Boeckx and Theofanopoulou (2015) today produced a commentary on ‘Labels, Cognomes and Cyclic Computation: An Ethological Perspective’ (Murphy 2015a; henceforth LCC). With care and instructive insights into the life sciences they expand the discussion of the computational capacities of non-humans, and note that the discussion of brain dynamics (the ‘dynome’; Koppell et al. 2014) in LCC is insufficient to act as a serious alternative to the Chomsky Hierarchy. This general omission was down to reasons of space and focus, so I would like to take this opportunity to further explore the topic.
Firstly, it should be stressed that LCC in fact acknowledges the limits of a purely formal approach to ‘computational ethology’, citing also Murphy (2015b). In this work, the extent to which brain rhythms are the suitable neuronal processes which can capture the computational properties of the human language faculty is considered against a backdrop of existing cartographic research into the localisation of linguistic interpretation. It follows Ramirez et al. 2015 in translating into rhythmic terms the operations of the human cognome. Motivations for this approach are not obscure: The ERP community has spent a great deal of time decomposing the major components, such as the P600 and N400. It is taken for granted that the level of analysis provided by these ‘large’ components does not suffice at the electrophysiological level to describe typically generic linguistic sub-operations. The urge to seek a finer level of granularity, then, is clearly manifested in the ERP community through EEG and MEG investigations (Lau et al. 2008), but this objective is not found in the vast majority of cartographic neuroimaging research.
The applications of narrow syntax must also be regulated, as Boeckx and Benitez-Burraco (2014: 5) put it, through ‘interfacing with and being embedded inside cognitive systems responsible for interpretation and externalization’. Reinterpreting their suggestions within a Label-based framework, possible physical correlates for Concatenate and Label are generic neural coding mechanisms within a globular cortical structure, with internally generated high frequency oscillations like the gamma range being ‘embedded inside an oscillation operating at a lower frequency such as the alpha range’ (2014: 5). Such lower frequencies are known to synchronise distant cortical regions; procedures which may represent the substrates of linguistic cross-modular mental transactions (Kinzler & Spelke 2007) being implemented via concatenation and labeling. Typically sidelined in the past, cortical oscillations are now understood to play ‘a potential role’ in speech processing, according to Poeppel’s ‘temporal view’ hypothesis (Poeppel 2014: 142). Oscillations have also been linked to the timing of cortical information processing (Klimesch et al. 2007).
Boeckx and Theofanopoulou (2015) also note the inadequacy of the syntactic concept ‘labeling’ in exploring cognitive phylogenies. Their alternative suggestion is to ground the cognome in the workings of brain dynamics, specifically oscillations – as is noted in LCC. The reason LCC introduced the notion of labeling at the behavioural and computational level was purely to keep within the current – though, as noted, inadequate – pace of ethological inquiry. Dedicating more of LCC to the dynome would not have given the paper the approachability initially sought. It should also be stressed that, by introducing (in LCC) and later discussing (in Murphy 2015b) the dynome-cognome relation, computational ethology is not incommensurable with neuroethology. In addition, LCC makes clear what kind of evidence is needed to falsify the Labeling Hypothesis at the behavioural level, even if the notion of labeling requires an adequate decomposition (Murphy 2015c) for it to be explored alongside the dynome.
Boeckx and Theofanopoulou justifiably attend to the long-term goals of a cognome-dynome reconciliation. But it seems to me that the short-term goals discussed in LCC are just as important; perhaps more so, considering the current gulf between computational and behavioural studies. LCC was mainly concerned with shifting ethology towards a finer grained computational analysis, and regardless of whether labeling is formulated at an adequate level of granularity for a computational-implementational settlement to be reached (which LCC acknowledges), delivering a more computationally rigorous science of animal cognition (‘computational ethology’) is a well-motivated goal.
The centrality of labeling effects in linguistic interpretation is also evidenced, it seems to me, in recent neuroimaging work. Santi et al. (2015), for instance, show that ‘the involvement of Broca’s area in processing syntactic movement is best captured by memory mechanisms affected by agrammatically instantiated type-identity (i.e., NP) intervention’. Regarding the goals of investigations into the dynome, even though this work is important and fruitful, currently not enough is known about how oscillations relate to cognitive operations. The topic is empirical by nature, and what is needed at the moment are experimental designs which can tease apart rhythms, demonstrating a correlation with particular syntactic phenomena. And so while the dynome adds a vital biophysical perspective, traditional cartographic concerns should not be sidelined.
To illustrate, consider briefly the role of the left inferior frontal gyrus (LIFG) or Broca’s area, the traditional language region of the brain. Far from LIFG being the seat of syntax, Bornkessel-Schlesewsky and Schlesewsky (2013) provide reasons to believe that Broca’s area processes syntactic representations assembled in other brain regions. Considering that syntax is ‘a relatively basic and early information source’, and the frontal cortex ‘constitutes the point of convergence between the [dorsal and ventral] streams and is thereby essentially the furthest possible point downstream from primary auditory cortex’, the idea that LIFG is crucially involved in structure-building ‘appears somewhat surprising’ (2013: 63). Their time-(in)dependent model instead leads them to predict that syntax is ‘processed in networks that are still relatively far upstream within the processing streams and … close to primary sensory cortices’. They ultimately settle on posterior temporal regions as candidates for syntactic computation (see Bemis & Pylkkänen 2011, but also Theofanopoulou and Boeckx forthcoming for an overview of the potential role of the thalamus), while complementary research has revealed significant anterior temporal activity during compositional ‘semantic’ interpretation (Westerlund & Pylkkänen 2014).
We could say, then, that the ventral stream uses the lexical information provided by the anterior and posterior temporal lobe (Hickok & Poeppel 2007) to build sentence-level semantic representations which are ‘labeled’ (assigned projections/heads) by the dorsal stream’s parallel role of establishing syntactic (constituent) structure via what LCC terms the ‘Labeling Assembly’, lending neurobiological validity to the separation of set-formation and labeling seen in LCC.
As noted, shifting our focus from neuroimaging to more recent investigations of brain oscillations may provide a welcome (but as yet tenuous) way of translating into neural terms the operations of theoretical syntax. The brain rhythms investigated by Ramirez et al. (2015) – θ, α, β, γ – in their attempt at such a translation are generated by various cortical and subcortical structures. It has by now been well established that neural oscillations are related to a number of basic and higher cognitive functions (Buzsáki and Freeman 2015), for example speech perception (Giraud & Poeppel 2012). As Vaas notes, ‘Intrinsic oscillatory electrical activities, resonance and coherence are at the root of cognition’ (2001: 86).
Ramirez et al. also claim that the interaction of the dynome’s rhythms yields the syntactic sub-operations of lexicalisation, set-formation, labeling and cyclic Spell-Out. Set-formation, for instance, appears to be achieved by ‘a cross-frequency coupling mechanism between higher order thalamic nuclei … oscillating at α frequency … and [supragranular layers of cortical regions of the Default Mode Network (Raichle et al. 2001)] oscillating at the γ range’ (2015: 7). Labeling is achieved by one basal ganglia-thalamic-cortical loop, ‘likely crossing the dorsolateral striatum, disinhibiting the thalamic medio-dorsal nucleus, by means β of the rhythm, retaining in working memory one of the objects generated by [lexicalisation]’ (8). Related to Balari and Lorenzo’s (2013) claim that the basal ganglia is the centre of their ‘Central Computational Complex’ (the Merge capacity), Ramirez et al. propose that this region holds one of the γ-supported items before slowing it down to the β frequency as a consequence of the conduction delays resulting from the surrounding neural regions. Thus ‘the β frequency fulfils the role of non-terminal symbols’ (8); that is, labels.
In addition, the common claim that LIFG is necessary for processing hierarchical structures can now be qualified with the observations that, (i) this is only one aspect of syntactic processing (though a crucial one), and (ii) LIFG appears to be involved in ‘comprehending’ syntactic structures only insofar as it is responsible for the aspects of cognitive control which select among alternative representations. LIFG is correspondingly not the centre of syntactic comprehension, though Broca’s area does play a critical role in processing hierarchical representations. It may therefore be vital to labeling, but not set-formation.
Having evaluated the prospects for inquiry into the role of the LIFG in syntactic comprehension, it should be noted that the capacities I have claimed this region possesses are likely not unique to language (as Boeckx and Theofanopoulou 2015 also note), being instead domain-general computations found in other cognitive faculties (see the hierarchical processing found in vision (Ursini 2011) and motor planning (Fujita 2009)), and indeed other species (Schlenker et al. 2014). The exception, however, may be labeling. Finally, the operations of set-formation and labeling are not to be found ‘in LIFG’ or ‘in the left ATL’, but may rather emerge from the way brain waves synchronise the activation of pathways storing discrete featural representations. While it could be said that this simply amounts to a special kind of localisation, understanding brain rhythms could on the contrary shed light on why language is restricted to set-formation and labeling, and not some other imaginable operations which fall outside electrophysiological constraints.
Balari, S., & Lorenzo, G. (2013). Computational Phenotypes: Towards an Evolutionary Developmental Biolinguistics. Oxford: Oxford University Press.
Bemis, D.K., & Pylkkänen, L. 2011. Simple composition: A magnetoencephalography investigation into the comprehension of minimal linguistic phrases. Journal of Neuroscience 31, 2801-2814.
Boeckx, C., & Benitez-Burraco, A. (2014). The shape of the human language-ready brain. Frontiers in Psychology 5, 282. doi: 10.3389/fpsyg.2014.00282.
Boeckx, C., & Theofanopoulou, C. (2015). Language, cognomes, and the challenges of building cognitive phylogenies. Frontiers in Psychology 6:784. doi:10.3389/fpsyg.2015.00784.
Bornkessel-Schlesewsky, I., & Schlesewsky, M. 2013. Reconciling time, space and function: A new dorsal-ventral stream model of sentence comprehension. Brain and Language 125, 60-76.
Buzsáki, G., & Freeman, W. 2015. Editorial overview: Brain rhythms and dynamic coordination. Current Opinion in Neurobiology 31, v-ix.
Giraud, A-L., & Poeppel, D. (2012). Cortical oscillations and speech processing: emerging computational principles and operations. Nature Neuroscience 15, 511-517.
Fujita, K. (2009). A prospect for evolutionary adequacy: merge and the evolution and development of human language. Biolinguistics 3(2), 128-153.
Hickok, G., & Poeppel, P. (2007). The cortical organization of speech processing. Nature Reviews Neuroscience 8, 393-402.
Kinzler, K. & Spelke, E. (2007). Core systems in human cognition. Progress in Brain Research 164, 257-264.
Klimesch, W., Sauseng, P., & Hanslmayr, S. (2007). EEG alpha oscillations: the inhibition/timing hypothesis. Brain Research Reviews 53, 63-88.
Kopell, N.J., Gritton, H.J., Whittington, M.A., & Kramer, M.A. (2014). Beyond the connectome: the dynome. Neuron 83(6), 1319-1328.
Lau, E., Phillips, C., & Poeppel, D. (2008). A cortical network for semantics: (de)constructing the N400. Nature Reviews Neuroscience 9, 920-933. doi:10.1038/nrn2532.
Murphy, E. (2015a). Labels, cognomes and cyclic computation: An ethological perspective. Frontiers in Psychology 6: 715. doi: 10.3389/fpsyg.2015.00715.
Murphy, E. (2015b). Labeling Effects, Rhythmic Cell Assemblies and the Brain Dynamics of Linguistic Computation: Towards an Explanatory Neurolinguistics. Ms. University College London.
Murphy, E. (2015c). Reference, phases and individuation: Topics and the labeling-interpretive interface. Opticon1826 17(5), 1-13. doi: http://dx.doi.org/10.5334/opt.cn.
Poeppel, D. (2014). The neuroanatomic and neurophysiological infrastructure for speech and language. Current Opinion in Neurobiology 28C, 142-149.
Raichle, M.E., MacLeod, A.M., Snyder, A.Z., Powers, W.J., Gusnard, D.A., & Shulman, G.L. (2001). A default mode of brain function. Proceedings of the National Academy of Sciences 98, 676-682.
Ramirez, J., Theofanopoulou, C., & Boeckx, C. (2015). A hypothesis concerning the neurobiological basis of phrase structure building. Ms. University of Barcelona.
Santi, A., Friederici, A. D., Makuuchi, M., & Grodzinsky, Y. (2015). An fMRI study dissociating distance measures computed by Broca’s area in movement processing: clause boundary vs. identity. Frontiers in Psychology 6: 654. doi: 10.3389/fpsyg.2015.00654.
Schlenker, P., Chemla, E., Arnold, K., Lemasson, A., Ouattara, K., Keenan, S., Stephan, C., Ryder, R., & Zuberbühler, K. (2014). Monkey semantics: two ‘dialects’ of Campbell’s monkey alarm calls. Linguistics and Philosophy 37(6), 439-501.
Theofanopoulou, C., & Boeckx, C. (Forthcoming). The central role of the thalamus in language and cognition. In Boeckx, C., & Fujita, K. (eds). Advances in Biolinguistics: The Human Language Faculty and its Biological Basis. London: Routledge.
Ursini, F-A. (2011). Space and the vision-language interface: A model-theoretic approach. Biolinguistics 5(3), 170-225.
Vaas, R. (2001). It binds, therefore I am! Review of Rodolfo Llinás’ ‘I of the Vortex’. Journal of Consciousness Studies 8(4), 85-88.
Westerlund, M., & Pylkkänen, L. (2014). The role of the left anterior temporal lobe in semantic composition vs. semantic memory. Neuropsychologia 57, 59-70.
Thanks for expanding on our commentary (nice title for the post, btw!). Allow me to address/correct a few points. I think the best way to do this is if I focus on specific passages of your text, since the general points were made in the commentary Constantina and I published today. I’ll be brief.
To recap and contextualize what follows: we stress two things in that commentary (more so than in previous publications of mine, btw; full credit to my co-author here). (a) computational ethology, as you outlined it in LCC, may not be feasible, and (b) the primacy and absolute necessity to take the ‘vertical dimension’ of Boeckx and Theofanopoulou (2014) [your Figure 5 in LCC] into account.
Given this, let me take issue with the following passages of your blog post.
“[Boeckx and Theofanopoulou] expand the discussion of the computational capacities of non-humans”: we crucially don’t do this. We in fact note that we cannot determine these comp. capacities at the level of granularity that you choose to formulate the Labeling claim. Not just in other species. In other cognitive domains for humans as well. Think of the range of tests linguists have developed over decades to determine constituency, locality, etc. And they still disagree on things like the necessity of labels! We have nothing like this for the rest of cognition/species. So, how can we begin to apply the comparative method in your computational ethology approach? You say that “LCC makes clear what kind of evidence is needed to falsify the Labeling Hypothesis at the behavioural level”. I am sorry, but I very much doubt this. Did you ever see linguists ‘falsifying’ each other’s favorite computational claims? Was that easy? [the last question is rhetorical]. So, I find it surprising that you can write that “the short-term goals discussed in LCC are just as important”, because short-term-wise, I don’t know how we can get computational comparisons off the ground.
I understand the “reasons of space and focus”, but this is not just a matter of focus. (Remember our lip-smacking example in the commentary: in terms of brain oscillations, lip-smacks and syllables are one and the same. But syllables are unique to humans. This should make you worry.) It seems to me that the fine-grained computational properties you stress is one of the things that make Poeppel’s granularity problem harder. We don’t believe that ” the discussion of brain dynamics (the ‘dynome’; Koppell et al. 2014) in LCC is insufficient to act as a serious alternative to the Chomsky Hierarchy” Rather, brain dynamics suggest a quite different comparative program (one much closer to Darwinian descent, I think).
The point is not to “ground the cognome in the workings of brain dynamics, specifically oscillations”, but to reconstruct the cognome from the bottom up, taking seriously the idea that computations will be elementary and generic (and so claims about human uniqueness unhelpful).
You write that “computational ethology is not incommensurable with neuroethology”. I beg to differ. And I’m not alone. Though he draws different conclusions from me, Gallistel stresses the gap between the computational and the neuro. Currently, you can’t marry these two. Especially, if you adhere to a rich notion of computation, of the sort linguists adopt. “Shifting ethology towards a finer grained computational analysis” won’t do, because it will simply make Poeppel’s granularity problem harder.
I will not comment too much on your speculations concerning Broca’s role, etc. I’m skeptical, especially about your take on Labeling and the dorsal stream, since labeling appears to matter a lot to construct semantic representations [this was the topic of Aritz Irurtzun’s nice PhD thesis from a few years ago] but I’ll wait until I see the account fleshed out. Concerning your use of my paper with Constantina and Javi on brain rhythms, let me stress that I think you fail to appreciate how much an analysis at the dynome level transforms our understanding of what one finds at the level of the cognome. The main difference for us is that labeling cannot be so exceptional as you claim (and we certainly don’t attribute linguistic sub-operations to specific brain frequencies; it is their conjunctions that matter). In fact, I think the dynome renders notion like labeling, syntax, language, etc. pretty vacuous in a biolinguistic context.
Thanks for the reply Cedric, the advice and guidance is much appreciated and I think I’m beginning to understand the real implications for the cognome, I’ll definitely address these considerations in future work. I suppose it’s hard to let go of notions like ‘labeling’ and ‘language’ when they seem to do so much explanatory work for you, but since we currently lack joint cognome-dynome terminology it seems inevitable that ‘set-formation’, ‘labeling’ and ‘atomization’ will be around – at least on some level of analysis – for a while. Given your point that the goal is to reconstruct the cognome from the bottom up, rather than freely import syntactic constructs, I’m now even more sympathetic to the multidimensional framework outlined in Boeckx & Theofanopoulou (2014), whilst also bearing in mind one of Poeppel’s original major concerns about the possibilities of computational studies directing and delimiting biophysical investigations. At the computational level, I’m still convinced of the FLN-esque centrality of labeling, something which I don’t think has been appreciated over the last five years of intensive work into the labeling algorithm (see Figure 3 in LCC), and reminding ourselves of labeling effects (and not just ‘concatenation effects’) when conducting animal behavioural studies will not only produce a more precise ethology, but will also encourage interdisciplinary discussion and, hopefully, collaboration. In addition, the project of relocating the universality of Universal Grammar to the dynome, and away from the genome – where there are surprising layers of variation – seems to me to be the appropriate task for biolinguistics, and as I said in the blog the importance of novel experimental paradigms, and not just theoretical commensurability, should not be underestimated.