Dysfunctional Tissue Correlates of Unrelated Naming Errors in Acute Left Hemisphere Stroke

Abstract

Most naming error lesion-symptom mapping (LSM) studies have focused on semantic and/or phonological errors. Anomic individuals also produce unrelated word errors, which may be linked to semantic or modality-independent lexical deficits. To investigate the neural underpinnings of rarely-studied unrelated errors, we conducted LSM analyses in 100 individuals hospitalized with a left hemisphere stroke who completed imaging protocols and language assessments. We used least absolute shrinkage and selection operator regression to capture relationships between naming errors and dysfunctional brain tissue metrics (regional damage or hypoperfusion in vascular territories) in two groups: participants with and without impaired single-word auditory comprehension. Hypoperfusion—particularly within the parietal lobe—was an important error predictor, especially for the unimpaired group. In both groups, higher unrelated error proportions were associated with primarily ventral stream damage, the language route critical for processing meaning. Nonetheless, brain metrics implicated in unrelated errors were distinct from semantic error correlates.

Introduction

The incidence of aphasia, a neurogenic communication disorder characterized by receptive and/or expressive language deficits, ranges from approximately 18–30% of the nearly 15 million individuals worldwide who experience a new stroke each year (Benjamin et al., 2019; Flowers et al., 2016; Laska et al., 2001; Pedersen et al., 1996). The specific deficits that individual patients exhibit vary from person to person, yet impaired word retrieval, otherwise known as anomia, is ubiquitous across all people with aphasia (Goodglass & Wingfield, 1997). In clinical settings, word retrieval is typically assessed with object picture naming tests. Theoretical models of lexical access (e.g., Dell et al., 1997; Dell & O’Seaghdha, 1992; Foygel & Dell, 2000; Indefrey & Levelt, 2004; Levelt et al., 1999; Rapp & Goldrick, 2000; Schwartz et al., 2006; Ueno et al., 2011) differ in the number and nature of processing stages, yet all models propose that picture naming includes semantic and phonological processes and that errors produced during naming attempts, along with patterns of performance across lexical tasks, can be used to determine locus of impairment. In addition to providing robust deficit information for treatment planning purposes, anomia profiles can be—and have been—leveraged to investigate the neurobiology of language through lesion-symptom mapping (LSM).

Most previous LSM studies of naming errors have focused on the origins of semantic and/or phonological errors. Semantic paraphasias are traditionally defined as single-word errors that are conceptually related to target items in taxonomy (e.g., “animal” or “horse” for cow) or by association (e.g., “milk” for cow). Some between-study differences in specific lesion foci exist (Cloutman et al., 2009; Fridriksson et al., 2018; Halai et al., 2018; Mirman, Chen, et al., 2015; Mirman, Zhang, et al., 2015; Schwartz et al., 2009; Stark et al., 2019; Walker et al., 2011), yet semantic naming errors are thought to originate from damage to anterior to mid temporal regions, damage to posterior temporal into parietooccipital regions, or anterior-posterior disconnection (see e.g., McKinnon et al., 2018), likely depending on the underlying component of the naming process that is impaired. Phonological errors—errors that have some degree of phonemic overlap with target items (e.g., “dard” for dart)—have been linked to dysfunctional brain tissue (i.e., frank damage or hypoperfusion) in the left precentral gyrus (LPrCG) as well as premotor cortex and left supramarginal gyrus (LSMG) (Chen et al., 2019; Cloutman et al., 2009; Halai et al., 2018; McKinnon et al., 2018; Schwartz et al., 2012; Stark et al., 2019; but cf. Fridriksson et al., 2018). The neural correlates of errors reported in these studies generally cohere with a dual-stream language network architecture (Hickok & Poeppel, 2000, 2004, 2007). Specifically, within the dual-stream model, regions primarily inferior to the Sylvian fissure comprise the ventral stream, involved in mapping auditory inputs to meaning, whereas regions mostly superior to the Sylvian figure comprise the dorsal stream involved in mapping speech signals to articulatory processes.

While the majority of LSM naming error studies have focused on semantic and phonological error types, people with aphasia often produce errors that are unrelated to target items, such as single-word unrelated verbal paraphasias (e.g., “boy” for helicopter), perseverations (i.e., production of a previous response), and omissions (i.e., naming attempt failures). Unlike semantic and phonological errors, unrelated errors—and omissions in particular—carry no information that elucidates their origins. Dell et al. (2004) found that the best theoretical and predictive model of omissions was a two-step semantic-phonological model (Foygel & Dell, 2000) that assumed omissions occur due to sub-threshold activation of any lexical item (Matti et al., 1998), resulting in a failure to correctly or incorrectly name the target. More recently, Chen et al. (2019) proposed different potential underlying causes of omissions, including 1) breakdown within the semantic system or disconnection of semantic and lexical layers, 2) executive dysfunction causing mis-selection of competing activated representations, 3) post lexical selection impairments in phonological retrieval, assembly, or motor speech skills, and 4) deficits outside the core naming processes (e.g., visual object recognition). In their study, Chen and colleagues (2019) found that omission error rates were highest in patients with impaired lexical-semantic mapping during naming, aligning with the first deficit profile they outlined. Furthermore, the lesion correlates of omissions—located in the left middle frontal gyrus (LMFG), left inferior frontal gyrus, pars triangularis (LIFGtri), the anterior to mid-left middle temporal gyrus (LMTG), and the temporal pole—overlapped with lesion clusters for semantic errors and nonverbal associative semantic deficits but not phonological errors.

In another chronic stroke cohort, Halai and coworkers (2018) implemented a data-driven approach to combine error and neuropsychological assessment data. They found significant lesion clusters associated with omission errors within the left anterior temporal lobe (LATL) that extended into temporal deep white matter and were distinct from phonological error clusters. Stark and colleagues (2019) investigated naming errors during object picture naming and connected speech tasks and discovered that lesion clusters of unrelated verbal paraphasias and semantic errors overlapped within the anterior to posterior LMTG. However, unrelated errors extended into anterior and more inferior temporal regions compared to the slightly more posterior and superior clusters for semantic errors. In all, these few studies implicate ventral stream left anterior to middle temporal lobe in the production of unrelated and omission errors, which may indicate these errors have a semantic origin.

Despite consistent findings between these few studies, further exploration into the origins of different types of unrelated errors is warranted. Replication is useful in LSM, especially considering the inherent variability in post-stroke samples in terms of lesion distributions and deficit profiles. Moreover, none of the studies investigated unrelated verbal paraphasias and omissions within the same sample. It is also noteworthy that prior LSM studies of unrelated errors were conducted using data from patients with chronic aphasia, which differ from LSM of acute post-stroke stage (e.g., < 2 weeks following stroke onset) in terms of theoretical motivations and methodological considerations (cf. Ding et al., 2020). By the chronic post-stroke stage, neural reorganization of language has occurred to compensate for brain damage and dysfunction (Cramer, 2008; Hillis, 2007; Hillis & Heidler, 2002; Kiran, 2012), and thus, behavioural deficit correlates identified using chronic-only lesion data may not align with the neural correlates of those same processes in a neurotypical population. Therefore, it may be more apt to interpret chronic LSM findings as indices of recovery (de Haan & Karnath, 2018). In acute stroke, such reorganization has not yet occurred, and lesion correlates may be more likely to reflect the undamaged language network.

Part of the aforementioned variability in post-stroke samples relates to stroke recurrence. Approximately 20% of stroke survivors have suffered more than one stroke, indicating that individuals with a history of multiple strokes are representative of a substantial portion of the larger stroke population (see Benjamin et al., 2019). Nonetheless, most investigations involving stroke patients exclude individuals with multiple strokes, resulting in greater sample homogeneity but potentially limited generalization of findings. In general, the relationship between stroke recurrence and language abilities is understudied. In a recent study by our group, Goldberg et al. (2021) found that cumulative brain damage (indexed by total acute + chronic lesion volume) was significantly related to overall language abilities at the acute post-stroke stage, but stroke recurrence as a dichotomous classification (i.e., history/no history of prior stroke) was not. In this sample (which overlaps with the present study sample), prior strokes were asymptomatic, and lesions were from small, lacunar infarcts for many participants within the recurrent stroke group. Given this fact, we believe that the results of the present study will likely reflect brain-behavior relationships before language reorganization (aligning with the advantages of LSM in acute stroke). The Goldberg et al. (2021) findings suggest that including acute stroke survivors with a history of prior stroke in LSM analysis may be reasonable as long as the contributions of new and previous stroke lesions are carefully accounted for in analyses. Therefore, while the focus of this study pertained to language deficits after acute stroke, some of our cohort (i.e., 35% of the sample) had a history of prior infarct as well.

In this study, we aimed to investigate the locus of omissions and other unrelated errors. The production of unrelated errors may be specific to expressive language or may reflect lexical deficits independent of modality. Individuals with multimodality access deficits to semantics or impaired semantic representations make semantic errors in receptive tasks in addition to errors on different types of production tasks (e.g., confrontation naming). On the other hand, individuals with intact semantic representations but deficient access to lexical representations during language production will show normal performance on single-word receptive tasks (e.g., auditory word-to-picture matching). Given that unrelated errors have a suspected semantic locus, patients with modality-independent lexical deficits (i.e., those who demonstrate impaired single-word production and comprehension) may exhibit a different error and neural correlate profile for unrelated errors than patients with single-modality (i.e., production-only) impairments. As such, we compared LSM results of unrelated errors in individuals with impaired single-word auditory comprehension skills to results from individuals with intact comprehension. We also conducted LSM analyses of semantic errors in these two groups to determine potential spatial overlap in unrelated and semantic error findings. If unrelated errors in both groups reflect a failure to access lexical representations from impaired semantic representations, irrespective of modality, the lesion correlates for these errors would be similar between the two groups. On this assumption, we hypothesized that both semantic and unrelated errors would be associated with damage to ventral stream structures inferior to the Sylvian fissure, consistent with the dual-stream model of language (Hickok & Poeppel, 2000, 2004, 2007) and the prior chronic-only stroke literature (Chen et al., 2019; Halai et al., 2018; Stark et al., 2019).

Materials and Methods

Participants

Patients

We reviewed records for 396 patients who were hospitalized within 48 hours of experiencing a suspected new left hemisphere stroke and subsequently enrolled in an ongoing project investigating post-stroke outcomes. We excluded participants who were ultimately diagnosed with an acute event other than left hemisphere stroke (e.g., transient ischemic attack), had uncorrected vision or hearing difficulties, lacked premorbid proficiency in English, could not complete testing protocols for medical (e.g., minimal consciousness) or other (e.g., early discharge) reasons, had contraindications for MRI, or were previously diagnosed with a neurological condition affecting the brain other than stroke. As previously referenced, prior stroke in itself did not warrant exclusion, nor did some lesioned tissue outside of the left hemisphere. However, we did exclude patients with substantial (> 20%) damage to any cortical right hemisphere region, given that our aims and hypotheses pertained primarily to left hemisphere dysfunctional tissue correlates of naming errors. Finally, only participants who produced errors during object naming were included in the study, resulting in a final patient sample of 100 individuals (48 women; 89 right-handed participants; average age: 60.74 ± 13.47 years; average education: 13.89 ± 2.97 years).

Within the patient sample, 35 patients had a history of prior stroke, mainly small lacunar infarcts in either cerebral hemisphere. Individuals with a history of multiple strokes did not differ from individuals with a single stroke event in any demographic, language, or stroke variable, p ≥ 0.200 (see Supplemental Table 1). Despite the lack of between-group differences in key stroke and language variables, we still took measures to mitigate the effects of stroke history on LSM results, as described in subsequent sections.

Controls

We also recruited a group of 83 neurologically-intact healthy controls to determine impairment cut-offs on our two measures of single-word auditory comprehension skills (see below). Control participants had no history of neurological conditions affecting the brain or uncorrected vision or hearing difficulties.

While our full control sample did not differ from patients in most demographic variables, controls had significantly more years of education than patients, t_177.35 = 5.21, p < 0.001. We found that higher education was associated with better single-word auditory comprehension on one test (i.e., the Boston Diagnostic Aphasia Examination-3 Word Comprehension) in the control group, r = 0.33, p = 0.003. Thus, we reduced the control sample until we obtained the largest possible group (n = 50; 26 women; 46 right-handed participants; average age: 62.90 ± 13.12 years; average education: 14.66 ± 2.23 years) that matched patients on all demographic variables (i.e., age, education, sex, and handedness, p ≥ 0.084) (see Supplemental Table 2).

Informed Consent

The Institutional Review Board of Johns Hopkins University approved all study procedures. Controls and patients who could make their own healthcare decisions provided their written informed consent to study protocols. For other patient participants, a medical proxy consented to study procedures.

Language Assessment

To assess object naming abilities, we used the shortened, 30-item version of the Boston Naming Test (BNT) (Fisher et al., 1999). For each item, participants saw a line drawing of an object and attempted to say aloud the name of the target. We first scored responses as being either correct or incorrect. Next, one of the study authors (CRK), a speech-language pathologist, coded patients’ errors on incorrect items as belonging to one of 10 error classifications from the Boston Diagnostic Aphasia Examination, 3^rd edition (Goodglass et al., 2001). As shown in Table 1, these narrow error classifications were aggregated into broad error types, i.e., semantic, phonological, or unrelated errors. Note that perceptual errors were coded as “Other” given that they arise due to breakdown outside of core semantic and phonological stages of processing. Error coding according to the broad error types was performed by another speech-language pathologist (EBG) who was blinded to the first coder’s classifications. Inter-rater reliability was excellent between coders, κ_w = 0.935, z = 26.60, p < 0.001. Error types with example errors are listed in Table 1.

Table 1.

Error classifications and examples

Broad classification	Narrow classification	Definition	Example
Semantic errors	Related verbal paraphasia	A real word that is semantically related to the target	“broccoli” for asparagus; “Egypt” for pyramid
	Circumlocution	A multi-word utterance that describes the target, typically a semantic feature of the target	“used them in school” for protractor; “bird that likes fish” for pelican
	Mixed paraphasia	A real word error that is both semantically and phonologically related to the target but likely arises to due to a breakdown in semantics	“elevator” for escalator
Phonological errors	Nonword phonemic paraphasia	A nonword that shares at least 50% of the target item’s phonology (including phonemic transpositions within one position of the target position)	/bivedɚ/ for beaver; /ɪblu/ for igloo
	Formal error	A real word that shares at least 50% of the target item’s phonology	“camera” for camel; “trial” for trellis
Unrelated errors	Unrelated verbal paraphasia	A real word that is neither semantically nor phonologically related to the target	“house” for pencil; “card” for “helicopter”
	Omission	A single- or multi-word utterance that does not include any target-relevant information and is not considered a paraphasia	“I don’t know”; “can’t do it”; “no no…no”
	Multi-word paraphasic error	A multi-word utterance including a paraphasic and/or paragrammatic error	“scares out of everyone in there” for volcano
	Perseveration	A production of a prior target or a previously-uttered paraphasia	“pencil” for whistle (which follows pencil)
	Neologism	A nonword paraphasia that retains less than 50% of the target item’s phonology	/bɔɪn/ for hanger; /tʌmsɪs/ for protractor
Other	Perceptual error	A real word of another item that resembles the target picture	“branch” for asparagus; “ghost” for octopus

Every BNT item could not be coded for every patient. Some missing data occurred due to transcription errors, but in other instances, assessments were terminated before all items were administered due to severe anomia (per the BNT administration rules). As a result, the mean percentage of missing data was 11.63 ± 17.94% of trials across the patient sample. To control for missing data as well as overall anomia severity, the proportion of each error type (i.e., the total number of trials/the total number of opportunities) was used as the dependent variable in LSM analysis. However, error type proportions are also influenced by the raw number of each type of error the patient produced and do not solely account for missing data. Therefore, we also included the percentage of coded errors for each individual as an additional regressor in LSM models.

We used the 14- or 15-item shortened form of the three pictures version of the Pyramids and Palm Trees Test (PPTT; Breining et al., 2015) to assess nonverbal semantic association abilities in patients. For each item, participants saw three line drawings (one picture at the top and two pictures situated below it) and selected the bottom picture that was related to the top drawing. We assessed single-word auditory comprehension abilities in both groups using either the Auditory Word Recognition subtest from the Western Aphasia Battery-Revised (WAB-R; Kertesz, 2007) or the Word Comprehension subtest from the Boston Diagnostic Aphasia Examination-3 (BDAE-3; Goodglass et al., 2001). In each of these tests, participants were required to select the target object or pictured item or point to the body part that matched a word spoken by the examiner. Note that testing for patients was completed in person, but healthy control data were acquired virtually due to restrictions imposed by the COVID-19 pandemic.

As expected, patients had significantly poorer naming and single-word auditory comprehension skills than controls, p < 0.001 (see Supplemental Table 2). Patients were classified as having impaired single-word auditory comprehension if they scored more than two standard deviations below the control mean (i.e., < 92.48% on BDAE-3 Word Comprehension and < 98.62% on WAB-R Auditory Word Recognition).

Neuroimaging Data

As part of routine clinical diagnostic imaging, patients completed an MRI scanning protocol that included diffusion-weighted imaging (DWI), fluid attenuated inversion recovery (FLAIR), and T1-weighted sequences obtained on either a 1.5T or 3T Siemens scanner. Using MRIcron (https://www.nitrc.org/projects/mricron), trained lab technicians demarcated acute lesions by manually tracing hyperintense areas on every slice of the DWI for each participant. For patients with a history of prior stroke, technicians completed slice-by-slice tracings of chronic lesions—appearing as dark areas of cavitation, often surrounded by hyperintense gliosis—on FLAIR images. All images and lesion maps were warped to MNI space using routines from SPM12 (Statistical Parameter Mapping; https://www.fil.ion.ucl.ac.uk/spm/software/spm12/) and the Clinical Toolbox (Rorden et al., 2012). We checked normalization accuracy by visually assessing the alignment of warped images and overlaid lesion maps to the template images.

We used NiiStat (https://www.nitrc.org/projects/niistat/) to calculate prior and acute stroke volumes and determine the percentage of damaged tissue within dorsal and ventral stream regions of interest (ROIs). Dorsal stream ROIs included LMFG, LIFGtri, LIFG, pars opercularis (LIFGop), left anterior insula (LaInsula), left posterior insula (LpInsula), LPrCG, left postcentral gyrus (LPoCG), LSMG, and the left arcuate fasciculus (LAF). Ventral stream ROIs included LIFG, pars orbitalis (LIFGorb), left superior temporal pole (LSTpole), left middle temporal pole (LMTpole), left superior temporal gyrus (LSTG), LMTG, left inferior temporal gyrus (LITG), the posterior portions of each temporal gyri (LpSTG, LpMTG, and LpITG), left angular gyrus (LAG), left fusiform gyrus (LFuG), and the left fronto-occipital (LIFOF), uncinate (LUF), and inferior longitudinal (LILF) fasciculi. See Supplemental Figure 1 for a visualization of study ROIs.

Given that a substantial percentage of our sample (35%) had a history of prior stroke, we took extra measures to ensure stroke history did not disproportionately affect the LSM results. In general, most prior stroke lesions were small and trended towards being smaller than acute lesion volumes, p = 0.076 (mean acute volume: 19.85 ± 52.58 cc; mean prior stroke volume: 5.07 ± 8.86 cc). To account for the extent and location of brain damage due to multiple strokes, we merged the normalized acute and prior stroke lesion maps into one total lesion volume map using MarsBaR (Brett et al., 2002). These lesion maps were used in the aforementioned NiiStat calculations of percent damage to ROIs. We also included total lesion volume (split by hemisphere) as a regressor in all LSM analyses.

Figure 1 shows the overlap map of all patients’ lesions (A) as well as separate overlap maps of patients with unimpaired (B) and impaired (C) comprehension. It should be noted that lesion location varied from person to person, not just for individuals with multiple strokes but across the entire sample. Not all participants experienced acute strokes confined to the left cerebral hemisphere, and some patients had prior infarcts in other regions of the brain. Within the full sample of 100 patients, 19 individuals had some damage in the right cerebral hemisphere (but less than left hemisphere involvement), and 17 individuals had damage to structures outside of the cerebrum (i.e., brainstem and/or cerebellum) in addition to the left hemisphere. Critically, BNT total accuracy and broad error type totals did not significantly vary by stroke location within the entire patient sample, p ≥ 0.194 (see Supplemental Table 3).

Figure 1.

Lesion overlap map. Overlay of lesions across (A) the entire sample of patients (n = 100) with minimal overlap indicated by cool colours and maximal overlap (n = 10) indicated by warm colours. Lesion overlap maps of the groups of patients with (B) unimpaired and (C) impaired single-word auditory comprehension skills are also shown with maximal overlap of n = 6 patient lesions per group.

A major additional consideration for LSM in acute stroke is the role of hypoperfusion in language impairment. In the aftermath of acute ischemia, low cerebral blood flow in areas surrounding the infarct core (i.e., the ischemic penumbra) renders those tissues dysfunctional. The ischemic penumbra can extend far beyond the infarct borders, and in some patients, may explain acute deficit severity better than the infarct core itself (Beaulieu et al., 1999; Hillis et al., 2002, 2004, 2006, 2008; Hillis, Kane, et al., 2001; Hillis, Wityk, et al., 2001; Sorensen et al., 1996). As such, a critical component of LSM in acute stroke is to account for the ischemic penumbra in some fashion. Hypoperfusion can be quantified by metrics such as delayed time-to-peak or reduced regional cerebral blood flow derived from perfusion-weighted imaging (PWI) sequences. However, PWI is not included in all clinical MRI scan protocols for different reasons, including time constraints and contraindications to intravenous contrast (e.g., allergy, renal failure) (Beckett et al., 2015; Clement et al., 2018; Schieda et al., 2018). This was true of a significant portion of our sample.

Therefore, we measured hypoperfusion not from PWI-derived metrics but instead by visual assessment of the vasculature on FLAIR images. Blood vessels that supply hypoperfused tissue are hyperintense on T2-weighted sequences because the flow void characteristic of blood-rich vessels has lessened. We used the NIH-FLAIR Hyperintense Vessel (FHV) rating system (Reyes et al., 2017) to determine the number of hyperintense vessels within six left hemisphere regions, including the anterior cerebral artery (ACA), posterior cerebral artery (PCA), and middle cerebral artery (MCA) territories, with the MCA split by lobe into frontal, insular, temporal, and parietal lobes. Using this scoring system, raters count the number of FHVs and provide a score for each territory on a scale from 0 to 2, where 0 = no FHVs, 1 = one to two FHVs within a given slice and no more than two slices contain FHVs, and 2 = three or more total FHVs or three or more slices that contain FHVs. Two raters each scored a portion of the patient sample (n = 25), resulting in a mean inter-rater reliability across territories of 0.829, range: κ_w = 0.643–1.00, all p ≤ 0.001.

Statistical Analyses

All analyses were conducted in RStudio (R Core Team, 2020). We first investigated relationships between error types across the sample using Spearman correlations and compared patients with intact and impaired auditory comprehension on a variety of factors using Wilcoxon Rank Sum tests for continuous variables and Fisher’s exact tests for categorical variables (correction done at the false discovery rate, q < 0.05). Because of the inclusion of the FHV ratings, traditional voxel-based LSM methods could not be done. Instead, we used Least Absolute Shrinkage and Selection Operator (LASSO) regression (Tibshirani, 1996) to determine the dysfunctional tissue correlates (captured by the percentage of damage to ROIs, FHV ratings, and right and left total lesion volumes) of error type totals within each patient group. LASSO is a multivariate regression approach that incorporates shrinkage parameters and variable selection to arrive at a sparse, optimal model. LASSO regression is advantageous for LSM analysis because it handles multicollinearity between variables—which is typical of adjacent ROIs that share variance because of co-occurring damage—by retaining only the best, most explanatory predictors and setting other predictors to zero (Meinshausen & Yu, 2009). We conducted LASSO regressions as implemented in the glmnet package in R (Friedman et al., 2010), using a leave-one-out cross-validation procedure with 5000 permutations to arrive at the lambda that resulted in the minimal mean error per model.

For each patient group, we ran one LASSO regression for each semantic and unrelated error type (broad and narrow classifications). Operating under the assumption that greater damage and/or hypoperfusion causes greater impairment (and not the inverse), we conducted one-tailed tests. As referenced previously, naming errors were not recorded for all BNT items for some individuals. As such, in addition to using error proportions (rather than total number of errors), we included the percentage of coded errors as a covariate in all analyses. We also included total left hemisphere lesion volume, total right hemisphere lesion volume, age, and education as covariates. To ensure the results were not driven by outliers, we only included ROI and FHV rating variables for which at least 10% of the sample had some value indicating damage or dysfunction. Similarly, we did not conduct LASSOs for error types that were produced by fewer than 10% of the patients in the sample.

Results

Of the total number of errors produced across the sample (n = 958), 41.65% were semantic, 12.00% were phonological, and 45.93% were unrelated errors. Fewer than 10% of participants produced mixed errors, multiword paraphasias, or perceptual errors. Figure 2A shows relationships between the remaining error classifications. Pertaining to unrelated errors specifically, we found that a greater proportion of unrelated verbal paraphasias and omissions was associated with higher proportions of perseverations and neologisms, r ≥ 0.326, q < 0.003. Production of most unrelated errors was not significantly related to the production of related verbal paraphasias or other phonological errors, q > 0.05. As the exceptions, greater proportions of circumlocutions and omissions were weakly related, r = 0.270, q = 0.023, and a higher proportion of neologisms was associated with a higher proportion of formal errors, r = 0.241, q = 0.044. In general, however, patients who produced unrelated errors tended to produce only unrelated error subtypes (i.e., omissions, unrelated verbal paraphasias, perseverations, or neologisms).

Figure 2.

Characterization of unrelated errors in the sample, including A) a correlation matrix illustrating relationships between unrelated error subtypes and other types of errors (red-shaded rho values indicate significant results at FDR-corrected q < 0.05) and B) differences in the proportion of unrelated errors produced by individuals with intact versus impaired single-word auditory comprehension skills.

Table 1 includes a summary of demographics, language abilities, and broad lesion characteristics for patients with spared (n = 71) versus impaired (n = 29) single-word comprehension skills. Patient groups did not significantly differ in age, years of education, or sex or handedness distributions. Patients with impaired auditory comprehension had significantly poorer confrontation naming and nonverbal semantic association skills compared to patients with intact word comprehension, q < 0.001. While patients with impaired single-word auditory comprehension had significantly larger lesions than the unimpaired group, q < 0.001, there was no between-group difference in the number of participants who had a history of prior stroke, q = 0.308.

Despite this finding, it is possible that lesions outside the left hemisphere and stroke recurrence differentially impacted naming deficits and the proportion of error types in the two comprehension groups. Therefore, we conducted a series of two-way analyses of covariance (ANCOVAs) to determine whether there was an interaction effect between auditory comprehension group status and stroke group status (either stroke location or stroke recurrence). The interaction of stroke location (i.e., left cerebral hemispheric damage only, bilateral cerebral hemispheric damage, or other [i.e., damage outside of the cerebrum in addition to left cerebral involvement]) and auditory comprehension (i.e., impaired or unimpaired) was not significantly related to overall naming or any broad error type proportions, p ≥ 0.383 across tests (see Supplemental Table 4). The interaction of recurrent stroke status (i.e., single acute stroke or history of prior stroke) and auditory comprehension status was not significant for the proportion of overall semantic errors, F_1,95 = 0.138, p = 0.711. In contrast, the interaction term was significant for total naming accuracy, F_1,96 = 5.490, p = 0.021, proportion of total phonological errors produced, F_1,95 = 5.267, p = 0.023, and most critically, proportion of total unrelated errors produced, F_1,95 = 7.670, p = 0.007 (see Supplemental Table 5 for full results). Because of these findings, we controlled for recurrent stroke status in all LSM analyses.

The proportion of unrelated errors produced by people with intact versus impaired single-word comprehension skills was significantly different, W = 536.5, q < 0.001 (see Figure 2B). Specifically, people with impaired auditory comprehension produced significantly greater proportions of unrelated verbal paraphasias, W = 686.5, q < 0.001, omissions, W = 764, q = 0.015, perseverations, W = 723.5, q < 0.001, and neologisms, W = 754.5, q = 0.005, than did people with intact single-word comprehension. Nonetheless, unrelated verbal paraphasias, omissions, and neologisms comprised a certain proportion of errors in at least 10% of participants with and without word-level auditory comprehension deficits, so we investigated the neural correlates of these errors within each group.

Areas of Dysfunction Associated with Errors in Each Group

Of note, damage to LMTpole and hypoperfusion in the left ACA territory were each noted in fewer than 10% of participants, and thus, these variables were excluded from all LASSO regression models. In LASSO regression, the combination of retained variables regardless of independent significance provides meaningful information about the predicted outcome; yet, for simplicity, the retained metrics significant at p < 0.05 are described below and depicted in Figure 3. The full LASSO regression results for all unrelated and semantic error proportions are reported in Table 3 for patients with unimpaired comprehension and Table 4 for patients with impaired comprehension.

Figure 3.

Visualization of dysfunctional tissue correlates of unrelated errors. Regions significantly associated (p < 0.05) with each error type proportion are shown, with results for individuals with unimpaired single-word auditory comprehension in the top row and results for individuals with impaired comprehension shown in the bottom row. Red and purple shading correspond to significant hypoperfusion and regional damage metrics, respectively. Not depicted are other variables retained within LASSO models that provide additional meaningful information. AG = angular gyrus, FHV = FLAIR Hyperintense Vessel ratings, IFGorb = inferior frontal gyrus, pars orbitalis, ILF = inferior longitudinal fasciculus, ITG = inferior temporal gyrus, L = Left, MCA = middle cerebral artery, MTG = middle temporal gyrus, p = posterior, PCA = posterior cerebral artery, STG = superior temporal gyrus, STpole = superior temporal pole, UF = uncinate fasciculus.

Table 3.

Dysfunctional tissue correlates in patients with intact single-word auditory comprehension (n = 71)

	UNRELATED								SEMANTIC
	All Unrelated Errors		Unrelated Verbal Paraphasias		Omissions		Neologisms		All Semantic Errors		Related Verbal Paraphasias
	Est.	p	Est.	p	Est.	p	Est.	p	Est.	p	Est.	p
% of coded errors	0.062	0.227	--	--	0.123	0.072	--	--	--	--	--	--
Stroke group (SE)	0.013	0.532	--	--	--	--	--	--	--	--	--	--
% damage LMFG	0.088	0. 078	--	--	0.044	0.082	0.017	0.057	--	--	--	--
% damage LIFGorb	0.258	0.059	--	--	0.310	0.029	0.029	0.091	--	--	--	--
% damage PrCG	0.013	0.086	--	--	--	--	--	--	--	--	--	--
% damage LpSTG	--	--	0.098	0.014	--	--	--	--	--	--	--	--
% damage LpMTG	0.384	0.003	0.322	0.003	0.227	0.013	0.230	0.002	--	--	--	--
% damage LITG	0.095	0.040	0.067	0.040	0.157	0.031	--	--	--	--	--	--
% damage LpITG	--	--	0.060	0.005	--	--	--	--	--	--	--	--
% damage LFuG	0.024	0.114	--	--	0.011	0.105	--	--	--	--	--	--
% damage LAG	--	--	0.163	0.009	--	--	--	--	--	--	--	--
FHV L PCA	0.128	0.070	--	--	0.167	0.045	--	--	--	--	--	--
FHV L MCA-frontal	0.150	0.048	--	--	0.169	0.027	--	--	--	--	--	--
FHV L MCA-insula	0.093	0.111	--	--	0.110	0.095	--	--	--	--	--	--
FHV L MCA-temporal	0.138	0.043	--	--	0.150	0.028	--	--	--	--	--	--
FHV L MCA-parietal	--	--	--	--	--	--	--	--	0.195	0.001	0.151	0.004

Notes: A dashed line indicates the predictor was not retained in the final LASSO model. Bolded text indicates a significant predictor at p < 0.05. Abbreviations: AG = angular gyrus, Est. = estimate, FHV = FLAIR hyperintense vessels, FuG = fusiform gyrus, IFGorb = inferior frontal gyrus, pars orbitalis, ITG = inferior temporal gyrus, L = left, MCA = middle cerebral artery, MFG = middle frontal gyrus, middle temporal gyrus = MTG, p = posterior, PCA = posterior cerebral artery, PrCG = precentral gyrus, SE = single event stroke group, STG = superior temporal gyrus. Null significance was observed for the Circumlocutions model.

Table 4.

Dysfunctional tissue correlates in patients with impaired single-word auditory comprehension (n = 29)

	UNRELATED								SEMANTIC
	All Unrelated Errors		Unrelated Verbal Paraphasias		Omissions		Neologisms		All Semantic Errors		Related Verbal Paraphasias
	Est.	p	Est.	p	Est.	p	Est.	p	Est.	p	Est.	p
% of coded errors	0.125	0.222	--	--	0.133	0.320	--	--	--	--	--	--
Age	--	--	--	--	--	--	--	--	0.203	0.009	0.225	0.038
Education	--	--	--	--	--	--	0.019	0.253	--	--	--	--
Stroke group (RS)	0.409	0.014	--	--	0.258	0.091	0.230	0.049	--	--	--	--
% damage L Putamen	0.221	0.046	--	--	0.155	0.073	0.436	0.006	--	--	--	--
% damage LSTG	0.207	0.005	--	--	--	--	--	--	--	--	--	--
% damage LSTpole	--	--	--	--	0.211	0.024	--	--	--	--	--	--
% damage LpSTG	--	--	--	--	0.053	0.078	--	--	--	--	--	--
% damage LMTG	--	--	--	--	0.090	0.106	--	--	--	--	--	--
% damage LpMTG	0.465	<0.001	--	--	0.478	0.002	0.117	0.055	--	--	--	--
% damage LAG	--	--	--	--	0.063	0.249	--	--	--	--	--	--
% damage LUF	0.076	0.038	--	--	--	--	--	--	--	--	--	--
% damage LILF	--	--	--	--	--	--	0.074	0.040	--	--	--	--
% damage LIFOF	--	--	--	--	--	--	0.015	0.057	--	--	--	--
FHV L PCA	--	--	--	--	--	--	0.017	0.359	--	--	--	--
FHV L MCA-frontal	0.034	0.127	--	--	0.069	0.105	--	--	--	--	--	--
FHV L MCA-insula	0.103	0.180	0.250	0.009	--	--	0.190	0.092	--	--	--	--
FHV L MCA-parietal	--	--	--	--	--	--	--	--	0.108	0.042	0.313	0.010
Right lesion volume	--	--	--	--	0.197	0.227	--	--	--	--	0.029	0.232

Notes: A dashed line indicates the predictor was not retained in the final LASSO model. Bolded text indicates a significant predictor at p < 0.05. Abbreviations: AG = angular gyrus, Est. = estimate, FHV = FLAIR hyperintense vessels, FuG = fusiform gyrus, pars orbitalis, IFOF = inferior fronto-occipital fasciculus, ILF = inferior longitudinal fasciculus, L = left, MCA = middle cerebral artery, MTG = middle temporal gyrus, p = posterior, PCA = posterior cerebral artery, RS = recurrent/multiple stroke group, STG = superior temporal gyrus, STpole = superior temporal pole, UF = uncinate fasciculus. Null significance was observed for the Circumlocutions model.

Unrelated Errors

For patients with intact auditory comprehension, a greater proportion of unrelated errors (across all subtypes) was associated with damage to LpMTG, p = 0.003, and LITG, p = 0.040, and hypoperfusion in the frontal, p = 0.048, and temporal, p = 0.043, portions of the left MCA territory. A greater proportion of unrelated verbal paraphasias was associated with damage to left posterior temporoparietal regions, specifically LpSTG, p = 0.014, LpMTG, p = 0.003, LITG, p = 0.040, LpITG, p = 0.005, and LAG, p = 0.009. Damage to LpMTG, p = 0.013, and LITG, p = 0.031, was similarly implicated in the production of omissions within this group, as was damage to LIFGorb, p = 0.029, and hypoperfusion within the left MCA frontal, p = 0.027, left MCA temporal, p = 0.028, and left PCA, p = 0.045, territories. A higher proportion of neologisms was significantly associated with greater LpMTG damage only, p = 0.002.

For patients with impaired single-word auditory comprehension, a higher proportion of unrelated errors (across subtypes) was significantly associated with history of prior stroke, p = 0.014, damage to the left putamen, p = 0.046, LSTG, p = 0.005, LpMTG, p < 0.001, and LUF, p = 0.038. A greater proportion of unrelated verbal paraphasias was associated with hypoperfusion in the insular cortex, p = 0.009. On the other hand, production of a greater proportion of omissions was related to greater damage to LSTpole, p = 0.024, and LpMTG, p = 0.002. A higher proportion of neologisms was associated with history of prior stroke, p = 0.049, and damage to the left putamen, p = 0.006, and LILF, p = 0.040.

Semantic Errors

For individuals with intact single-word auditory comprehension, a greater proportion of all semantic errors (i.e., semantically-related verbal paraphasias, circumlocutions, and mixed errors combined) was associated with greater parietal hypoperfusion, p = 0.001, as was production of semantically-related verbal paraphasias alone, p = 0.004. A null LASSO model resulted when we separately analyzed circumlocutions within this group.

For patients with impaired auditory comprehension, we also found a null model when analyzing proportion of circumlocutions. A greater proportion of overall semantic errors was associated with older age, p = 0.009, and left parietal hypoperfusion, p = 0.042. Age, p = 0.038, and parietal hypoperfusion, p = 0.010, were also associated with the proportion of semantically-related verbal paraphasias. Interestingly, these results indicate no spatial overlap in metrics implicated in proportion of semantic and unrelated errors in either patient group.

Discussion

In an effort to build upon previous LSM findings in chronic stroke, we investigated the relationships between naming errors and dysfunctional brain tissue in individuals with and without intact single-word auditory comprehension skills following a new acute stroke. Across the sample, we found that the production of unrelated error subtypes co-occurred more often with each other than with phonological or semantic errors. Individuals with impaired single-word auditory comprehension produced a greater proportion of unrelated errors than individuals with intact comprehension, but unrelated errors existed in both groups. With a few exceptions, unrelated error proportions were associated with damaged ventral stream tissues for both patient groups. Unlike previous studies (Chen et al., 2019; Halai et al., 2018; Stark et al., 2019), we found no spatial overlap in regions implicated in semantic and unrelated errors within either group.

Before turning to the main LSM findings, we briefly address the relationships between error subtypes. Most strikingly, we found that patients who made unrelated errors most often produced only different types of unrelated errors. These findings are fairly consistent with results from Halai et al. (2018), who reported that omissions did not load onto principal components with commission errors. These results also make some theoretical sense, given that errors with phonological or semantic overlap likely reflect at least partially correct and thus intact retrieval of the target semantic features, lemma, and/or phonological codes. Neologisms and unrelated verbal paraphasias are also characteristic of fluent, Wernicke-like speech output and of phonological and semantic jargon, respectively (Marshall, 2006). People with jargon-like output can produce both types of errors (Marshall et al., 1996a, 1996b), which was likely the case of some individuals in our study.

Broadly speaking, the pattern of retained predictors (Tables 3 and 4) for overall unrelated errors and specific unrelated error subtypes (unrelated verbal paraphasias, omissions, and neologisms) predominantly map onto regions within the ventral stream for both patient groups, with some similarities and differences between groups in the specific metrics implicated in each analysis. Similar to prior LSM studies in chronic aphasia (Chen et al., 2019; Halai et al., 2018; Stark et al., 2019), we found that damaged tissue within mid-anterior to posterior temporal cortex was most consistently implicated for different types of unrelated errors within each group. The findings regarding omissions specifically are also compatible with studies citing that omissions are common in patients with LATL epileptogenic lesions (Campo et al., 2016) and individuals with the semantic variant of primary progressive aphasia (Budd et al., 2010; Mesulam et al., 2013; Reilly et al., 2011). Consequently, it is likely that temporal lobe damage plays some role in the production of unrelated errors, irrespective of other lexical deficits.

That being said, obvious differences exist between our findings of different unrelated error subtypes and between groups. For patients with intact auditory comprehension skills, damage to LpMTG and LITG was implicated in the production of unrelated errors (across subtypes), unrelated verbal paraphasias, and omissions. These areas have been implicated in many previous studies of impaired naming with intact word comprehension (e.g., Foundas et al., 1998; Hillis et al., 2006; Price et al., 2003). Moreover, within this group (but less so in the impaired group), several hypoperfusion metrics were retained and significant in LASSO models, a finding that aligns with prior work from our group (Hillis et al., 2002, 2004; 2006, 2008; Hillis, Kane, et al., 2001; Hillis, Wityk, et al., 2001) that hypoperfusion can explain language deficits after an acute stroke event, particularly in patients with small lesions. Dysfunctional/damaged tissue in the frontal cortex was also associated with proportions of total unrelated errors and omissions in the patients with intact (but not impaired) auditory comprehension skills. LIFGorb, which is linked to the anterior temporal lobe via the uncinate fasciculus, is classically considered part of the ventral stream. Whether hypoperfused frontal tissues were primarily within ventral anterior (ventral stream) or dorsal posterior (dorsal stream) areas cannot be determined given the poor localization capacity of the FHV ratings (discussed in greater detail below).

In the impaired (but not unimpaired) comprehension group, subcortical regions (left putamen, LUF, LILF) were retained in certain LASSO models, as was right hemisphere lesion volume. Total lesion volume is often cited as the main driver of language impairments in acute stroke and subsequent recovery (e.g., Plowman et al., 2012; Watila & Balarabe, 2015), indicating it is an important variable to measure and model in stroke studies. Interestingly, right hemisphere lesion volume did not reach the traditional significance threshold in the impaired group models, and left hemisphere lesion volume was not retained in any model in either group. We found significant differences in total lesion volume between our patient groups, which aligns with the between-group differences in overall naming and nonverbal associative semantic skills. Nonetheless, the retention of ROI and FHV measures indicates that metrics reflecting lesion and hypoperfusion topology are better predictors of naming errors than left or right hemisphere total lesion volume.

Another between-group difference is that distinct traditionally dorsal stream regions were implicated in the production of unrelated errors in patients with impaired single-word auditory comprehension but not in individuals with intact comprehension (excluding frontal hypoperfusion). In those with impaired word comprehension, insular hypoperfusion was associated with unrelated verbal paraphasias, and putamen damage was related to the production of omissions and neologisms. Given that the insula and putamen are considered critical for coordination of complex articulatory movements (Baldo et al., 2011; Dronkers, 1996; Wise et al., 1999), it is possible that deficits in post-lexical processes were the root of some unrelated errors in this group. Alternatively, the left insula is also considered a crucial hub within the salience network (Chand et al., 2017; Menon & Uddin, 2010; Raichle, 2011, 2015), which is critical for maintaining goal-directed behaviour during cognitive tasks, and the left striatum—including the putamen—has been implicated in executive control during lexical retrieval tasks (Badre et al., 2014; Crosson, 1985). As such, we cannot definitively draw any conclusions about the specific role of either region in the production of unrelated errors without more extensive neuropsychological assessment.

For each patient group, we found associations between hypoperfusion in the parietal lobe and the production of semantic errors. Given that parietal regions—and LAG in particular—have been implicated repeatedly in the fMRI literature as critical for semantic processing (Binder et al., 2009; Binder & Desai, 2011; Humphreys & Lambon Ralph, 2015; Noonan et al., 2013; Seghier, 2013; Xu et al., 2017), semantic errors are a logical consequence of parietal tissue dysfunction. Despite our semantic error findings cohering with activation studies, they differ from certain prior LSM studies (Mirman, Chen, et al., 2015; Mirman, Zhang, et al., 2015; Schwartz et al., 2009; Walker et al., 2011), which consistently implicated anterior LMTG damage and LATL in the production of semantic errors. Excluding the link between LSTpole damage and omissions errors in the impaired patient group, our findings are more consistent with other investigations that reported posterior lesion correlates of errors with a semantic origin (Cloutman et al., 2009; Fridriksson et al., 2018; Stark et al., 2019). These results may reflect lesion distribution (e.g., lack of anterior LMTG involvement) or time since stroke onset differences between prior samples and ours.

It should be noted, though, that our semantic error findings also do not perfectly align with those from another acute stroke cohort published by our group. Specifically, Cloutman et al. (2009) found that semantic errors arising from semantic impairments and lexical access deficits were associated with damaged or hypoperfused tissue in BAs 22 and 37, respectively. Neither of these BAs falls within the parietal lobe. One reason for the discrepant findings could be due to methodological differences between studies in identifying hypoperfused tissue. The PWI tracings used in the LSM from Cloutman et al. (2009) allow for precise localization of hypoperfusion-deficit correspondence, whereas the FHV ratings do not. Yet, within our study sample, greater parietal hypoperfusion (per FHV ratings) was associated with greater damage to LpSTG, r = 0.228, p = 0.022, and LpMTG, r = 0.276, p = 0.005, regions that partially overlap with those implicated in Cloutman et al. (2009). The retention of FHV parietal ratings over ROI damage metrics in LASSO models indicates the hypoperfusion ratings were stronger predictors of semantic errors in both patient groups. Again, it may be that acute hypoperfusion is the best indicator of acute deficits, particularly in patients with smaller lesions on DWI (Hillis et al., 2002).

Another explanation for distinct lesion sites associated with semantic errors in different studies is that semantic errors can occur as a result of disruption of different components of the naming process. They can arise from impairment in accessing semantics from vision (e.g., in optic aphasia), degraded conceptual semantics (e.g., in semantic variant primary progressive aphasia), semantic access or retrieval deficits due to damage to the lexical-semantic system, or impaired connections between the semantic system and lexical representations (Caramazza & Hillis, 1990; Foygel & Dell, 2000; Hillis et al., 1990; Hillis & Caramazza, 1995; Jefferies et al., 2007, 2008; Jefferies & Lambon Ralph, 2006; Mirman & Britt, 2013). Because the loci of impairment can vary from person to person, it is possible that the deficit profiles of patients who made semantic errors differed between studies. Alternatively, a network perspective of semantic processing assumes that different processing levels are mediated by different brain regions and that communication between hubs is essential (Binder & Desai, 2011; Lambon Ralph et al., 2017; Xu et al., 2017). In theory, semantic errors could arise from damage to the main semantic hubs (e.g., LATL, LAG, LpMTG) or the white matter fibers connecting them. Indeed, some researchers found this to be the case: McKinnon et al. (2018) reported a link between semantic errors and axonal volume loss in LILF, and Fridriksson et al. (2018) revealed several relationships between semantic errors and disconnection of anterior and posterior temporal/temporoparietal areas.

Regardless, we split our sample according to single-word auditory comprehension deficits to illuminate such differences in breakdown locus. Although dysfunctional tissue correlates of semantic errors were dissociable from unrelated errors in this study, semantic error correlates were the same in each patient group. Unrelated error correlates included temporal metrics in both groups, but additional frontal hypoperfusion metrics (in the unimpaired comprehension group) and subcortical regions (in the impaired comprehension group) were noted. For patients with less severe impairments, frontal (and possibly temporal) hypoperfusion may result in impaired selection of the target from semantically-related lexical representations, resulting in an omission error (the most common type in this group). Ultimately, the fact that neural correlates of neither error type overlapped to a great degree with prior findings renders concise interpretations difficult.

Lesion and language impairment characteristics of our sample may also explain the lack of overlap between unrelated and semantic error dysfunctional tissue metrics. We determined inclusion based on presence of left hemisphere stroke rather than necessitating an aphasia diagnosis, which ensured that, compared to prior LSM chronic-only stroke studies, we included a wider range of behavioural profiles, from patients with minimal deficits to individuals with severe language impairments. Similarly, lesions within our sample also ranged from small to large, but on average, they tended to be smaller and more circumscribed than lesions in patients with chronic aphasia. Across this sample and in a sister study that included a larger cohort of acute stroke patients (Meier et al., 2020), we noticed a pattern in which individuals who produced a greater proportion of semantic errors tended to have smaller lesions and better overall naming skills compared to individuals who produced a high proportion of unrelated errors. Semantic errors—and circumlocutions in particular—are characteristic of naming failures in older age and individuals with pure anomia, and likely reflect only minor naming difficulties in our sample (Albert et al., 1988; Grima & Franklin, 2017; Hodgson & Ellis, 1998; Nicholas et al., 1989; Tallberg, 2005). In this way, our sample likely diverged from the chronic aphasia cohorts in most other LSM studies.

Furthermore, a key motivation of this work was to investigate understudied naming errors in acute stroke before the onset of functional reorganization of brain-behaviour relationships. We likely achieved this goal in the group with unimpaired single-word auditory comprehension abilities, but there is less certainty regarding the impaired group’s results. Specifically, contrary to our expectations based on findings from Goldberg et al. (2021), we found differences in anomia profiles (i.e., overall naming accuracy and proportions of phonological and unrelated errors) between auditory comprehension groups that varied by recurrent stroke group status. These findings cannot be used to infer a causal relationship that stroke recurrence causes worse anomia or production of a greater number of unrelated errors. However, one might expect that greater difficulty accessing semantic and/or phonological representations during naming attempts will occur with accumulation of damage across different parts of the brain from multiple stroke events. Within the impaired auditory comprehension group (but not the unimpaired group), a history of prior stroke was significantly related to greater proportions of unrelated errors (across subtypes) and neologisms within LSM analyses.

While the majority of our patient sample had either a single, primarily left hemisphere acute stroke (n = 65) or a single acute stroke plus a prior stroke with minimal to no cortical damage (n = 31), four patients presented with a prior stroke that incurred > 20% damage to at least one left hemisphere cortical ROI. Three of these individuals had intact auditory comprehension skills, and the proportion of unrelated errors produced was low for each patient (P1: 8.33%, P2: 3.33%, P3: 0%). The remaining patient, a member of the impaired auditory comprehension group, had a single prior stroke that occurred just three weeks prior to the new stroke event that coincided with her study inclusion. A partially-completed testing battery immediately after her first stroke revealed a moderately-severe Broca’s aphasia and anomia (19/30 on the BNT), deficits which worsened following her new stroke. Of the six other members of the impaired auditory comprehension group with a history of multiple stroke, four had lesioned tissue outside the left cerebral hemisphere (either in the right hemisphere, cerebellum, and/or brainstem). For these seven individuals in particular, it is possible that some functional reorganization of language occurred in the time after the prior stroke that impacted brain-behavior relationships in the acute phase of the new stroke event.

Study Limitations

The current study had several limitations. First, despite our efforts to control for other lexical impairments with likely ventral stream origins, our testing battery was limited in this capacity. Prospective studies are needed in acute stroke in which error types on receptive tasks are coded, compared to, and combined with errors on production tasks to fully classify patients as having modality-dependent or independent lexical access deficits. Furthermore, as Chen and colleagues (2019) outlined, unrelated errors may also result from deficits in phonological retrieval or assembly, consequences of motor impairments further downstream, executive function deficits, or impairments outside of core naming processes. Time constraints limit the length and depth of acute post-stroke speech-language assessment batteries. Therefore, only a small portion of the sample completed other semantic and phonological tasks or motor speech assessments. As such, we could not disentangle the contributions of other impairments on the production of unrelated errors; doing so remains an important avenue of future investigation.

Other limitations pertained to our dysfunctional tissue measures. First, as previously highlighted, the lack of PWI data for many of our participants meant that it was not feasible to localize hypoperfusion correlates within the MCA territory beyond lobar divisions. Second, we acknowledge that our methods for accounting for prior stroke and lesions outside the left cerebral hemisphere may have been insufficient. Furthermore, an issue that is problematic for LSM in general is lesion coverage and mislocalization of brain-behaviour relationships as a result (Mah et al., 2014; Mirman, Zhang, et al., 2015; Pustina et al., 2018; Sperber et al., 2018). In this study, very few of our participants had damage to LMTpole, unlike the samples included in the series by Schwartz, Mirman, and colleagues (Chen et al., 2019; Mirman, Chen, et al., 2015; Mirman, Zhang, et al., 2015; Schwartz et al., 2012; Walker et al., 2011) or damage localized to dorsal brain regions. At this time, there is no gold standard approach in LSM nor methods to completely mitigate coverage effects, but it is necessary for researchers to acknowledge this limitation to situate their results within the greater LSM literature.

Another difference between our study and many other LSM and naming error studies is the type of LSM statistical analyses used. Specifically, we used region-wise LSM (RLSM) and LASSO regression to identify brain-behaviour relationships. While other studies have used RLSM (Cloutman et al., 2009; Fridriksson et al., 2018), many of the studies most relevant to our own (Chen et al., 2019; Halai et al., 2018; Mirman, Chen, et al., 2015; Mirman, Zhang, et al., 2015; Schwartz et al., 2009, 2012; Walker et al., 2011) included voxel-wise LSM (VLSM) methods, either mass-univariate VSLM or support-vector regression (SVR)-LSM. As referenced previously, we could not perform voxel-based analyses given the importance of accounting for hypoperfusion via the FHV ratings. This approach was supported by the fact that FHV were retained in several LASSO models in this study. Similar to SVR-LSM and other multivariate methods, though, LASSO regression considers covariance between damaged regions, and variables that remained within each model were the most likely unique prediction candidates. It should be noted that this approach does not identify univariate relationships between error types and dysfunctional tissue metrics that may exist. Last, it is possible that other regions within a larger set of ROIs could be associated with naming errors but were excluded due to our approach.

Conclusions

From the era of Wernicke and Broca to the present day, LSM has played a pivotal role in advancing cognitive neuroscience. Modern LSM techniques have evolved our understanding of the neurobiology of language beyond what the founding aphasiologists could have envisioned. Nonetheless, debate persists regarding optimal approaches for characterizing lesion-language relationships, and exploring the neural correlates of different types of naming errors in post-stroke samples remains an open area of research.

In our study, we aimed to investigate the neural bases of unrelated error subtypes which have received relatively little attention in the LSM literature thus far. We found that unrelated errors—and omissions specifically—were most consistently related to damage or hypoperfusion in ventral stream regions spanning anterior to posterior temporal cortex. Although our findings regarding semantic errors diverge from certain studies, we propose that the collective semantic error findings reflect the distributed nature of semantic processing or possibly differences in the loci of semantic impairments in individuals included in different studies. Moreover, the lack of spatial overlap in the neural correlates of semantic and unrelated errors in our study likely reflects lesion and impairment differences between our stroke sample and the cohorts of individuals with chronic aphasia included in prior work. In all, these findings complement prior naming error LSM work done primarily in chronic stroke and provide some initial work regarding unrelated error types in individuals following an acute left hemisphere stroke event.

Table 2.

Demographic, language, and lesion volume comparisons between patients with unimpaired versus impaired single-word auditory comprehension

	Unimpaired Comprehension (n = 71)	Impaired Comprehension (n = 29)	p-value, q-value
Age (in years)	60.96 (13.14)	60.01 (14.27)	0.952, 0.952
Sex (n Women:Men)	36:35	13:17	0.509, 0.572
Education (in years)	13.68 (2.91)	14.53 (3.13)	0.246, 0.370
Handedness (n Right:Left)	61:9	29:1	0.484, 0.573
Single-word AC (z-score)	−0.62 (1.43), −3.67 – 0.52	−37.79 (38.48), −147.99 – −5.22	< 0.001, < 0.001
BNT (total correct, /30)	22.04 (6.07), 1 – 29	9.62 (8.82), 0 – 27	< 0.001, < 0.001
*PPTT (% correct)	97.85 (4.93), 71.43 – 100.00	78.51 (23.87), 6.67 – 100.00	< 0.001, < 0.001
Prior stroke (n Yes:No)	28:43	7:22	0.171, 0.308
Total lesion volume (cc)	21.32 (41.20), 0.11 – 299.94	52.79 (50.79), 2.92 – 228.26	< 0.001, < 0.001

Notes: One participant reported being ambidextrous. For continuous variables, group means and standard deviations are reported, M (SD), and comparisons between comprehension groups were performed using Wilcoxon rank sum tests. Fisher’s exact tests were used to compare handedness, sex, and presence of prior stroke between groups. BNT = Boston Naming Test, PPTT = Pyramids and Palm Trees Test, AC = auditory comprehension per either Auditory Word Recognition from the Western Aphasia Battery-Revised or Word Comprehension from the Boston Diagnostic Aphasia Examination-3. Note that 17 patients (n = 13 patients with unimpaired comprehension) did not complete the PPTT due to time restrictions or differences in language battery administration over time.

Data Availability Statement

Upon reasonable request, deidentified study data can be made available to interested parties by contacting the first author.