Sergio Balari and Guillermo Lorenzo have a paper in the current volume of Biolinguistics, which is dedicated to celebrating the 50th anniversary of Lenneberg’s Biological Foundations of Language. The paper includes a number of unusual claims about computational approaches to neurobiology which I want to briefly address here.
They begin their discussion of neuolinguistics by claiming that “it is our contention that most self-declared biolinguistic approaches … have systematically misapplied the [Marrian] notion of ‘level’ in their attempts at solving the unification problem [of unifying linguistic computation with neurobiology]”. They claim that the connectome (the set of neural connections) and dynome/oscillome (brain dynamics) “are clearly not levels in any possible sense” purely because there is an ongoing project (that the authors are not involved in) to map how brain regions are dynamically connected. But this project crucially is far from over, and so even though at some point direct connectome-oscillome connections should be made, at the moment they clearly are fundamentally distinct levels of description – unless Balari and Lorenzo can explain how they are unified.
The following four paragraphs proceed to repeat the claim that we do not currently understand neural computation. This is true, but only insofar as a physics paper containing four paragraphs lamenting the lack of a Grand Unified Theory is also true. It is common knowledge in the field that neural computation is not understood, yet there are in fact a number of theoretical attempts to solve this conundrum, which Balari and Lorenzo do not critically engage with but rather dismiss out of hand:
“We raise these issues hopefully not for provoking a paralyzing effect, but to caution against an excessively enthusiastic reading of certain recent proposals concerning the computational character of brain oscillations (e.g., those of Murphy 2015, 2016) which do not seem to have taken into account the complications we just alluded to. To repeat, this is not to deny the potential relevance of brain oscillations in an eventual account of neural computation, but evidence so far is only correlational, in the sense that oscillations do play some role in linguistic tasks (e.g., Lewis et al. 2015, Lewis & Bastiaansen 2015, Ding et al. 2016), but we have so far been unable to disentangle the computational role they purportedly play.”
Had they read the papers they cite, they would have found that Murphy (2016a: 16) provides precisely the kind of analysis they claim is lacking in the field, discussing recent indications that oscillations play a causal role in the perceptual segregation of sound patterns – a topic expanded on in Murphy (2016b) which explores other recent tACS experiments into the theta-gamma code for working memory lending these oscillations a causal role in explaining the physical limitations of cognition. Citing existing work into the causal role of oscillations in behaviour is not an “excessively enthusiastic” thing to do, it is simply a way of providing evidence for one’s claims, a procedure seemingly alien to the authors.
More generally, as Uriagereka has already pointed out, expecting a one-to-one mapping between higher-order computational or psycholinguistic theories and neurobiology is similar to expecting a one-to-one mapping between cosmic background radiation and the Big Bang – certainly possible, but highly unlikely, and not even the goal of the neurolinguist attempting to rebuild our understanding of linguistic computation from the bottom-up.
In short, Balari and Lorenzo claim to provide a bold critique of the oscillation literature but in fact do no such thing.
Cedric Boeckx has a new paper out in the Journal of Neurolinguistics, “A conjecture about the neural basis of recursion in light of descent with modification”. The central thesis is summarised as follows: “I argue that the expansion of the parietal region associated with the globularization of the neurocranium in our species contributed to the transformation of the connection between Broca’s and Wernicke’s region via Geschwind’s territory, and enabled the pairing of evolutionary ancient networks that together became capable of constructing and processing not just sequences, but sequences of sequences”.
Boeckx writes that recursion “rests firmly on primate cognition and neural circuitry”. More precisely, the paper claims that “the neuroanatomical reconfiguration of the parietal lobe brought about by globularization extended the Broca-Wernicke connection (to be precise, its dorsal dimension) into “Geschwind’s territory”, forming a fronto-parieto-temporal circuit that provides the basis for richer representational capacities, viz. recursive capacities”.
Here, he discusses and agrees with the Friederici-inspired story that the development of the fronto-temporal dorsal stream somehow brought about a shift from single-instance concatenation operations to hierarchically organised sequences. The additional detail Boeckx provides is to claim that the expansion of the parietal lobe – a consequence of a more globularised braincase – served to bring about the strengthening of dorsal stream connections. This in turn resulted in “the pairing of two evolutionary ancient networks (one fronto-parietal, the other fronto-temporal), both of which build and process sequences”. This idea has been expressed in Berwick and Chomsky’s recent book Why Only Us, and who were themselves re-articulating the findings of primatologists from the past couple of years. However, this story says nothing about how the brain actually implements what Boeckx refers to as “the nature of the computation (and algorithm) involved”.
Boeckx goes on to claim that this pairing of two streams (each capable of finite-state computations) “could have the effect of boosting computational possibilities. Instead of operating on one-dimensional sequences, one now operates on two-dimensional, ‘tree’ representations”. Boeckx believes that if the fronto-parietal dorsal stream were to be “combined (integrated) with another sequencing machine, sequences of sequences would naturally emerge as a result”. But this appears to be something of a magic step: Pairing two finite-state machines does not produce a higher-order device, and pairing a sequence with another sequence does not necessarily produce “sequences of sequences”. While it may be true that syntactic labeling plus a form of “spell-out” provides the human-specific features of language, and that both labeling and spell-out can be loosely (but only partly) attributed to the hierarchically organised fronto-parietal structures and the fronto-temporal loop connections to externalisation, the jump from single-instance concatenation to recursion likely required more than the pairing of two sequence processors.
Still, the paper is filled with attractive ideas, such as the following: “Perhaps the fact that the parietal lobe has long been associated with numerosity … may help us understand why syntactic structures exhibit a spontaneous ‘logicality’ … intimately related to quantification, processed by the fronto-parietal network”. There is almost certainly something to this, and as Boeckx is well aware the field could certainly do with more of these sorts of multidisciplinary gestures.
There are some very concrete proposals about the neurocomputational properties of brain waves with respect to language, for instance here, here, here and here. There does, however, seem to be much confusion concerning this oscillatory approach to language. Goucha, Zaccarella and Friederici (2016), for instance, make the following claim:
“Alternative mechanisms based on brain oscillations have been proposed as a crucial element for the emergence of language (Murphy, 2015b) … However, those mechanisms seem to already be in place in other species. For example, despite the crucial brain expansion that took place in primates and especially humans compared to other mammals, the rhythmical hierarchy of oscillations is mainly kept unchanged (Buzsáki et al., 2013). As Friederici and Singer (2015) pinpoint, the basic neural mechanisms behind cognition through the hierarchical embedding of oscillations are transversal across animals.”
However, while the hierarchy of brain rhythms themselves may be preserved, it is crucially their cross-frequency coupling relations which are human-specific – a major topic for future research.
Goucha, T., Zaccarella, E., & Friederici, A.D. 2016. A revival of the homo loquens as a builder of labeled structures: neurocognitive considerations. Ms. Max Planck Institute for Human Cognitive and Brain Sciences, Germany.
New paper published in Biolinguistics on the nature of pragmatic unarticulated constituents and the syntax-semantics interface.
“This paper explores the prospect that grammatical expressions are propositionally whole and psychologically plausible, leading to the explanatory burden being placed on syntax rather than pragmatic processes, with the latter crucially bearing the feature of optionality. When supposedly unarticulated constituents are added, expressions which are propositionally distinct, and not simply more specific, arise. The ad hoc nature of a number of pragmatic processes carry with them the additional problem of effectively acting as barriers to implementing language in the brain. The advantages of an anti-lexicalist biolinguistic methodology are discussed, and a bi-phasal model of linguistic interpretation is proposed, Phasal Eliminativism, carved by syntactic phases and (optionally) enriched by a restricted number of pragmatic processes. In addition, it is shown that the syntactic operation of labeling (departing from standard Merge-centric evolutionary hypotheses) is responsible for a range of semantic and pragmatic phenomena, rendering core aspects of syntax and lexical pragmatics commensurable.”