Introduction

Speech production is one of the most clearly lateralized functions in the human brain. Control is dominated by the left hemisphere (LH) in 88% to 96% of right-handers [1–3]. Cases of atypical speech lateralization in the right hemisphere (RH) or bilateral patterns can be found in large-scale neuroimaging studies with functional transcranial Doppler Sonography or functional Magnetic Resonance Imaging (fMRI), but they require a serious investment before a group of 20 participants is obtained. Clear RH dominance is mainly found in left-handers with estimates ranging from 6.5% to 27% compared to 0 to 4% in right-handers [1–4]. Particularly in healthy student populations the prevalence of RH speech dominance seems to be limited to some 10% of the left-handers [2–4].

Because of their larger variability, left-handers should be the first choice of study in experiments that aim to study correlations of functions. Surprisingly, that is not the case. Most studies are either limited to right-handers [5] or include a small number of left-handers (because the prevalence of left-handedness is limited to some 10–14% of the population). A small number of left-handers is a problem because the majority of left-handers is still LH dominant (see also[6, 7]) so that the number of participants with non-LH dominance in an unselected sample can be expected to be very small (some 20% of 10% or 2 out of 100 participants). Indeed, [8] showed that left-handedness by itself does not predict a reversed ear asymmetry in the dichotic listening task; such reversal was only observed when the lefthanders were limited to a group with established atypical speech dominance. The typical right ear advantage (REA) pointing at LH speech perception dominance was observed both in left- and right-handers, but an atypical left ear advantage (LEA) was only present in a subgroup of left-handers previously identified as being RH speech dominant.

Language processing and specifically speech laterality based on fMRI measurements is an interesting starting point to classify subjects in (a)typical categories for brain organization because of its strong lateralization and known incidence of RH dominance by large scale neuroimaging studies [9]. It is however also important to know how other functions are lateralized in left-handers to investigate cognitive relationships in healthy individuals. Hemispheric specialization characterizes many functions such as emotional processing, spatial attention, memory etc. [10]. The present article takes a first step by relating indices of face and language processing, together with different measures of handedness, to the laterality of language (speech production, reading, and speech perception).

The present study contributes to the existing literature in five ways. First, we provide the distribution of lateralization indices for the three main functions of language in the same sample of participants. Previous (neuroimaging) studies combined maximally two language-related tasks in right-or left-handers (e.g.[4, 11]). They concluded that language subprocesses can occasionally lateralize to different hemispheres or, more likely, to a different degree. In particular, evidence for unilateral control was stronger for the inferior frontal gyrus involved in speech production than for the ventral occipito-temporal region involved in reading. The current study additionally looked at the correlations with speech perception.

Second, all current tasks are behavioral computer tasks. Lateralized behavioral tasks have extensively been used in the past [12–15]. For vision, the visual half field (VHF) task is used most often. In this task better performance to stimuli presented in the left visual half field (LVF) indicates RH functional dominance and better performance to stimuli in the right visual half field (RVF) indicates LH dominance, because the optic fibers partially cross on their way from the eyes to the brain. We used methodologically optimized versions of the VHF task following guidelines from [16–17]. A new lateralized paradigm was developed to test reading lateralization (see below). For auditory modality, dichotic listening is the preferred technique. We used the standardized version of the Bergen group, which has been used in many research projects (e.g.[18]).

Behavioral tasks are interesting, because they require fewer resources than neuroimaging studies. They are particularly interesting for the initial screening of participants with potential atypical brain dominance [4]. Computer tasks are less time-consuming, expensive and restrictive for participant inclusion than neuroimaging techniques (e.g. they do not suffer from the problems of irremovable ferromagnetic material in the participants, noisy environment, and claustrophobia often encountered in fMRI research). We therefore wanted to explore whether a combination of behavioral tasks could provide us with consistent individual lateralization patterns that can have predictive value for the relationship between lateralized neural functions.

Behavioral tasks are unlikely to be specific enough for high-stake assessment (e.g., to determine speech lateralization in patients who have to undergo brain surgery), but they can serve as a quick first screening. In particular, we want to know whether VHF tasks can give a quick first indication of the lateralization of a wide variety of functions related to language, memory, emotional processing, face processing and so on. For this we need validated tasks, preferentially versions that do not take too much time. [19] gives an example of a smartphone app based on dichotic listening, which can be used for screening ear preferences in large groups of the population.

The third contribution of this study is that we added a face VHF task to the language tasks. This allowed us to further examine the many-to-many view on brain organization proposed by [20]. According to this view, face processing elicits a LVF/RH advantage in people who have learned to read because face processing and reading compete for neural resources in the ventral visual cortex (see for example [13] for first behavioral evidence in right- and left-handers). As a result, learning to read (which critically depends on the language dominant hemisphere) pushes face processing to the less taxed homolog brain region in the non-dominant hemisphere. Given the available preliminary evidence for this view we aimed to validate the contralateral dominance of reading and face processing in our large sample of left-handers containing a wide variety of lateralization patterns.

The fourth contribution of the present study is that we related laterality information to language performance, as measured with tests of naming speed and vocabulary size. Mixed results have been reported in the literature. [21] argued that the issue is not yet resolved because researchers rarely measure overlapping functions. However, even in studies that carefully matched lateralization function and performance test, inconsistent results have been reported. [14] for example used word and face VHF tasks and reported an inverted U-shaped curve, with optimal performance in the case of a bilateral VHF advantage and worse performance with more extreme LVF and RVF advantages. In contrast, [22] used data from the Bergen dichotic listening database and reported a U-shaped curve, with better performance for clear right or left ear advantages than for cases with no clear asymmetries. [23] also found a U-shaped curve when relating speech production fMRI LIs to a dozen of verbal and spatial performance tests. This study included a high number of left-handers from the BIL&GIN dataset [24], so that more variability was present in the participants. An aphasia patient study by [25] confirmed the finding of better language performance with more extreme lateralization indices in the case of rapidly progressing brain lesions. Their patients were all right-handed and showed LH lateralization in a verb generation task. Optimized lateralization thus seems to differ across language processes according to these studies. In the current study we further investigated the relationship between language laterality and performance by comparing LIs from behavioral tasks in a large left-handed sample.

Finally, the interpretation of correlations between laterality indices critically depends on the reliabilities of the indices. Two indices will not correlate if they are not reliable. Therefore, absence of correlation cannot be interpreted as evidence for independence of function unless the indices are shown to be reliable. The best way to assess the reliability of an index is to test it twice. For a well-designed behavioral lateralization task performance should vary more between participants than within participants tested twice. Lateralization indices in the sample should in addition have a broad distribution range to be able to represent individual variation [26]. We therefore invited 50 of our 100 left-handed participants for a second session one week after the first to redo the language and face test battery, so that we could evaluate test-retest reliability.

To meet the above aims, the following tasks were included. A first set consisted of two lateralized language tasks validated by [4] as a measure of speech production laterality. In these VHF tasks, participants are asked to name words (task 1) or pictures (task 2). Different words/pictures are presented in LVF/RVF and participants have to name the word/picture indicated by a centrally presented arrow as fast and accurately as possible. The LVF/RVF advantages correlated up to r = .76 with fMRI LIs in a word generation task (i.e. mentally think of as many words as possible starting with a target letter) in the study of [4]. Everyone who named RVF stimuli at least 25 ms faster than LVF stimuli was confirmed to be LH speech dominant in fMRI and all but one participant with an LVF advantage of at least 60 ms turned out to be atypically RH lateralized.

A lateralized lexical decision task based on [27] was developed as a behavioral predictor for reading lateralization (task 3). In that study, the authors showed that participants with typical language dominance named words faster when they fixated the first letter of a word than when they were forced to look at the last letter. The optimum for RH dominants lied more towards the word end. Words fixated at the first letter are transmitted almost entirely to the LH whereas words fixated at the last letter initially go to the RH because of the partial crossing of optic fibers. This pattern is part of the optimal viewing position effect (OVP; [28]) and indicates that there is no bilateral representation of the fovea but that letters in central vision are contralaterally sent to the brain instead [27]. An advantage of the OVP technique is that the words can be presented in central vision, rather than in parafoveal vision, where visual acuity is lower (and words are more difficult to identify). In the current lexical decision task, we presented six-letter words and pseudowords that could be fixated either at the first of last letter because naming latency differences between laterality groups were largest for this word length in [27].

The fourth task was the dichotic listening paradigm developed at Bergen University (e.g.[18]). We only used the non-forced condition of the paradigm, in which participants are asked to report which of two binaurally presented sounds they heard best or first. More reports from the left ear point at a RH dominance for auditory speech perception, whereas more reports from the right ear point at an LH dominance [29]. We chose to double the number of trials in the standard Bergen version from 36 to 72 and to skip the forced conditions in which participants have to report the stimuli from one ear and ignore the other side, because the forced conditions serve as an attention control for which we did not expect any differences between LH and RH dominant participants (confirmed in unpublished data from [8] who reported clear left/right ear advantages in RH/LH speech dominants respectively in the unforced condition and overall left/right ear advantages in the left/right forced conditions).

In addition to the language tasks, we also presented a face VHF task to the participants. It was based on [13, 20, 30], in which it was argued that word and face processing compete with each other because of overlapping brain tissue. As a result, [30] predicted stronger RH laterality of face perception when the fusiform gyrus in the LH is increasingly used for reading. In the paradigm of [13], a target face is presented centrally, followed by a face in the LVF or RVF. Participants are asked to indicate whether the two faces are the same or not by pressing buttons. We slightly adjusted the paradigm by using another face database [31], aligning fixation and stimulus presentation durations with the other VHF tasks we ran, and adding a face to the non-target side in order to avoid attentional biases. In [13] the VHF face task resulted in a group-level LVF advantage in terms of accuracy and reaction times in right-handers (N = 24) and also shorter reaction times for LVF stimuli in left-handers (N = 24). The same sample showed RVF/LH advantages in a word version of the paradigm. In our study, we wanted to explore how the distribution of VHF asymmetries looked like for a large group of left-handers containing individuals with atypical language (in particular reading) lateralization.

Handedness was measured in three ways. [32] found reduced LH story listening lateralization in participants with left-handers among their first-degree relatives and/or weak manual preferences as measured with the Edinburgh Handedness Inventory [33]. So, we included both the Edinburgh Handedness Inventory and a questionnaire about left-handedness among relatives in our test battery. In addition, we added a finger tapping task in which the participants were asked to press a button as many times as possible with their left or right indices during ten trials of ten seconds each (as in [34] cited in [24]).

The final part of our test battery consisted of three short language proficiency tests. Reading speed for words and pseudowords were measured with the Dutch één-minuut-test (EMT; one-minute-test; [35]) and the Klepel [36] in which participants named as many words or pseudowords as possible within one or two minutes respectively. These tests were included as brief indicators of reading ability because [37] showed that scores on a word identification test similar to the one-minute-test predicted more efficient eye movement behavior (e.g. reduced fixation durations) in sentence reading. [38] also showed that these tests made the biggest difference between control and dyslexic readers. We added the Lextale vocabulary test to the two naming speed tests [39]. In this test, participants are given a list of 40 words and 20 nonwords and they have to indicate which words they know (the nonwords are used to correct for guessing). We used Lextale, because [40] related higher scores on this questionnaire to asymmetries in a VHF lexical decision task in bilinguals. The current study did not take into account bilingualism even though most Flemish students can be considered as late bilinguals as they are taught French from fifth grade in primary school and English from second grade in high-school. The Lextale mainly served as a standard vocabulary test that can be related to differences in language proficiency.

Method

Results

The raw data of the VHF tasks and dichotic listening were first cleaned before we ran the main analyses. This involved the exclusion of the following trials for reaction time (RT) analyses: errors (Picture VHF: 2.5%, Word VHF: 19.4%, OVP: 9.9%, Face VHF: 24.0%), voice key failures (i.e. no trigger, triggered by environmental noise, participant saying uh, coughing etc.; Picture VHF: 14.6%, Word VHF: 3.5%), RTs less than 200 ms or greater than 1500 ms (Picture VHF: 0.5%, Word VHF: 0.0%, OVP: 0.3%, Face VHF: 10.7%) and for the remaining trials latencies above/below 2.5 SDs from the participant's mean RT (Picture VHF: 1.9%, Word VHF: 8.5%, OVP: 2.0%, Face VHF: 0.6%). Only words and no pseudoword conditions were taken into account in the OVP task RT and accuracy analyses. In the face VHF task, only Same responses and no Different conditions were included. A few data files were lost due to technical problems (e.g. computer failure, a badly timed fire drill etc.; Picture VHF: N = 1, Word VHF: N = 1, OVP: N = 2, DL: N = 1, Face VHF: N = 1, One-minute-test: N = 4, Klepel: N = 1). Finally, seven data points were excluded upon visual inspection of the trade-off between individual mean RT and accuracy (see S1 Table; Picture VHF: N = 1, OVP: N = 3, Face VHF: N = 3). Data points were only removed for the outlier task. Other data from these participants were kept, except for the same task in the second session. One more data point in the Word VHF task was removed because the participant only named 3% trials correctly in the LVF. Table 1 shows the mean (and SD) RTs and accuracy scores for the LVF and RVF in the first session of each task together with the effect sizes of the VHF difference (based on the means, SDs and correlation of both VHFs; [44]) because they may be of interest for future meta-analyses. Analyses reported below were based on LIs calculated with formula (1) to enhance the comparability across tasks (with for DL right ear reports–left ear reports instead of LVF-RVF for the VHF tasks with reaction times so that a higher score always represents a higher LH advantage), unless otherwise mentioned. All individual raw and LI data can be found in the S1 Table.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Mean reaction times, accuracy scores (in % errors) and their standard deviations in the left and right visual field of the visual half field tasks.

The mean (and standard deviation) lateralization index is calculated based on formula 1 (see main text). Cohen’s d is provided as effect size measure of the visual half field differences and is based on the means, standard deviation and correlation of both visual half fields. For the dichotic listening task, the means and standard deviations of left and right ear matches are reported instead of reaction times.

https://doi.org/10.1371/journal.pone.0208696.t001

Prevalence of (a)typical lateralization per task

Fig 1 shows the reaction time LI distributions of the participants for the picture VHF, word VHF, OVP, DL and face VHF tasks.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Reaction time lateralization indices for the laterality tasks.

Reaction time lateralization index (LI) distributions for the picture visual half field (VHF), word VHF, Optimal Viewing Position (OVP), dichotic listening and face VHF tasks. LIs are sorted from most negative to most positive scores per task. Note that only the dichotic listening chart has a deviating scale on the y-axis to improve visibility because LIs were based on left/right ear matches and not reaction times.

https://doi.org/10.1371/journal.pone.0208696.g001

Fig 2 shows the same information when the laterality indices are categorized as typical, bilateral, and atypical. Panel A is based on individual raw differences of mean RT LVF–mean RT RVF. For the DL task, which does not require the participant to respond as fast as possible, the LI based on accuracy scores was taken because DL results are usually reported in this form in the literature. Negative differences below -10 ms were considered as an indication of RH dominance, differences above 10 ms as an indication of LH dominance and values in between as an indication for an unclear difference or bilateral lateralization pattern. The percentages below the graph confirm the expected LH dominance for language and RH dominance for face processing in the majority of participants, ranging from 65 (Face) to 74% (OVP). Only the DL task classified slightly less than half of the participants (46%) as clearly LH dominant with this LI = 10 criterion. Clear atypical dominance is present in 9% (OVP) to 29% (face VHF). The prevalence of non-clear typical dominance (i.e. unclear patterns and clear RH dominance together) is around 30% for most tasks, except for the DL task that elicited more bilateral patterns (in 33%) than the other tasks. Note however that the testing procedure and scoring of DL deviates from the other tasks which may lead to a different distribution across laterality types. We chose to report the raw data in this way, because it is the most commonly used method and thus enhances comparison with previous studies. DL LI scores are often cut off at a value of 0 instead of 10 which in this sample results in 65% participants with a typical right ear/LH advantage.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Percentage participants categorized as showing typical, bilateral or atypical lateralization based on their reaction times.

Reaction time distributions expressed as percentages and divided in a typical, bilateral and atypical category for the picture visual half field (VHF), word VHF, Optimal Viewing Position (OVP), Dichotic Listening (DL) and face VHF tasks. Panel A classifies the participants based on raw differences between the left and right visual field or ear. The data in Panels B and C are based on lateralization indices (LIs) that represent a mean reaction time difference of 10 ms (same values as in Panel A for the DL task) and 0 ms respectively.

https://doi.org/10.1371/journal.pone.0208696.g002

Panel B shows the distribution based on LI scores for each task. We further chose to calculate an LI limit for the three categories based on the current dataset, i.e. LI scores that represent a mean RT difference of 10 ms between the VHFs in our sample (absolute LI boundary value for Picture VHF: 0.71, Word VHF: 1.13, OVP: 0.97, Face VHF: 1.03). A 10 ms/items boundary stays an arbitrary choice, but recalculating the LIs in this way optimizes the comparability across tasks within this study. Moreover, an LI of 1 equaled a 14 ms (Picture VHF), 9 ms (Word VHF), 11 ms (OVP) and 10 ms (Face VHF) RT VHF difference indicating that an overall 10 ms boundary is a reasonable choice for this dataset. This is confirmed by the fact that the percentages in each laterality category barely change from Panel A to B. A difference of 10 reported consonant-vowel syllables from the left and right ear in the DL equaled an absolute LI boundary value of 21.1. This again clearly deviates from the other values because the maximum difference is 60 items compared to 1300 ms in the other tasks and seems not appropriate given that this results in only 23% participants with a clear LH dominance. This is unrealistic given that we know that speech perception lateralizes to the LH in the majority of people. It is however impossible to determine an LI limit that is comparable to the LI boundaries in the RT experiments. Panel B therefore kept the original 10 LI boundary for DL. In addition, we show the prevalence of (a)typical lateralization per task based on a LI cut-off score of 0 in Panel C. We believe leaving out bilateral patterns is the only way to compare laterality tasks using a different method given the difficulty of determining universal LI limits. A classification based on a LI 0 ms/0 ear reports difference mainly increased the number of typical LH dominants in the DL and OVP tasks. This brings the overall image of laterality categories to a comparable level across all included tasks, i.e. typical patterns ranging from 65% (DL) to 86% (OVP) and atypical patterns ranging from 14% (OVP) to 35% (DL). The OVP lexical decision task elicited the highest percentage of typical LH lateralization but bear in mind that no less than 20% participants fell in the bilateral category in Panel B. Finally, if we count how many participants belong to the same laterality category for all task LIs based on a 0 LI cut-off score 26% was typically lateralized overall and none was atypical for all tasks. These percentages increase when only taking the language tasks into account without the face VHF (Typically LH dominant: 39%, atypically RH dominant: 5%). These percentages are still far below what can be expected based on previous studies and the percentages per task in Fig 2, especially in the typical group.

Relationship of degree of lateralization between behavioral laterality tasks

Exploring Pearson correlations between the LIs of the laterality tasks, handedness measurements and language performance tests can shed further light on how these variables are linked in left-handers on a continuous scale. These results tell more about the influence of degree of lateralization than the categorical approach reported in 3.1. In this section, we will present correlations that test the relationship between the abovementioned variables based on RT LIs and accuracy LIs together with correlations between RT and accuracy for each task where applicable, and scatter plots to contrast two tasks against each other.

Reaction time, accuracy and reaction time-accuracy correlations.

Table 2 displays the Pearson correlations between all variables tested, with the LIs based on RT for the VHF laterality tasks. First, the language tasks correlate significantly and positively with each other as expected even though the correlations are rather low (between r = .26, p < .05 for VHF picture-DL and r = .38, p < .001 between VHF word-OVP), except for the VHF picture with OVP task (r = .17, p = .12). The VHF face task does not correlate significantly with any other variable (ps > .09). With respect to handedness, no correlation was found between the overall Edinburgh Inventory score and any of the lateralized tasks (ps > .30), nor with the language performance one-minute, Klepel or Lextale tests (ps > .11) and not even with finger tapping or familial sinistrality (ps > .21). The total handedness score only correlated positively with the separately questioned lateral preferences for hand, ear, eye and foot use (rs between .37 and .94, ps < .001). Note however that this sample only contained left-handers to increase variability in laterality so the handedness scores represent weak to strong preferences without covering both negative left-handed and positive right-handed scores. The one-minute and Klepel reading tests strongly correlated (r = .69, p < .001), but otherwise the performance tests did not show any relationship with other variables except for a few rather weak correlations (EMT-VHF picture: r = .25, p = .02; Klepel-OVP: r = .22, p = .04; Lextale-Hand: r = .21, p = .04; Lextale-Eye: r = .33, p < .01). A polynomial curve such as a U-shaped relationship in which bilateral VHF or ear differences go together with a lower performance would however not become visible in the Pearson correlations. We therefore provide the scatter plots contrasting the laterality and language performance tests in S1 Fig. As can be seen, there is no polynomial relationship either despite quite some variability in the EMT and Klepel scores (but a limited range in the Lextale vocabulary test).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Pearson correlations between all variables tested, with the lateralization indices based on reaction times for the visual half field tasks and left/right ear matches for the dichotic listening task.

https://doi.org/10.1371/journal.pone.0208696.t002

Table 3 contains the correlations based on accuracy in the VHF laterality tasks. The included DL LIs are the same as those in Table 2. Only two positive significant accuracy correlations were observed: Between DL and VHF word (r = .31, p < .01) and between DL and OVP (r = .26, p < .05; all other ps > .08). Four weakly significant, positive correlations were found between the questionnaire data of lateral preferences and laterality tasks (Ear-VHF picture: r = .21, p = .05; Ear-VHF word: r = .26, p = .01; Eye-VHF word: r = .24, p = .02; DL-Eye: r = .22, p = .03) and three stronger correlations with the VHF face accuracies (Total EHI: r = -.29, p < .01; Hand: r = -.32, p < .01; Foot: r = -.31, p < .01).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Pearson correlations between all variables tested, with the lateralization indices based on error rates for the visual half field tasks and again left/right ear matches for the dichotic listening task.

https://doi.org/10.1371/journal.pone.0208696.t003

Third, Table 4 presents the correlations between RTs and accuracy for the laterality VHF tasks (not for DL as there is no time pressure for participants to respond) to investigate the speed-accuracy trade-off in each task and evaluate whether both latency and accuracy data could serve as an LI. The RT and accuracy LIs did significantly correlate in all four tasks despite big differences in accuracy rates (e.g. on average only 2.5% errors made in the VHF picture task compared to 24.0% errors in the VHF face task, see section 3; rs between .32 and .59, ps < .01).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Pearson correlations between reaction times and error rates for the four visual half field tasks.

The speed-accuracy trade-off correlations per task are highlighted in bold.

https://doi.org/10.1371/journal.pone.0208696.t004

Scatter plots contrasting two laterality tasks.

Fig 3 contains scatter plots showing the most interesting LI contrasts between two tasks based on RTs for the VHF tasks. Panel A contrasts the two speech production VHF tasks. Most LIs fall in the upper right quarter of the scatter plot, i.e. the quarter with participants that have a positive LI for both the word and picture task as expected based on the literature and the significant correlation reported in 3.2.1 (r = .34, p < .01). We plot data from the same tasks tested in 250 left-handers in [4] in Panel B for a comparison with the current study. For this study, the VHF picture and word LIs recalculated with formula (1) correlated with a value of r = .46, p < .001. 77% participants fell in the LH/RH category for both tasks when taking an LI cutoff of 0 compared to 63% in the current study. The LIs correlated most highly with LIs measured in an fMRI word generation task when only taking into account participants that showed clear (i.e. with a 10 ms VHF cut-off) and consistent VHF asymmetries (i.e. r = .76 and .74, ps < .001 for the picture and word task respectively). Indeed, 85% fell in the same laterality category when taking LI limits based on an average 10 ms VHF difference in [4]. A similar criterion in the current study also increases the number of consistent participants to 75%. The remaining panels of Fig 3 will mainly be evaluated based on an LI cutoff score of 0 because of the abovementioned complexity to calculate LI limits in the DL task.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Scatter plots contrasting two tasks.

Data points represent individual lateralization indices (LIs) based on reaction times. VHF = Visual Half Field; OVP = Optimal Viewing Position; DL = Dichotic Listening. Note that only the dichotic listening chart has a deviating scale on the y-axis to improve visibility because LIs were based on left/right ear matches and not reaction times.

https://doi.org/10.1371/journal.pone.0208696.g003

Panels C and D contrast the VHF word task against OVP and DL respectively. We only include the VHF word and not picture results because both tasks measure the same underlying function (i.e. speech production), correlate (see 3.2.1 and previous paragraph) and the word task is slightly more stable than the picture task (see 3.3). A cutoff value 0 classified 74% in the LH or RH laterality category for both the word and OVP task with as expected the majority of participants in the upper right quarter of Fig 3C. Leaving out participants with a bilateral VHF difference based on a 10 ms LI again helped to increase the incidence of consistent participants to 85%. Panel D (VHF word vs. DL) only placed 66% of the participants in the upper right (consistent LH dominance) or lower left (consistent RH dominance) panel. Combining OVP and DL 0 LI cutoff classifications resulted in 71% consistent participants, mainly driven by the many LH dominant patterns in the OVP task. Finally, typical or atypical lateralization for reading in the OVP and face processing in the VHF face task was found in 63% of the participants. This contrast shows the least consistent patterns, presumably again because of the large majority of participants showing an RVF advantage in the OVP task in addition to a low reliability of the currently used face task (see 3.3). Using a stricter 10 ms LI cutoff did not improve the consistency this time as expected because only 7% VHF face participants had an unclear bilateral VHF difference as reported in 3.1.

Test-retest reliability of laterality tasks

A laterality score from one test can maximally correlate with a laterality score from another test to the degree that it correlates with itself. Tables 5 (RTs) and 6 (accuracy) provide the correlations between the first and second session for each laterality task and the finger tapping task (only included in Table 6), tested in about half of the participants. The RT LIs of all dominant hemisphere tasks correlated more across the two sessions than the correlations between tasks described in 3.2.1, ranging from r = .49, p < .01 (OVP task) to r = .83, p < .001 (VHF word). In contrast, the laterality indices from the VHF face task did not correlate across the two sessions (r = .22, p = .12), pointing at a weak reliability of the task in its current form.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 5. Pearson correlations between the lateralization indices based on reaction times for session 1 (variables in rows) and session 2 (variables in columns, indicated by task name 2).

The test-retest correlations per task are highlighted in bold.

https://doi.org/10.1371/journal.pone.0208696.t005

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 6. Pearson correlations between the lateralization indices based on error rates for session 1 (variables in rows) and session 2 (variables in columns, indicated by task name 2).

The test-retest correlations per task are highlighted in bold. Note that correlations with fingertapping lateralization indices are only included in this table, not in Table 5.

https://doi.org/10.1371/journal.pone.0208696.t006

The accuracy reliability correlations in Table 6 were somewhat lower but still highly significant for the VHF word, OVP, DL and finger tapping task (Range: r = .38, p < .01 for OVP to r = .88, p < .001 for finger tapping that was not included in the previous paragraph). The test-retest correlation for accuracy in the VHF picture task was not significant (r = .07, p = .67) but note that only 2.5% errors were made on average. The accuracy LIs of the VHF face task did correlate in contrast to the latency LIs, though still rather weakly (r = .35, p = .02), presumably because of the high error rates (on average 24.0%) that can be informative to which side face processing lateralizes.

As in 3.2.2 we will also express the stability of the tasks in terms of percentage participants that belong to the same laterality category based on 10 ms LI without bilateral patterns for all tasks except DL and a 0 cutoff LI value for all tasks including DL. The percentages for the 10 ms LIs were 85 (VHF picture), 90 (VHF word), 91 (OVP) and 68 (VHF face). In other words, even though the correlated LIs were significant but moderate to high, categorizing participants as being clearly LH or RH dominant was rather stable across two sessions except for the VHF face task (a 10% error LI limit for accuracy in the face task resulted in 79% stable classifications). The percentages based on a 0 cutoff LI score were 83 (VHF picture), 83 (VHF word), 82 (OVP), 78 (DL) and 67 (VHF face; 69% for accuracy).

Discussion

The current study provides for the first time an overview of the distribution of lateralization indices for language processes (VHF word and picture task for speech production, lateralized lexical decision task for reading and dichotic listening for auditory speech perception) and face processing in a large sample of left-handers. The test battery only included behavioral computer tasks that are easy to run and half of the subjects performed the tasks twice to assess the test-retest reliability. Language proficiency and handedness measurements were included as well.

Left-handers were chosen because of their larger variability. This makes them a more interesting group to study atypical dominance and correlations between functions. Indeed, if none of the participants have atypical dominance (which may very well be the case for a group of righthanders tested), a correlation between two laterality indices says nothing about the co-lateralization of functions (as they are all lateralized to the same hemisphere), but only informs us about the correlation of the lateralization strengths of these (see also [45]).

Dividing participants in a typically and atypically lateralized group based on positive and negative LIs resulted in about 70% typical dominance and 30% atypical dominance in the VHF and DL tasks. These percentages agree reasonably well with the often cited incidence of about 27% atypical speech dominance in left-handers found by [1] who also used a cut-off value of 0 for their word generation task in functional Transcranial Doppler Sonography. fMRI large-scale studies introduced a third category with participants who show an unclear dominance pattern [2–4]. They only report 6.5 to 10% clear speech RH dominance in left-handers. The current tasks however resulted in about 20% clear atypical lateralization when leaving out bilateral patterns based on a 10 ms LI difference (except for the 8% in the OVP lexical decision task). This leads to two interesting points of discussion. First, behavioral VHF tasks seem to overestimate the prevalence of clear atypical dominance at the group level compared to fMRI studies. The method chosen seems to mainly affect the prevalence of atypical patterns. Second, the importance of choosing an appropriate way to define categories based on LIs should not be underestimated (cfr. the discussion of LIs that should be independent of arbitrarily chosen thresholds in fMRI laterality studies; [46]) even if that means that the same LI calculation method cannot be used for all tasks in the same test battery. The dichotic listening task in particular seemed to evoke more bilateral patterns when using a 10 LI boundary. This does not mean VHF tasks are better predictors of neural language lateralization because the DL task has been convincingly validated to reflect speech perception lateralization (e.g. [47]) when using a 0 cut-off value. It stresses that behavioral tasks using a different method should be treated in a different way when creating laterality categories. Asking participants to report which sound they heard best does not involve an accuracy-speed trade-off and LI boundaries are restricted by the number of items instead of milliseconds. The lower percentage of typical DL dominance may also be related to the fact that the task assesses phonological processing in speech perception (which may be more bilaterally distributed) rather than lexical and semantic processing (which may be more lateralized, as argued by [48]).

Our data further showed that LIs based on reaction times and accuracies do not lead to major differences. The picture of prevalence of (a)typical lateralization remained the same. The relationships between functions also showed the same global pattern for reaction time and accuracy LIs: language laterality tasks correlated with each other, but not with the face task (meaning that in contrast to [13] no evidence could be found for the many-to-many view on brain organization which may be due to the low reliability of the current VHF face task, see below). There were also significant correlations between reaction time and accuracy LIs within the tasks that measured both variables (i.e. VHF word, picture, face and OVP). Of course, reporting all results should be encouraged in studies but the large differences in task difficulty with performances at ceiling level in the VHF picture task and only two significant correlations between the language tasks based on accuracy LIs seem to suggest that using latency LIs for further analyses optimize the comparability across VHF studies. RT-based indices are best when accuracy is high, whereas accuracy-based measures are optimal when accuracy is low.

We also notice that in our study language lateralization did not correlate with naming speed (words in the one-minute-test, pseudowords in Klepel) or vocabulary knowledge (Lextale). This may partly be due to the low reliability of the laterality indices (the language tests are known to have reliabilities over .8), although it may be interesting to keep in mind that previous studies described in the Introduction also found limited or mixed relationships between degree of lateralization and performance (e.g. [14, 22, 23]). This is an issue that is likely to require a study with high power or a meta-analysis to be settled. It would also be interesting to see whether age or educational level influences the correlation between language lateralization and language performance. We could not test this in the current study because these two factors were too homogeneous in our sample and almost all participants scored very high on the speeded naming and vocabulary tests. More variability due to age or educational level could reveal a closer relationship with language lateralization. The finding that the handedness measures (Edinburgh Inventory combined with the Porac and Coren questionnaire, familial sinistrality and finger tapping) did not correlate with language laterality is interesting as well, although it should be kept in mind that all our participants were left-handed. This limits the range of the handedness variable. Still, it is interesting to know that for lefthanders there is little covariation between the degree of laterality and language performance.

Despite the abovementioned expected laterality patterns at the group level, the results also made clear that a behavioral laterality test battery does not yield consistent patterns across tasks at the individual level. If all five tasks were considered with a 0 LI cut-off score, only 27% of the participants would be typical dominant overall. One can debate over the exact prevalence, but it is known that the majority of people is typically lateralized. In other words, even methodologically well-controlled behavioral tasks cannot replace the more accurate laterality estimates measured in neuroimaging yet. This is particularly true for the indices close to 0. In [4] we showed that RT-based VHF differences become more useful for interpretation when they are larger, because the vast majority of participants with an RVF/LVF advantage of over 25/60 ms on the VHF naming tasks were LH/RH dominant for speech in the scanner respectively. As a result, behavioral test batteries do not seem to be useful (yet) in clinical settings. Even for a quick pre-surgery screening testing multiple cognitive functions the individual patterns are too inconclusive at this moment. In the future, patients with brain damage could reveal unique information about the relationship between functions when tested with optimized test batteries, because this population also shows a wide variety in laterality indices. Similar to [25] who reported better language performance with more extreme lateralization, we could for example speculate that performances are better if a clear separation of language and face lateralization is preserved because that is the standard organization in healthy subjects.

The low individual stability may in part be explained by the significant but sometimes rather low test-retest reliability of the tasks. The VHF picture (r = .77) and word task (r = .83) are close to the clinical standard of excellent reliability of .8, and the DL task had a good reliability above .6. The reliabilities of the new paradigms we tried out show that they are open to improvement with r = .49 for the OVP task and even non-significant reliabilities for the VHF face task (r = .22 for the RT-based laterality index, r = .35 for accuracy; see for example [49–50] for other lateralized face paradigms). In particular, finding reliable lateralized non-speech tasks turns out to be a challenge.

In order to optimize paradigms we think it is important researchers develop detailed protocols and share their findings, even when these are less neat and convincing than they had hoped. [51] advised to administer more trials in VHF tasks to obtain more reliable VHF asymmetries and thus more individual stability. They asked their participants to name five series of 100 four- and five-letter words. However, the time needed for such a test would cancel most of the benefits behavioral tasks have over neuroimaging studies (even in terms of money if one must pay for the tester and the testing room). We could not foresee this, but the lack of reliability turned out to be the main limitation of the current study. If reliability had been better, we could have concluded more about the relationship between functions tested by computer tasks. We advise future studies to first optimize the individual tasks in terms of test-retest reliability before implementing them in a test battery. Our experiences with the face recognition task further point to the importance of following existing procedures (protocols) strictly. Every change introduced may have (usually negative) consequences on the reliability and should be tested in advance.

Linking low individual consistency to the test-retest reliability raises a final interesting point of discussion. [52] recently argued that cognitive tasks robust at the group level may differ from tasks useful to reveal individual differences. Experimental psychologists try to find stable, reproducible effects at the group level with a minimum of variance between participants (therefore often excluding left-handers, see above), whereas psychologists interested in individual differences require tasks with a maximum of systematic variance between individuals. These may be different tasks, with test-retest reliability a particularly important aspect for research into interindividual differences. Stable individual differences are also important in studies looking for the degree of correlation that can be expected between tasks. If individuals cannot be distinguished on one dimension measured by task A, there cannot be a correlation with another dimension measured by task B ([53], cited in [52]).

Alternatively, it could be that behavioral laterality tasks show low intercorrelations because large networks in the brain contribute to task performance (measured with speed or accuracy). This may be too blunt an instrument to measure the laterality of functions that are limited to a very small brain region. Laterality of faces, for example, is limited to a tiny region in the fusiform area [49, 54]. It may be that this area contributes only a small part of the VHF differences observed in a behavioral laterality task. For such functions, neuroscientific techniques with a higher resolution are a better measure.

Should we conclude that behavioral tasks have to be banned from laterality research? Not yet. Behavioral laterality tasks based on speech production can serve as a valid, quick screening method for (a)typical speech dominance [4, 49] and are stable enough to reveal overall laterality patterns when the tasks are methodologically optimized. There may be other tasks of similar use, which we have not found yet. For such tasks it is not so much important that they show a big difference between left and right hemisphere processing (e.g., a big VHF advantage), but that they are related in a reliable and valid way to individual differences in brain laterality. It will be interesting to see whether such behavioral tasks can be found. Inspiration from neuroscientific findings is likely to be important in this search.

Supporting information

Acknowledgments

We would like to thank Bruno Rossion for sharing the stimuli of his face database [31] and René Westerhausen for programming the Flemish version of the dichotic listening task [8].