., 2024; Dosenbach et al., 2008). Although our results suggest that the right inferior frontal region supports accuracy across domains, they do not imply that central monitoring is exclusively right lateralized. This is an important consideration given that our lesion coverage in the inferior frontal region was lower in the left hemisphere - likely due to exclusion criteria related to aphasia - so a similar role for the left inferior frontal region cannot be ruled out. Furthermore, after controlling for age, sex, and education, the significance of the right inferior frontal cluster for spatial errors was lost, raising the question of whether its contribution to accuracy reflects a truly domain-general, central function.

While the right inferior frontal region may support domain-general accuracy, the observed domain-incongruent white matter lateralization suggests that neural networks for accuracy diverge across task domains and cerebral hemispheres. Specifically, lesions in left white matter were associated with spatial accuracy, whereas lesions in right white matter were associated with verbal accuracy. This reverse lateralization is consistent with the error-monitoring hypothesis, which proposes that the nondominant hemisphere monitors dominant hemisphere processing for errors (Hochman and Eviatar, 2004, 2006). Thus, our results suggest that reverse lateralized error-monitoring processes contribute to the maintenance of accuracy, although overt errors and error monitoring are clearly distinct phenomena. The fact that lateralized lesions did not directly modulate overt spatial or verbal error rates further underscores the multidimensional nature of overt error generation.

Taken together, our results suggest that cognitive accuracy are supported both by a domain-general, possibly right-lateralized network and by domain-incongruently lateralized networks that are widely distributed, as described in detail in the Results. This view is consistent with previous research (Cipolotti et al., 2020; Mandal et al., 2020; Robinson et al., 2021; McCall et al., 2022). However, clinical studies using go/no-go and stop-signal tasks have reported inconsistent lateralization patterns - some finding increased commission errors in right lateral prefrontal patients (Aron et al., 2003; Molenberghs et al., 2009), while others report no effects or increased omission errors (Krämer et al., 2013; Picton et al., 2007). Similarly, task-switching studies have yielded mixed results, with left PFC lesions associated with rule-switching difficulties and right PFC lesions associated with inhibitory deficits (Mayr et al., 2006), while other studies find increased errors with inferior medial frontal lesions (Shallice et al., 2008, with lesions not being clearly confined to either the left or the right hemisphere) or increased switching costs in both hemispheres (Aron et al., 2004). In addition, results from the Trail Making Test, Part B (TMT-B) are mixed, with some studies reporting slower performance and more errors in frontal patients, especially with bilateral lesions (Stuss et al., 2001), while others find no significant differences between frontal and non-frontal patients (Chan et al., 2015, regardless of whether the frontal damage was in the left or right hemisphere). Lesion-behavior mapping studies further implicate both left (Varjacic et al., 2018) and right (Kopp et al., 2015) lesions in TMT-B performance. These inconsistencies are likely due to methodological differences and small sample sizes, highlighting the complexity of delineating the neural bases of cognitive accuracy.

Studies suggest that stroke, neurodegenerative disease, and aging (Hoffmann and Schmitt, 2006; Hämmerer et al., 2011; Kopp et al., 2014; Al Banna et al., 2016; Giannouli and Tsolaki, 2023) are all associated with declines in performance monitoring - as evidenced, for example, by age-related declines in error-related negativity (Falkenstein et al., 2001). However, stroke studies offer the particular advantage of linking focal brain lesions to specific aspects of performance monitoring, as shown in the present study. For example, Hochman et al. (2015) found a double dissociation: damage to the basal ganglia impairs error correction (recognizing and correcting errors), while damage to the dorsomedial prefrontal cortex impairs error inhibition (preventing errors before they occur). This underscores the critical contributions of clinical neuropsychology to our understanding of the relationship between neural structures and behavioral performance. Clinical studies using lesion-behavior mapping clearly provide a complementary method to electrophysiology and functional brain imaging.

While our findings are consistent with the error-monitoring hypothesis, alternative models remain viable. One such model is the cognitive interference hypothesis (Kopp, 2025). It proposes that the left hemisphere controls proactive interference (PI), preventing old information from interfering with new processing. The right hemisphere controls retroactive interference (RI), protecting previously stored information from new input. PI sensitivity, which may be reflected, for example, in the Stroop effect due to the prior availability of semantic information that interferes with the later availability of color information, is associated with the left PFC in both verbal and spatial task versions (Cipolotti et al., 2016; Perret, 1974; Tsuchida and Fellows, 2013; Ambrosini and Vallesi, 2017). RI sensitivity is associated with the right PFC, perhaps as assessed by the Hayling effect, in which later available congruent closing words interfere with earlier prepared incongruent closing words (see Introduction; Volle et al., 2012; Cipolotti et al., 2016). PI may also gradually accumulate in spatial fluency tasks due to their high self-structuring demands, explaining why left hemisphere lesions correlate with spatial repetition errors. Conversely, RI may be more prominent during vowel selection in verbal completion tasks, potentially linking right hemisphere lesions to verbal errors. To disentangle these two models, future research should systematically manipulate PI and RI across domains and examine lesion effects. The error monitoring hypothesis predicts that left lesions will be associated with spatial accuracy and right lesions will be associated with verbal accuracy, regardless of the type of interference, whereas the interference hypothesis predicts that left lesions will be associated with PI-enriched accuracy and right lesions will be associated with RI-enriched accuracy, regardless of the domain.

Among the most significant confounds is the high-dimensional nature of the differences between our spatial fluency and verbal completion tasks. Although we designed these tasks to be both structurally parallel and highly reliable, achieving comparable validity has been more challenging due to inherent differences beyond the intended domain distinctions. A key factor is the self-structuring demands of each task, an essential aspect of assessing executive functioning (Lezak, 1995). Spatial fluency places high demands on self-structuring, requiring participants to plan and organize unique designs in a self-ordered sequence, adding complexity beyond mere production. In contrast, verbal completion involves minimal self-structuring, which may reduce its sensitivity to executive functions. Another important difference is the measurement of errors: spatial fluency allows for a more nuanced classification of errors (i.e., repetitions, rule violations), whereas verbal completion lacks this granularity, introducing an additional confound when comparing performance accuracy across tasks.

Our study distinguishes between productivity and accuracy measures. It also addresses a common limitation of previous research, namely the use of tasks that are too easy, often resulting in low error rates (see Introduction). By increasing task difficulty, we ensured consistently high error rates on both tasks (~25 percent), resulting in similarly reliable accuracy measures (see Methods). However, productivity and accuracy differ in their validity. Productivity, defined as the percentage of items completed, is a relatively process impure measure. It is affected by confounding factors such as overall processing speed, including sensory and motor speed, which can be related to stroke, and task engagement. In contrast, error rate is a more process pure measure because it is calculated as the ratio of two behaviorally derived measures: incorrect responses (numerator) divided by total responses (denominator). This ratio (error rate = incorrect responses/total responses) inherently controls for individual differences in such confounding factors and better isolates the cognitive process of interest: monitoring. In support of this argument, our pilot study found that while productivity did not correlate with standard neuropsychological measures, accuracy did show meaningful associations: spatial accuracy correlated with phonemic fluency, again suggesting left hemisphere involvement, while verbal accuracy correlated with visuoconstructive abilities, again suggesting right hemisphere involvement (see Appendix 1). These considerations highlight the importance of further refining performance measures in future studies to increase the specificity of cognitive assessments.

A note of caution regarding white matter disconnection and the limitations of voxel-level lesion mapping seems to be in order. The goal of our analysis was to provide an initial indication of which white matter tracts correspond to the lesion clusters identified by SCCAN, rather than to perform a detailed tract-level lesion mapping analysis aimed at associating specific tracts with cognitive deficits. Our approach superimposes the significant voxel clusters on a white matter atlas to provide general insight into the potential involvement of tracts, but is not intended to establish definitive tract-behavior relationships. However, our goal here was to provide a first step toward understanding the broader anatomical correlates of cognitive accuracy in production tasks, and future studies may benefit from more specific tract-level analyses.

Although lesion location was the primary focus, supplementary analyses showed that covarying for basic demographics provided additional insight into lesion-deficit relationships (see Appendix 5). Specifically, while the initially reported right inferior frontal cluster associated with accuracy in the spatial fluency task did not remain significant after controlling for age, sex, and education, the left frontal white matter cluster showed a robust and significant association with spatial fluency accuracy. These findings highlight the importance of controlling for demographic factors when interpreting lesion-deficit relationships and suggest that the left frontal white matter cluster may play a particularly critical role in spatial fluency. The common brain regions involved in performance monitoring across tasks remain intriguing and warrant further investigation.

Two methodological decisions regarding patient recruitment deserve brief comment. First, we excluded patients with large cortical lesions to minimize confounding variability from individuals who may be unable to perform the production tasks. Second, although our simple neglect measure may not detect subtle or complex forms of neglect, undiagnosed neglect cannot account for our overall findings. Furthermore, excluding the one patient with neglect would not meaningfully alter the results, as this individual’s performance was not a critical outlier. Consequently, asymmetry-related measures were not included as covariates in our analysis.

Despite its limitations, this study provides a basis for further investigation of the neural mechanisms underlying accuracy maintenance using clinical neuropsychological methods. The sample size of 110 patients, while substantial, limits generalizability and underscores the need for replication with larger samples to increase statistical power and broader applicability. Other limitations include task validity and lesion coverage, which warrant further improvements, as discussed above.

Our study identifies a multidimensional neural architecture for performance monitoring that blends domain-general and domain-specific processes. A central, domain-general monitoring system - anchored at least in the right inferior frontal region - appears to be associated with accuracy across cognitive domains. In addition, we found robust evidence for a distinct, reverse-lateralized monitoring system: lesions in the left hemisphere were associated with spatial accuracy, whereas lesions in the right hemisphere were associated with verbal accuracy. These findings go beyond traditional lateralization models and suggest more complex hemispheric interactions in performance monitoring. If replicated and validated, these findings could advance our understanding of how the human brain organizes performance monitoring and ultimately lead to targeted rehabilitation strategies for stroke survivors.