Eighty-four control subjects of ages between 18 and 90 years (M = 50.6) participated in the study, as well as twelve participants of ages between 44 and 92 years (M = 58.5) suffering from different types of chronic aphasia, due to acquired brain injury in the left hemisphere. All the participants were bilingual speakers of Catalan and Spanish, though Catalan was their dominant language (and they spoke various dialects of Catalan), and they had different educational backgrounds (Table 1). According to a self-report, none of the control participants had a history of brain injury, cognitive disorders, or learning disabilities.

The participants with different types of aphasia were selected from the patient pools of several hospitals in the Barcelona area, namely the Hospital Sant Rafael, the Escola de Patologia del Llenguatge de l'Hospital de Sant Pau and the Grup de Suport Neuropsicològic. The inclusion criteria involved self-reported language and cognitive skills within normal range before the stroke. Details of age, education, handedness, etiology, aphasia subtype, and time post-onset collected from the medical health record of the PWA appear in Table 2. The participants with aphasia were diagnosed by speech pathologists and neurologists in the abovementioned centers according to clinical criteria and formal or informal tests mostly carried out in Spanish, since there are no available standardized tools in Catalan to diagnose aphasia. Since one of the goals of the study was to determine the feasibility of cognitive testing with CAT-CAT, the aphasic sample represents the variability found in clinical practice (The CAT does not aim at discriminating among aphasia types and does include an aphasia severity scale; Swinburn et al., 2004).

A chi-square goodness of fit test revealed that the proportions in both groups differed by dialect use (χ² = 10.57, p = 0.005), but not by education level (χ² = 5.76, p = 0.056). On the other hand, a Wilcoxon-Mann-Whitney and a Fisher exact tests indicated that there were no significant differences between the two groups with respect to age (U = 658.5, p = 0.399) and sex (two-tailed p = 1).

The participants were administered the Catalan version of the Comprehensive Aphasia Test (CAT; Swinburn et al., 2004) adapted by Salmons et al. (2021), which consists in a battery of twenty-seven subtests that evaluate both cognitive and linguistic skills, including the four behavioral tasks reported in the present study. Three of these tasks assessed visual and verbal short-term memory, and required the storage and maintenance of information (Baddeley, 2007). These type of tasks have been shown to be appropriate for use with subjects with brain damage (Wright and Fergadiotis, 2012). The last task evaluated the comprehension of different sentence types. The words and structures tested are not subject to variation between Eastern and Western Catalan, and therefore the tasks were suitable for all participants independently of the dialect they spoke.

The procedure of the study was approved by the UAB Ethics Committee (CEEAH 5656) and followed the Declaration of Helsinki ethical standards. All the subjects provided their written informed consent prior to participation. Participants were tested in medical institutions or at home by trained examiners. The experimental tasks were conducted in a quiet room to minimize distractions. Control subjects completed the CAT-CAT (Salmons et al., 2021) in a 40 min session, whereas subjects with aphasia needed up to 1 h sessions. Prior to the task, the participants were interviewed to collect relevant personal and medical information. Then the examiner explained the task. All the tests included one practice item in order to ensure that the participants comprehended the task. Once the practice item was performed correctly, the experimental items were presented and the responses were collected manually.

Descriptive statistics were presented as mean, standard deviation, median, range and interquartile range. Distribution for normality was assessed with the Shapiro–Wilk test. The Mann–Whitney-Wilcoxon test was used to compare the distributions of the two groups given the non-normal distribution of the data. Since the sample size for the aphasic group was small, a fixed effects model for binomial data was performed to determine whether there was a significant effect of group, to model the probability of success, and to calculate the odds ratio and 95% confidence intervals. Cutoff values were also calculated for each subtest in order to discriminate impaired and unimpaired performance. The cutoff points corresponded to the score that the 95% of control participants exceeded, following the methodology used in other versions of the CAT (Swinburn et al., 2004; Zakariás and Lukács, 2021). Scores below the cutoff represent impaired performance. A logistics regression analysis was conducted to estimate and compare the percentages of participants at or above the cutoff point in both groups. Finally, Pearson and Spearman correlation coefficients were used to measure lineal and monotonic associations between the four tasks. The correlation coefficients were corrected using the Bonferroni correction method. Statistical significance was set at p < 0.05. The statistical analysis was performed using SAS software, version 9.4 (SAS Institute Inc., USA).

Both groups obtained high scores on the visual STM task, but the control participants' mean of correct responses (M = 9.69, SD = 1.15) was slightly higher than the aphasics' (M = 9.08, SD = 1.51). The responses of both groups were not normally distributed (W_Control = 0.29, p < 0.001; W_Aphasic = 0.68, p = 0.001), and their distributions differed significantly, U = 437, p = 0.016. The fixed-effects model revealed a significant main effect of group, F = 9.47, p = 0.003, and indicated that the estimated success rate for the control group was of 96.9% (95% CI [95.5, 97.9]), and of 90.8% (95% CI [84.1, 94.9]) for PWA. According to the parameters of the model, the probability of success of the control participants was three times higher than the aphasics' (OR = 3.26, 95% CI [1.5, 6.64]). This difference was statistically significant, t = 3.08, p = 0.003.

If we look at the individual responses as shown in Table 4, nine out of the twelve PWA scored at or above the cuttoff value. Two of the three participants who did not obtained a very close score of 8/10, while only one patient scored very low (5). However, the difference between both groups' percentages of participants that reached the cutoff point was significant (Fisher's exact test = 8.23, p = 0.025). In addition, the estimated percentages of passing the cutoff point of control and PWA was of 96.4% (95% CI [89.4, 98.9]) and 75% (95% CI [44.4, 91.9]), respectively. The difference was significant, t = 2.47, p = 0.015, the odds of the control group being nine times higher than the aphasics' (95% CI [1.54, 52.57]).

The mean correct responses of the subjects with aphasia was lower than the controls': 13.67 (SD = 2.71) and 17.61 (SD = 0.68) on the 0–1 scale scoring and 22.17 (SD = 5.83) and 34.39 (SD = 1.87) on the 0–2 scale scoring, respectively. The differences between groups were significant: S = 114, p < 0.001 and S = 97, p < 0.001 for scores based on the 0–1 and the 0–2 scales, respectively. With regard to the scores based on the 0–1 scale, which only takes account of correct and incorrect responses, the results of the fixed effects model revealed a main effect of group, F = 16, p < 0.001. The estimated percentage of success of PWA was only of 50% (95% CI [24.1, 75.9]), much lower than that of control subjects, 96.4% (95% CI [89.4, 98.9]). The difference was significant, t = 4, p < 0.001, the odds of success of the control group being 27 times higher than the aphasics' (95% CI [5.26, 138.65]). In this case, only two PWA reached the cutoff threshold. The percentage of individuals who scored at or above the cutoff point was significantly lower than that of the control participants (Fisher's exact test = 52.04, p < 0.001). The fixed effects model revealed an effect of group (F = 24.59, p < 0.001), the estimated percentage of exceeding the cutoff being of 16.7% (95% CI [4.1, 48.2]) and 95.2% (95% CI [87.9, 98.2]) for subjects with and without aphasia, respectively. The difference was significant, t = 4.96, p < 0.001, and the odds of success of the control group was 100 times the odds of the experimental group (95% CI [15.82, 632.16]).

If we consider the scoring system based on the 0–2 scale, only one participant scored above the cutoff point, which differed significantly from the percentage of control participants that scored at or above the cutoff point (Fisher's exact test = 65.42, p < 0.001). As previously stated, this scoring system is more sensitive as it takes into account variables such as delayed response, request for repetition and self-correction. Not only a main effect of group was observed, F = 22.57, p < 0.001, but the magnitude of the difference between the control and aphasic group was greater than the one observed on the 0–1 scale: OR = 297, 95% CI [27.49, 3208.4]; t = 4.75, p < 0.001.

In order to establish the relationship between language comprehension and STM deficits in Catalan-speaking subjects with aphasia, correlations between the scorings on the four tasks were calculated (Tables 5, 6). Spearman correlation coefficients of 0.69 (p = 0.013) and 0.76 (p < 0.001) were obtained between the two scores of the comprehension subtest for aphasic and control subjects, respectively, indicating a moderate level of association. With regard to the performance of PWA, the score based on the 0–2 scale was strongly associated with their score on the digit span task [r₍₁₀₎ = 0.85; p < 0.001; see Figure 2]. The results of the two verbal STM scores of the aphasic group were also moderately correlated, r(10) = 0.62; p = 0.03. After applying the Bonferroni correction test, the only significant association obtained was between the score of PWA based on the 0–2 scale in the comprehension task and digit span (Figure 2). Regarding control subjects, the only significant correlation observed was between the two scores of the comprehension subtest.

These results were in line with the Pearson correlation analysis (Tables 7, 8): both scores for the comprehension task were positively associated in the aphasic [r₍₁₀₎ = 0.76; p = 0.004] and control [r₍₈₂₎ = 0.85; p < 0.001] groups. Like before, the digit span of PWA was correlated with their score based on the 0–2 scale in the comprehension task [r₍₁₀₎ = 0.83; p < 0.001; see Figure 2] and their listening span [r₍₁₀₎ = 0.6; p = 0.04]. In addition, two weak associations were observed in the control group responses between their score based on the 0–2 scale in the comprehension task and their digit span [r₍₈₂₎ = 0.27; p = 0.012], on the one hand, and their listening span [r₍₈₂₎ = 0.26; p = 0.017] on the other. After applying the Bonferroni correction test to the Pearson correlation analysis, significant associations were observed between the two scores of the comprehension task in both groups, and between the score of PWA based on the 0–2 scale in the comprehension task and their digit span (Figure 2).

Our results are in line with the literature reviewed above, according to which visual and verbal STM is often impaired in aphasia, but subject to individual variability (Gonzalez et al., 2020). Interestingly, the performance of PWA on the visual STM task was very high despite being significantly lower than the controls' and, as already pointed out, most participants reached the cutoff point. Other studies have already reported mild or no impairment in visuospatial STM (e.g., Seniów et al., 2009; Potagas et al., 2011).

In contrast, the results of the aphasic group on the verbal STM subtests were much lower and, if we look at the individual scores, fewer participants scored at or above the cutoff point on the sentence repetition task. In fact, the probability of PWA of scoring at or above the cutoff point decreased from 9 to 28 and 116 times less in visual, digit and sentence STM tasks, respectively.

The difference in performance between visual and verbal STM tasks has also been found in studies that assessed visuospatial and verbal STM through other measures (Seniów et al., 2009; Potagas et al., 2011). In our study, no significant correlations between visual and verbal STM subtests were found. These findings rule out the possibility that the deficit observed in STM tasks is due to a disruption of a common memory mechanism, more specifically the central executive, and rather suggest a selective impairment of the phonological loop. It could also be argued that the difficulties shown by the PWA in verbal span tasks are due to a deficit in repetition, a function that can be selectively impaired in aphasia (Geschwind, 1970). Yet, most of the subjects that participated in our study presented very high scores on a task that assessed their ability to repeat simple words (Table 4), and no significant correlations between the scores on this task and those on the STM tasks were observed. The only exception is subject 2, whose mistakes involved mainly low-frequency words (five out of six errors) and words with more than two syllables (four out of six errors).

The scores on the two verbal STM subtests did correlate significantly, which reflects the fact that both tasks evaluate the same subtype of memory but, as already mentioned, fewer PWA reached the cutoff threshold on the sentence repetition task than on the digit span subtest. Previous studies reported similar results (Martin and Ayala, 2004; Lang and Quitz, 2012). The possibility that the poor performance on the digit repetition task may be due to a problem with numerical skills can be ruled out based on the PWA's results on the arithmetic task (see Table 4). The lower performance on the listening span subtest could be attributed to the higher complexity of sentence repetition as a task, since it requires both production and comprehension skills. Based on the results on the word repetition task, the possibility of a word length effect (Baddeley, 2007) can be ruled out.

An anonymous reviewer pointed out that verbal spans may have reflected different types of production deficits, grammatical deficits in particular. We analyzed the sentences produced by PWA in this task, and found that PWA made a total of 28 errors involving content words: 24 omissions and 4 substitutions of content words. All the substitutions except one were synonyms or semantically related words (e.g., llegir el conte “to read the story” instead of explicar el conte “to tell the story”). In two cases, the error consisted in repeating the sentence in a different word order. Except for one case in which the participant produced a sentence with an infinitive as the main verb, as well as unfinished sentences (El gat... “The cat...” instead of El gat atrapa el ratolí “The cat catches the mouse”), the resulting utterances were grammatical (e.g., La dona tanca la porta “The woman closes the door” instead of La dona marxa i tanca la porta “The woman leaves and closes the door”). The PWA omitted and substituted 17 and 9 grammatical morphemes, respectively, and one patient produced one extra preposition, but these omissions and substitutions were ancillary of the changes in content words, since they did not result in ungrammaticality. Therefore, even if function words were not taken into consideration for the purpose of evaluating short-term memory, it does not seem that the 12 subjects with aphasia that participated in the study produced ill-formed sentences in this particular task. Finally, most mistakes affected longer sentences, which suggests problems with PWA's memory capacity rather than agrammatism or difficulties with syntactically complex sentences, as Caplan and Waters (1999) argued.

The aphasic performance on the sentence comprehension task was significantly worse than the controls', which is in line with the results on the other versions of the CAT (see Table 9) and the other studies reviewed above. This was actually the task where the least participants scored at or above the cutoff point, this being 100 times less likely for PWA than for controls. In fact, their performance was more consistent than the results on the other tasks despite the heterogeneity of the sample. The deficit observed in this subtest cannot be due to an impairment in word comprehension (Baddeley, 2003), as the results of the aphasic group on a word-to-picture matching task revealed (Table 4). Also, no significant associations were found between word and sentence comprehension.

The sentence comprehension task of the CAT does not aim at evaluating the comprehension of specific syntactic structures (Swinburn et al., 2004), yet it includes a wide range of constructions. If we look at the error distribution, the PWA made a total of 52 mistakes (see Table 10) that mainly involved semantically reversible sentences with non-canonical word orders, like passives or clitic left dislocations (2b), and with PPs such as The glass is under the plate which corresponded to the 34.6 and 36.5% of the total errors, respectively. On the other hand, fewer mistakes were found with irreversible (2a) and reversible canonical sentences (The postwoman is chasing the cook), which only represented 5.8 and 19.2% of all errors, respectively, or on irreversible object relatives (3.8%). These findings suggest that the problems to understand the sentences arose when their interpretation relied on particular syntactic operations. There is actually extensive crosslinguistic evidence that PWA present impairments in comprehension that affect specific syntactic structures like passives, object relatives or clitic left dislocations (among others, Caramazza and Zurif, 1976; Hickok et al., 1993; Grodzinsky, 2000; Luzzatti et al., 2001; Friedmann, 2008; Salmons and Gavarró, 2013; Varlokosta et al., 2014; Adelt et al., 2017; Terzi and Nanousi, 2018; Garraffa and Fyndanis, 2020). Moreover, the fact that the comprehension task of the CAT, despite being a general comprehension task, is sensitive to the specific linguistic deficits suggests that it is an efficient diagnostic tool for aphasia.

We also took into consideration the results on the comprehension task based on the two scoring systems which, as expected, were correlated. The scoring system based on the 0–2 scale provides additional information as it penalizes response delays, repetition requests and self-corrections, and the individual scores were low. These low scores probably reflected the greater complexity of the task, given that it demanded not only sentence comprehension, but also spatial search, scene analysis, word retrieval and sentence-to-picture matching (Grodzinsky, 2010; Caplan et al., 2013). The cutoff point based on the controls' scores based on the 0–2 scale favors this view, since their results in sentence comprehension were very poor compared to the other tasks (Table 4). Interestingly, Sung et al. (2018) carried out a study on the production abilities of a group of Korean-speaking aphasics, and the results showed that task type related to individual WM capacity, but not to sentence type (active vs. passive).

The present study also aimed at testing the hypothesis that the difficulties of PWA in STM and sentence comprehension tasks are due to a common cause, whether it be an attention (McNeil et al., 2011), memory (Caplan and Waters, 1999; Aboitiz et al., 2006) or language deficit (Christensen and Wright, 2010). If that was the case, we would expect strong correlations between STM and comprehension subtests. However, the observation that no significant associations were found between the scores on these subtests rules out this possibility and suggests that a single-system deficit cannot explain the impairments observed in these tasks. Rather, we hypothesize that the disrupted tasks reflect deficits in different cognitive systems, namely, short-term memory and language.

Other studies have reported significant associations between forward digit and listening spans and sentence comprehension tasks (among others, Sung et al., 2009; Lee and Pyun, 2014; Choinski et al., 2020), interpreted as evidence that a common underlying mechanism was affected, usually a component of the STM (see Wright and Fergadiotis, 2012 for a review). This view also finds support from the fact that short-term memory is needed in sentence comprehension and, hence, that a deficit in this domain probably impacts on comprehension (Baddeley, 2003). Under this view, it could be argued that the impairments observed in sentence comprehension were triggered by a memory disorder, given that sentences with syntactic dependencies require greater memory and impose higher processing demands (Caplan and Waters, 1999). However, this cannot account for the results from behavioral studies on the comprehension of other types of syntactic dependencies that, despite requiring STM, are well-comprehended by PWA (Santi and Grodzinsky, 2007).

Hence, even if memory and language measures correlated, a STM disorder could not explain the pattern of syntactic deficits observed in the comprehension tasks in the present study and in previous literature (see Grodzinsky, 2000 for a review).

While admittedly it is difficult to assess different cognitive systems in isolation and to identify the impact of memory in sentence comprehension tasks, we believe that in the present study the impact of STM disorders was better captured by the scoring system based on the 0–2 scale. In fact, significant correlations were found between the score on the sentence comprehension task based on the 0–2 scale and the digit and listening spans. Hence, even though higher spans did not translate into higher accuracy rates, the 0–2 scoring system reflected greater difficulties with the comprehension task correlating with lower spans (Figure 2). This, however, does not hold for all participants (e.g., 1, 7, and 10) which, in turn, highlights the variability of STM deficits in aphasia. As we have argued before, we believe this is due to the higher task demands of the comprehension subtest. In fact, some of the studies that have reported significant correlations between memory and sentence comprehension used tasks that could have been more sensitive to STM deficits. This is the case of studies that used complex span tasks to assess WM, which are cognitively more demanding (Ivanova et al., 2017), or sentence comprehension tasks that involved items that gradually increased in length requiring a greater memory load and, hence, relied heavily on short-term memory. To illustrate, Choinski et al. (2020) used both the Token Test (De Renzi and Faglioni, 1978) and a sentence comprehension task with 30 PWA, and they obtained lower scores in the first task. These findings are in line with Caplan and Waters (1999)'s hypothesis, since the scores based on the 0–2 scale could actually reflect post-interpretive processes, such as remembering and using the meaning of the sentence to perform the task of analyzing the images and pointing to the correct one. According to these authors, a contrast between an active and a passive sentences involves mainly interpretive processes, while contrasts like the ones included in the Token Test rather reflect post-interpretive processing.

Our findings are therefore compatible with the assumption that the primary deficit in aphasia is linguistic; this in itself does not exclude the presence of other cognitive impairments or less direct associations between STM and sentence comprehension (Caplan and Waters, 1999; Varkanitsa and Caplan, 2018), in line with the SSIR hypothesis. This would account for the fact that, despite the heterogeneity of the sample, the deficits observed in sentence comprehension were more consistent than the ones observed in STM tasks. In fact, as argued before, there is wide evidence that the pattern of performance of aphasics in sentence comprehension tasks is robust across languages (Drai and Grodzinsky, 2006a,b). In contrast, the deficits in STM tasks were subject to individual variability as several participants performed within the normal range, specially on the visual and digit span tasks (e.g., participants 1, 3, 6, and 10). Other studies have also observed dissociations among participants in non-linguistic cognitive measures (Seniów et al., 2009; Potagas et al., 2011; Caplan et al., 2013; Lee and Pyun, 2014; Gonzalez et al., 2020).

In our view, the conflicting results reported in some of the literature, as well as the discrepancies with regard to the correlation between memory and comprehension measures in aphasia, favor the idea that aphasia does not always entail memory disorders, and that memory deficits alone cannot explain the linguistic impairment, as others have already argued (Murray, 2012; Fonseca et al., 2016; Marinelli et al., 2017). The non-linguistic cognitive disorders that often co-occur with aphasia could be the result of different factors that have already been pointed out in the literature, including the presence of brain damage (Wright and Fergadiotis, 2012) and other medical conditions like dementia or depression (El Hachioui et al., 2014). This is consistent with the literature on cognitive abilities that states that brain-damaged patients present different deficits regardless of the hemisphere damaged or the presence of aphasia (see Fonseca et al., 2016 for a review), as well as with the observation that aphasia severity is often related to the STM capacity (Gonzalez et al., 2020) as more severe aphasias often involve larger lesions.

One of the arguments used to argue that a short-term memory impairment underlies the comprehension deficits observed in aphasia is that both functions have been located in the same brain areas (Hickok and Poeppel, 2000; Aboitiz et al., 2006; Choinski et al., 2020), including the inferior frontal gyrus (IFG). Neuroanatomical data from aphasic and healthy subjects support the co-occurrence of STM and comprehension disorders in aphasia. However, as argued by Santi and Grodzinsky (2007), these deficits reflect the multimodal nature of the brain damaged areas rather than a working memory deficit. Also, general cognitive impairments have been claimed to explain linguistic deficits in several clinical conditions such as schizophrenia (see Kuperberg, 2010 for a review), Down syndrome (Baddeley and Jarrold, 2007) or primary progressive aphasia (Gorno-Tempini et al., 2004). Yet, the relationship between linguistic and other cognitive abilities may vary across clinical populations. Little et al. (2019), for example, compared the performance on verbal and nonverbal cognition tasks of two groups of twenty subjects with aphasia and thirty subjects with schizophrenia. The results showed that while both groups presented significant linguistic deficits, a significant nonverbal cognitive impairment only occurred in schizophrenia. An association between verbal and nonverbal cognition tests was only observed in the schizophrenia group, in line with previous studies (Kerns, 2007). This also implies that, while there is clearly an association between verbal and nonverbal cognitive skills, it is not a direct one.

Taken together this lead us to the conclusion that non-linguistic short-term memory disorders often coexist with language deficits in aphasia. Yet, the present study and previous research show that non-linguistic cognitive deficits are subject to wide individual variability while linguistic deficits remain constant. This emphasizes the need to assess people with aphasia exhaustively in order to precisely describe each patient's profile and plan an adequate intervention program (see van Mourik et al., 1992; Hinckley et al., 2001; Helm-Estabrooks, 2002; Seniów et al., 2009; Murray, 2012; Kuzmina and Weekes, 2017; Marinelli et al., 2017; Fonseca et al., 2018; Zakariás et al., 2018).

There are some standardized tests in different languages that include both verbal and nonverbal cognitive tasks and are specifically designed to evaluate individuals with aphasia, such as the Cognitive Linguistic Quick Test (Helm-Estabrooks, 2001) or the Comprehensive Aphasia Test (Swinburn et al., 2004) in English. The CAT is one of the most used tools for that purpose (Salis et al., 2018), as it aims at evaluating both verbal and nonverbal cognitive skills in a relatively short administration time, which makes it suitable for clinical use. In the present study we presented four subtests of the Catalan version of the test (Salmons et al., 2021) that evaluate visual and verbal STM, and sentence comprehension, and we have shown that it can be a useful tool to evaluate Catalan-speaking subjects with aphasia. Yet, further research is needed to enlarge the PWA sample of the present study. Also, this study does not allow us to investigate whether different aphasia types correlate with different patterns of comprehension and STM impairment, given that the CAT does not distinguish among aphasia types; furthermore we were lacking in information about the participants in our sample. Similarly, the CAT is a tool developed to be used in clinical practice, and the sentence comprehension task does not allow to investigate whether there are significant correlations between the comprehension of different syntactic structures and STM. Therefore, establishing links, on the one hand, between aphasia types, and comprehension and STM and, on the other hand, between syntactic structures and STM deficits, remains a topic for future research.