Language bootstrapping the brain

7 minute read

Marina Bedny and colleagues Bedny:2011 show that, to a remarkable degree, the visual cortex of blind subjects takes on language-specific processing tasks.

I think the paper makes a nice occasion to consider just how language-specific areas of the left hemisphere may have evolved. The fact that one of the most domain-specific cortical regions of the brain can, to some degree, be reprogrammed to support language processing suggests that language itself is surprisingly voracious in its ability to consume brain resources and redirect development.

I’m a little surprised that we didn’t already know that blind subjects use visual cortex in language. It has the ring of previous scholarship. And actually the authors discuss a boatload of previous studies that appear to show precisely that: blind subjects relying upon visual cortex for language processing. The visual cortex increases activity during language tasks in blind subjects; blind subjects who have their occipital lobes zapped with transcranial magnetic pulses have problems performing language tasks, and visual cortex activity in blind subjects appears to be correlated with verbal memory. But Bedny and colleagues discuss several reasons why the previous results were not fully convincing; the visual cortex might be taking on domain-general or sensory cognitive tasks instead of language processing proper.

Bedny and colleagues devised a series of tests involving different language tasks, showing that the visual cortex in blind subjects responded not merely to difficult or memory-intensive tasks, but specifically to those tasks that most tax the language regions of normal subjects. The simplest interpretation is that the visual cortex has indeed taken on language-specific functions in blind subjects.

Below: How language eats brains, and why it matters to language evolution.

Plasticity, canalization, and self-organization

This is a unique kind of cortical plasticity. Take a piece of the brain that in most adults seems to be highly specialized at the neural level for visual processing, remove visual stimuli during development, and observe as the same area takes on apparently a very different function.

Jonah Lehrer covers the story, putting it in context of earlier work in animals:

In the late 1990s, a team of neuroscientists at MIT led by Mriganka Sur undertook an audacious experiment: they rewired the brain of a ferret, so that the information from its retina was plugged into its auditory cortex. The assumption was that the animal would be blinded, unable to make sense of all the incoming pixels. To Surs astonishment, however, the ferrets could still see. Furthermore, their auditory cortex now resembled the typical ferret visual cortex, complete with spatial maps and neurons tuned to detect certain slants of light. At the time, Michael Merzenich, a leading plasticity researcher at UCSF, called this experiment The most compelling demonstration you could have that experience shapes the brain. Our mental hardware wasnt hard at all.

Cortical plasticity is not itself a surprising story. If neurons couldn’t bootstrap cortical networks in atypical ways, then we wouldn’t see any recovery of function in victims of strokes or other brain injuries. Left hemispherectomy patients – people who have had the left half of their neocortex surgically removed – can develop language abilities localized on the right side. The brain can sometimes adapt itself to correct for serious deficits. To some families, this plasticity can seem like a miracle.

But the visual cortex seems like it should be canalized – developmentally constrained to express a specific neural structure. Within area V1, for example, there is a spatial field map corresponding to the visual field, which in humans exhibits a “magnification effect” in which the central visual field takes up a disproportionate cortical area. Meanwhile different neurons in V1 exhibit tuning to visual stimuli of different kinds, giving them the ability to filter fine-scale visual information. It sure looks like a specialized circuit that would be poorly suited for processing anything other than visual information. That seems like the kind of thing that would require a very specialized developmental process with some strong genetic control. So how could it be rewired on the fly, to effectively support language processing?

Kaschube and colleagues Kaschube:2010 showed that the apparent canalization of the visual cortex might emerge as a natural consequence of cortical development in space and with exposure to visual stimuli. For example, V1 is highly similar among distant groups of mammals, which would ordinarily point us to a deep homology in which these mammals shared a common ancestor with the same visual processing layout. But Kaschube and colleagues showed that the apparent developmental robusticity of the visual cortex could be maintained by simple rules of self-organization. It doesn’t take specialized genetic control to create a visual cortex, it just takes information structured in the right way to flip a few genetic triggers.

What's up with language processing?

Let me suggest a couple of informed speculations – which I’m happy to call speculations because they’re running far in advance of citations tonight.

Suppose that the neurons of the occipital cortex have few genetic switches affecting the way they organize their functions. Recycling of regulatory genes is very common during development, because the specific combination of these genes and positional or other environmental information interact to direct developmental events. The specific visual inputs that shape the development of the visual cortex would never ordinarily be present in other brain areas. So the genetic switches may be widely recycled – these genes wouldn’t have negative epistases with visual inputs elsewhere in the brain. I would expect likewise for other brain functions – the genetic switches will often be the same, the environmental inputs make the difference during development.

In a visual cortex deprived of visual stimuli, then, a subset of neurons would be likely to respond usefully to non-visual inputs. Language is one of the most demanding cognitive tasks faced by humans early in their development.

Still, I think it should take more than sheer cognitive demand to plant language-specific processing in the occipital cortex. The classic language areas have their own developmental biases emerging from the functions of nearby cortical areas. The frontal cortex area immediately rostral to Broca’s area develops earlier as a center for processing action sequences. Other areas involved in language processing lie usefully near auditory and association areas. The visual cortex seems like it should be outside the loop.

So I would hypothesize that language has a bit more oomph. We know that in developmentally compromised brains, some language facility will develop in atypical places (e.g., right hemisphere areas) at the partial expense of the functions that would ordinarily be localized there. In such cases, the language facility may itself be compromised – unlike the case of the blind subjects who have partial visual cortex language localization.

Language has sharp elbows. It muscles its way into the brain, crowding out other neural functions. Language has the most powerful weapons at hand – a baby’s first word prompts an entire language community to pull the dopamine and serotonin levers of emotion and attention.

A function that was strongly specified by genetics, patterned early in brain development, would not plant itself in spare neurons like a weed in a vacant lot. Only a system that bootstraps itself upon experiencing language inputs could have such plasticity. The structure of the language environment fosters the development of the classic language areas, biased to appear in those particular places by prenatal developmental trajectories, but not built according to a genetic blueprint.

The blind subjects tell us that the ground for language processing is almost as fertile elsewhere in the cortex. Many brain areas have the genetic equipment to recruit and organize neurons into useful circuits for language processing. Language development is developmentally robust because it can rely on a rich language environment, not because of genetic standardization. The basic problems of language evolution must be explained by showing how robust language communities emerged. I don’t preclude genetics, far from it – weaker language environments may have become stronger because of evolutionary change. But that evolution must have been substantially domain-general, because language processing is not specifically canalized by genetics.

I like this scenario because it means we shouldn’t be looking for lots of language-specific genetic changes in the last few hundred thousand years. The Neandertal genome suggests that there may not have been any at all.

My second speculation: If the language environment determines the instantiation of language processing, then brains must be substantially different in the way they process language. Children experience different language environments – not only different languages, but different microenvironments within language communities. Only strong genetic controls could canalize brains despite the differences in their language environments. In brains where language processing emerges readily in the visual cortex, genetic controls cannot possibly synchronize brains in the face of environmental variation.

I have much more to write on this topic, but it will have to wait for another time and more references. What anyone familiar with my thinking should anticipate: Rich language communities with strong environmental variation have imposed selection pressures on many other aspects of cognition.