Hierarchy processing in human neurobiology: how specific is it?

Abstract

Although human and non-human animals share a number of perceptual and cognitive abilities, they differ in their ability to process hierarchically structured sequences. This becomes most evident in the human capacity to process natural language characterized by structural hierarchies. This capacity is neuroanatomically grounded in the posterior part of left Broca's area (Brodmann area (BA) 44), located in the inferior frontal gyrus, and its dorsal white matter fibre connection to the temporal cortex. Within this neural network, BA 44 itself subserves hierarchy building and the strength of its connection to the temporal cortex correlates with the processing of syntactically complex sentences. Whether these brain structures are also relevant for other human cognitive abilities is a current debate. Here, this question will be evaluated with respect to those human cognitive abilities that are assumed to require hierarchy building, such as music, mathematics and Theory of Mind. Rather than supporting a domain-general view, the data indicate domain-selective neural networks as the neurobiological basis for processing hierarchy in different cognitive domains. Recent cross-species white matter comparisons suggest that particular connections within the networks may make the crucial difference in the brain structure of human and non-human primates, thereby enabling cognitive functions specific to humans.

This article is part of the theme issue ‘What can animal communication teach us about human language?’

1. Introduction

The question ‘What makes us human?’ has been asked many times over the past centuries, without resulting in a clear answer. Cognitive and social abilities, where humans surpass what is observed in non-human animals, are usually listed as possible answers. This leads to a number of cognitive abilities that consensus suggests are uniquely human. The most obvious is the human capacity of language. Music has also considered to be an ability specific to humans, as has mathematics, at least beyond counting and simple calculations. More recently, the capacity to attribute mental states to others—e.g. beliefs, intents, desires, emotions, perspectives— and to understand these, a capacity called Theory of Mind, has been proposed to be human-specific. Each of these higher-order cognitive abilities may be rooted in the human capacity to process the structural hierarchy underlying the sequential surface [1]. Such underlying hierarchies are assumed to be present in language, music, mathematics and Theory of Mind. If this assumption holds, the questions arise to what extent the ability to process such hierarchies is specific to humans, and what is its neurobiological basis.

A first step in reducing the search space in defining human-specific functions is the observation of the presence or absence of a particular ability in human compared to non-human primates. Pure observation, however, is not sufficient to argue for or against the species-specificity of a particular capacity. Rather, one would need strong empirical grounds, preferably based on behavioural and brain evidence. In the following, I will discuss the available data on language and these other higher-order cognitive abilities, considering both human and non-human primate studies.

It appears pretty straightforward to make the claim that the capacity for human language is specific to our species. Non-human primates, as our closest relatives, do not speak, nor do they understand speech. However, the matter is not that simple.

First, language does not depend on speech. This is clearly seen in sign language, with different versions across the world. These clearly count as full natural languages as they fulfil the three obligatory domains of any natural language: semantics, syntax and even phonology. A detailed description of the sign language system can be found in Sandler & Lillo-Martin [2]. Second, the fact that non-human primates do not speak is not dependent on an inability to produce the sounds comprising the vowels and consonants in natural spoken language. In fact, it has been shown that the articulatory space in non-human primates is similar to that of humans [3]. Third, the fact that non-human primates do not produce or comprehend language does not necessarily mean they are incapable of learning words or meaningful sound patterns. For example, it has been demonstrated that non-human primates, as well as dogs, can learn to associate words with various objects, shapes and colours [4–7].

Why then, are non-human animals incapable of processing our language? This inability rather depends on their incapacity of combining words into phrases and sentences. This incapacity is demonstrated in a very impressive study by Pearce [8] comparing a chimpanzee to a normally hearing child and a deaf child in their ability to combine words in an utterance. The hearing child already starts to produce multi-word utterances by the age of 2 years, and the deaf child learning sign language somewhat later. Both children dramatically increase the length of their utterances between the ages of 3–4 years. However, the chimpanzee, who had learned a visual symbol language, never learned to combine words into a multi-word ‘utterance’ even when trained up to 4 years [8] (see also [9]). Although these are descriptions of single cases, they illustrate well the difference between non-human primates and humans: it is the ability to combine words into phrases and sentences.

Following these early case studies on language production numerous studies were conducted with non-human primates in the domain of speech perception and understanding, either looking into the combination of symbols in the semantic domain [4] or in the domain of rule-based sequences [10–15]. The presently available data suggest that the non-human primate's ability to process structured sequences is very limited but more elaborated than initially thought. For example, non-human primates are able to learn rule-based auditory sequences involving adjacent and non-adjacent dependencies of low complexity (i.e. AxB grammars, in which the elements A and B have a rule-based relation and x is a variable element; go le ku, go zu ku). They seem to fail, however, when confronted with hierarchically structured sequences (i.e. (AB)² with embedded relations of A and B (A₃(A₂(A₁B₁)B₂)B₃)) [10]. Such latter sequences, however, are at the core of any natural grammar processed by humans, for example when considering the relation between (A) the subject-noun (man) and (B) the verb (was running) in the sequence with a centre-embedded part (she saw) (i.e. the man [she saw] was running).

2. Language

The ability to process hierarchically structured sequences is at the core of the human language system, which consists of words and a set of syntactic rules that govern the combination of words allowing us to form phrases and sentences [16]. According to a classical linguistic theory, this rule system is rooted in one single computation called Merge. Merge is defined as the most basic syntactic computation that binds two elements together into a syntactic hierarchy, resulting in a new higher-order element [17]. For example, a determiner (D) the and a noun (N) man are combined to create a higher-order element, a determiner phrase (DP), the man. This DP can combine with a verb (V) run to make up a next higher-order element, a sentence (S), the man runs. By applying Merge recursively, hierarchically complex sentences can be built. As Merge is assumed to be the most basic operation of hierarchy building, it functions as the basis of the human-specific capacity of language.

The computation Merge has empirically been located in a constrained region in the human brain, namely in a subpart of Broca's area, that is Brodmann area (BA) 44 [18]. Broca's area, located in the inferior frontal gyrus (IFG), consists of a more posterior part BA 44 and a more anterior part BA 45, each of which is part of a different neural network. BA 44 is part of a fronto-temporal language network connecting language-relevant regions via dorsally located white matter fibre tracts, whereas BA 45 is part of a ventrally located network (for a review see [19]). For a neuroanatomical model of the language network, see figure 1.

Figure 1.

Neuroanatomy of language. Anatomical and cytoarchitectonic details of the left hemisphere (LH). Top: The different lobes (frontal, temporal, parietal and occipital) are marked by coloured borders. Major language-relevant gyri (IFG, STG and MTG) are colour-coded. The temporo-parietal junction is marked by an asterisk. Numbers indicate language-relevant BAs that Brodmann [20] defined on the basis of cytoarchitectonic characteristics. Broca's area consists of the pars opercularis (BA 44) and the pars triangularis (BA 45). Located anterior to Broca's area is the pars orbitalis (BA 47). The frontal operculum (FOP) is located ventrally and medially to BA 44, BA 45. The premotor cortex (PMC) is located in BA 6. Wernicke's area is defined as BA 42 and BA 22. The primary auditory cortex (PAC) is located in a lateral to medial orientation in the temporal lobe. White matter fibre tracts, i.e. the dorsal and ventral pathways connecting the language-relevant brain regions, are indicated by colour-coded arrows (see bottom right for key). The dorsal fibre tracts consist of the superior longitudinal fascicle (SLF) (connecting the frontal cortex and the parietal cortex) and the arcuate fascile (AF) curving posteriorly into the temporal cortex. Reprinted from [19].

The localization of Merge, as the most basic operation of hierarchical structure building in BA 44, fits well with a large number of studies on syntactic processing as indicated by a respective meta-analysis [21]. Many of these studies used relatively complex sentences, but the only variable tested systematically was a structural hierarchy. Consistent with the anatomical location of Merge, the major activation hub for the processing of syntactic hierarchy cross the different studies was found to be located in left Broca's area [22,23]. Broca's area is part of a neural network that also involves the posterior portion of the superior temporal gyrus (STG). When processing syntactically complex sentences, this temporal area is activated together with the posterior part of Broca's area, BA 44 [18]. The cooperative nature of these two brain regions, when processing syntactically complex sentences, has been shown with functional connectivity studies [24,25] as well as oscillation studies [26–29].

It has been argued that the functional connectivity between BA 44 and the STG is based on a structural connection, namely a white matter fibre tract connecting the two regions. This dorsally located fibre tract consists of the superior longitudinal fascicle and the arcuate fascicle (SLF/AF) (figure 1). The evidence that this fibre tract is relevant for the processing of syntactically complex sentences comes from combined behavioural and structural magnetic resonance imaging studies. In a developmental study, it was demonstrated that the maturation of the dorsal fibre tract connecting BA 44 and the STG during development was highly correlated with the accuracy and speed with which syntactically complex sentences were understood [30]. Additional evidence comes from a patient study showing that a deficient dorsal fibre tract between Broca's area and the temporal cortex leads to deficits in processing syntactically complex sentences [31]. Thus, this dorsal fibre pathway connecting Broca's area with the STG seems to be critically involved in processing syntactically complex sentences, which are characterized by their underlying structural hierarchy. This fibre tract is present in the adult human brain but is not well developed in the newborn prelinguistic human [32]. It remains plastic through childhood, thereby allowing it to be formed by language input. With its plasticity, it stands in contrast with three other language-relevant fibre tracts connecting the frontal and temporal cortices, which are already considerably mature and myelinated at birth.

These other three fibre tracts are responsible for non-hierarchical processes in the larger speech and language network. One of these fibre tracts is a second dorsal fibre tract, and the two others are ventral fibre tracts (figure 1), briefly mentioned here. The second dorsal fibre tract targets BA 6 in the premotor cortex [32] and has been discussed to be functionally relevant for auditory-to-motor mapping [33,34]. The two ventral white matter pathways link the ventral part of the frontal cortex to the temporal cortex. These two ventral fibre tracts are neuroanatomically hard to differentiate as they run partly in parallel, but they can be distinguished by their respective target regions in the IFG [19]. One branch targets BA 45, the anterior part of Broca's area, whereas the other branch targets the frontal operculum (FOP), located ventrally to Broca's area [35]. While the former is suggested to support semantic processing [34], the latter appears to subserve the processing of simple rule-based sequences [36]. These three non-hierarchy-related fibre tracts are well myelinated at birth and may be functional already early during development. During the early developmental stages, the dorsal fibre tract targeting BA 6 might support babbling, which requires auditory–motor mapping [33]. The ventral branch projecting to BA 45 may support early word learning present as early as in the first year of life [37–40]. The ventral branch targeting the FOP possibly supports the processing of simple rule-based sequences—an ability also already present in the prelinguistic infant [41,42]. Thus, it seems that during ontogeny those brain systems processing semantic aspects and simple rule-based sequences have different developmental trajectories from those processing hierarchical sequences.

A cross-species view on the ability to process hierarchical structured sequences and its brain basis may add to the present discussion. When considering the evolutionary trajectory of the ability to process simple rule-based sequences and structural hierarchies, there are a few, but crucial studies comparing human and non-human primates. In a seminal behavioural study by Fitch & Hauser [10], it has been shown that non-human primates (cotton-top tamarins) can learn auditory sequences following a simple finite-state grammar rule (AB)ⁿ, but not sequences that follow a phrase structure grammar with a more complex AⁿBⁿ rule leading to non-adjacent hierarchical dependencies. Adult humans, by contrast, learn both types of grammar easily [10], though a subsequent functional magnetic resonance imaging study revealed that the neural basis underlying the processing of these two types of grammar differs [36]. Processing the simple finite-state grammar activated the FOP, a phylogenetically older cortex than Broca's area [43–45], whereas the more complex phrase structure grammar additionally activated left Broca's area, in particular, BA 44 [36]. A functional imaging study comparing the non-human and human ability to process adjacent relationships in sequences directly reported a strong activation of the frontal opercular cortex in both species, but more pronounced activations in areas 44 and 45 in monkeys compared to humans [12]. This again suggests that simple rule-based sequences in humans recruit the FOP, whereas non-human primates may additionally recruit Broca's area even for processing the simple sequences.

Brain structural data revealed that the FOP and BA 44, the posterior portion of Broca's area, belonged to two different neural networks [35]. Consistent with the dorsal and ventral pathways described above, the FOP was connected to the temporal cortex ventrally, while Broca's area (BA 44) was connected to the temporal cortex via the dorsal pathway (i.e. SLF/AF). These brain-based data from adult humans suggest two separate pathways for the processing of simple sequences and hierarchically structured sequences, respectively. These two pathways, moreover, seem to have different ontogenetic and phylogenetic trajectories, with the ventral pathway targeting the FOP developing prior to the dorsal pathway targeting BA 44.

Bringing together behavioural data from human and non-human primates, and brain-based data from human and non-human primates, the following picture emerges. Given the finding that the ventral fibre tract to the FOP is well developed in non-human primates [46] and in prelinguistic infants [32], and given that non-human primates [14] and prelinguistic infants [42] are able to process simple rule-based sequences, the available data suggest that the ventral pathway targeting the phylogenetically older FOP is sufficient to process simple finite-state grammars. This stands in contrast with the dorsal fibre tract targeting BA 44 responsible for the processing of hierarchically structured sequences in humans, which is less developed in non-human primates and in prelinguistic infants [32]. Thus, it appears that the processing of simple rule-based sequences and the processing of hierarchically structured sequences as well as their neural bases differ both developmentally and evolutionarily.

In sum, studies from numerous laboratories and disciplines suggest that the human language capacity is rooted in the ability to process hierarchically structured sequences. This ability is neurobiologically grounded in a brain system consisting of BA 44 as the posterior part of Broca's area and its connection to the temporal cortex via the dorsal pathway, i.e. the SLF/AF. These brain structures constitute a phylogenetically younger system, possibly unique to humans.

3. Non-language domains

The question of whether the structural hierarchies characterizing natural languages can also be found in non-language domains of human higher cognition such as music, mathematics or Theory of Mind is heavily debated. This is the case in particular with respect to the involved brain structures, notably Broca's area [1,13,47–51]. The answer to this question strongly depends on the definition of hierarchy and its possible applications across domains. In the linguistic domain, ‘hierarchy’ in its most basic terms is defined as the computation Merge, which requires that the elements to be bound together carry word category labels and once merged create a new element (with a new label). These labels indicate whether the set of rules governing the application of Merge can apply or not [52]. For example, a determiner (the) and a noun (man) can be combined (the man) to make up a DP; whereas a determiner (the) and another determiner (a) cannot be combined (the, a). This definition cannot easily be applied to non-linguistic domains as discussed by Moro [53] and Goucha et al. [52].

Stimulus sequences in non-language domains, however, can and have been described as bearing hierarchies even though these do not fulfil the requirements of the linguistic definition [54]. With this possible caveat in mind, the available literature in these non-language domains, concerning the neural basis of processing complex sequences, will be discussed. The focus will be primarily on the involvement of Broca's area (BA 44), as a possible domain-general processor of structural hierarchy, but moreover on dorsal white matter structures connecting Broca's area to the temporal cortex. It has been claimed that Broca's area functions as a general processor of hierarchy across domains [1]. The elegance of this claim lies in its simplicity, but the evidence for this claim is sparse and partly non-supportive. In the following, I will review the respective studies on the processing of complex structures for the domains of music, mathematics and Theory of Mind in turn. This will be done for human and whenever possible for non-human primate studies, both with respect to function and brain structure.

The musical harmonic and melodic structure have been defined as musical syntax, and the brain correlates evaluated. Complex auditory sequences of music can be described as containing hierarchical structures [55,56]. Functional imaging studies have repeatedly reported both right and left hemispheric activation for the processing of music, and specifically left Broca's area and its right-hemispheric homologue for the processing of hierarchical structures of music [57–59]. Recent work has shown that nested non-adjacent dependencies in music clearly involve the homologue of Broca's area in the right hemisphere—in particular, the posterior portion of Broca's area (BA 44) [60]. This finding, together with the previous findings in the language domain, suggest that aspects of hierarchical processing rely on lateralized, domain-selective, neural populations—language being lateralized towards the left hemisphere and music towards the right hemisphere. In the latter study, the term ‘domain-selective’ was proposed since a strong, domain-specific brain functional representation would stand in contrast with earlier studies showing an interaction between syntactic processing in music and language [61–64]. This notion of domain-selective lateralization is compatible with the idea of relative functional lateralization and the observed interaction between music and language is partly compatible with the view of left Broca's area as playing a domain-general part in processing structural hierarchy.

In addition to the brain functional findings of a right hemisphere homologue of Broca's area that processes hierarchy in music, analyses of white matter tracts in the right hemisphere have been conducted [65]. The functional relevance of different right-hemispheric fibre tracts for music processing has been demonstrated in respective white matter studies. Studying persons with acquired amusia, it was found that structural damage and subsequent degeneration of a number of white matter fibre tracts in the right hemisphere were correlated with this syndrome [66]. These fibre tracts included the ventral fibre tracts, as well as the AF as part of the dorsal pathway. Moreover, it was shown that an abnormally reduced AF connectivity leads to tone deafness [67]. A study analysing the relative strength of the SLF/AF in the left and right hemispheres, in musicians and non-musicians, reported a stronger AF in the right hemisphere for musicians, suggesting the relevance of this fibre tract for music processing [68].

To conclude, for the domain of music the left BA 44 plays a certain role when processing musical hierarchies, but it is the right homologue of Broca's area and its connection to the temporal cortex that plays the more important role.

When considering the specificity of processing music in humans compared to non-human primates, the database is more than sparse as systematic studies on the processing of hierarchical structures in music so far have not been conducted. With respect to music processing in general, a striking difference between human and non-human primates was reported in that humans preferred listening to music compared to silence, whereas the reverse was found for non-human primates [69].

Mathematics is certainly a cognitive domain that is often discussed to be human-specific, at least when considering mathematics at a certain level of complexity. There are several studies that have investigated the processing of hierarchical structures in mathematics and compared the resulting brain activity to that of hierarchical processing in language. A study with patients with extended left perisylvian lesions, including language areas, reported severe impairment in processing linguistic syntactic structures, while these patients' ability to solve embedded mathematical expressions was well preserved [70]. This indicates independence of linguistic and mathematical performance but, moreover, independence of mathematics from Broca's area [70]. Focusing on the processing of hierarchical structures in mathematics, in particular, a functional imaging study found that the activations for processing mathematics are located outside the core language network, suggesting independence of mathematics from the core language brain areas, in particular, Broca's area [71]. This also holds for logical inference making [72]. A brain imaging study comparing linguistic and algebraic hierarchies directly also revealed a dissociation of activation patterns between processing linguistic hierarchy and algebraic hierarchy, suggesting that hierarchy in these two domains is not processed by the same cortical region [49]. Another study comparing hierarchy in mathematics and language, in mathematicians and non-mathematicians, again revealed distinct activations for the processing of linguistic and mathematical hierarchy—even when the degree of automaticity was comparable between domains [73]. Mathematical processing rather activated the left middle frontal gyrus and the inferior parietal cortex known to be part of a network supporting high-level cognitive processes [74]. These parietal activations are compatible with previous work reporting the involvement of the parietal cortex and the intraparietal sulcus for numerical and arithmetic processes [75–77]. From these studies in the domain of mathematics, one can conclude that Broca's area, which processes hierarchy in language, does not subserve hierarchy processing in the domain of mathematics, and therefore supporting the view of Broca's area as a domain-general processor of hierarchy.

A recent study compared white matter tracts involved in mathematics with those involved in language [78]. The branch of the dorsal pathway targeting BA 44 again was found to support hierarchical processing in language, but not the processing of mathematical hierarchy. Rather, processing mathematics appeared to be supported by the branch connecting the precentral gyrus and the temporo-parietal cortex. This is in general agreement with prior studies investigating structural connectivity in numerical-cognition (for a review, see [79]).

To conclude, for the domain of mathematics, the data indicate no involvement of left Broca's area nor of the dorsal fibre tract connecting BA 44 and the temporal cortex.

A cross-species comparison in the domain of mathematics demonstrated that monkeys can perform basic arithmetic calculations such as addition, and the qualitative similarity between their performance and that of humans suggests that numerical addition as a primitive component of cognition has a common evolutionary origin among primates [80]. Brain structural data indicate that monkeys have strong white matter connections from the frontal cortex to the parietal cortex via the superior longitudinal fascile, whereas the connection to the temporal cortex is not strongly developed [81].

Theory of Mind is the third domain described to involve hierarchy building, and proposed to be human-specific. This ability allows humans to attribute mental states such as intentions, beliefs and desires to others and to thereby understand and predict the behaviour of other people. Theory of Mind has long been taken as a cognitive domain involving hierarchy. False beliefs, in particular, seem to rely on hierarchical processing, because they require one to keep two diverging perspectives at different levels: I know X (level 0) and I know that you know Y (level 1) [82–84]. Theory of Mind is a crucial ability in humans with a clear developmental trajectory. False belief paradigms have been instrumental in Theory of Mind research measuring when, and how, participants acquire knowledge that other individuals also have views of the world, which are often divergent. Implicit false belief paradigms suggest that Theory of Mind is present, although not explicitly, in children around the age of 2 years [85–87]. Explicit false belief tasks, on the other hand, can only be achieved successfully by the age of 4 years [88,89]. There has been an extensive discussion of whether this ability is human-specific or not [90,91].

For humans, a meta-analysis of functional brain imaging data of Theory of Mind studies has revealed a large, bilateral network involving the temporo-parietal junction, superior temporal sulcus, middle temporal gyrus (MTG) as well as the precuneus and medial prefrontal cortex [92], whereas the IFG is usually not considered to be part of the Theory of Mind neural network [93]. The Theory of Mind regions are connected by various white matter fibre bundles and it appears that deficits in the behavioural performance of classical Theory of Mind tasks correlate with the integrity of the most anterior subpart of the dorsal fibre tract [94,95]. In children, stronger myelination of this subpart predicted the ability of explicit false-belief understanding, present at the age of 4 years but not at the age of 3 [96]. The relation of Theory of Mind performance with this subpart of the dorsal pathway was independent of children's linguistic ability, thereby indicating certain independence of the two domains. Thus, having explicit knowledge of false beliefs appears to depend on the maturation of the dorsal pathway.

These findings concerning the neural basis of Theory of Mind as a cognitive domain involving hierarchical processing demonstrate that hierarchy is not processed in a domain-general manner in one brain region. Thus it is not Broca's area itself, but rather the dorsal white matter fronto-temporal connectivity that supports processing hierarchy in Theory of Mind.

A cross-species comparison shows that apes do not succeed on tasks that measure false-belief understanding on explicit behaviour [97–99]. However, more recently, it has been shown that great apes (chimpanzees, bonobos and orang-utans) can operate, at least on an implicit level, with an understanding of false beliefs [100]. This implicit level was measured with the use of an anticipatory look test as similar to the one used in human infant studies. Thus, it appears that in both apes and human infants there is a priority for implicit false-belief understanding. Neurobiologically this raises the possibility that explicit false-belief understanding and thereby the full capacity of Theory of Mind are dependent on brain structures that are less well developed in apes and human infants compared to human adults.

4. Summary across domains

Considering the available functional and structural brain data on processing hierarchy in language and non-language domains, we can conclude that there is little direct overlap in the functional brain areas supporting hierarchy in language and other higher-order cognitive domains, except partly for music. Music shows some functional activation overlap with language in the left hemisphere, but with a clear involvement of the right hemisphere. Only language hierarchy is solely processed in the left hemisphere in a confined fronto-temporal network including Broca's area and posterior temporal cortex, connected via a dorsal fibre tract (SLF/AF). Processing hierarchy in other cognitive domains includes non-language brain areas in the left and the right hemispheres. These activation patterns associated with processing hierarchy in non-language domains are in contrast with the language domain, with a clear lateralization of the language system in the left hemisphere.

Left Boca's area, in particular, has been claimed not only to function as a hierarchy processor independent of the cognitive domain [1]. Left Broca's area, however, has also been proposed as a domain-general working memory system [101]. Since hierarchical sequences may tax working memory in general, this view was put to test in a functional imaging study varying the factors syntactic hierarchy and working memory independently [102]. It was found that syntactic hierarchy activated BA 44, whereas working memory during sentence processing activated the left inferior frontal sulcus located dorsally to Broca's area. Investigating the domain-specificity of Broca's area, it was found that Broca's area with its two sub-regions (BA 44 and BA 45) is specifically engaged in language processing, but that it is surrounded by other regions that are engaged across a variety of tasks and domains [103]. Given these data, it appears that the claim of a domain-general system in Broca's area cannot be upheld, at least in its strong version. Activation of left Broca's area has clearly been demonstrated for language [19], but for music only in synchrony with its right-hemispheric counterpart [104]. For mathematics, Broca's area is not involved when language and mathematics are mastered with an equal level of expertize [73] and it does not seem to be involved in Theory of Mind processing [105].

A common aspect of the neural basis for higher cognitive abilities may rather be the white matter connections between frontal and temporal or parietal brain regions that constitute domain-selective neural networks for language, music, mathematics and Theory of Mind. For language, the left dorsal fibre tract from Broca's area to the temporal cortex plays a crucial role for language [30,31]. For Theory of Mind, it is the most anterior sub-part [96], and for mathematics, it is the portion connecting the frontal cortex to the parietal cortex. In music processing, the right dorsal fibre tract may play the crucial functional role [68]. Given that the left dorsal fibre tract is weak in non-human primates [46,106], it is not yet well myelinated in human infants and matures late [30,32,107], this fibre tract suggests itself as a possible brain basis for—at least—some of the human-specific, higher-order, cognitive abilities.

5. Conclusion

Broca's area in the left hemisphere is crucial for the processing of hierarchy in language, but not for hierarchy processing in other higher-order cognitive domains, and can thus not be viewed as domain-general. This area rather appears to be selective for language. Broca's area (BA 44) and the full maturation of its dorsal fibre connection to the temporal cortex seem to be necessary conditions for processing structural hierarchy in language—an ability only observed in human adults. Without the full functioning of this dorsal fibre tract, the brain is not able to process hierarchical structures, as indicated by white matter structural findings of non-human primates and prelinguistic infants. To what extent white matter differences between human and non-human primates can explain functional between-species differences observed in non-language domains remains to be seen.

Data accessibility

This article has no additional data.

Competing interests

We declare we have no competing interests.

Funding

This work was funded by the German Research Foundation (DFG) in the Research Unit FOR 2253: Crossing the borders (grant no. FR 519/23-1).