Introduction
Sign languages are the primary mode of communication in Deaf communities across the world. Similar to spoken languages, signed languages have syntactical, lexical, and sub-lexical structures that differ between languages across geographical regions (Mathur and Rathmann, 2014). However, sign language is expressed in the manual-visual domain, whereas speech is formed in the oral-aural modality. In spite of the modality differences, the existing evidence suggests that neurobiological correlates of language processing overlap to a great deal across modalities (MacSweeney et al., 2008; Malaia and Wilbur, 2010; Cardin et al., 2020a; Trettenbrein et al., 2021). Most previous studies have applied task-based functional magnetic resonance imaging (fMRI) to identify brain regions that are associated with the processing of sign language. These studies have improved our understanding of neural structures involved in sign language perception and understanding. However, functional connectivity can advance our understanding of how different language-relevant brain regions work together for optimal language processing (Fedorenko and Thompson-Schill, 2014; Hagoort, 2019). In the present study, we investigate associations between sign language proficiency and resting-state functional connectivity (rsFC) in deaf early signers. Thus, we explore whether individual differences in sign language skills are associated with how brain regions are intrinsically and functionally related.
Hickok and Poeppel (2007) proposed a neural model of language processing with core language regions in the superior temporal cortex bilaterally, and dorsal and ventral processing streams representing different functional operations. The dorsal stream is left-hemisphere biased and includes the parieto-temporal intersection region, as well as premotor and inferior frontal cortical nodes. The ventral stream, on the other hand, covers bilateral posterior middle and inferior temporal cortical regions, and the left anterior middle temporal gyrus and inferior temporal sulcus. The ventral stream is related to the mapping of an incoming language signal to its meaning, whereas the dorsal stream deals with production. However, it should be noted that consensus does not exist regarding the exact functions and anatomical distribution of the streams, and alternative accounts (e.g., Friederici, 2012; Bornkessel-Schlesewsky and Schlesewsky, 2013; Ullman, 2016; Hagoort, 2019) have a somewhat different emphasis than the model proposed by Hickok and Poeppel (2007). Nevertheless, models overlap with similar functions located across temporal, parietal, and frontal regions, and a dual-stream model finds support in meta-analytic work on speech processing (Adank, 2012; Walenski et al., 2019).
Several studies have reported shared neural activation across spoken and signed language (Söderfeldt et al., 1994b,1997; Petitto et al., 2000; MacSweeney et al., 2002; Sakai et al., 2005; Emmorey et al., 2007, 2014; Courtin et al., 2011; Zou et al., 2012; Johnson et al., 2018; Moreno et al., 2018; Finkl et al., 2020), although a recent study by Evans et al. (2019) indicated that only semantic, not form-based, representations share neural activation patterns. Due to modality-specific operations needed for processing of manual-visual language, some suggest that certain brain regions (e.g., the superior parietal lobule) might be specifically engaged for sign language (Söderfeldt et al., 1994a; Zou et al., 2012; Emmorey et al., 2014). It has also been suggested that the involvement of the right hemisphere might be more prominent for sign language than for speech (Newman et al., 2002, 2010; Emmorey et al., 2005, 2014; Sakai et al., 2005). The role of the right hemisphere has further been proposed to be dependent on proficiency (Malaia and Wilbur, 2010) and the age of acquisition (AoA) of sign language (Neville et al., 1997; Newman et al., 2002; Mayberry et al., 2011). In addition, Stroh et al. (2019) suggested that deaf signers compared to hearing signers might recruit regions in the right hemisphere to a greater degree for certain linguistic tasks.
In a recent meta-analysis of the neural underpinnings of sign language processing, Trettenbrein et al. (2021) proposed that the right inferior frontal gyrus (IFG) and right middle temporal gyrus are specifically recruited for sign language processing. These two regions are typically not regarded as “language regions”, but based on a recent meta-analysis by Walenski et al. (2019), temporal lobule activation for spoken language processing was described as bilateral, in line with the dual-stream model proposed by Hickok and Poeppel (2007). Trettenbrein et al. (2021) further identified critical nodes for sign language processing in the left IFG and precentral/middle frontal gyrus, corroborating reports from spoken language (Walenski et al., 2019). Trettenbrein et al. (2021) also noted that previous literature indicates a role of left middle gyrus, superior temporal gyrus (STG), supramarginal gyrus, and bilateral superior parietal lobules in sign language processing. These regions did show an effect of sign language processing when compared to rest conditions, but the activation overlapped with activation from non-linguistic sign-like actions from an independent set of studies. Thus, these regions might not be critically involved in linguistic aspects of sign language processing although the studies were not designed to test this directly. Due to the limited number of studies, Trettenbrein et al. (2021) could not differentiate between regions that might be specific for production versus comprehension, or syntactical, lexical, and phonological levels of processing. It thus remains unclear how different regions relate to the type of linguistic processing in sign language. Some propose that the left supramarginal gyrus is particularly important for phonological analysis in sign language (Corina et al., 1999; Emmorey et al., 2003; MacSweeney et al., 2008), and activation of the inferior frontal and superior and middle temporal cortices is associated with sentence level processing of speech (Walenski et al., 2019). In addition, language comprehension seems to be bilaterally distributed to a larger degree than language production (Walenski et al., 2019).
The literature indicates that sign language proficiency might influence neural responses. For example, AoA, which likely influences proficiency (Twomey et al., 2017), seems to produce lateralization effects (e.g., Newman et al., 2002; Mayberry et al., 2011). Newman et al. (2002) compared neural activation for real American Sign Language sentences compared to sentence-like pseudo-sign utterances, and reported a right hemisphere effect of AoA, with activation of angular gyrus in early, but not late, learners. AoA effects have further been reported to be associated with the level of activation in other regions. Mayberry et al. (2011) regressed neural activation on AoA and showed that, with earlier AoA, there was stronger activation in bilateral dorsolateral prefrontal cortex, left anterior insula/frontal operculum, left IFG, left ventral premotor region, and bilateral STG. Later AoA, on the other hand, was associated with stronger activation of left lingual gyrus and left middle occipital gyrus. Twomey et al. (2020) reported that late as compared to early signers, regardless of hearing status, had stronger activity in the occipital segment of the left intraparietal sulcus. Further, early deaf signers showed greater activation in the left posterior superior temporal cortex in response to real sign language sentences as compared to made-up signs. In yet another study, Cheng et al. (2019) reported that language deprivation in three cases of deaf individuals was associated with altered structural connectivity in the left dorsal arcuate fasciculus pathway, a fiber tract connecting superior temporal to frontal regions. Thus, early access to sign language seems to produce effects on neural activation in occipital, parietal, temporal, and frontal regions, but also influences the development of language-relevant structural pathways. It is likely that level of proficiency is associated with differences in neural activation patterns. Emmorey et al. (2015) reported that the regional neural activation that correlates with behavioral performance on linguistic tasks differs depending on the specific linguistic operation of the task. For example, they saw that fingerspelling ability was negatively associated with neural activation in right frontal regions, whereas sentence processing of sign language was negatively associated with activation in angular gyrus and middle temporal gyrus.
Apart from the above-described association between sign language skills and region-specific activations, associations between language skills and brain activity have also been investigated using rsFC. For example, Qian et al. (2016) saw associations between reading skills and rsFC of visual dorsal stream regions (i.e., involved in spatial processing) and regions invoked by reading tasks, such as the fusiform gyrus. In another study, Koyama et al. (2011) reported an association between reading skills and rsFC between the left fusiform gyrus and left frontal and inferior parietal regions. In addition, Chai et al. (2016) investigated associations between second language acquisition and rsFC based on two regions of interest, the anterior insula/frontal operculum and the visual word form area in the fusiform gyrus. They reported positive associations between degree of second language acquisition and connectivity between anterior insula/frontal operculum and left posterior STG as well as the dorsal anterior cingulate cortex. In a study on associations between rsFC and language skills in deaf adults, Li et al. (2013) investigated rsFC within and between superior temporal regions and how connectivity was related to written language skills in congenitally deaf adults, adults with acquired deafness, and hearing adults with no knowledge of sign language. Connectivity between the middle and anterior parts of the superior temporal cortex was associated with written language skills in participants with deafness, but not in hearing participants. However, Li et al. (2013) did not report any associations between written language skills and rsFC outside superior temporal regions. In summary, there are studies showing associations between spoken language skills and rsFC in both hearing (Chai et al., 2016) and deaf (Li et al., 2013) individuals, but as far as we know, associations between sign language proficiency and rsFC have hitherto not been described in the literature.
Given the available evidence, there is a multitude of potentially relevant brain regions to include in an analysis of associations between rsFC and sign language proficiency. For the purposes of the present study, we based the selection of seeds of interest in our connectivity analysis on two of the most studied regions in relation to language processing, i.e., the IFG and STG (e.g., Trettenbrein et al., 2021). To restrict the number of statistical tests performed in our exploratory analysis, only four nodes were included as seeds: bilateral IFG and posterior STG. This selection of regions does not include all potentially relevant regions proposed by dual-stream models, such as the one by Hickok and Poeppel (2007), but it overlaps with such models. For the selected regions, we estimated seed-to-voxel based connectivity for each individual and correlated with behavioral performances. We included two separate measures of sign language proficiency, one that taps onto phonological skill and another that represents sign language sentence processing skill. This was because there is reason to believe that the type of linguistic operation might affect which specific neural networks that are involved (Hickok and Poeppel, 2007; Friederici, 2012; Bornkessel-Schlesewsky and Schlesewsky, 2013; Hagoort, 2017).
Discussion
In the present, explorative study, we investigated how individual variability in sign language proficiency, at phonological and sentence levels, is associated with resting-state functional connectivity (rsFC). More specifically, we investigated associations between two different types of sign language skills: phonological skill and sentence processing skill, and rsFC from bilateral IFG and posterior STG to the rest of the brain. Faster phonological processing was positively associated with stronger connectivity between right IFG and left middle frontal gyrus/frontal pole; however, this association did not remain after controlling for non-verbal cognitive ability. Worse sign language sentence processing ability was associated with stronger positive connectivity from left IFG to sensorimotor regions. Thus, rsFC between prefrontal and sensorimotor language regions seems to co-vary with sign language reproduction skill.
Resting-State Functional Connectivity and Sign Language Processing
We saw that rsFC from the right IFG to left middle frontal gyrus/frontal pole is negatively associated with performance on a speeded phonological awareness task, which might suggest that the strength of communication between these regions at rest is indicative of phonological skill. However, this association was not statistically significant after control for non-verbal cognitive ability (as measured on the Visual Puzzles sub-test from WAIS-IV, Wechsler, 2008). Thus, it is possible that the association we see is driven by non-linguistic, cognitive skills. Due to the small sample in the present study, we had limited power to detect associations, and controlling for multiple covariates (i.e., age, non-verbal cognitive ability), as we did here, reduces the degrees of freedom even more. Thus, associations might exist that we could not detect. On the other hand, a small sample might produce spurious and random results that do not replicate. That is, the association we saw in the first place might have occurred by chance. In their recent meta-analysis, Trettenbrein et al. (2021) noted that the right IFG is involved in sign language processing. However, based on the available literature we see no particular reason why rsFC between this region and left middle frontal gyrus/frontal pole should be associated with phonological processing of sign language. Instead, such connectivity might reflect intrinsic activation within a lateral frontoparietal network used in the processing of executively demanding tasks (Uddin et al., 2019; Witt et al., 2021). This could possibly explain why the association we first saw disappeared when controlling for non-verbal cognitive ability.
We further saw that those who struggle more with reproducing sign language sentences correctly have stronger rsFC between left IFG and a cluster peaking in the left precentral gyrus. In the context of language processing, left IFG is typically described as a control region involved in language production and complex linguistic analysis (Hickok and Poeppel, 2007; Friederici, 2012; Bornkessel-Schlesewsky and Schlesewsky, 2013; Hagoort, 2017), and Corina et al. (1999) concluded that left IFG is critically involved in sign language production. Precentral regions have been proposed to be involved in the processing of movement of sign language (Emmorey et al., 2014). However, in recent meta-analytic work on both signed and spoken language, motor regions have been reported to be invoked also by linguistic processing (Walenski et al., 2019; Trettenbrein et al., 2021). In the dual-stream model proposed by Hickok and Poeppel (2007), left IFG and premotor cortex are included in the dorsal language stream, which also includes temporoparietal junction regions. In speech, this stream is assumed to integrate language representations with motor representations, and it is therefore critical for language development. The STS-SRT task involves the forming of linguistic output, and our finding might thus be interpreted as support of the idea that left IFG and sensorimotor regions work together to support integrative language processes. In extension, our finding suggests that the proposed neurocognitive underpinnings of spoken language production might apply also to sign language production. This is not to say that all neurocognitive mechanisms are shared across language modalities (cf., Evans et al., 2019), but as suggested by many before us (e.g., Cardin et al., 2020a; Rönnberg et al., 2021; Trettenbrein et al., 2021), we believe that mechanisms are not unique.
Our results point to that the intrinsic connectivity between left IFG and left sensorimotor regions is sensitive to sign language proficiency and might be stronger in individuals with poor language skills as compared to better skill. Flinker et al. (2015) reported that the left inferior frontal region (Broca’s Area) is typically not co-activated with motor regions in language production. Instead, when activation goes up in motor regions during production, activation in Broca’s Area goes down, and when activation increases in Broca’s Area in pre-articulatory stages, motor regions are relatively silent. As displayed in Figure 2, individuals with stronger sign language sentence reproduction skill tend to have negative intrinsic connectivity between left inferior frontal and sensorimotor regions, whereas individuals with worse skill have positive connectivity between these regions. Thus, intrinsic positive functional connectivity between these regions in deaf adults might be a marker of a language network that is sub-optimally (dis)-integrated, and this might be what we see evidence of in the present study. Proficiency is linked to different developmental trajectories, and the idea that we propose here is thus loosely related to the notion that the dorsal stream is important for language development via input–output matching mechanisms (Hickok and Poeppel, 2007). Sub-optimized network integrity might mean that individuals with weaker proficiency use non-linguistic motor representations to compensate for poorly defined language representations, or that access to language-based motor representations requires greater involvement of language-control functions. Based only on the present study, any definitive conclusion is of course premature. Nevertheless, our results suggest that the strength of intrinsic connectivity between dorsal stream left inferior frontal and sensorimotor regions might be a marker of the level of ability to reproduce sign language sentences.
Individual Differences in Sign Language Proficiency and Functional Connectivity
Banaszkiewicz et al. (2021) reported a sign language-specific change in task-invoked connectivity between left IFG and left lateral superior occipital cortex in hearing adults who were beginning learners of Polish Sign Language. The task was a sign-based lexical decision task (deciding whether visually presented signs were real or not), which represents an intermediate level of linguistic processing compared to the behavioral measures used in the present study. In our case, we saw a connection that suggests an effect related to effective perception-production processing (the STS-SRT), whereas the results reported by Banaszkiewicz et al. (2021) might reflect improved effectiveness of a perception-identification interface. Another important difference is that Banaszkiewicz et al. (2021) investigated task-based connectivity and not rsFC. Despite the methodological differences between the present study and the study by Banaszkiewicz et al. (2021), both studies suggest that the left IFG might be a critical node in functional networks related to sign language proficiency. How this region is connected functionally and structurally to other regions of the brain should thus be further studied in relation to (sign) language proficiency in future research. To our knowledge, no previous study has investigated task-based functional connectivity in relation to different types of sign language processing skills. Such a study could reveal, by experimental manipulation, what the rsFC-behavior associations revealed in the present study reflect.
In contrast to Li et al. (2013), which was the only previous study on rsFC-behavior associations including deaf participants that we found, we did not see that rsFC within the STG was associated with language proficiency. Although Trettenbrein et al. (2021) noted that the STG might not respond to linguistic sign-based stimuli per se, the broader literature indicates that this region is involved in language processing (Hickok and Poeppel, 2007; Friederici, 2012; Bornkessel-Schlesewsky and Schlesewsky, 2013; Cardin et al., 2020a). The lack of statistically significant associations between any of our behavioral measures and connectivity from this region might thus be surprising, although different methodology across studies is a likely explanation. The previous study most like the present study, conducted by Li et al. (2013), reported a correlation between story writing ability and connectivity between right middle superior temporal sulcus and left superior temporal sulcus/STG. Li et al. (2013) did however not observe any statistically significant association between their measure of language skill and rsFC from the superior temporal cortex to other regions of the brain. The lack of similar associations in the present study and the study by Li et al. (2013) might be explained by that the type of language skills investigated were different, or that different seeds were used in the analyses. In the present study, we used the posterior STG language network nodes in Conn as our seeds. This seed is relatively large, and it might therefore be difficult to capture meaningful connectivity in a small sample. On the other hand, using a pre-defined seed makes the design more transparent and facilitates replication. Both we and Li et al. (2013) failed to find that inter-regional rsFC from the STG predicts language skill in deaf adults, and it might be the case that it does not. However, a future study aiming at capturing this might fare better than us by including a larger sample and more precisely delineated seeds.
Since the connectivity observed here is based on rsFC, that is, co-activation between brain regions when the participants have no specific task to perform, the associations between regions are not invoked by a language task and therefore might not have anything specific to do with language processing. As mentioned earlier in the Discussion, one risk with a small sample is that findings are random. We tried to minimize the risk for this by restricting the number of statistical tests we performed. However, we also wanted to maximize statistical power and therefore applied a liberal approach when correcting for multiple statistical tests. Since the one association that we saw fits reasonably well with the existing literature, we believe that it is a meaningful association. At the same time, it does probably not represent the only relevant association. Based on previous empirical findings (Trettenbrein et al., 2021) and theoretical considerations (Hickok and Poeppel, 2007), we selected four language-relevant nodes as our seeds, but several potentially relevant regions were not included (e.g., supramarginal gyrus, anterior temporal regions, lateral occipital cortex). Another limitation is that we were only able to detect effects that were strong. This is reflected in the effect sizes of observed statistically significant effects, ranging from R2 0.75 to 0.87. Thus, in addition to the risk of finding random association, the small sample we included also carries a risk of missing out on meaningful effects that are small in magnitude.
Our results indicated an association between only one of the behavioral measures and rsFC. This might be because the measures tap onto different aspects of linguistic processing, or that task demands differ. We cannot determine whether the results we see are driven by a specific linguistic skill, or by any other skill or task-dependent factor. C-PhAT is performed by mentally converting orthographic input to sign-based representations, and then comparing representations before a decision is made explicit by a timed button press. In the STS-SRT task, on the other hand, the participant views and repeats a sign language sentence, with no further decision to make. The STS-SRT demands that a sentence is produced and the C-PhAT includes covert production of signs, as the individual mentally represents handshapes and compares them. Additionally, the STS-SRT includes phonological, lexical, and syntactical knowledge-structures, combined into a coherent expression, whereas C-PhAT taxes isolated phonological processing ability, and, given its design, orthographic-phonological mapping. On top of this, the STS-SRT has a social component (i.e., viewing another person producing a sentence) that the C-PhAT does not. Taken together, it is difficult to identify a specific origin of differences in associations observed between these tasks. We saw that better performance on C-PhAT had a positive association with non-verbal cognitive ability, and we controlled for scores on the Visual Puzzles task in connectivity analysis on C-PhAT. Thus, the influence of general cognitive factors in the association for that task was controlled for. We did not make the same control for STS-SRT, since no association with non-verbal cognitive ability was observed and adding a covariate to the analysis would then only reduce statistical power. However, non-linguistic processing skills might explain the observed relationships and differences in associations between tasks in the present study. In summary, our study has some methodological issues that future studies should correct, including the small sample size and the selection of seeds. In addition, future work should also carefully consider which behavioral measures to include. However, given that the studied population is a unique group and our approach here is novel we believe that the present findings represent an important contribution to the field.
Both structural and functional plasticity as a result of deafness and sign language use has been reported in the literature (Sadato et al., 2004; Cardin et al., 2013, 2018; Olulade et al., 2014; Ding et al., 2015; Twomey et al., 2017; Benetti et al., 2018; Trumpp and Kiefer, 2018; Bonna et al., 2020; Finkl et al., 2020; Andin et al., 2021; Dell Ducas et al., 2021; for reviews see Alencar et al., 2019; Cardin et al., 2020b). Thus, brain connectivity patterns that underlie linguistic operations in deaf sign language users might not be the same as for hearing individuals who use speech (or sign language). In our design, we did not compare across groups and previous studies on hearing individuals with a similar design as the present study are lacking. Although the general pattern of the available studies is that there is a great deal of overlap in the neural underpinnings of language processing regardless of modality (Walenski et al., 2019; Cardin et al., 2020a; Trettenbrein et al., 2021), a few regions might be modality-specific (Trettenbrein et al., 2021) and differences might exist at a form-based representational level (Evans et al., 2019). In addition, the type of linguistic operation might interact with modality effects. Studies with a design that allows for comparisons across language modality and different types of linguistic tasks are well needed to improve our understanding of how language processing is represented neurally, both in terms of modality–specificity and modality–generality.