Authors: Mackenzie E. Fama (1Department of Speech, Language, and Hearing Sciences, The George Washington University, Washington, DC), Joshua D. McCall (2Department of Neurology, Georgetown University Medical Center, Washington, DC), Andrew T. DeMarco (2Department of Neurology, Georgetown University Medical Center, Washington, DC; 3Department of Rehabilitation Medicine, Georgetown University Medical Center, Washington, DC), Peter E. Turkeltaub (2Department of Neurology, Georgetown University Medical Center, Washington, DC; 3Department of Rehabilitation Medicine, Georgetown University Medical Center, Washington, DC; 4Research Division, MedStar National Rehabilitation Hospital, Washington, DC)
Categories: Article, Executive function, inner speech, language, aphasia, stroke
Source: Neuropsychologia
Authors: Mackenzie E. Fama, Joshua D. McCall, Andrew T. DeMarco, Peter E. Turkeltaub
Research over the past several decades has revealed that non-linguistic cognitive impairments can appear alongside language deficits in individuals with aphasia. One vulnerable cognitive domain is executive function, an umbrella term for the higher-level cognitive processes that allow us to direct our behavior towards a goal. Studies in healthy adults reveal that executive function abilities are supported by inner speech, the ability to use language silently in one’s head. Therefore, inner speech may mediate the connection between language and executive function deficits in individuals with aphasia.
Here, we investigated whether inner speech ability may link language and cognitive impairments in 59 adults with chronic, post-stroke aphasia. We used two approaches to measure inner one based on internal retrieval of words and one based on internal retrieval plus silent manipulation of the retrieved phonological forms. Then, we examined relationships between these two approaches to measuring inner speech and five aspects of executive function response inhibition, conflict monitoring/resolution, general task-switching ability, phonological control, and semantic control. We also looked for dissociations between inner speech ability and executive function ability.
Our results show tentative relationships between inner speech (across multiple measurement approaches) and all aspects of executive function except for response inhibition. We also found evidence for a double many participants show intact executive function despite poor inner speech, and vice versa, so neither process is strictly reliant on the other. We suggest that this work provides preliminary evidence of a bidirectional relationship between inner speech and executive inner speech supports some aspects of executive function via internal self-cueing and certain aspects of executive function support performance on complex inner speech tasks.
Aphasia is an acquired language disorder with chronic impacts on communication and quality of life (Hilari & Byng, 2009). Recent work reveals that non-linguistic cognitive processes including executive function (EF) are also commonly impaired in individuals with aphasia (Fonseca et al., 2017). Given the demonstrated relationship between EF, functional communication, and overall therapy outcomes (Gilmore et al., 2019; Olsson et al., 2019), it is crucial to refine current theories of cognitive impairments in aphasia in order to improve clinical decision-making and therapeutic planning.
EF abilities are essential in daily life, as they enable individuals to direct their behavior towards a relevant goal. EF is not a singular cognitive process, but rather is an umbrella term that encompasses a number of specific subprocesses (e.g., inhibiting, task-switching) (Miyake & Friedman, 2012). Three of these subprocesses will be examined within this inhibition, task-switching, and conflict resolution. Inhibition refers to the ability to override a prepotent response and is experimentally measured by tasks such as stop-signal or Go-NoGo. Task-switching refers to the ability to change between tasks in a flexible manner and can be measured via tasks with alternating demands (e.g., alternating between selecting two or more different target stimuli). Conflict resolution refers to the ability to adjust behavior in order to resolve incompatibility among representations or stimuli (Hussey & Novick, 2012) and can be measured by tasks with mismatched representations in verbal (e.g., syntactic ambiguity) or non-verbal (e.g., color) stimuli.
There is evidence to suggest that EF abilities are closely linked to language processing. First, in all language users, production and comprehension tasks occur in the context of semantically and phonologically related competitors wherein a speaker/listener must select the target from these near competitors (Nozari & Novick, 2017). In order to achieve communicative success, a speaker must resolve this competition by activating EF subprocesses such as conflict resolution and inhibition. Furthermore, EF abilities can be impaired after stroke in individuals with aphasia, a language disorder (Gilmore et al., 2019; Murray & Ramage, 2000). While some studies suggest that EF impairments in aphasia are most evident when the cognitive tasks have high inherent verbal demands (Dutta et al., 2023), there is also substantial evidence that individuals with aphasia can struggle with EF tasks even when language demands are minimal (Meier et al., 2022).
There is some debate in the literature as to whether impairments in non-verbal cognitive processes including EF are wholly separate from language processing in aphasia or whether there is a mechanistic link between them. Evidence for a link between language and EF processing in aphasia comes from research showing an association in severity of deficits across the two domains. For instance, more severely impaired cognitive performance has been associated with more severe deficits in semantic processing, naming, and auditory comprehension (Baldo et al., 2005, 2010). Prior work has also demonstrated significant relationships between cognitive processing (i.e., verbal working memory and other aspects of EF) and functional communication abilities in stroke survivors with aphasia (Gilmore et al., 2019; Schumacher et al., 2020). In some prior work, overall severity of language impairments relates to performance on some but not all EF tasks (Dutta et al., 2023). A small case series also showed that language and EF abilities can dissociate over time during stroke recovery (Meier, Kelly, & Hillis, 2022). Finally, the neural substrates for EF deficits in this population are also under debate, as studies have shown EF deficits linked to extrasylvian regions typically associated with domain-general processing (Meier et al., 2022) as well as to Broca’s area within the language network (Akkad et al., 2023). In the latter finding, the authors remain agnostic as to whether this colocalization of deficits suggests a mechanistic link or simply a co-occurrence of both language and EF deficits (Akkad et al., 2023).
Some studies have suggested that there may be a distinction between domain-general and language-specific uses of EF (Jacquemot & Bachoud-Levi, 2021; Kuzmina & Weekes, 2017). Recent work supports an even finer distinction within language-specific EF by showing a dissociation between phonological control, which allows one to manage phonological interference during a task, and semantic control, which allows one to manage semantic interference (McCall et al., 2022). Semantic and phonological control likely help to resolve the day-to-day linguistic competition mentioned above. There is evidence that these forms of cognitive control/EF^1^ can be impaired in individuals with aphasia, especially semantic control (Lambon Ralph et al., 2017). Overall, the above evidence suggests a potential link between EF and language processing in aphasia, but the underlying mechanism of this relationship has not been fully elucidated and further research in this area is warranted.
There is a relevant interaction between language and non-language domains in healthy adults, wherein EF abilities (e.g., task-switching) are supported by inner speech (IS), the ability to use language silently in one’s head (Emerson & Miyake, 2003). IS is a phenomenon that is well-known to most healthy language users and can be defined most simply as the mental imagery of having a little voice in one’s head (Sokolov, 1972). Over the past century, IS has been studied relatively widely for its supporting role in a number of mental processes, including language learning (Vygotsky, 1962), silent reading, and EF abilities (for a recent review, see Alderson-Day & Fernyhough, 2015).
Functionally, IS is thought to support EF ability by serving as a self-cue to help retrieve and activate the goals of a given task (Emerson & Miyake, 2003). With respect to cognitive models, IS is thought to serve as the basis for verbal rehearsal, i.e., the articulatory/phonological loop in Baddeley’s seminal model of working memory (Baddeley & Hitch, 1974). Working memory is an aspect of short-term memory that allows us to temporarily store and manipulate a limited amount of information. The relationship between working memory and EF is complex, but it is generally agreed that working memory serves as an underlying process that supports many of the EF subprocesses described above (e.g., inhibiting and task-switching) (Baddeley et al., 2001). Importantly, several researchers have proposed that cognitive deficits in individuals with aphasia arise from an impairment of working memory/verbal rehearsal, implicitly acknowledging the role of IS in cognition in this population (Baldo & Dronkers, 2006; Martin & Reilly, 2012; Murray, 2012a). There are also a few studies that investigate whether treatment of attention or working memory can improve language recovery (see Murray, 2012b for a review). However, published studies that directly investigate the specific relationship between IS and cognitive deficits in individuals with aphasia are limited.
Furthermore, there is a significant literature on spoken language processing in aphasia, but there is limited evidence about the nature of IS in this population in general (see Fama & Turkeltaub, 2020 for a scoping review). The overall number of studies on IS in aphasia is very small, and inconsistent definitions of IS make comparisons across studies challenging. There are two overarching definitions of IS across the literature; notably, these definitions are found throughout the interdisciplinary literature on IS, but are also applicable within the specific studies examining IS in post-stroke aphasia.
One approach to defining IS focuses on the subjective nature of IS, i.e., the ability to generate language silently in one’s head (Hayward et al., 2016; Fama et al., 2017; Fama, Snider, et al., 2019; Fama, Henderson, et al., 2019). This subjectively-focused definition may be measured in an abstract way, via investigation of the everyday experience of using language internally in one’s mind, or it may be measured more concretely through silent picture-naming, where an individual is asked to report yes/no whether they can say/hear the name of a picture inside their head. The concrete, experimental approach to defining a subjective experience of IS has been shown in adults with aphasia to relate closely to the process of lexical/phonological retrieval (Fama, Snider, et al., 2019; Fama, Henderson, et al., 2019). Individuals with aphasia often reveal a mismatch between inner and overt speech by reporting sense of successful IS even when they cannot say words aloud. Our prior work demonstrates that these reports are meaningful, i.e., when a stroke survivor reports being able to say a word in their head, they are likely to demonstrate accurate phonological knowledge about that word (Fama, Snider, et al., 2019; Fama, Henderson, et al., 2019).
The other approach to defining IS focuses on the more objectively measurable aspects of IS, encompassing both the ability to generate internal language and also to perform a computation or manipulation on that mental imagery, which likely engages some domain-general cognitive processes (Geva et al., 2011; Stark et al., 2017). This objective definition can be measured experimentally through tasks such as silent rhyme or homophone judgments, in which a pair of written words or pictures is presented and the individual is asked to report whether the two words rhyme or sound the same, without speaking aloud. Importantly, regardless of the operational definition being used, most prior research has shown that IS ability differs across individuals with aphasia (Fama & Turkeltaub, 2020; Stark et al., 2017).
Given the evidence for variability in IS and EF in adults with aphasia, coupled with studies showing a role for IS in EF ability in the general population, a natural question Does inner speech support executive function ability in aphasia? To date, no studies have examined whether impairments of IS have a direct relationship to EF deficits in aphasia. The primary objective of this project is therefore to establish and characterize a link between IS ability and EF ability in individuals with aphasia. Participants with chronic, post-stroke aphasia completed a behavioral battery of language, IS, and EF tasks, most of which were normed in healthy age-matched adults. Our general hypothesis is that IS supports EF abilities, so we predicted that we would find significant correlations between measures of IS and EF.
Importantly, a correlation between IS and EF will not provide insight into causation or directionality of the relationship, so we also tested whether either ability requires the other by looking for dissociations between IS and EF at the individual participant level. We predicted that we would find a double dissociation in this analysis. For the first direction (good IS but poor EF), we predicted a dissociation due to the while IS may support performance during an EF task, it is not sufficient for adequate performance, so impairments in domain-general processes could still result in poor performance on EF tasks despite intact IS. For the second direction (good EF but poor IS), we anticipated that some participants would perform relatively well on all EF tasks despite IS impairments. There would be at least two potential explanations for this direction of a dissociation. First, one might judge the word retrieval demands of our EF tasks to be simpler and more repetitive (i.e., the same items repeatedly) compared to our IS tasks, so participants may actually be able to use some limited IS successfully during the EF tasks despite poor performance on the IS tasks. Alternatively, perhaps some participants may have adapted alternative compensatory strategies for carrying out EF tasks without the use of IS (e.g., visualizing task goals vs. using IS to verbally rehearse them), given that they are all in the chronic stage of recovery from aphasia and may be aware (consciously or subconsciously) of limited access to IS.
Participants included 59 adults in the chronic (>6 months) phase of recovery from left-hemisphere stroke. All participants were enrolled in the Brain-based Understanding of Individual Language Differences after stroke (BUILD) study (R01 DC014960 to Peter E. Turkeltaub), a crosssectional investigation of stroke recovery at Georgetown University Medical Center. Informed consent procedures were approved by the Georgetown University Institutional Review Board and a data sharing agreement was established between Georgetown University and the George Washington University. Participants were 36 males and 23 females, 52 right-handed and 7 left-handed, with average age of 60.7 years (SD=11.0), average years of education 16.7 (SD=2.7), and average time since stroke of 52.6 months (SD=61.3). Race and ethnicity were reported separately, as 38 White participants, 21 Black participants; one participant of Hispanic/Latino origin, 58 participants of non-Hispanic/Latino origin.
Control participants were 68 adults with no prior history of stroke, brain injury, or other neurologic 32 males and 36 females, 63 right-handed and 5 left-handed, with average age of 60.9 years (SD=11.6) and average years of education 17.0 (SD=2.5). Self-reported race/ethnicity was as 44 White participants, 24 Black participants, and no participants of Hispanic/Latino origin. This control group provided normative data on the EF tasks (section 2.2.2). Control norming for the IS tasks (section 2.2.1) was completed as part of a previous study; see Fama, Snider, et al., 2019 for details. These control data will serve as the basis for the analysis examining dissociations between IS and EF abilities (see section 2.4).
All computer-based tasks were developed and presented in E-Prime 3 (https://pstnet.com/products/e-prime/). Participants completed the tasks on Dell Inspiron 17” 2-in-1 touchscreen in tented position.
Participants completed the Western Aphasia Battery-Revised (WAB) (Kertesz, 2006) as a measure of overall aphasia severity and general language processing ability. Average WAB Aphasia Quotient was 79.8 (SD=19.7, range 22.4 – 99.6). WAB testing also produces four domain Spontaneous Speech & Fluency, Naming & Word-Finding, Auditory Verbal Comprehension, and Repetition. They also completed a spoken naming task with black-and-white line drawings of common items (N=60).
Participants also completed five IS
After participants completed these tasks, we analyzed three retrieval-based measures of IS: Inner Speech Self-Report in addition to a subtraction measure for Rhyme Judgment and a subtraction measure for Syllable Counting. The picture-based measures require participants to retrieve the name of the image(s) and perform a manipulation on the retrieved name(s), whereas the auditory versions require only the manipulation. Thus, for both measures, a proxy measure of word retrieval was isolated by subtracting the auditory version score from the picture-based scores. We also analyzed two retrieval-and-manipulation measures of IS, consisting of the accuracy scores for picture-based Rhyme Judgment and Syllable Counting.
Participants completed three tasks assessing different aspects of EF:
We used the Antelopes & Cantaloupes task to measure three aspects of EF. General task-switching ability was measured on the Standard and Nonverbal versions of the task (using switch cost as described above). We measured semantic control and phonological control via subtraction scores of performance (TTS) on different versions of the task (S1 only; no switching): Semantic minus Unrelated and Phonological minus Unrelated.
We predicted that IS would relate to performance on both Simon and the Go-NoGo task. For the Antelopes & Cantaloupes task, we predicted that IS would relate to the Standard version but not the Nonverbal version, as the Standard version uses easily recognizable and nameable shapes. Finally, we predicted that IS would relate to phonological control but not to semantic control, as IS is understood to be a primarily phonological phenomenon.
First, we computed descriptive statistics in order to determine mean, standard deviation, and range of performance for all measures of interest in both stroke survivors and control participants. Then, in stroke survivors, we used Spearman’s correlations to examine the relationships between tasks within each domain (IS and EF) as well as between each domain and language, as measured by the four WAB subscores. We then used Spearman’s correlations to examine the relationships between IS and (1) conflict resolution (Simon), (2) response inhibition (Go-NoGo), (3) task-switching ability (Standard and Nonverbal versions of Antelopes & Cantaloupes), and (4) phonological/semantic control (subtraction scores from Antelopes & Cantaloupes). Finally, we examined this last set of relationships between IS and EF as a set of nonparametric partial correlations, controlling for overall language ability as measured by the WAB Aphasia Quotient. We evaluated significance with a critical alpha of .05. Results will show uncorrected p-values and will also denote significance at a Bonferroni-corrected significance threshold to adjust for multiple comparisons, as appropriate. Three participants did not complete one or both of the syllable counting tasks (see Supplementary Table 1), so to account for this missing data they were excluded from all analyses involving those specific scores.
Our primary analyses described above are correlational in nature, so while our results may show that IS and EF are related, we will not be able to draw conclusions regarding causality or directionality of the relationship based on correlations alone. Thus, we next examined dissociations between IS and EF at the individual participant level in order to evaluate whether either ability is strictly necessary for the other. Specifically, we evaluated the presence of (1) participants who showed poor IS (either retrieval-based or retrieval-and-manipulation-based) but performed within normal limits on any EF tasks, and (2) participants who showed poor EF ability on any task but were able to perform within normal limits on any IS tasks. In order to distinguish between “poor” and “within normal limits” performance, we used an interquartile range (IQR) outlier analysis (Tukey, 1977) based on control performance, as all measures except IS report were normed in healthy, age-matched controls (see Section 2.1). This nonparametric approach was used due to non-normal distribution of control performance. The outlier analysis used interquartile range (Q3-Q1) to identify objective cutoffs for scores that are outside the normal range of performance in controls. For IS report, which was not administered to control participants, we identified typical range of performance within stroke survivors themselves using a parallel IQR analysis approach as described in the following paragraph.
To establish the boundaries of normal performance, we used two Lower boundary = Q1 - (Q3-Q1)3 and Upper boundary = Q3 + (Q3-Q1)3. For most tasks, participants were identified as poor performers if their scores were below the lower boundary. For the tasks where higher scores represent worse performance (i.e., Simon task, switch cost on Antelopes & Cantaloupes), participants were identified as poor performers if their scores were above the upper boundary. Notably, IQR outlier analyses can be performed using the strict criterion given above (multiply by 3, the “outside fence” per Tukey, 1977), or by a less stringent criterion (multiply by 1.5, the “inside fence”). Since we are using this analysis to examine dissociations between IS and EF, we used the strict criterion in order to increase confidence that participants identified using this method are “far out outliers” (Tukey, 1977) and are in fact performing outside of the expected range. Once we identified outliers in one domain (i.e., either EF or IS) using the strict criterion (IQR3), we assessed whether any participants performed within normal limits on tasks in the other domain (i.e., the either IS or EF) using the less strict criterion (IQR 1.5). This specific approach allowed us to identify whether participants who are “extreme outliers” (strict criterion) in one domain can show adequate performance (less strict criterion) in the other domain. By using the less strict 1.5 criterion in the second step of the analysis, we aimed to be less stringent regarding what level of performance would qualify as impaired in the second domain. Therefore, dissociations identified via this analysis reflect participants with extremely poor performance in one domain who very clearly perform within normal limits in the other domain.
Stroke survivors (N=59) showed a range of ability across language domains (Figure 1) as measured by the Western Aphasia Battery-Revised (WAB) (Kertesz, 2006) and an in-house picture naming task. Average aphasia severity was in the mild range (WAB AQ = 79.8) but individual participant severity ratings ranged from mild to severe according to WAB scores. Importantly, all participants received a score of at least 4.3/10 on the Auditory-Verbal Comprehension subscore of the WAB, suggesting adequate comprehension for following instructions for the task. Note: six participants scored above the AQ cutoff of 93.8 to qualify as aphasic according to the WAB but are still considered to have aphasia based on self-report and clinician expertise.
Participants also showed a range of abilities in the areas of IS (Figure 2) and EF (Figure 3). Control participants were near ceiling on most tasks. Many of the stroke survivors showed the expected mismatch between inner and overt speech, wherein IS Report scores were often higher than Spoken Naming scores in individual participants (see Supplementary Table 1). Note that IS Report was not administered to controls due to expected ceiling effects.
Notably, several individual stroke survivors showed poor performance on the auditory-based Rhyme Judgment and Syllable Counting tasks. Low scores on these tasks suggest that participants cannot perform the necessary manipulations even when word retrieval is not required. Since we would be unable to meaningfully interpret their performance on the picture-based tasks, we developed the following approach for excluding participants who did not perform above chance on the auditory tasks. Chance-level performance on the yes/no Rhyme Judgment task is 50%, and a binomial distribution reveals that a minimum score of 25/40 (62.5% accuracy) is needed to differentiate performance from chance at a significance value of p < .05. Five participants scored below this cutoff, so were excluded from all subsequent analyses involving the Rhyme Judgment tasks. Chance-level performance on the Syllable Counting task is 33.3% (three choices), and a binomial distribution reveals that a minimum score of 14/30 (46.67% accuracy) is needed to differentiate performance from chance at a significance value of p < .05. Three participants scored below this cutoff, so they were excluded from all subsequent analyses involving the Syllable Counting tasks.
Relationships between IS, EF, and general language ability in stroke survivors are shown in Figure 4, which also reveals intercorrelations among many of the various IS and EF tasks. Importantly, IS self-report (a purely subjective measure of word retrieval) related significantly to the two WAB subscores that emphasize word retrieval (Naming & Word-Finding and Spontaneous Speech) but not to Auditory Comprehension or Repetition, which have limited or no retrieval demands. The relationship between IS Report and WAB Naming & Word-Finding remains significant at a Bonferroni-corrected threshold. The manipulation-and-retrieval measures of inner speech (picture-based rhyme and syllable counting) appear closely related to general language processing on the WAB, with most relationships significant even at a strict Bonferroni-corrected threshold ofp < .0011. Relationships of objective retrieval-based measures to general language processing are also present, but some do not survive correction for multiple comparisons. For EF, the only comparison to reach significance at a strict Bonferroni-corrected threshold of p < .0009 was between conflict resolution (Simon task) and spoken naming ability. However, there were additional significant relationships between general language abilities (including spoken naming and several WAB subscores) and conflict resolution, response inhibition, and task-switching (Standard version only, not Nonverbal).
Conflict resolution (Simon task) was significantly correlated with the three measures of inner speech (Figure 5): two retrieval-based IS measures (IS Report rs = −.270, uncorrected p = .039; Rhyme Judgment subtraction score rs = −.361, uncorrected p = .007) as well as the Picture-based Rhyme Judgment task, which requires both internal phonological retrieval and manipulation (rs = −.351, uncorrected p = .009). Note: a Bonferroni-adjusted significance threshold for this set of five comparisons is p < .01. Additionally, the negative direction of these relationships is due to the fact that lower scores are better on the Simon task measure. In a set of partial correlations controlling for overall language ability via the WAB Aphasia Quotient (Table 1), there are no significant relationships between IS and conflict resolution.
There were no significant relationships between response inhibition (Go-NoGo task) and any measures of IS (Figure 5). In a set of partial correlations controlling for overall language ability via the WAB Aphasia Quotient (Table 1), there are no significant relationships between IS and response inhibition.
As predicted, several of the IS measures related to task-switching in the Standard version of Antelopes & Cantaloupes, which includes traditional, easily nameable shapes as stimuli (Figure 6). No IS measures were significantly related to performance on the Nonverbal version, which includes unfamiliar shapes that are not easily nameable. The two significant relationships between IS and task-switching are approaching significance at a Bonferroni-adjusted threshold of p < .005 (IS Report: rs = −.360, p = .006; Picture-based Syllable Counting rs = −.359, p = .008). Note: the negative direction of these relationships is due to the fact that lower scores reflect better performance across all Antelopes & Cantaloupes measures. In a set of partial correlations controlling for overall language ability via the WAB Aphasia Quotient (Table 1), there are no significant relationships between inner speech and task-switching ability.
No measures of IS related to semantic control as measured on the Antelopes & Cantaloupes task (Figure 7). Two measures of IS related to phonological control (Picture-based Rhyme Judgment rs = −.349, uncorrected p = .01; Rhyme Judgment subtraction score rs = −.268, uncorrected p = .047). In a set of partial correlations controlling for overall language ability via the WAB Aphasia Quotient (Table 1), the Picture-Based Rhyme Judgment task, a manipulation-based measure of inner speech, is still significantly related to phonological control (rs = −.380, uncorrected p = .005).
Results of the outlier analysis (section 2.4) showed that a number of participants scored below normal limits compared to control performance on retrieval-based and retrieval-and-manipulation IS tasks, the Go-NoGo task, and the Simon task. No participants scored outside of normal limits (using the outside fence/extreme outlier criterion) on any measure from the Antelopes & Cantaloupes task; note that a prior task validation analysis in McCall et al., 2021 showed that most adults with post-stroke aphasia can perform the Antelopes & Cantaloupes task with high overall accuracy. Note: the participants who performed below chance performance on the auditory Syllable Counting and Rhyme Judgment tasks were excluded from the dissociation analysis, so N=54 for these analyses.
Overall, 31 participants (57.4%) showed significantly impaired performance on at least one retrieval-based measure of IS (Figure 8; Supplemental Table S2). Of those poor performers, 21 (67.7%) performed well on every measure of EF, demonstrating a dissociation between retrieval-based IS and EF. Forty-seven participants (87%) showed poor performance on at least one manipulation-based measure of IS. Of those poor performers, 28 (59.6%) performed well on every measure of EF, demonstrating a dissociation between manipulation-based IS and EF. Overall, these results show that participants can display adequate EF performance despite frank impairments of retrieval-based and/or manipulation-based IS.
Overall, eight participants (14.8%) showed extremely poor performance on at least one measure of EF: seven on the Go-NoGo task and one on the Simon task (Figure 9; Supplemental Table S3). All eight participants with evidence of significantly impaired EF performed well on at least one measure of retrieval-based IS, and two participants also performed well on at least one measure of manipulation-based IS. These results show that participants can display adequate IS performance despite impairments of EF; however, this dissociation is more robust for retrieval-based IS than for manipulation-based IS, as shown in Figure 9. These results should not be overinterpreted given the small number of participants in each group.
The most basic results from this study show variable performance across language, IS, and EF tasks in this group of individuals with chronic, post-stroke aphasia. These findings align with prior evidence that IS ability varies among individuals with aphasia (see Fama & Turkeltaub, 2020 for a review). Additionally, performance varied across IS tasks for the group as well as within individual participants, which underscores the importance of careful operational definitions of IS in research design. Our results also align with prior work showing that various aspects of EF can be vulnerable in this population and that EF ability can relate to general language abilities (Dutta et al., 2023; Gilmore et al., 2019; Meier et al., 2022), while illustrating that there is individual and task-level variability in this domain as well. Crucially, given the patterns observed in the results, this variability is likely due to inherent differences between specific tasks, i.e., each IS and each EF task is measuring slightly different abilities, as opposed to random variability.
Our primary research question was, “Does inner speech support executive function ability in aphasia?” This question arose from a substantial literature demonstrating that IS supports EF ability in healthy adults, and our results indeed suggest that there is a relationship between IS and various aspects of EF among stroke survivors with aphasia. While some of our findings appear to confirm the role of self-cueing in EF, we also establish consistent relationships between EF and IS when the IS tasks require not simply the generation of internal language (e.g., self-reported accuracy on silent picture-naming) but also some level of manipulation of that internally generated language (e.g., picture-based rhyme judgment). By design, the IS tasks that involve manipulation of internal representations inherently require some elements of EF for successful performance, for example, inhibiting close competitors and/or maintaining and updating working memory when deciding whether two words rhyme. Therefore, we speculate that the relationship between IS and EF is bidirectional (Figure 10): (1) IS supports task-switching and conflict resolution abilities through verbal self-cueing, and (2) phonological control and other EF abilities supports performance on certain IS tasks, particularly those that are commonly used in experimental settings. In this discussion, we will address our specific findings regarding each type of IS and EF task, describe the clinical implications, identify limitations, and suggest future directions for this work. Please note that the relationships that we discuss below should be considered tentative, given that some did not survive multiple corrections comparison. Additionally, many of these relationships between IS and EF did not remain significant in analyses controlling for overall language ability using the WAB AQ. This is unsurprising given that our prior work suggests that inner speech relates strongly to phonological retrieval ability, which is a prerequisite for performance on the Spontaneous Speech and Naming and Word Finding WAB subtests, which together account for 60% of the WAB AQ score. Thus, controlling for the WAB AQ should mask some true relationships between IS and EF, and this finding does not necessarily negate those relationships.
First, we showed that the integrity of IS related to conflict resolution (Simon task) but not to inhibitory control (Go-NoGo). We found relationships between conflict resolution and three measures of IS spanning both retrieval-based and retrieval-and-manipulation definitions. Relationships between conflict resolution and the retrieval-based IS measures suggest that there may be a role for self-cueing during the Simon task. The relationship with Picture-based Rhyme Judgment may also reveal some overlap in the cognitive processes needed to perform conflict resolution and some of our IS tasks. Rhyme Judgment requires one to retrieve and hold two distinct phonological forms in working memory while comparing them, revealing a role of domain-general processes (Classon et al., 2013). A prior study showed that performance on the Simon task did not predict auditory rhyme judgment ability in aphasia (Obermeyer et al., 2020). Here, we found a correlation between the Simon task and the picture-based rhyme judgment task, suggesting that perhaps the relationship between IS and conflict resolution requires the combination of both retrieval and manipulation elicited by our picture-based vs. auditory rhyme judgment task. Importantly, the relationships between IS and conflict resolution were not found to be significant when controlling for overall severity of language impairment, so an alternative interpretation is that IS does not directly support conflict resolution or response inhibition, but rather has an indirect effect due to its role in more general language processing. We did not find any significant relationships between IS and inhibitory control (Go-NoGo). One interpretation may be that the inhibition task requires processing speeds that are too fast to benefit from internal self-cueing, although to our knowledge this has not been empirically tested.
Next, we showed that IS relates to non-linguistic task-switching ability, which is supported by prior research (Emerson & Miyake, 2003). All EF tasks in our study can be considered non-linguistic in nature, as the stimuli are entirely visual/picture-based with no linguistic representations. However, the nature of the stimuli in a non-verbal task can vary in ways that influence one’s ability to apply verbal labels. In this study, we directly compared a task version with familiar, easily nameable shapes (e.g., circle, star) to a version with abstract, unfamiliar shapes. As predicted, we found that IS measures related to task-switching performance only when verbal labels could easily be used to describe the stimuli. These tentative findings do align however with prior suggestions that IS supports such tasks via self-cueing. Emerson & Miyake (2003) varied the level of explicit cueing provided during their arithmetic switching task, with results indicating that inner speech self-cueing is most beneficial when explicit cues are not given, i.e., when the individual must cue themselves to the task. Our results add to this prior finding by suggesting that the nature of the stimuli themselves can also influence whether self-cueing is easily accessible during a given task or not. In other words, our work combines with prior findings to demonstrate that non-linguistic tasks may vary in their likelihood of eliciting IS, due both to the need and accessibility of IS given the specific stimuli and cues.
Next, we showed that IS ability relates to phonological control, but not semantic control, in adults with aphasia. Overall, there is more prior evidence for deficits in semantic control than for phonological control in this population, with some prior work suggesting that phonological control impairments are subtle overall (McCall et al., 2022). However, our finding regarding phonological control aligns with the general understanding of IS as a primarily phonological phenomenon (Alderson-Day & Femyhough, 2015). In a recent paper, McCall et al. (2022) investigated the interactions between switching ability and phonological/semantic control in healthy adults. This study found that inner speech is supported by the ability to manage phonological interference but not semantic interference, which is in line with our findings. The most robust relationship in our findings was between phonological control and picture-based rhyme judgment, which survived multiple comparisons correction and was also significant when controlling for overall severity of language impairment. Rhyme judgment tasks are a common experimental paradigm used as a proxy for measuring IS. Again, our correlational results cannot speak to directionality, but we can speculate that phonological control may contribute directly to the ability to perform rhyme judgment tasks. This finding builds on prior literature suggesting that rhyme judgment tasks require other domain-general cognitive processes such as working memory (Classon et al., 2013).
In our final analyses, we examined dissociations between IS and EF at the individual participant level. The dissociation analysis provided clear evidence that individuals with aphasia can demonstrate adequate EF performance despite impairments of IS. This dissociation held true in analyses using the retrieval-based IS measures as well as the retrieval-and-manipulation IS measures, suggesting that neither form of IS is strictly necessary for EF function (at least as each mental process is operationally defined in this study). Furthermore, this dissociation was true across all EF measures in our study, i.e., for every EF measure there was at least one participant who performed within normal limits despite doing poorly on at least one measure of IS. When looking for dissociations in the opposite direction, we found relatively strong evidence for spared IS ability despite impairments of EF. Notably, several participants with poor EF showed adequate retrieval-based IS, but only two participants who showed poor EF displayed adequate retrieval-and-manipulation IS ability. From these results, one might feel relatively confident that EF is not necessary for retrieval-based IS and that relatively intact IS is not sufficient for EF ability. However, the evidence for whether EF is necessary for manipulation-based IS appears inconclusive based on a single dissociation in one participant.
Generally, the evidence for good EF despite poor IS appears to be stronger than the evidence for good IS despite poor EF. There are several ways to interpret this difference. First, there was only a small number of participants who qualified as having poor EF performance as we defined it, so we had less opportunity to examine the range of IS performance in this group of outliers. It may be the case that in a larger sample or in a sample with more participants who have poor EF ability, we would have found a different result. Additionally, though, the lack of strong evidence for a dissociation between poor EF and intact performance on the manipulation-based IS measures requiring manipulation of internal representations is not actually surprising given that these experimental measures of IS are understood to require some EF ability by design (e.g., working memory, phonological control, decision-making). Finally, one minor additional limitation in interpreting the dissociation analysis is that our approach revealed disparities not only between IS and EF but also within IS and within EF, as a participant could be categorized as a poor performer on only one or a subset of tasks within each of the two domains. This potential concern is minimized (1) the common understanding of EF as an umbrella term encompassing many related but distinguishable subprocesses, and (2) our own prior work and the work of others revealing the breadth of approaches to defining/measuring IS across the literature, suggesting that IS too is a multidimensional phenomenon rather than a single process. The variability of performance across tasks within the IS and EF domains, therefore, should not detract from our ability to interpret the implications of the dissociations we have revealed.
The study findings, although preliminary in nature, have interesting clinical implications. As speech-language pathologists become more aware of concomitant impairments in non-linguistic cognition among people with aphasia, there is growing attention to addressing these issues in therapy alongside more “traditional” language intervention. Clinical decision-making in this area would be improved by a better understanding of the mechanism linking language and EF deficits in aphasia. For example, identifying that deficits in EF (e.g., inhibiting semantic or phonological competitors) were independently contributing to measurable language impairments might lead a clinician to treat the EF skills directly, in the hopes that certain language skills would improve as a result (Bontemps et al., 2024; Frankel et al., 2007; Kuzmina & Weekes, 2017; Ramsberger, 2010). In contrast, the knowledge that a client’s EF impairments were actually caused (or at least influenced) by deficits in IS and other language processes might lead clinicians to continue focusing primarily on language treatment. Importantly, the interpretation of our findings applies most directly to the way we have operationally defined EF and IS in our experimental design, and further research would be needed in order to understand the relationship between IS and EF in more naturalistic, everyday contexts both for individuals with aphasia and all language users.
Overall, the results from these analyses show a relationship between IS and various aspects of EF ability (including task-switching, response inhibition, and conflict monitoring) in adults with aphasia. However, there are some limitations that should be addressed in future studies. One potential limitation is the use of motor-based tasks for measuring EF, as many of our stroke survivors presented with right-sided hemiparesis and used their non-dominant left hand for task completion. The use of a weak or non-dominant hand may have impacted reaction time on our tasks, but our EF measures are primarily within-task comparisons (e.g., standard vs. nonverbal switch cost on Antelopes & Cantaloupes), providing an internal control for hand use in our measures of interest. Another limitation of the study design that we have noted throughout the discussion is that these analyses are all correlational, so the findings cannot speak to the directionality of the relationship between IS and EF. Our work does contribute to prior literature showing that IS supports various EF processes via articulatory suppression paradigms (e.g., Baddeley et al., 1984; Emerson & Miyake, 2003; Wallace et al., 2017). Finally, our use of subjective reports as one measure of successful inner speech may lead to some concern about the accuracy of participants’ insights. In this study, we required that participants demonstrate an adequate level of auditory comprehension ability for understanding task instructions. Our prior work does show that most people with aphasia can reliably report their inner speech (Fama, Snider, et al., 2019), but it remains possible that some individual participants may still have over-reported the accuracy of their inner speech.
Additional open questions in this line of research include relationships between IS and other aspects of EF not examined here, as well as relationships between EF and IS processing based on written word stimuli rather than pictures or auditory stimuli, as are used in these experiments. It would also be interesting to incorporate a self-report component regarding the use of IS during these experimental tasks, in order to examine conscious use of IS as a strategy and whether some individuals may not use IS in a given task even if it is potentially accessible to them. A questionnaire about IS in everyday life may also elucidate the role of IS in supporting EF during day-to-day tasks. Additionally, we used a theory-driven approach to focus on individual relationships between various aspects of EF and IS, but future analyses of these (or similar) data could utilize summary scores or a principal components analysis in a more data-driven approach. Overall, future studies examining these types of open questions will continue to clarify the mechanistic relationship between language, IS, and EF. These findings have clinical implications for post-stroke aphasia and may ultimately improve our understanding of (and ability to enhance recovery from) the challenges faced by this population.
IS and EF abilities vary within and across individuals with post-stroke aphasia. There is consistent evidence for a relationship between these two processing domains that extends beyond the relationship between EF and general language processing, as well as evidence that IS and EF ability can dissociate from one another in both directions. Although our results are correlational in nature and there is some evidence for indirect effects, we suggest that the relationship between IS and EF may be, to some extent, IS can support non-linguistic EF ability via self-cueing, and EF can support IS ability when the IS task requires domain-general processes such as working memory and/or phonological control. Future studies will continue to elucidate the relationship between language, IS, and EF abilities, which has the potential to advance cognitive theory as well as clinical practice.