Towards a unified understanding of lateralized vision: A large-scale study investigating principles governing patterns of lateralization using a heterogeneous sample

While functional lateralization of the human brain has been a widely studied topic in the past decades, few studies to date have gone further than investigating lateralization of single, isolated processes. With the present study, we aimed to arrive at a more unified view by investigating lateralization patterns in face and word processing, and associated lower-level visual processing. We tested a large and heterogeneous participant group, and used a number of tasks that had been shown to produce replicable indices of lateralized processing of visual information of different types and complexity. Following Bayesian statistics, group-level analyses showed the expected right hemisphere (RH) lateralization for face, global form, low spatial frequency processing, and spatial attention, and left hemisphere (LH) lateralization for visual word and local feature processing. Compared to right-handed individuals, lateralization patterns of left-handed and especially those who are RH-dominant for language deviated from this 'typical' pattern. Our results support the notion that face and word processes come to be lateralized to homologue areas of the two hemispheres, under influence of the RH- and LH-specializations in global form, local feature, and low and high spatial frequency processing. As such, we present a more unified understanding of lateralized vision, providing evidence for the input asymmetry and causal complementarity principles of lateralized visual information processing. The absence of correlations between spatial attention and lateralization of the other processes supports the notion of their independent lateralization, conform the statistical complementarity principle.


Introduction
Functional lateralization, or the differential specialization of the two cerebral hemispheres, enables the human brain to process a multitude of different types of information in an efficient and optimized manner (Hellige, 1993). At the population level, this division of labor is expressed in 'typical' patterns of lateralization, such as left-hemisphere (LH) dominance for most language-related processes, and righthemisphere (RH) dominance for face processing . At the same time, individuals can still differ in direction and strength of lateralization to such an extent that some people show RH-dominance for language, whereas others show no clear evidence for either hemisphere being dominant for language-related tasks (Mazoyer et al., 2014;Rasmussen & Milner, 1977). To understand the principles underlying the distribution of functions across the two hemispheres in both typical and reversed or atypical lateralization, the current study examined the relationships between hemispheric specializations for an array of processes including and subserving language and face perception, using a sample of participants that could be expected to show considerable heterogeneity in their direction and strength of lateralization based on variability in their handedness and/or known hemispheric dominance for language. To date, only few studies have examined the relationships between different lateralized processes eoften including measures of language and face processing, and the results of these studies have led to different views on the existence and nature of these relationships. Specifically, previous studies have suggested a number of hypotheses about the principles that may govern patterns of lateralization, which we here summarize as the statistical complementarity (Bryden, H ecaen, & DeAgostini, 1983), causal complementarity (Bryden et al., 1983), and input asymmetry (Andresen & Marsolek, 2005) principles. Each of these three principles assumes that various processes are lateralized, and attempt to explain patterns of lateralized processing.
According to the statistical complementarity principle, each process has a certain probability of being lateralized to one hemisphere, which is independent of the probability that other processes are lateralized to the same or the other hemisphere. Consequently, certain brain processes may show consistent lateralization to contralateral hemispheres at the population level, but there is no causal relation underlying this division of labor. In line with this view are the results of a factor-analytic study of cortical activity during rest, supporting the independence of lateralized brain systems involved in vision, internal thought, attention, and language (Liu, Stufflebeam, Sepulcre, Hedden, & Buckner, 2009). This claim was further corroborated by a review by Badzakova-Trajkov, Corballis, & H€ aberling (2016), showing that most evidence supports independent lateralization of different processes, especially with regard to processes that operate on information from different domains. In support of the conclusions from their literature review, the authors additionally present the results of a factor-analysis on neuroimaging data, suggesting independently lateralized systems governing spatial attention, word generation, and face processing (Badzakova-Trajkov et al., 2016).
In contrast to the statistical complementarity principle, the causal complementarity and input asymmetry principles both assume that lateralization of one process does depend on lateralization of others, with the former accounting for functional segregation (i.e., lateralization of different functions to opposite hemispheres) and the latter for co-lateralization (of different functions to the same hemisphere) . According to the causal complementarity principle, once a certain process is lateralized to a specific cortical area in one hemisphere, there is limited room for specialization of other processes in this area (Andresen & Marsolek, 2005;Badzakova-Trajkov et al., 2016;Cai, Van der Haegen, & Brysbaert, 2013;Gerrits, Van der Haegen, Brysbaert, & Vingerhoets, 2019). As a consequence, other types of information that may initially have been processed by the now occupied area, will become lateralized to homologous areas in the contralateral hemisphere. Support for this notion has been provided by Dundas, Plaut, & Behrmann (2015). In their study, Dundas and colleagues presented word and face stimuli (both assumed to recruit the middle fusiform gyrus) to a group of 7e12 year-olds, who varied in their word recognition competence, while measuring the electro-encephalography (EEG) response. The results showed that the more LHlateralized the children were for word processing, the more RH-lateralized they were for face processing, as reflected by the differing magnitudes of the measured event-related potentials (ERPs) in response to word and face stimuli. Aside from this evidence for causal complementarity between face and visual word processing, a number of studies have suggested causal complementarity for face processing and the production of language during speech. For example, Gerrits et al. (2019) showed a correlation between LH-lateralization of brain regions that were active during language production and RH-lateralization of brain regions that were active during face perception.
The question then arises why visual word processing becomes lateralized to the LH and face processing to the RH, rather than the other way around. Behrmann and Plaut (2013) suggest that in order to arrive at efficient word and face processing, there is pressure for intrahemispheric connectivity to c o r t e x 1 3 3 ( 2 0 2 0 ) 2 0 1 e2 1 4 areas governing the representation of information necessary for such processing. While word and face processing demand similar resources (such as central vision, a possible reason they both engage the fusiform gyrus [Hasson, Levy, Behrmann, Hendler, & Malach, 2002]), they also differ in the types of information necessary from lower levels in the processing hierarchy. In the case of visual word and face processing, this difference would concern cortical areas devoted to language, which are necessary for processing of the former but not the latter type of visual information. Indeed, previous studies have shown a positive correlation between lateralization for language production (i.e., verbal fluency) and visual language perception (i.e., word reading) (Gerrits et al., 2019;Van der Haegen & Brysbaert, 2018) As such, hemispheric dominance for speech can be seen as a candidate for driving the direction of the complementary lateralization of word and face processing.
The input asymmetry principle captures this colateralization principle more generally, in proposing that lower-level processes subserving higher-level processes will drive ipsilateral lateralization of the latter (Andresen & Marsolek, 2005). This principle is, for example, reflected in a theory by Ivry and Robertson (1998), which has its basis in the assumption that the LH selectively processes relatively high frequency information, while the RH selectively processes relatively low frequency information. Any higher-level visual process that operates on a specific range of spatial frequencies, therefore, would also be lateralized to the hemisphere specialized for lower-level processing of that frequency range. As the holistic processing of a face has been shown to be affected by removing low spatial frequency (LSF) but not high spatial frequency (HSF) information (Goffaux & Rossion, 2006), the strength of RH-lateralization for face processing would thus be expected to depend on the strength of RH-lateralization for LSF processing. Conversely, word processing has been shown to rely on HSF information (Ossowski & Behrmann, 2015), and the strength of LH-lateralization would thus be expected to depend on the strength of LHlateralization for HSF processing. This idea has been supported by findings of differential sensitivity to spatial frequency information in the LH and RH fusiform gyri (Woodhead, Wise, Sereno, & Leech, 2011). Specifically, they used sine-wave gratings to show that the LH fusiform gyrus ean area specialized in word processinge responds more strongly to the presentation of HSFs, while the RH fusiform gyrus, especialized in face processinge, responds more strongly to the presentation of LSFs.
In summary, previous studies examining the relationships between different instances of hemispheric specialization have resulted in diverging claims about the existence and nature of these relationships. Specifically, the statistical complementarity principle assumes no relation between lateralization of different processes, while the other two principles do. The causal complementarity principle explains how different processes become functionally segregated to the two hemispheres. Furthermore, the input asymmetry principle proposes that cortical areas devoted to different processes within a processing hierarchy benefit from intrahemispheric connectivity and thus promote co-lateralization of these processes to the same hemisphere. As such, the causal complementarity and input asymmetry account for two sides of the same coin: the former proposing contralateral specialization of processes recruiting similar resources (e.g., faces and words), the arrangement of which in turn is driven by ipsilateral specialization of processes within a processing hierarchy (e.g., faces and low spatial frequencies), as proposed by the latter. As such, the causal complementarity and input asymmetry principles are not mutually exclusive, while both are mutually exclusive with the statistical complementarity principle.

Present study
In the present study, we aim to shed light on these relationships by investigating the lateralized processing of different types of visual stimuli, using a large sample of participants (n ¼ 122) who would be expected to show heterogeneity in both strength and direction of lateralization because of variation in, amongst other things, handedness. Specifically, we examined the relationships between behavioral indices of lateralized processing of visual words, faces, global and local elements, high and low spatial-frequency information, and the distribution of spatial attention, using tasks that we had previously found to produce replicable lateralization indices for population-typical lateralization in a sample of righthanded participants (see Brederoo, Nieuwenstein, Cornelissen, & Lorist, 2019). Using this series of tasks we aimed to test previous claims proposing causal complementarity between the processing of words and faces, and to determine whether any such complementarity might relate to hemispheric specialization for lower-level perceptual processes that rely on similar information (i.e., high spatial frequencies and local elements in the case of visual words vs low spatial frequencies and global form in the case of faces), as proposed by the input asymmetry principle. Furthermore, we will test whether lateralization of these several types of visual information is statistically independent from lateralization of spatial attention, which is often measured in the visual domain with the landmark task (e.g., Badzakova-Trajkov, H€ aberling, Roberts, & Corballis, 2010;Cai et al., 2013).
In examining the relationships between lateralized processes, we also aimed to determine whether certain processes are consistently mediated by the same or by different hemispheres, irrespective of which hemisphere this might be. That is, we investigated whether people who show populationtypical lateralization for one process (e.g., LH-dominance for recognizing visual words) also show population-typical lateralization for another (e.g., LH-dominance for high spatial frequencies), and whether people with reversed lateralization for one process then also show reversed lateralization for the other processes. To be able to address this question, we included a number of participants (all left-handed) whom were known to show RH-dominance for language processing. Such RH-dominant individuals are difficult to find in random samples, which is why we recruited them from a sample of left-handed participants whose language dominance had previously been assessed using functional magnetic resonance imaging (fMRI) and behavioral methods in a study by Van der Haegen, Cai, Seurinck, & Brysbaert (2011). By including a sample of this rarely studied group of participants, our study offered a unique opportunity to determine if c o r t e x 1 3 3 ( 2 0 2 0 ) 2 0 1 e2 1 4 consistent patterns of contralateral and ipsilateral specialization can be found for participants who differ in terms of which hemisphere is dominant for language. Previous studies indeed suggested that individuals with RH-dominance for language can show absent or reversed lateralization of other processes, such as face processing (Gerrits et al., 2019) and spatial attention (Cai et al., 2013).
Furthermore, we actively sought to include as many lefthanded participants (generally known to be more variable with regard to their lateralization of language [Knecht et al., 2000]) as possible so as to obtain a participant sample that could be expected to be heterogeneous with regard to hemispheric dominance for language. This resulted in subgroups of right-handed, left-handed (for whom language dominance was unknown), and (left-handed) RH-dominant participants.
As such, the present study deviated from many earlier lateralization studies in that the gathering of lateralization indices of a multitude of within-domain processes and the aimed for heterogeneity of our sample allowed for a thorough evaluation of predictions following the three principles of lateralized processing (Table 1). The statistical complementarity principle predicts the absence of correlations between lateralization indices of different processes, and based on this principle there is no reason to assume lateralization patterns other than the 'typical' one to occur. If, to the contrary, the lateralization of the investigated processes is not independent, the causal complementarity principle predicts negative 1 correlations between processes governed by homologue areas (i.e., the stronger LH-lateralization for words, the stronger RHlateralization for faces). Following this prediction, lateralization patterns should be reversed for individuals for whom language dominance is reversely lateralized to the RH. The causal complementarity principle does not allow any predictions regarding lateralization of processes that do not become lateralized to homologue areas. As processing of global form and local features (Chechlacz, Mantini, Gillebert, & Humphreys, 2015) and LSFs and HSFs (Peyrin, Baciu, Segebarth, & Marendaz, 2004) have both been proposed to recruit differing cortical areas, no predictions regarding correlations between lateralization of these processes can be made based on the causal complementarity principle. Furthermore, based on the causal complementarity principle we cannot make any predictions regarding processes that are lateralized to the same hemisphere (i.e., ipsilateral processes such as face and global feature processing). The input asymmetry principle fills this gap by predicting both the direction of correlations and the lateralization patterns of higher-level ipsilateral processes to simply mirror those of lower-level processes. In addition, the input asymmetry principle predicts positive correlations between ipsilateral processes within a processing hierarchy (e.g., the stronger LHlateralization for local processing, the stronger LHlateralization for word processing). Based on previous studies on the relation between spatial attention, language production, face processing, and vision more generally (Badzakova-Trajkov et al., 2016;Liu et al., 2009), lateralization of spatial attention is predicted to be statistically independent from that regarding other processing domains.

Participant recruitment
Right-and left-handed participants were recruited at the University of Groningen, while only left-handed participants were recruited at the University of Ghent, from an existing left-handed participant pool. Part of the participants in this Ghent participant pool had undergone fMRI scanning in a previous study, establishing their RH-dominance for language (Van der Haegen et al., 2011). The final sample was based on pragmatic considerations, testing as many participants as we could get. Participants in Groningen who had participated in a previous study using the same tasks (Brederoo et al., 2019) were excluded from participation in the current study, to prevent unwanted practice effects or familiarity with the stimuli to influence the results.  1 We here adhere to the common calculation of lateralization indices (see Methods), resulting in values smaller than 0 for RHlateralization, and larger than 0 for LH-lateralization. Consequently, correlations describing a positive relation in terms of lateralization strength between two processes lateralized to opposite hemispheres, will numerically become negative correlations.

Participant sample
In total, 122 (69 women and 53 men) were tested 2 ; 99 at the University of Groningen and 23 at Ghent University. Mean age of the participants was 21.3 years (range 17e35 years). All participants were native speakers of Dutch, German, or English, and reported normal or corrected-to-normal vision. Participants were classified as right-handed when they had a positive score on the Flinders Handedness Questionnaire, and as left-handed when they had a negative score on this questionnaire (Nicholls, Thomas, Loetscher, & Grimshaw, 2013). Twenty-three people from the left-handed Ghent participant pool signed up to participate, of whom 13 had known RHdominant for language as verified with a verbal fluency task during fMRI scanning (Van der Haegen et al., 2011). Accordingly, our participant sample could be grouped into righthanded participants (n ¼ 69), left-handed participants of whom hemispheric dominance for language was unknown (n ¼ 40, including the 10 Ghent participants who had not undergone fMRI scanning), and left-handed RH-dominant participants (n ¼ 13). Participants received course credit or a monetary compensation for their participation. The ethical committee of the Psychology Department of the University of Groningen approved the experimental procedure, and all participants gave informed consent before the start of the experiment.

Tasks
Over the past decades of lateralization research, a wide variety of tasks have been devised to measure lateralization of information processing. For the current study, we used a series of tasks that we have previously shown to produce reliable evidence for population-typical visual lateralization in righthanded participants (Brederoo et al., 2019). A detailed description of these tasks and their methods can thus be found in our earlier study. The only general difference to the earlier study is that in the present study an in-house manufactured button box was used to collect responses in all tasks. A short description of each of the tasks will now follow, and the minor differences to the earlier study (Brederoo et al., 2019) will be mentioned. In the face similarity task (Brederoo et al., 2019) eassessing lateralized face processinge, participants were presented with a neutral face image and two symmetrical composites of that same image: one consisting of twice the left side, the other of twice the right side of the original image. Participants then had to judge which of the two composite faces resembled the original image most. In the lexical decision task (Hausmann et al., 2019;Willemin et al., 2016) eassesing lateralized word processinge, participants saw strings of letters to the left and right of a central fixation point. Participants had to indicate whether the left, right, or neither of the letter strings was a valid word. In the picture matching task (Peyrin, Mermillod, Chokron, & Marendaz, 2006) eassessing lateralized spatial frequency processinge, participants were presented with images of natural scenes. A first centrally presented image (S1) was followed by the same or a different image (S2) to the left or right of central fixation, the latter one being filtered to contain only relatively low or high spatial frequencies. Participants indicated whether the S1 and S2 had depicted the same natural scene. Of note is the fact that in our previous replication study, the picture matching task produced less convincing results than the other tasks used in the current study (Brederoo et al., 2019). We opted to use it nevertheless, as we were unaware of a more suitable task to measure lateralization of spatial frequency processing. In the hierarchical letter task (Brederoo, Nieuwenstein, Lorist, & Cornelissen, 2017) eassessing lateralized global form and local feature processinge, participants were shown so-called Navon letters, one to the right and one to the left of a central fixation point. A prespecified target letter could appear as the local elements making up one of the Navon letters, as the global Navon letter, or be absent. Participants indicated whether the target letter had been present. In contrast to the earlier study (Brederoo et al., 2019), only bilateral presentation was used and the presentation durations were slightly different: a trial started with a blank screen, lasting 280 msec, followed by a centrally displayed fixation asterisk for 500 msec, and the Navon letters were presented for 100 msec. In the landmark task (Cai et al., 2013;Linnell, Caparos, & Davidoff, 2014) eassessing spatial attention biase, participants were presented with a horizontal line that was transected by a vertical line at .2 , .6 , or 1 to the left or right from the midpoint. 3 Participants had to judge whether the transection occurred to the left or right of the midline.

General procedure
The experiments took place in a darkened and soundattenuated room. Distance to the monitor (22 00 , 1280 Â 1024, 100 Hz in Groningen; 24 00 , 1920 Â 1080, 100 Hz in Ghent) was kept fixed using a chin rest to ensure stability of the visual angle. All experimental tasks were run in E-Prime (E-prime Psychology Software Tools Inc., Pittsburgh, USA), and were preceded by the Flinders Handedness Questionnaire to measure handedness and the Dolman Method to measure eye dominance. As eye dominance is not a focus of this study, we will not elaborate further on it.
As the strength of RH-lateralization in the landmark task has been suggested to decrease over the course of an experimental session (Manly, Dobler, Dodds, & George, 2005), participants always completed this task at the start of the session. After that, the participants performed the remaining tasks (face similarity; lexical decision; picture matching; and hierarchical letter tasks), the order of which was randomized and counter-balanced over participants. 4

2.4.
Data pre-processing Before analyses, we inspected the data per task to remove the data of participants who performed at chance level. To do so, error rates (ERs) were computed separately for left visual field (LVF) and right visual field (RVF) trials in the visual half-field tasks. A participant's data were removed only when he or she performed at chance level in both the LVF and RVF, assuming that when performance is at chance level in one but not the other visual field this could be considered to reflect lateralization rather than poor performance. Removing data when performance in both visual fields was at chance resulted in missing data of 1 participant for word processing, 1 for LSF processing, 1 for HSF processing, 10 for local processing, and 1 for global processing. Due to a coding error, 2 participants' data were lost for spatial attention.
We used the outlier removal procedure as described by Van Selst and Jolicoeur (1994) to remove outliers in the RT data. This resulted in the removal of 1.64% of the trials in the hierarchical letter task, 2.23% in the lexical decision task, and 2.58% in the picture matching task.
For each of the tasks, we first conducted analyses to confirm that they produced the expected, population-typical lateralization effects. LVF and RVF performance were compared using paired t-tests on ERs and RTs to assess lateralized processing of LSF and HSF, global and local, and word processing. Lateralization of face processing and spatial attention were assessed with a one-sample t-test, testing the difference of the visual field bias against zero.
The results of these analyses, reported in Appendix A, showed that, except for the picture matching task, all of the tasks indeed produced the expected lateralization effects, thus corroborating the findings of our earlier study (Brederoo et al., 2019).

Outcome variables for correlational analyses
The main analysis of interest examined the correlations between lateralization indices of the different processes. To assess the degree of lateralization for each process, we derived a scaled index for the extent to which performance and judgments differed for stimuli shown in the left and right visual fields. For performance-based measures (i.e., error rates and reaction times), this index was computed by subtracting a participant's RVF-performance from his or her LVFperformance, and dividing it by the sum of both. Accordingly, positive values for these indices indicate the presence of an RVF-advantage, suggestive of LH-dominance for the task in question, whereas negative values indicate the presence of an LVF-advantage. For the face similarity task, an analogue to a performancebased index of lateralization was derived by subtracting the proportion of choice for the right composite face (indicating that participants judged the face based on the facial information on the right side) from the proportion choice for the left composite face (indicating that participants judged the face based on the facial information on the left side). Thus, negative values on this measure of the face similarity task indicate an LVF-advantage in processing faces, while positive values indicate an RVF-advantage.
Lastly, for the landmark task, the index of lateralization was defined as the point of subjective equality (PSE), with 0 being assigned to the veridical point of equality. Consequently, negative values for the outcome of the landmark task indicate an LVF-bias suggestive of RH-dominance in allocating spatial attention, whereas positive values indicate an RVFbias indicative of LH-dominance in spatial attention.

Statistical analysis of correlations
To assess the extent to which our data supported the presence (H1) or absence (H0) of the hypothesized correlations (see section 1.1), we used Bayesian analyses. An advantage of Bayesian statistics over null hypothesis significance testing (NHST) is that it provides information about the likelihood of the null hypothesis being true, given the data. In contrast, NHST only allows for rejection of the null hypothesis. Specifically, in Bayesian analyses, evidence is based on the relative plausibility of the data under the alternative (H1) versus the null hypothesis (H0) (Wagenmakers et al., 2017). A Bayesian analysis produces a Bayes factor (BF 10 ), where BF 10 > 3 indicates moderate, BF 10 > 10 indicates strong, BF 10 > 30 indicates very strong, and BF 10 > 100 indicates extreme evidence for H1, while BF 10 < 1/3 indicates moderate, BF 10 < 1/10 indicates strong, BF 10 < 1/30 indicates very strong, and BF 10 < 1/100 indicates extreme evidence for H0 (Jeffreys, 1961). When the BF 10 ranges between 3 and 1/3, the data are said to be inconclusive with regard to the hypotheses. We computed pairwise correlations for each possible pair of lateralization effects using the libDienesBayes package in R, and used a uniform prior (making no specific predictions with regard to the strength of the correlations) with the lowest split-half reliability of the two correlated lateralization indices as upper bound. The reported correlations were corrected by accounting for the split-half reliabilities of both measures (r corrected ¼ r/√[reliability 1 * reliability 2 ]). These choices for an upper bound of the lowest split-half reliability in the Bayesian analyses, and for correction of the correlations with the splithalf reliabilities, were made on the grounds that a measure cannot show a larger correlation with another measure than it can with itself. Split-half reliability was computed as the correlation between lateralization indices for odd and even trials in each task, corrected for halving the length of the task by means of the Spearman-Brown formula, and can be found in Appendix A. In each of the tasks, there were equal numbers of LVF-and RVF-trials within blocks, and randomized presentation order per participant, warranting the calculation of split-half reliabilities.

Subgroup-comparisons
We contrasted the lateralization effects for right-, lefthanded, and RH-dominant participants. All these paired contrasts were made using one-sided Bayesian t-tests, based on the associated hypotheses (see section 1.1).

Results
In all tasks, there was substantial evidence for the expected RVF-and LVF-advantages, with the exception of those in the picture matching task, which showed no evidence for the presence of an RVF-advantage for HSF processing (see Appendix A). Nevertheless, we did include this non-significant RVF-advantage in all following analyses, as there could still be a relation of HSF-lateralization to lateralization of the other processes and/or differences between subgroups. Split-half reliability of each of the measures was good (all BF 10 > 38), ranging from r ¼ .45 for HSF processing to r ¼ .86 for word processing (see Appendix A). In the following, for all derived laterality indices and statistics, positive values indicate RVF/LH-lateralization, and negative values indicate LVF/RH-lateralization.

Correlations between lateralization indices
We here report the correlations for which we found at least substantial evidence in favor of their presence (see Fig. 1) or absence. An overview of all correlations, including those for which the data were inconclusive, can be found in Appendix A.
There were no relations between spatial attention bias and lateralization of the other visual processes, with inconclusive evidence with regard to positive relations with lateralization of global processing (BF 10 ¼ .629, t[112] ¼ .945) and that of word processing (BF 10 ¼ .532, t[112] ¼ 1.16), and support for the absence of any relations with lateralization of the other processes (all BF 10 < .25, |t| < .529).
In addition to lateralization indices, proportions of participants who show typical lateralization within a subgroup can be informative on the direction of lateralization for different processes. A table reporting these proportions can be found in Appendix A.
In addition to the presence of these subgroup differences, we found support for the absence of a number of differences. 5 We deem ERs and RTs to be reflective of the same lateralized processing elateralization effects in ERs and RTs of the same processes were always in same direction (see Appendix A) and did not correlate in an opposing manner with other lateralization effectse, and here report the highest of correlations in ERs and RTs, when lateralization of a process correlates substantially with lateralization of another in both these measures. Fig. 1 e Correlations between scaled lateralization indices of right-handed (blue), left-handed (yellow), and RH-dominant (orange) participants. In each diagram, the grey-colored area depicts the locus of typical lateralization patterns. Larger positive and negative values indicate larger RVF-and LVF-advantages, respectively. ** BF 10 > 10; * BF 10 > 3.16. Fig. 2 e Lateralization indices and accompanying probability densities for right-handed, left-handed, and RH-dominant participants. Represented values are the scaled indices by dividing by the root mean square. White diamonds represent the means, where larger positive and negative values indicate larger RVF-and LVF-advantages, respectively. *** BF 10 > 100; ** BF 10 > 10; * BF 10 > 3.16;~BF 10 > .316 < 3.16; x BF 10 < .316, where H 1 is that the mean is higher or lower (depending on the hypothesis) than zero.
Right-handed participants did not differ from left-handed participants with regard to lateralization of local processing in ERs (BF 10  Evidence was inconclusive for all other pairwise comparisons between the subgroups (BF 10 > .33 < 2.57, |t| < 1.74).

Summary of results
Before evaluating our findings in light of the previously proposed principles underlying patterns of lateralization, we present a short summary of the results. As predicted, grouplevel analyses indeed gave rise to a 'typical' pattern of lateralization: left hemisphere (LH) processing of words and local features; right hemisphere (RH) processing of faces, global form, and low spatial frequencies (LSF), and a RH spatialattention bias. The evidence for the expected LHlateralization of high spatial frequency (HSF) information processing was not substantial, confirming neither its presence nor its absence in the group as a whole.
In addition to the group-level analyses, we investigated possible differences between right-handed, left-handed, and RH-dominant participants. We found that, as a group, righthanded participants showed the typical lateralization pattern, with the exception of LH-lateralization for HSF processing. RH-dominant participants did not show the expected reversed lateralization pattern, but their results were characterized by an absence of lateralization effects, with the exception of inconclusive data with regard to RHlateralization for global processing and RH spatial attention bias. Left-handed participants showed results more similar to right-handed participants than did the RH-dominant participants, but still deviated from the typical pattern. For lefthanded participants, RH-lateralization for LSF processing and RH spatial attention bias were absent, and the data were inconclusive with regard to RH-lateralization for face processing and LH-lateralization for HSF. Left-handed participants did show typical lateralization for word, local, and global processing.

4.2.
Principles governing patterns of lateralized processing When considering the implications of the current findings for the previously proposed principles underlying patterns of lateralization, we can conclude that the input asymmetry and causal complementarity principles are best supported (Table  2). These principles are not mutually exclusive, but rather complement each other in explaining how lateralization of related processes comes about.

Mixed support for causal complementarity
In accordance with the causal complementarity principle (Bryden et al., 1983), which proposes that different processes recruiting similar brain regions will come to be lateralized to homologue areas in opposite hemispheres, our correlational analyses support a relation between LH-lateralization for word processing and RH-lateralization for face processing, where an increase of one co-occurs with an increase in the other. This is in line with a similar correlation found between the strength of LH-lateralization for visual word processing and that of RH-lateralization for face processing in a group of children who were learning to read (Dundas et al., 2015). Our results furthermore corroborate those reported by Badzakova-Trajkov et al. (2010) and Gerrits et al. (2019). In both these neuroimaging studies, LH-lateralization of brain regions activated during language production (i.e., letter fluency task) correlated with RH-lateralization of brain regions activated during face perception.
While the causal complementarity principle further predicts reversed lateralization patterns (i.e., LH-lateralization for face processing and RH-lateralization for word processing) for individuals who are RH-dominant for language, our data did not support such a pattern. Instead, the subgroup of participants who were RH-dominant for language showed an absence of RH-lateralization for face and LH-lateralization for word processing. One possible interpretation of the absence of reversed typical lateralization patterns in RHdominant individuals is that causal complementarity is not a very strong driving force for functional segregation, resulting in subtle lateralization patterns. As a case in point, Badzakova-Trajkov et al. (2016) suggest that the mirroring of different functions to homologue areas does not take place as absolutely as the causal complementarity principle would dictate. Their results show that while eas predicted by the causal complementarity principlee one part of Broca's homologue in the RH is activated by face stimuli (i.e., the pars opercularis), another part of Broca's RH-homologue (i.e., the pars triangularis) is not. Relating to this, H€ aberling, Corballis, & Corballis (2016) entertain the possibility that lateralized brain systems can evolve following the causal complementarity principle (i.e., by competing pressure for cortical space), but once instantiated will go on to develop more independently. As a result, a directly observable relation between the functionally segregated functions will dissipate over the course of evolution.
Alternatively, it is possible that causal complementarity underlies the functional segregation of face and word processing, and that RH-dominant individuals in fact do tend to display reversed patterns, but that our sample of n ¼ 13 was too small to detect this. What we can conclude based on our Bayesian analyses, is that in case of the RH-dominant participants the data supported the absence of typical lateralization of face and word processing. Due to practical limitations, we were unfortunately unable to enlarge our sample of RHdominant participants.
Finally, these findings could indicate that lateralized processing in RH-dominant individuals does not adhere to the same principles as it does in LH-dominant individuals. Given the scarcity (±6%, Knecht et al., 2000) of RH-dominance for language, it is unsurprising that our understanding of lateralization in this group of individuals is limited as of yet (see also . Our results can be taken as encouragement for future research to further explore lateralized processing in its atypical as well as typical form.

Support for input asymmetry
In support of the input asymmetry principle, we found the predicted correlations between lateralization of low-level processes and lateralization of associated higher-level processes in the same hemisphere (Andresen & Marsolek, 2005). The only such predicted relation for which we did not find conclusive support was that between lateralization of face and LSF processing. Specifically, as predicted by the input asymmetry principle, we generally found that stronger LHlateralization for local feature processing was associated with stronger LH-lateralization for word-and HSF-processing. Complementing this, stronger RH-lateralization for face processing was associated with stronger RH-lateralization for global form processing. We further found that stronger RHlateralization for face processing was associated with stronger LH-lateralization for local feature processing. However, the input asymmetry principle also predicted positive relations between RH-lateralization for face processing and LSF processing, and between LH-lateralization for word processing and HSF processing; relations which were absent in the present study. The input asymmetry principle further predicted the relation of lower-level processes to mirror those of higher-level processes. Indeed, in our data the stronger RHlateralization for global processing was associated with stronger LH-lateralization for local processing, which is in accordance with the relation between the higher-level face and word processing.
Furthermore, rather than making predictions with regard to typical and atypical lateralization patterns in individuals varying in handedness and hemispheric dominance for language, the input asymmetry principle predicts that whatever is the pattern found for lower-level processes, should be the pattern found for higher-level processes. Indeed, our results support this notion in that (1) typical lateralization of spatial frequencies (with the exception of HSF processing) and global and local processing co-occurred with typical lateralization of face and word processing; and (2) the absence of lateralization of spatial frequencies co-occurred with absence of lateralization of face and word processing in RH-dominant individuals. As such, lateralization patterns, or lack thereof, of higher-level processes mirrors that of lower-level processes.

Statistical complementarity for attention and vision
According to the statistical complementarity principle, the distribution of lateralization of different processes arises by chance (Bryden et al., 1983). Based on a factor analysis of neuroimaging data, Liu et al. (2009) suggested that such independent lateralization is the case for the domains of vision, language, attention, and internal thought. In line with Liu et al. (2009), we showed that spatial attention bias does not relate to lateralization of any of the other processes, and as such can be considered to be statistically independent from lateralization of the remaining visual processes under study. This is in line with the factor-analysis reported in Badzakova-Trajkov et al. (2016), which also suggested the existence of independently lateralized brain systems for face processing and spatial attention. Furthermore, in showing that our handedness groups differed in language lateralization but not in spatial attention bias, we corroborated earlier findings (Badzakova-Trajkov et al., 2010;Karlsson, Johnstone, & Carey, 2019), and provided further support for statistical independence of spatial attention and language lateralization. However, our results differ from two earlier studies showing support for causal complementarity of spatial attention and language production (Cai et al., 2013;Zago et al., 2015). Specifically, these studies showed reversed typical patterns consisting of LH-lateralization for spatial attention and RH-lateralization for verbal fluency in a group of RHdominant individuals (Cai et al., 2013), and a correlation between RH-lateralization for spatial attention and LHlateralization for language production in a group of lefthanded participants (Zago et al., 2015). In accommodating these differences with our results, it is important to note that our sample included right-handed as well as left-handed participants, while these results by Cai et al. (2013) and Zago et al. (2015) are based on groups consisting solely of lefthanded participants. Zago et al. (2015) separately tested a group of right-handed participants, and found no evidence for c o r t e x 1 3 3 ( 2 0 2 0 ) 2 0 1 e2 1 4 a relation between spatial attention and language within this group. In the study by Zago et al. (2015), right-handed and lefthanded participants were not pooled together for the analyses, unfortunately precluding a direct comparison with our results. In relation to this, Gerrits et al. (2019) eusing a sample of only left-handed participants, failed to show a relation between lateralization for face processing and visual word processing (with p ¼ .065), while in our data this correlation was evident. These observations may be taken as a reminder of caution in selecting participants; while in laterality research it is important and fortunately good custom to include lefthanded participants, the exclusion of right-handed participants may come with its own price. To prevent the emergence of incomplete or distorted depictions of lateralization patterns, future research on this topic should use participant samples that are maximally heterogeneous with regard to handedness and hemispheric dominance for language. Finally, we showed that lateralization of processes within the domain of vision are not statistically independent: lateralization of each of the measures of visual information processing was correlated to lateralization of another. Furthermore, the statistical complementarity principle predicts there to be no difference in lateralization patterns between individuals based on their handedness and/or hemispheric dominance for language. As described above, right-handed, left-handed, and RH-dominant participants were in fact shown to differ in terms of lateralization patterns, further discrediting the notion of independent lateralization of the visual processes under study. Taken together, our results support statistical independence of the lateralization of attention and vision, but not within the domain of vision.

Conclusion
In sum, the typical and deviating patterns of lateralization of face, word, global form, local feature, low and high spatial frequency processing can best be explained by the governing principles of input asymmetry and, to a lesser extent, causal complementarity. Our results are partly in line with the notion that processes recruiting similar resources will come to be lateralized to homologue areas in a manner that promotes intra-hemispheric proximity to other cortical areas within their processing hierarchies. In the case of word and face processing, the former will be driven to the LH because of its specialization in language, local feature and high spatial frequency processing, while the latter will be driven to the RH because of its specialization in global form and low spatial frequency processing. We further suggest that statistical complementarity applies to the relation between lateralization of attention and vision.
In the present study, group-level analyses gave rise to typical lateralization patterns, while separate analyses for subgroups differing in terms of handedness and RHdominance for language provided a more nuanced view. Future research should keep studying lateralized processing in individuals who are expected to deviate from typical patterns, so as to increase our understanding of hemispheric specialization in all its diversity and complexity.