Broca's area and the processing of hierarchically organised sequences pt.1

ResearchBlogging.orgEver since its discovery in 1861, Broca’s area (named after its discoverer, Paul Broca) has been inextricably linked with language (Grodzinsky and Santi, 2008). Found in the left hemisphere of the Pre-Frontal Cortex (PFC), Broca’s region traditionally[1] comprises of Broadmann’s areas (BA) 44 and 45 (Hagoort, 2005). Despite being relegated in its status as the centre of language, this region is still believed to play a vital role in certain linguistic aspects.

Of particular emphasis is syntax. However, syntactic processing is not unequivocally confined to Broca’s area, with a vast body of evidence from “Studies investigating lesion deficit correlations point to a more distributed representation of syntactic processes in the left perisylvian region.” (Fiebach, 2005, pg. 80). A more constrained approach places Broca’s area as processing an important functional component of grammar (Grodzinsky and Santi, 2007). One of these suggestions points specifically to how humans are able to organise phrases in hierarchical structures[2].

In natural languages, “[…] the noun phrases and the verb phrase within a clause typically receive their grammatical role (e.g., subject or object) by means of hierarchical relations rather than through the bare linear order of the words in a string. [my emphasis]” (Musso et al., 2003, pg. 774). Furthermore, these phrases can be broken down into smaller segments, with noun phrases, for example, consisting of a determiner preceding a noun (Chomsky, 1957). According to Chomsky (1957) these rules exist without the need for interaction in other linguistic domains. Take for example his now famous phrase of “Colourless green ideas sleep furiously.” (ibid, pg. 15). Despite being syntactically correct, it is argued the sentence as a whole is semantically meaningless.

The relevant point to take away is a sentence is considered hierarchical if phrases are embedded within other phrases. Yet, examples of hierarchical organisation are found in many domains besides syntax. This includes other language phenomena, such as prosody. Also, non-linguistic behaviours – such as music (Givon, 1998), action sequences (Koechlin and Jubault, 2006), tool-use (cf. Scott-Frey, 2004) and tool-production (Stout et al., 2008) – are all cognitively demanding tasks, comparable with that of language. We can even see instances of non-human hierarchical representations: from the songs of humpback whales (Suzuki, Buck and Tyack, 2006) to various accounts of great apes (McGrew, 1992; Nakamichi, 2003) and crows (Hunt, 2000) using and manufacturing their own tools[3].

With this in mind, we can ask ourselves two questions corresponding to Broca’s area and hierarchical organisation: Does Broca’s area process hierarchically organised sequences in language? And if so, is this processing language-specific? The logic behind this two-part approach is to help focus in on the problem. For instance, it may be found hierarchical structures in sentences are processed by Broca’s area. But this belies the notion of other hierarchically organised behaviours also utilising the same cognitive abilities.

2. Language studies, hierarchical organisation and Broca’s Area

2.1 Natural Language Studies

Finding evidence of grammatical processing, let alone the processing of hierarchical structures, is evidently difficult in language studies. Some of the more obvious problems include: neurological differences in comprehending and producing grammar (Greenfield, 1998); processing of written and spoken forms of grammar (Bahlmann et al., 2008); and lest we forget, phonological, semantic and other linguistic domains are inextricably enmeshed with syntax (cf. Bookheimer, 2002).

Subverting such problems requires a number of creative methods. One way is to vary level of grammatical complexity within sentences. For example, sentences with a centre-embedded, object-relative clause are considered more complex than those consisting of a right-branching, subject-relative clause. In theory, these grammatical variations should show differences in neural activation (i.e. increased blood flow); and is the basis for several studies using positron emission tomography (PET) (Stromswold et al., 1996; Caplan et al., 2001) and functional magnetic resonance imaging (fMRI) (Friederici et al., 2004).

For the most part, the results of these neuroimaging studies “[…] have been interpreted as supporting the notion that the pars opercularis/BA 44 of the left hemisphere is critical for processes that structure the incoming linguistic input.” (Fiebach et al., 2005, pg. 80). Still, this perceived activation is far from conclusive, and other reports suggest syntactic complexity is also associated with temporal and parietal areas (Ben-Shachar et al., 2003; Fiebach et al., 2005).

Providing some clarification are two studies (Tettamanti et al., 2002; Musso et al., 2003) testing grammatical and non-grammatical syntactic rules. The basis here is simple: “Given that language rules follow a specific set of principles, what happens if the brain is confronted with a nongrammatical rule to learn?” (Tettamanti et al., 2002, pg. 701). In particular, Musso and her colleagues (2003) perform an experiment in which German native speakers learn small sections of vocabulary from a different language (either Italian or Japanese). Participants are then shown either grammatical or non-grammatical rules, in correspondence to their learnt vocabulary, while fMRI scans for activity.

Interestingly, both forms of rules produce activation in Broca’s area. However, BA 44 shows a progressively increased activation when grammatical rules are processed. Musso et al. are careful in their concluding remarks, but they suggest that in regards to grammar Broca’s region is “[…] specialized for the acquisition and processing of hierarchical (rather than linear) structures, which represent the common character of every known grammar.” (ibid, pg. 778).

Now, it is clear these studies are far from a definitive crux for hierarchical structures being processed in Broca’s area. Natural-language studies elicit a whole host of explanations for the increase of activations in Broca’s area (Grozinsky and Santi, 2007). For instance, both Cooke et al. (2001) and Fiebach et al. (2005) claim Broca’s area processes Working Memory (WM), with varying levels of sentence complexity merely reflecting differences in storage demands. While Santi and Grozinsky (2007) opt for a slightly alternative, albeit related, approach and state Broca’s area is focused on syntactic movement in filler-gap dependency relations.

Syntax is not the only language phenomenon engaging Broca’s area. The hierarchical organisation of speech into suprasegmental acoustic features, known as prosody, also shows activation in BA 44. But in an interesting twist, it is the right hemisphere homologue of Broca’s area doing the prosodic processing, specifically in fundamental frequency (F0) modulations (Hesling et al., 2004). Even more interesting is that “High degrees [complexity] of prosodic information seem to trigger right specific activations in a wider neuronal network… such as the right inferior prefrontal cortex.” (ibid, pg. 945).

Again, it seems a greater level of complexity requires certain processing abilities found in the prefrontal cortex. But this avenue of research needs to be furthered before any meaningful inferences can be made. For now, the significance of such a finding shows that the right and left hemispheres of Broca’s area are engaged in two linguistic domains, both of which display hierarchically organised characteristics.

2.2 Artificial Grammar Studies

On the basis of natural-language studies examined, it can be confidently said that Broca’s area processes certain aspects of grammar. But just because grammar displays hierarchical organisation, does not necessarily mean Broca’s area is the place in which it is taking place.

Moving away from these inherent problems in natural-languages are several studies into artificial grammar (cf. Bahlmann et al., 2008). Apparently, by using meaningless[4] syllables (for example: di, yo, gu), and then ordering these to form a structure, researchers can specifically focus on syntax while minimising the influence of other linguistic elements – such as semantics, phonology and morphology.

In two related papers by Friederici and colleagues (Friederici et al., 2006; Bahlmann et al., 2008), the role of neurological processing in adjacent dependency and hierarchical dependency rules[5]. The idea is that these rules reflect non-embedded ((AB)n) and embedded structures (AnBn ), and therefore require different processing capabilities.

In the experiment, participants are split into two groups: one learns the (AB)n rule and the other learns the AnBn rule. Two days later the groups, while undergoing fMRI scans, are shown either correct or incorrect sequences corresponding to their learnt rule. Interestingly, the AnBn rule activates a phylogenetically younger portion of the brain (BA 44/6) than the (AB)n rule (the frontal operculum)[6]. On this basis, the authors conclude that “[…] Broca’s area subserves the processing of hierachical structures in the domain of grammar.” (ibid, pg. 533).

But these results and interpretations may not be as clear cut as first believed. Perruchet and Ray (2004), for example, argue a different heuristic strategy, such as counting, can explain how the AnBn rule is processed. In another study, European starlings are taught to process the AnBn rule (Gentner et al., 2006). More importantly, starlings do not have a pre-frontal cortex and by extension lack any neurological structure approximating Broca’s area (ibid).

To this end, is AnBn rule processing really reflective of hierarchical processing? In the team’s later study (Bahlmann et al., 2008) additional methodological controls are applied by defining positions within a sequence as dependency relations. Dependency relations for both the (AB)n and AnBn rules removes the possibility of employing a counting strategy, and in the case of AnBn rules ensures “[…] sequences are processed by participants in an embedded manner.” (ibid, pg. 526).

Employing dependency relations might negate one problem, but it potentially raises another: artificial grammar might now be considered context-sensitive. Through the application of a dependency relation, artificial grammar reflects natural grammar in that “[…] both require a match between two constituents.” (ibid, pg. 532). By extension, the AnBn rule may not necessarily equate with grammatical complexity, but rather a different process such as the aforementioned working memory.

Of course this is not an outright rejection of Bahlmann et al.’s conclusions. For a start, their study activates BA 44 while syntactic movement tests for working memory load engage BA 45 (Santi and Grodzinksy, 2007). But as Gentner et al. cautiously, and sensibly, point out: “Whether the different neuroanatomical location of the activation… is due to different language types (artificial vs. natural) or can be attributed to different dependency types (hierarchical vs. movement) will have to be clarified” (ibid, pg. 533).

To help reconcile these accounts of grammatical processing in Broca’s area, Hagoort (2005) proposes different levels of structural processing are represented by a posterior-to-anterior gradient, taking place across the left inferior frontal gyrus (LIFG): “LIFG is thus involved in at least three different domains of language processing (semantic, syntactic, phonological), with, presumably, a certain level of specialization in different LIFG subregions.” (ibid, pg. 420). Looking at it this way, the syntactic subsregions in Broca’s area could reflect both processing of hierarchical sequences (BA 44) and syntactic movement (BA 45).

Hagoort also makes room for non-linguistic operations in Broca’s area, which fittingly leads us to our next question: Does grammar share Broca’s region with other non-linguistic, hierarchically organised domains?

[1] However, there is some contention over this issue in regards to the exact boundary definition. For example, based on sulci and differences in cytoarchitecture Hagoort (2005) asks what justification is there to, “[…] subsume these two cytoarchitectonic areas under the overarching heading of Broca, rather than, say, Brodmann's areas 45 and 47.” (pg. 419).

[2] Hierarchical structure is sometimes, incorrectly, equated as a synonymous term for recursion. This is not the case, as Parker (2006) succinctly explains: “Importantly, a structure may be hierarchical without being recursive. While hierarchy involves phrases embedded within other phrases, recursion involves identical phrases embedded inside each other.” (pg. 3).

[3] Though as Conway and Christiansen (2001) point out, non-human primates “[…] are limited in their ability to learn and represent the hierarchical structure of sequences.”(pg. 539).

[4] Does using syllables with a standard English typography necessarily negate meaning?

[5] It should be noted that the (AB)n and AnBn rules in the 2006 study are referred to as finite state grammar and phrase structure grammar.

[6] The AnBn rule also activated the frontal operculum.

Main References

Bahlmann J, Schubotz RI, & Friederici AD (2008). Hierarchical artificial grammar processing engages Broca's area. NeuroImage, 42 (2), 525-34 PMID: 18554927

Musso M, Moro A, Glauche V, Rijntjes M, Reichenbach J, Büchel C, & Weiller C (2003). Broca's area and the language instinct. Nature neuroscience, 6 (7), 774-81 PMID: 12819784

Hagoort, P. (2005). On Broca, brain, and binding: a new framework Trends in Cognitive Sciences, 9 (9), 416-423 DOI: 10.1016/j.tics.2005.07.004