Geometry in the brain optimized for sign language – A unique role of the anterior superior parietal lobule in deaf signers

Geometry has been identified as a cognitive domain where deaf individuals exhibit relative strength, yet the neural mechanisms underlying geometry processing in this population remain poorly understood. This fMRI study aimed to investigate the neural correlates of geometry processing in deaf and hearing individuals. Twenty-two adult deaf signers and 25 hearing non-signers completed a geometry decision task. We found no group differences in performance, while there were some differences in parietal activation. As expected, the posterior superior parietal lobule (SPL) was recruited for both groups. The anterior SPL was significantly more activated in the deaf group, and the inferior parietal lobule was significantly more deactivated in the hearing group. In conclusion, despite similar performance across groups, there were differences in the recruitment of parietal regions. These differences may reflect inherent differences in brain organization due to different early sensory and linguistic experiences.


Introduction
Language and mathematical processing are intimately related, but diverse mathematical domains rely differently on verbal support.For example, arithmetic computation requires verbal scaffolding for the manipulation of arithmetic facts, whereas geometry is more dependent on visuo-spatial perception.Some studies suggest that aspects of the modality of language may selectively support or impair mathematical ability (Andin, Elwér, & Mäki-Torkko, 2023;Marcelino, Sousa, & Costa, 2019;Pagliaro & Kritzer, 2013).The biological basis of such differences has rarely been examined, and therefore it is unclear if biological processing networks differ between individuals that use languages in different modalities.In this study, we examined neural correlates of geometry tasks in adult deaf signers and hearing non-signers.By doing so we could investigate possible reasons for behavioral differences between these groups.From a broader perspective, such findings can contribute to the understanding of other types of mathematical difficulties (i.e., different types of dyscalculia) as well as inform teaching strategies for deaf and hard-of-hearing individuals.Supporting the development of geometry in schools is important as such abilities have been identified as reliable predictors of academic, as well as occupational, success in the fields of science, technology, engineering, and mathematics (Wai, Lubinski, & Benbow, 2009).

Mathematical abilities in deaf signers
Deaf individuals commonly show lower results compared to hearing peers on mathematical tasks, although the gap differs across mathematical domains (Ansell & Pagliaro, 2006;Hyde, Zevenbergen, & Power, 2003;Titus, 1995).In a research synthesis, Gottardis, Nunes, and Lunt (2011) found that deaf or hard-of-hearing children across the ages of 4-16 lagged in mathematics with a combined effect size of − 0.7 across a variety of tasks.Differences between hearing and deaf individuals may appear as consequences of several different factors as suggested by previous studies (Alcock et al., 2016;Pagliaro & Kritzer, 2013); domain-specific factors such as the inherent support that language modality lends to mathematical processing (Pagliaro, 2010), domain-general factors such as differences in working memory capacity (Marcelino et al., 2019), and differences in learning experiences between deaf and hearing children (Swanwick, Oddy, & Roper, 2005).
Differences between hearing non-signers and deaf signers are most pronounced in tasks that require verbal support such as arithmetic fact retrieval, comparing quantities using words like "more" or "less," as well as fractions and multiplication (Andin, Rönnberg, & Rudner, 2014;Kelly, Lang, Keith, & Davies, 2003;Nunes et al., 2009;Titus, 1995).These results suggest that deaf signers may use the verbal system to solve these tasks differently.The performance of deaf signers on mathematical tasks that rely on visuospatial systems such as perceiving distance, direction, length, and angles are commonly in the typical range (Hauser, Cohen, Dye, & Bavelier, 2007;Zarfaty, Nunes, & Bryant, 2004).

Processing of geometry
The processing of geometry can be divided into three different subcategories (Bonny & Lourenco, 2015).Euclidean constructions include principles central to Euclidean geometry such as parallelism, straight lines, and shapes.Sense and transformation cover understanding of symmetry by paying attention to change and relations, such as up/down and left/right, i.e., processes typically included in mental rotation tasks.Metric properties deal with quantitative comparisons such as determining differences in length and proportions.
The biological base of processing geometry tasks has been examined in a few imaging studies, showing robust activation in posterior superior and inferior parietal regions (posterior SPL and IPL) across a wide range of spatial/geometrical tasks (Buening & Brown, 2018).One specific type of spatial task is mental rotation.This type of task is representative of the sense and transformation subcategory, as it examines the ability to imagine the appearance of objects if they were rotated in space.In a meta-analysis, Zacks (2008) found that the posterior parietal cortex, mainly corresponding to the posterior SPL, extending down to the posterior superior occipital cortex was consistently activated during a wide range of different mental rotation tasks.Further, several studies have shown sensitivity to shape information in the posterior SPL (Freud, Culham, Plaut, & Behrmann, 2017).According to the two-streams hypothesis, two major cortical streams are involved in visuospatial cognition.Functionally, these streams have been dissociated into the dorsal where/how and ventral what pathways.In geometry tasks, the dorsal stream is of specific interest.Processing concerned with spatial properties of objects and information used to manipulate objects derives from the dorsal stream.The dorsal stream projects from the visual area 1 (V1) to the posterior SPL and IPL.The anterior regions of the SPL might also engage in visuospatial processing, yet research indicates that their primary function lies in attentional processes (Alahmadi, 2021).For example, the anterior SPL does not appear to be involved in the processing of visual shapes.Instead, this area in connection with IPS and the frontal eye field, is involved in the attentional network.Although both hemispheres are involved in the processing of geometry tasks there seems to be right hemisphere lateralization (Fink et al., 2000;Sack et al., 2007).In general, there appears to be an inverse correlation between strength of neural activity and performance, i.e. neural efficiency (Kozhenikov & Blazhenkova, 2013), with lower activation as skill increases.

Geometry in deaf signers
Some researchers have suggested that the processing of geometry can be a relative strength in signing populations (Marcelino et al., 2019;Pagliaro & Kritzer, 2013) as it taps into processes such as mental rotation (Emmorey, Klima, & Hickok, 1998;Talbot & Haude, 1993).These are processes that sign language speakers practice in their daily lives, as perspective-taking, and the ability to represent objects in space and to change perspectives are vital in signers (Emmorey, 2015).This is further reflected in the role that the superior parietal lobule has in sign language comprehension and production (for a review see Emmorey, 2021).Processing of sign language shows a stronger involvement of bilateral superior parietal regions, compared to speech.The increased activation in the left SPL has been connected to an increased need for somatosensory monitoring for planning how to place signs in space, whereas the right SPL seems to have a specific role for classifiers that are used for spatial relations in sign language (Emmorey, McCullogh, Mehta, Ponto, & Grabowski, 2013;Shum et al., 2020).In addition, the right SPL has been shown to be more strongly activated for perspective-dependent constructions (Emmorey, 2021), which is connected to better mental rotation skills in fluent signers (e.g.Emmorey et al., 1998;Kubicek & Quandt, 2021).There is also an association between sign language proficiency, understanding of perspective-taking in sign language, and mental rotation tasks (Secora & Emmorey, 2020).
Studies that have found advantages for deaf individuals, e.g. on spatial array problems (Zarfaty et al., 2004), mental rotation (Emmorey et al., 1998), and non-symbolic approximation problems lend further support to a geometry processing advantage in deaf populations (Masataka, 2006).However, the pattern is not consistent; Marschark et al. (2015) found better performance on tasks examining spatial relations and embedded figures for hearing individuals compared to deaf signers.
In one of few fMRI studies of geometry in deaf individuals, partly different processing networks were found for deaf and hearing individuals.Le et al. (2018) compared deaf signers and hearing nonsigners in mental rotation tasks and found decreased activations in brain regions such as the bilateral parahippocampal gyrus, the left posterior cingulate cortex, the right anterior cingulate cortex, and the right IPL when compared to the hearing non-signers.The authors also suggested a more exact anatomical network for the deaf signers, lateralized to the right hemisphere, an effect suggested to evolve due to daily exposure to similar tasks (perspective taking in sign language communication).In the present study, we are not explicitly investigating mental rotation, but the tasks employed here share aspects of rotation and/or shape recognition with a classical mental rotation task.

The present study
In sum, deaf individuals exhibit lower results on mathematical tasks in general and especially in domains that rely heavily on verbal support compared to hearing individuals.Consequently, language is considered an important factor influencing mathematical ability in deaf signers.Mathematical tasks that are more spatially oriented are a relative strength in deaf persons as such tasks require cognitive strategies that are practiced in everyday life (i.e.perspective-taking).Only a few studies have examined the processing of geometry in the brain.We know of only one study that has examined geometry in connection to the modality of language (Le et al., 2018).Thus, the aim of this fMRI study was to investigate the behavioral and neural correlates of three different geometric competencies (Euclidean constructions, sense and transformation, and metric properties) in deaf and hearing individuals.We made the following predictions: -Both deaf and hearing individuals will show activation of bilateral SPL (at least in the posterior portion), with stronger activation in the right hemisphere.-Deaf individuals will show stronger activation in right compared to left SPL for the sense and transformation tasks since right SPL is involved in perspective taking.-Hearing individuals will show stronger activation, compared to deaf individuals, in the right IPL.
For differences between the three task types, the study is explorative, as there are no previous studies to base predictions on.

Participants
Thirty deaf signers and 29 hearing non-signers were recruited to participate in the study.Eight deaf and four hearing individuals were excluded (see section 2.2.1), hence the final study group consisted of 22 deaf signers and 25 hearing non-signers.Participants in the deaf group had used Swedish sign language as their first language since at least age 3 as they exhibited prelingual deafness.They had participated in bilingual schooling and none of the participants had cochlear implants.

Elwér and J. Andin
Table 1 presents the inclusion and exclusion criteria used in the study.As we included deaf participants between the ages of 28 and 45, all participants had attended school after the introduction of the bilingual curriculum in Swedish deaf schools in 1983.This population is unique as the development of sign language is expected to be optimal compared to most other deaf populations.The control group was recruited to match the group of deaf signers on demographic and cognitive factors, such that there were no group differences in age, general non-verbal intelligence (as measured by the visual puzzles subtest from Wechsler adult intelligence scale; fourth version WAIS (Wechsler, 2008)), visuospatial working memory (Corsi, 1972), gender or level of education (Table 2).Gender was defined by the participants themselves as they answered the question "What is your gender?" in a questionnaire.They were given the options; Woman, Man, or Other (or prefer not to disclose).
The hearing participants were not familiar with Swedish sign language and had Swedish as their first language.All participants in both groups were right-handed and had no history of neurological och psychological conditions.
Instructions were given orally for hearing and in Swedish sign language for deaf participants by a signing research assistant.Written informed consent was signed by all participants.Information about the project and the rights of the participants was given in Swedish, Swedish sign language and written form before the consent form was signed.The project has been approved by the Swedish Ethical Review Authority (Dnr: 2019-00896).

Procedure
The participants were recruited by advertising on bulletin boards, online forums, contacts from previous studies, and suggestions from participants already included in the project.Before testing, the participants filled out an online questionnaire which included questions about age, gender, education, work situation, and handedness.All participants were asked about their knowledge of sign language and age of acquisition.The persons who fulfilled the inclusion criteria were contacted and invited to the Stockholm University Brain Imaging Centre where testing of math ability, non-verbal intelligence, and working memory, as well as fMRI of arithmetic (reported in Andin et al. (2023)) and geometry took place.The test sessions lasted about 2 h.Before scanning, the participants were informed about the protocol regarding any incidental findings on acquired brain images.No such deviations were found when scanning or analyzing the data.

Brain imaging and analysis
Stimulus material.The material was developed based on the intuitive geometry test by (Dehaene, Izard, Pica, & Spelke, 2006).The test has been used on adults and children (Bonny & Lourenco, 2015;Izard, Pica, Dehaene, Hinchey, & Spelke, 2011) and examines sensitivity to geometric properties, concepts, and transformations such as symmetry, mental rotation, and parallelism.The stimulus pictures were grouped by task types previously suggested by (Bonny & Lourenco, 2015) into Euclidean constructions, sense and transformations, and metric properties.Euclidean constructions entail principles central to Euclidean geometry such as parallelism, straight lines, and shapes like squares and triangles.Sense and transformations examine the understanding of change and relations (up/down, left/right) that influence symmetry.In this category, mental rotation is required for some items.The last category is metric properties where items include quantitative comparison of for example length and position.For examples of tasks see Fig. S1in the appendix.
Each item consisted of four figures and the task was to decide which figure that did not belong with the other three.For instance, three pictures could show parallel lines of different lengths and the fourth showed lines that were not parallel.A baseline task was also included.The baseline items were in the same format as the other tasks but consisted of three pictures taken from the same category of stimulus material and one large cross in the fourth picture.All categories of stimulus material were represented (Fig. S1).In the baseline condition, the participants were instructed to press the button that corresponded to the picture with the cross (i.e. the task was the same as for the experimental task; which figure does not belong with the rest).There were 40 unique trials for each task type.The benefit of using this test is that it required no verbal knowledge about geometry, so we were able to capture the neurobiological pattern of geometry rather than the verbal naming of shapes and figures which is common in other types of geometry tests.
Experimental Design.The experiment was set up as a 2x3 factorial design with group (deaf, hearing) as the between-subject factor and condition (Euclidean constructions, transformations, metric properties, baseline) as the within-subject factor.The blocked design included forty trials for each of the four conditions assigned to eight blocks of five trials per block in two separate runs.During each trial, the stimulus was displayed for 5000 ms preceded by a 500 ms fixation cross (¤), such that each block lasted for 27500 ms.The participants answered by pressing one of four buttons to indicate the picture that deviated most from the others.Responses were given using a diamond-shaped (HHSC-2x4-D) button box.Between blocks, there was a 10-second rest period.All 40 trials per condition were randomized into 8 different blocks.Block order was pseudo-randomized and permuted for each participant, such that every block type was equally likely to appear as the starting block in a run and such that every block type was presented once within each epoch (i.e., all block types appeared once before any block was repeated).There were as many unique compositions of blocks during the two runs as there were participants in one group, i.e., the same block order was used for the first participant in both groups, and a new order was used for the second participant in each group.To further avoid order effects there was a block order for one participant that was the reverse of the block order of another participant (i.e., participant 2 in each group was given the reverse block order of participant 1 and participant 4 was given den opposite order of participant 3).
Two deaf participants were excluded: one due to generally low performance overall (below 2 sd) and one due to extremely low performance on the baseline task.In addition, three participants (two deaf and one hearing) were excluded as no answers were recorded due to technical problems.
Statistical analyses of imaging data.Preprocessing of data, wholebrain analyses, and ROI analyses were performed using Statistical Parametric Mapping packages (SPM12; Wellcome Trust Centre for Neuroimaging, London, UK), including SPM Anatomy toolbox (version 2.2b; (Eickhoff et al., 2005) running under MatLab R2018a (The Math-Works Inc., Natick MA).Preprocessing was performed following SPM12 standard procedures, including realignment, co-registration, segmentation, functional and structural normalization, and spatial smoothing.Runs with movement above 3 mm in x, y, and z or more than 3 degrees in pitch, yaw, and roll, were excluded from further analyses.In total four deaf and two hearing individuals were removed completely, and another three deaf and five hearing individuals had one of two runs removed.Hence the final number of participants in the analyses were 22 deaf signers and 25 hearing controls.
Whole brain analyses.Whole brain data was processed by fitting a general linear model with regressions representing each of the three task types and the baseline task as well as the six motion parameters derived from the realignment step of the preprocessing.At the first level, each of the three task types was contrasted with the baseline task.At the second level, these individual statistical maps were brought into 1) withingroup second-level analyses, where one-sample t-tests were performed separately for each group, and 2) a 3 × 2 analysis of variance (task type × group).Activation was considered significant for FWE-corrected pvalues < .05 at peak level and cluster level.Analyses were also run with age and non-verbal intelligence (WAIS) as covariates.Since the pattern of results did not differ, only analyses without covariates are reported here.
ROI analyses.To investigate activation within the ROI:s (left and right anterior and posterior SPL and IPL), mean ROI values were entered into a 3 × 2 × 2 (task type × hemisphere × group) analysis of variance.One sample t-tests were used to evaluate whether activation differed significantly from 0. Outliers were removed if they were 1.5 × interquartile range above the third quartile or below the first quartile.The mean ROI values were also correlated with the accuracy for each task type to evaluate whether there was a correlation between performance and ROI activation.Analyses and graphic representations for both ROI and behavioral analyses were performed in RStudio (version 2023.06.1) (R Core Team, 2022) using packages ggplot2 (Wickham, 2016), tidyr (Wickham, Vaughan, & Girlich, 2023), dplyr (Wickham, François, Henry, & Müller, 2023), apaTables (Stanley, 2023), lme4 (Bates, Maechler, Bolker, & Walker, 2015), sjPlot (Lüdecke, 2023) and report (Makowski et al., 2023).

Behavioral tests and analyses
In addition to fMRI results, measures of working memory, non-verbal intelligence, and in-scanner responses were collected.Non-verbal intelligence and working memory testing were used to avoid potential bias in brain activation due to differences in general cognitive skills.
Visual puzzles.The visual puzzle subtest from the WAIS (Wechsler, 2008) was used to assess non-verbal intelligence.This task is highly correlated with general intelligence.We did not find any differences between groups in non-verbal intelligence (see Table 2).
Corsi block.Working memory was tested using a version of the Corsi-block-tapping test (Corsi, 1972).Psychtoolbox running under MatLab 2019a (The MathWorks Inc., Natick, MA) was used to computerize the task.In the computerized version, nine white squares appear on the screen.Sequences of squares that change color from white to green one at a time are presented to the participants.Each target remains green for 500 ms, and the interval between presented targets is 500 ms.At the end of the sequence, participants are asked to click the squares in the same order as they were presented using a computer mouse.The first sequence is of length 2. If the participant reproduces the sequence correctly the next sequence is increased by one square.If the participant fails to give the correct sequence, the participant gets a second attempt at the same sequence length.The test is terminated after two failed attempts at a sequence length.The span length is defined as the length of the longest sequence at which the participant gave a correct answer.The results of this task did not indicate any differences between the groups regarding working memory (see Table 2).
In-scanner responses.Assessment of sensitivity to geometric properties, the task that the participants performed in the scanner was analyzed using a logistic mixed model with group (deaf, hearing) and task type (Euclidean, sense and transformation, metrics) as fixed variables, and item and subject as random variables.Analyses were performed in RStudio (version 2023.06.1) (R Core Team (2022), 2022) using packages described in section 2.2.1.

In-Scanner results
We fitted a logistic mixed model (estimated using ML and BOBYQA optimizer) to predict accuracy (whether the given answer was correct or not) with group (deaf, hearing), task type (Euclidean, sense and transformation, and metrics) and their interaction as fixed variables.Both variables were centered based on the grand mean total of the variables for ease of interpretation of the results.The model included item and subject as random effects.Descriptives of the group's performance split by stimulus category are presented in Table 3.
The model's total explanatory power was substantial (conditional R 2 = 0.34), however, the part related to the fixed effects alone (marginal

Whole-Brain analyses
Significant brain activation for the two groups separately is presented in Fig. 1 (see also Table A1 for peak and cluster activation for all task types).
Analyses of variance showed no significant peak activation as an effect of group or type and no interaction between the two.There was, however, a significant effect of task type at cluster level in the left middle occipital gyrus (p fwe = .016,− 28 to -94 6); the sense and transformation task had stronger activation than metrics.The results were similar when age and non-verbal intelligence were added as covariates.We also compared across groups for all task types combined but found no significant activation at either peak or cluster level.
In the anterior SPL, deaf signers had significant activation for all task types in both hemispheres, while the hearing non-signers' activation differed from baseline only for metrics in the right hemisphere.Hence, there was a main effect of group, F(1, 234) = 4.98, p = .027,partial eta squared = 0.02, with stronger activation for deaf (m = 0.372, sd = 0.645) compared to hearing individuals (m = 0.167, sd = 0.759).There was also an effect of hemisphere, F(1, 234) = 4.88, p = .028,partial eta squared = 0.02, with stronger activation in the right (m = 0.358, sd = 0.753) compared to left (m = 0.156, sd = 0.667) hemisphere.There was no significant effect of condition and no significant interaction effects (ps > .349).
There were no significant correlations between accuracy and mean ROI activation for either group in any of the ROIs.

Discussion
In this study, we used a geometry decision task to examine intuitive judgments in visuo-spatial processing in deaf signers and hearing nonsigners.The results showed similar performance but partly different brain activation patterns across groups.The fMRI results suggest a (minor) neural reorganization, possibly related to the requirement for  specific spatial abilities in sign perception.However, this variance in neural depiction does not seem to be important for performance.
Although the tasks used in this study shared features with the mental rotation task (i.e.sense and transformation), it had an additional focus on patterns of geometric properties, such as the Euclidean construction.
As both groups performed at comparable levels, the results suggest that the task used in this study may not tap into the refined spatial strategies of the deaf signers suggested by studies showing an advantage in spatial array problems (Zarfaty et al., 2004), non-symbolic approximation problems (Masataka, 2006) and mental rotation for sign language speakers (Emmorey et al., 1998;Kubicek & Quandt, 2021).We found activation in the expected regions for processing of geometry covering primarily posterior parietal regions, overlapping with the dorsal visual stream, for both groups.The activation appeared more widespread for hearing compared to deaf individuals and for the metrics task type compared to Euclidean and sense and transformation (see Fig. 1).However, there were no group differences or interactions between groups or task types at the whole brain level.In the ROI analyses group and task-type effects, as well as lateralization effects were found.The ROI analyses showed the expected right-lateralized activation across both groups in the posterior SPL.Further, we found bilateral anterior SPL activation for the deaf group and IPL deactivation across both groups, with stronger deactivation for the hearing group.
In the ROI analyses, we focused on three main geometry-related regions in the parietal lobule, i.e., the anterior SPL, the posterior SPL, and the IPL.The posterior SPL was significantly activated across all task types for both groups.Activation was significantly stronger for metrics and to some extent for sense and transformation compared to the Euclidean task type.However, this was not related to difficulty or performance level as we found no performance difference between tasks and no significant correlations between performance and mean ROI values.Hence, we found no support for neural efficiency, i.e. decreased activation (or more efficient use) with increased performance (Kozhenikov & Blazhenkova, 2013).The posterior SPL has previously, primarily been shown to be involved in geometrical and spatial tasks related to mental rotation (Zacks, 2008), which is closely related to the sense and transformation task type used in the present study.Hence, this region may be less involved in the processing of geometrical forms, such as those included in the Euclidean task type here.No group differences in level of activation were found in this region.Hence, together with the lack of correlation between performance and level of activation, this study renders no evidence of posterior SPL activation being differentiated based on performance.Further, there was no evidence of differential activation between groups, despite the regional overlap with regions generally engaged during sign language processing.
In line with previous studies (Fink et al., 2000), there was a right lateralization present for both groups in the posterior SPL.However, contrary to Fink et al. (2000), the significant activation, compared to baseline, was present in both hemispheres, and contrary to Le et al. (2018), we did not find stronger activation in the right hemisphere for the deaf compared to hearing individuals.As for the posterior SPL, there was a general lateralization effect with stronger activation in the right hemisphere for anterior SPL.However, contrary to the posterior SPL, there was a significant group difference in the anterior SPL, where the deaf signers showed significantly more activation compared to the hearing non-signers.The activation, compared to baseline, was significant for all task types in both hemispheres for the deaf signers, while the only task to elicit significant activation in the hearing group was the metric task type, and then only in the right hemisphere (see Fig. 2).It has previously been suggested that the dorsal visual stream is affected by deafness (Armstrong, Neville, Hillyard, & Mitchell, 2002;Weisberg, Koo, Crain, & Eden, 2012).For example, Armstrong et al. (2002) used EEG to show that, for visual motion, deaf individuals activate the anterior SPL more extensively than do hearing individuals.This is further supported by the generally wider distribution of activation in response to visual motion in deaf compared to hearing groups, which in turn may be driven by enhanced responsivity for motion in relation to sign language processing (Armstrong et al., 2002).Hence, it could be hypothesized that the dorsal visual stream extends further into anterior SPL in deaf individuals, which might be related to the involvement of spatial relations in sign language.Although the present study does not employ visual motion stimuli, it is possible that the extended dorsal stream in deaf signers can be generalized to spatial processing used for geometrical tasks, which then explains the stronger activation of anterior SPL in the deaf group in the present study.
The IPL has previously been demonstrated to be activated in geometrical and spatial processing, spanning both hearing (Zacks, 2008) and deaf (Le et al., 2018) individuals.In the present study, we instead found the IPL to be significantly deactivated across all task types in both groups and stronger so in the hearing group.When investigating the nature of the effects in the IPL in other mathematical tasks studies from our lab (Andin et al., 2019(Andin et al., , 2023) ) as well as from others (Wu et al., 2009), deactivation rather than activation has been found.Similar findings of deactivation in the IPL have also been found for nonmathematical tasks, for example, face matching and Stroop color-word interference tasks (Harrison et al., 2011).An explanation for this is that IPL is a node in the default mode network (DMN; Raichle, 2015), which is thus deactivated during the task.Deactivation of the DMN has been interpreted as representing increased effort, which would indicate higher effort in the hearing group in the present study.Thus, the present findings suggest that the hearing individuals invested more effort to perform the tasks at the same level as the deaf group.This is contrary to the results from Le et al. (2018), where a decrease in activation was interpreted as increased efficiency in the deaf group.These group differences in IPL may also represent organizational differences in the DMN between deaf and hearing individuals, as previously suggested in several studies (Bonna et al., 2021;Dell Ducas et al., 2021;Malaia, Talavage, & Wilbur, 2014).An alternative explanation for the group differences in anterior SPL and IPL could be related to different allocation of resources.While the stronger deactivation in IPL for the hearing group might reflect an increased effort, the stronger activation in anterior SPL may reflect a stronger allocation of attentional resources (Alahmadi, 2021), ultimately resulting in similar performance across the group but based on different underlying mechanisms.
Interestingly, we show different neural underpinnings of geometry processing between deaf and hearing individuals, despite comparable performance.These results may reflect the growing body of evidence showing how early sensory and linguistic experiences shape brain organization, and subsequently brain functions.Several large-scale brain networks have been shown to be shaped differently between hearing and non-hearing populations, including the DMN, the salience network, the language network (Dell Ducas et al., 2021), the attention network (Andin & Holmer, 2022), which may explain why similar performance can be supported by different neuronal networks in different populations.In the larger project to which this study belongs, with both cognitive ability and education level being equal, we found differences between deaf signers and hearing non-signers in arithmetic (Andin et al., 2023), but not in geometry.Hence, as expected, geometry is a relative strength of deaf individuals (Hauser et al., 2007;Zarfaty et al., 2004).These results further support that the lag previously identified for deaf populations, is restricted to tasks that rely on the verbal system (Pagliaro, 2010).Although geometry is a relative strength for deaf individuals, we did not find that the advantage in visuospatial skills of the deaf signers led to better geometry performance.However, it should be noted that in a study on the same cohorts, we found the deaf group to perform at lower levels not only on multiplication tasks but also, unexpectedly, on subtraction tasks (Andin et al., 2023).Hence, there may be a bias in the recruitment with either the deaf group performing at lower levels than the deaf population or the hearing group performing better than expected.Previous results on visuospatial advantages of deaf signers are mixed (Emmorey et al., 1998;Marschark et al., 2015;Zarfaty et al., 2004), and the tasks used in this study may have been too specific for the transfer of visuospatial skills from sign language exposure.

Limitations
Although the current sample size is reasonably large for a study on deaf signers, there is a risk of skewed recruitment.Since there are no previous brain imaging studies on geometry processing in deaf groups, it was not possible to perform regular power calculations before recruitment.We thus based our sample size on power calculations adapted for arithmetic processing as the data collection was performed parallel to Andin et al. (2023).Due to the low sample size, we were not able to perform any meaningful sex-or gender-based analyses.We have no reason to believe that sex or gender play out differently across the groups, but to guard against possible gender effects, the same proportion of men/women were enrolled in both groups.
A further limitation of the present study is the relatively unexplored nature of the task used.Although some studies have used similar tasks in behavioral studies, few studies have investigated geometry using brain imaging techniques.With few fMRI studies using similar tasks, it is difficult to optimize the design for investigating geometry.More studies are, thus, required to understand the biological basis of geometry processes in both deaf and hearing populations.

Conclusion
During the processing of geometry tasks, deaf and hearing individuals recruit the posterior SPL, as expected given its known role in spatial processing.However, the anterior SPL was more strongly engaged in the deaf group, possibly indicative of an extended dorsal stream for deaf signers or differences in the allocation of resources.There was no evidence of an association between performance and strength of neural activation, as no discernible correlations emerged between task performance and level of activation.Further, in the IPL, a central component of the DMN, we found stronger deactivation in the hearing compared to deaf individuals.The stronger deactivation potentially implies a more effectively directed allocation of cognitive effort.The present study provides initial knowledge about how geometry processing is differently affected by spoken and signed language; however, future studies are needed to fully understand the relationship between geometry and the modality of language.In sum, there are differences in activation patterns between the deaf and hearing groups despite comparable performance, which might reflect inherent differences in brain organization as an effect of early sensory and linguistic experiences.Further studies are needed to discern the effect of early sensory and linguistic experience on geometry processing.

Funding
Vetenskapsrådet [Swedish Research Council], Grant/Award Number 2016-02337; Stockholms universitet, Grant/Award Number: SU FV-5.1.2-1035-15.The funders had no role in the study design, in the collection, analysis, and interpretation of data, in the writing of the report, and in the decision to submit the article for publication.Elwér and J. Andin

Fig. 1 .
Fig. 1.Whole-brain activation.Significant activation for the three different task types for deaf signers (red) and hearing non-signers (green).Displayed at FWE-corrected p < .05 at peak level.(For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Table 1
Summary of inclusion and exclusion criteria for participation.
1 Participants had to be at least 18 years old, i.e., those tested during 2019 were born before 2001 and those tested during 2020 were born before 2002.2If used to access spoken language.Participants using hearing aids for alarm purposes were included.

Table 2
Group demographics and general cognitive ability.

Table 4
Summary of results from the logistic mixed model for accuracy in the geometry task.