The study was part of a larger project consisting of several experiments at the Institute of Music Physiology and Musicians Medicine of the University for Music, Drama and Media, Hannover. Two further experiments and EEG recordings were conducted within the same two sessions in the lab and are reported elsewhere (Wenhart et al., 2018a,b). For this reason, pitch adjustment assessment as well as cognitive tests from previous publications were also used as control variables here. Therefore, all subjects participated in three parts: an online survey and two appointments in the lab. The online survey was used for pitch identification screening and diagnostic as well as demographic questionnaires (see below). General intelligence tests, a musical ability test, a pitch adjustment test (Dohn et al., 2014) and two experiments assessing local-to-global processing both in vision and audition were conducted in the lab (see Table 1).

In total, 31 absolute pitch (AP) musicians (16 female) and 33 relative pitch (RP) musicians (15 female) participated in the study. The above-mentioned online survey (UNIPARK software¹) was used to recruit participants. They primarily were students or professional musicians at the University for Music, Drama and Media, Hanover; four AP and two RP were amateur musicians. As part of the online survey, a pitch identification screening test (PIS), consisting of 36 categorical, equal-tempered sine tones over a three octave range between C4 (261.63 Hz) and B6 (1975.5 Hz) was used to allocate the participants to groups (AP: >12/36 tones named correctly, else RP). There is currently no consensus about a cutoff in terms of percentage of tones named correctly to be defined as absolute pitch possessor. We chose a cutoff of 12/36 tones named correctly and not a higher, e.g., 50 or 80% cutoff, as some of our subjects reported to have absolute pitch (from our experience, professional musicians usually know whether they themselves have absolute pitch or not) and performed like absolute pitch possessors in the absolute pitch adjustment test in the lab [see section “Pitch Adjustment Test (PAT)”], despite their comparable weak performance in the pitch identification test online. The latter might have resulted from problems with tone presentation on some of the subjects’ personal audio-devices (reported during personal communication) or from uncontrolled experimental conditions (online study). To verify this decision, scatterplots visualizing the relation between pitch naming and adjustment have been inspected. Similarly, one participant yielding 21 correct in the online pitch naming test has been re-assigned to the RP group because of weak performance in the pitch adjustment test (and reporting not to have absolute pitch). Above chance performance of musicians without “real” absolute pitch is also occasionally seen in pitch naming tests, as professional musicians can have several experience based strategies for use in the test (e.g., having pitch of empty strings in string players or starting tones of famous melodies as comparison in mind), which they frequently reported to us. The lenient threshold of 12/36 therefore is due to technical issues and does not reflect a suitable cutoff for the assessment of other samples. However, the clear difference between AP and RP in pitch adjustment test, where RP strategies do not seem to help (personal reports after test), confirms the separation into the two groups in our sample (see Table 1). Non-native German speakers had the choice between a German and an English version of the experiments (four AP subjects). All participants but one reported no regular medication or drug intake. None of the participants reported any history of severe psychiatric or neurological condition. The AP group consisted of 15 pianists, nine string players, three woodwind instrumentalists, two singers and two brass players; the RP group consisted of 13 pianists, four string players, six woodwind instrumentalists, three bassists/guitarists/accordionists, three singers, one drummer, and three brass players. The Edinburgh Handedness Inventory (Oldfield, 1971) was used to assess handedness. Apart from one subject all AP were consistently right handed, whereas three RP were left-handed and two RP ambidextrous. This study was approved by the local Ethics Committee at the Medical University Hannover. All participants gave written consent.

Two standardized tests were used to assess general nonverbal intelligence and information processing speed: Raven’s Standard Progressive Matrices (Raven et al., 2004) and “Zahlen-Verbindungs-Test” (ZVT; Oswald, 2016). AMMA (Advanced Measures of Music Audiation; Gordon, 1989), Musical-Sophistication Index (GOLD-MSI; Müllensiefen et al., 2014) and estimated total hours of musical training within life span (internal online questionnaire) served to control for musical ability and musical experience.

All statistical analyses were conducted using the open-source statistical software package R (Version 3.5³) and the R packages car, effsize, psych, sjstats and stats. Regression plots of Figure 4 were created with the R package ggplot2.

All statistical analyses have been performed on three different dependent variables per experiment: total number of correct trials (correct), reaction time medians (RT), and a combined score “Speed-accuracy-composite-score” (SACS).

Analyses revealed a main effect of hierarchical level for RT, F_RT(1,53) = 45.33, p = 1.21e-08, ηp2 = 0.75, [F_SACS(1,53) = 0.17, p = 0.69; F_correct(1,53) = 1.39, p = 0.24] and a main effect of congruency for all scores [F_RT(1,53) = 34.65, p = 2.74e-07, ηp2 = 0.55; F_SACS(1,53) = 33.30, p = 4.19e-07, ηp2 = 0.30; F_correct(1,53) = 36.76, p = 1.44e-07, ηp2 = 0.19]. Furthermore, there was a marginally significant main effect of group on RT [F_RT(1,53) 3.33, p = 0.07, ηp2 = 0.54]. Significant interactions for hierarchical level and congruency (see Figure 2) were also found for all scores [F_RT(1,53) = 7.43, p < 0.009, ηp2 = 0.12; F_SACS(1,53) = 25.27, p = 6.05e-06], ηp2 = 0.32; F_correct(1,53) = 23.31, p = 1.21e-05, ηp2 = 0.31], while a significant interaction of group and congruency was only found for total number of correct trials [F_correct(1,53) = 4.21, p = 0.04, ηp2 = 0.06] and marginally for SACS [F_SACS(1,53) = 3.53, p = 0.07, ηp2 = 0.04]. There were no three-way interactions. For means and standard deviations see Table 2.

The two-way ANOVA within the local condition revealed a main effect of congruency [F_congruency(1,53) = 55.02, p = 9.61e-10, ηp2 = 0.51] but not of group, nor was there any interaction [F_group(1,53) = 1.58, p = 0.21, ηp2 = 0.13; F_{congruency x group} (1,53) = 1.72, p = 0.19, ηp2 = 0.03]. The global condition yielded no significant main effects or interactions [F_group(1,53) = 0.39, p = 0.53, ηp2 = 0.01; F_congruency(1,53) = 0.06, p = 0.81, ηp2 < 0.01; F_{congruency x group} (1,53) = 1.13, p = 0.29, ηp2 = 0.02]. Figure 3 shows differences between conditions per group and experiment. No post hoc tests were calculated as for huge amount of comparisons in our three 3-way-ANOVAS (inflation of alpha error) and the already inconsistent picture of main effects and interactions.

Analyses yielded main effects of hierarchical level for all scores [F_RT(1,53) = 139.19, p = 2e-16, ηp2 = 0.93, F_SACS(1,53) = 94.85, p = 2.1e-13, ηp2 = 0.93; F_correct(1,53) = 119.69, p = 3.31e-15, ηp2 = 0.89] and a marginal main effect of congruency for total number of correct trials [F_correct(1,53) = 3.76, p = 0.06, ηp2 = 0.04]. Significant interactions were found for hierarchical level and congruency (see Figure 2) on RT [F_RT(1,53) = 5.56, p = 0.02, ηp2 = 0.09], for group and hierarchical level on all scores [F_RT(1,53) = 9.58, p = 0.003, ηp2 = 0.49; F_SACS(1,53) = 4.50, p = 0.04, ηp2 = 0.39; F_correct(1,53) = 6.01, p = 0.02, ηp2 = 0.28), and marginally for group and congruency on RT [F_RT(1,53) = 3.86, p = 0.05, ηp2 = 0.05]. There were no three-way interactions. For means and standard deviations see Table 3.

Two-way ANOVAs yielded a main effect of group [F_group(1,53) = 3.98, p = 0.05, ηp2 = 0.44] for the local condition [F_congruency(1,53) = 1.79, p = 0.19, ηp2 < 0.03; F_{congruency x group} (1,53) = 0.33, p = 0.57, ηp2 = 0.01] and no effects for the global condition [F_group(1,53) = 0.18, p = 0.68, ηp2 = 0.02; F_congruency(1,53) = 0.84, p = 0.36, ηp2 < 0.02; F_{congruency x group} (1,53) = 0.62, p = 0.43, ηp2 = 0.01]. No post hoc tests were calculated to avoid a huge amount of comparisons in our three 3-way-ANOVAS (inflation of alpha error) which would contribute to the already inconsistent picture of main effects and interactions.

In general, higher values for local-to-global interference or vice versa indicate higher interference by the local (respectively, global) level on incongruent trials. As smaller RT’s indicate better performance, RT interference effects are reversed (lower values indicating higher interference).

Analysis of local-to-global interference revealed negative correlations between absolute pitch performance and RT local-to-global interference for the auditory domain (MAD: r = −0.295, p = 0.05; SDfoM: r = −0.421, p = 0.001). Therefore higher accuracy on absolute pitch tests (lower values MAD and SDfoM) is associated with weaker local-to-global interference in audition (see Figure 4). No local-to-global interference effects were found for the visual domain.

In the auditory domain, better performance (pitch template tuning, consistency) on absolute pitch tests (SDfoM) was furthermore correlated with reduced global-to-local interference in audition (correct: r = 0.300, p = 0.05; SACS: r = 0.242, p = 0.075, marginally significant). Higher autistic traits were associated with marginally lower global-to-local interference in the visual domain (r = −0.231, p = 0.090). However, all other correlations remained non-significant (see Tables 4, 5). Furthermore RTs did not correlate with pitch accuracy among AP musicians for congruent AGLT.

Performance on auditory or visual hierarchically-constructed stimuli frequently used to assess cognitive style (Navon, 1977; Bouvet et al., 2011) did not reveal strong group differences between AP and RP musicians (see Figures 2, 3). As expected, global as well as congruent stimuli revealed a processing advantage both in terms of speed (reaction times) as well as accuracy (correct, speed-accuracy composite scores) independent of group. In general, AP and RP were similar in the degree of performance difference between local and global congruent and incongruent trials (three-way interaction). If anything, the groups might differ in performance on congruent vs. incongruent trials independent of hierarchical level, or vice versa. Reaction times in both audition and vision furthermore showed a tendency for slower responses of AP independent of experimental conditions (congruent vs. incongruent stimuli), which was especially prevalent in the local condition of the visual experiment. Interestingly, groups did not differ in basic information processing speed measured as a confounding variable by ZVT (“Zahlen-Verbindungs-Test”; Oswald, 2016), so general information processing ability cannot account for these differences. Possibly, absolute pitch possessors might be slower in judging relative information among tones (intervals, melodies) because of their automaticity of tone labeling, which has also been described with respect to interval labeling especially of out-of-tune reference notes in some (Miyazaki, 1995), but not all absolute pitch possessors (Dooley and Deutsch, 2011). However, this interpretation cannot explain slower reaction times in the vision. In summary, while frequent effects of congruency and local vs. global condition were independent of groups, there was no consistent tendency towards an advantage for particular processing levels (local vs. global congruent vs. incongruent trials) for the two groups, which would have been reflected in three-way interactions.

Correlation analysis revealed higher interference of local percept on global performance (lower local-to-global interference, see Figure 4A) is associated with higher accuracy in pitch adjustment test, but only for RT and only in audition. As we were expecting more detail-oriented perception for AP possessors (Chin, 2003; Mottron et al., 2012), this result actually stands against our hypothesis, as here RP are more affected by details in perceiving a global auditory percept. However, this was only present for RT measures, which alone might not comprise clear evidence in our experiments. Musical stimuli by their nature unfold over time and participants’ response latencies might differ according to their listening strategy. For example some individuals may listen to the whole stimulus, before deciding whether global or local changes were presented, whereas others may choose to press the button as soon as the crucial 4th tone is played (which allows them to notice the difference between global and local stimuli). In line with our hypothesis, reduced global-to-local interference in audition (total number of correct trials, speed-accuracy composite scores) is correlated with higher AP accuracy (see Figure 4B). In vision, however, higher autistic traits are associated with lower global-to-local interference [speed accuracy composite scores, see Figure 4C)]. Therefore, in audition, people who have a more accurate AP are less affected by the global shape when concentrating on local details, as are people with more autistic traits (in the same sample) in vision. However, we have to admit that this is a weak relationship as it is selective for certain performance measures and sensory domains. In contrast, prior research has shown that cognitive style is quite similar within subjects across audition and vision (Justus and List, 2005; Sanders and Poeppel, 2007; Bouvet et al., 2011). A possible explanation could be that our sample only consists of professional musicians and students at music universities. This is a highly auditorily trained population, a fact which might increase the likelihood of obtaining differing effects in audition and vision. Difference in audition might be caused by the differences in specific musical abilities (absolute pitch) whereas differences in vision maybe more domain general effects reflecting autism relevant mechanisms. We also have to draw attention to the problem of the confounding nature of absolute pitch and autistic traits, which was also present in our sample (see Table 1). Group differences and correlations with absolute pitch ability can therefore always also in part be confounded by autistic traits and vice versa. Other samples (e.g., autistic subjects with and without absolute pitch) would be necessary to analyze the reciprocal influence of those two. As for additionally controlling for musicality, this might be a very difficult task. Further limitations of our study are the absence of a non-musical control group as well as of a direct comparison to an autistic sample. However, as we tried to keep precisely with the experimental setup of Bouvet et al. (2014), we argue, that our results at least of the auditory experiments should technically be comparable to the autistic sample of Bouvet et al. (2014), apart from differences in musicality. In general, inconsistent and weak effects might also be due to subgroups within AP musicians, whereby not all AP musicians might exhibit heightened autistic traits and/or a detailed cognitive style. This view receives support from a range of research on AP showing various influences on the acquisition of the ability, including genetics (Baharloo et al., 1998; Gregersen et al., 1999, 2001, 2013), an early start of musical training (Baharloo et al., 1998; Chin, 2003; Gregersen et al., 2001; Bermudez and Zatorre, 2009; Gervain et al., 2013), a sensitive period (Saffran and Griepentrog, 2001; Saffran, 2003), musical education method (Gregersen et al., 2001) and nationality or mother tongue (Deutsch et al., 2006; Deutsch et al., 2009). However, larger sample sizes are needed to uncover subgroups in such a heterogeneous population.

Despite the popularity of the weak-central-coherence account (Happé, 1999; Happé and Frith, 2006) and similar theories of autism (Baron-Cohen, 2009; Mottron et al., 2012, 2006) as well as of the global-local paradigms (Navon, 1977), a few authors have already raised criticism concerning these hypothetical concepts. First, global-local paradigms in the sense of Navon (1977), exhibit a huge variability of results across previous studies even in healthy populations. For example, various studies yielded heterogeneous results that might be caused by the relative size and the number of local elements used to construct the hierarchical stimuli in the respective studies (Kimchi and Palmer, 1982). Kimchi (1992) further emphasizes that global-local paradigms using hierarchically constructed stimuli might not even measure the degree of holistic perception, as being holistic (i.e,. properties that depend on the interrelations between component parts) is not necessarily the same as involving global precedence (i.e., processing of the higher level preceding that of the lower one). Therefore, not all global-to-local paradigms might be adequate to measure holistic perception in terms of Gestalt principles (Wertheimer, 1925). Furthermore, even evidence on a reduced global precedence effect as a result of a more detail-oriented perception in autism is contradictory (Ozonoff et al., 1994; Mottron et al., 1999, 2000, 2003; Foxton et al., 2003; Germain et al., 2018).

Future studies should therefore address holistic vs. detailed perception using adapted paradigms (e.g., Kimchi, 1992; List et al., 2007; Sanders and Poeppel, 2007) to overcome restrictions of classical global-to-local paradigms (Navon, 1977). Furthermore, a consideration of neurophysiological or – anatomical correlates, especially asymmetries in hemispherical contributions, promises to offer a new contribution to the debate of detail-oriented processing style of AP musicians. Seminal work by Peretz and Morais (1987) and Peretz (1990), on patients with unilateral brain lesions (Peretz, 1990) and healthy non-musicians (Peretz and Morais, 1987) has shown a processing bias of local information by the left and global by the right hemisphere. This is especially interesting, as research from both fields, autism and absolute pitch, often reveals hemispherical associations (e.g., Keenan et al., 2001; Hyde et al., 2008; Brancucci et al., 2009; Wilson et al., 2009; Wengenroth et al., 2014; Dohn et al., 2015; Floris et al., 2016).

To sum up, the correlation analysis of global-to-local interference effects in particular revealed results in accordance with the hypothesis of a more detailed-oriented cognitive style in AP possessors, which is also associated with autistic traits within our sample. However, the inconsistency of the results – and the dissociation of a correlation of AP accuracy with auditory performance versus autistic traits with visual performance – remains to be understood.