I just finished reading Stanislas Dehaene’s Reading in the Brain, and I’m already scrambling to figure out how I’m going to pitch it to the several people in wildly different areas who I think will find it fascinating. The book has two mutually supporting strengths that make for only a rare dull moment: meticulousness and willingness to speculate on the questions that really matter to us.
The basic claims underlying Dehaene’s neurological work are meticulously and transparently backed up with oodles of experimental results from different labs and different schools of neuroscience. Some of the details are really nitty-gritty ones, but I’m left with little doubt about the validity of his main claims vis a vis the brain: that it has a “letterbox” region that activates during reading only and is culturally invariant, and that this region is divided into cortical areas that together demonstrate a neurological basis for the psychologist’s model of top-down and bottom-up, hierarchical networks of visual processing that allow us to read.
In particular, Dehaene outlines a slow grapheme-to-phoneme route for decoding and a lexical route that develops eventually in expert readers, and describes how such things as invariant processing of multiple fonts and sizes and levels of abstraction in feature detection can arise from networks of neurons operating in parallel, the “bottommost” ones in the tree coding for line segments presented on the retina, middle ones that act as letter-detectors, and top ones that code for entire words. This perhaps wishy-washy model based on potentially naive neural network models that are currently in vogue is backed up with exhaustive anatomical information that confirms that these types of neurons do in fact exist in the “letterbox area,” the left occipito-temporal cortex, and they behave in a way consistent with the model. Further, this letterbox is culturally acquired and specific: Hebrew words don’t activate the letterbox area for English readers.
More surprising is the precision of this correspondence between written language categories and the letterbox’s division: there seem to be particular areas that code for different categories of words, within the area of the letterbox that processes meaning: different cortical regions for faces, people, animals, etc. And within verbs, the area of the premotor cortex that maps to the particular body part in question is activated when reading the verb: so reading “kiss” activates the area for the mouth, reading “kick” activates the area for the leg–words literally activate the motor areas they refer to. So the image of reading in the brain is simultaneously extraordinarily specific–as the reader learns, the cortical areas begin to be more and more subdivided and specialized in what they react to (one model of what learning looks like in the brain in general)–and extraordinarily diffuse, with the letterbox area connecting to the corresponding region in the other hemisphere, to the frontal cortex where we “synthesize” ideas and inputs, and all around the brain.
Dehaene is exacting about his brain science: he describes in detail what PET, fMRI, diffuse MRI, and magnetoencephalography (direct tracking of cerebral activation on the cortical surface itself, which lets us make movies in slow motion that show how a single read word travels through the brain) and lesion analysis have told us about the brain’s functioning, and what their limits are. But for me, the most interesting science was the slightly more speculative stuff, which came in (1) the research that indicates how the brain comes to specialize in reading for expert readers (development), and (2) the comparative studies of the region corresponding to the letterbox in primates.
The entire chapter on “Learning to Read” is worth keeping in mind for its pedagogical implications in the teaching of reading. In brief, the child first has a “pictorial” stage where the visual system attempts to recognize words as though they were objects or faces; next is the grapheme-to-phoneme mapping; and last is the acquisition of “a vast lexicon of visual units of various sizes” that includes frequency information about these units and their (written) neighbors. Two pedagogical points: grapheme-to-phoneme conversion MUST be explicitly taught and emphasized (phonics), and literacy rewires almost everything in the child’s brain. Moreover, there aren’t lots of ways to learn to read: the reading brain looks pretty much the same across cultures.
What stuck with me was a strikingly brain-based description of synesthesia, and an amazing hypothesis about “normal” reader’s brains that resulted. Synesthesia is real; it occurs exclusively for learned cultural objects; and it may be due to a mutation of the genes involved in synapse pruning during development. The experimental finding was that in synesthetes, the areas activated by letters and by color patches overlap noticeably, and they don’t in non-synesthetes: the synesthete is “stuck” with his cortex in an intermediate state of specialization. The amazing hypothesis is that all children may be synesthetes: during development, as cortical specialization unfolds, there may be a period of incomplete specialization that leads to effects comparable to those in adult synesthetes!
The chapter on “Inventing Reading” outlines several features common to all writing systems, but Marc Changizi’s stands out as a quantitative and highly suggestive commonality that links our reading to the apes’ “reading of nature.” He found that the frequency distribution of certain shapes was constant, considering only their topological attributes: L and T were most common, followed by X and F, then Y and (Delta). This distribution was the same as for natural scenes! For example, a T or L shape often occurs during object occlusion, and a Y can occur at the corners of objects.
Working in apes, Tanifuji found that monkey neurons respond to a remarkably sparse representation of some object exactly as they do when the object is fully rendered–as long as certain “proto-letters” that encode “non-accidental properties” are kept in the picture. If the Y, T, X, O and J’s in an image are left, the monkeys (and humans) still recognize the object: not so if an equivalent amount is erased from the image that includes some of those proto-letters. That provides strong evidence for how we manage invariance in object recognition: we have certain neurons (or at least, monkeys do) that respond to the presence of certain shapes on the retina that probably correspond to “non-accidental” features: shapes that arose because of some object or relation between objects, not the random placing of lines. In particular, these shapes are the ones that largely look the same when scaled, rotated, or lighted differently.
Lastly, Tanaka found some neurons that code for a black dot on a white background–posited to be an “eye detector,” which is essential for a social species–and there are other detectors corresponding to biologically useful features: but most correspond to simple geometric shapes.
Taken all together, we have here in Dehaene’s book material useful for all teachers, for neuroscientists, but also for linguists, anthropologists and scholars of the history and development of the English language. Exciting! Most importantly, we have an excellent model of how to do “soft” science rigorously and honestly, working on hot areas like culture without puffing up your claims. And a critical call to educators to experiment in the classroom: “Every teacher bears the burden of experimenting carefully and rigorously to identify the appropriate stimulation strategies that will provide students’ brains with an optimal daily enrichment.” (233)
There’s a point beyond which Dehaene goes into speculation: but it’s fascinating and appropriately qualified, so I went with him there, to some startling findings. One on symmetry: at the neuronal level, there’s a right-left brain symmetry, so that every neuron in the left region of the visual cortex goes through the corpus callosum and maps to another neuron in the right cortex. Before processing into the letterbox region, visual stimuli presented to one half of the visual field activate BOTH neurons in the pair: initial processing is symmetric. This lends credence to the claim that initial processing is focused on object identification, where spatial orientation (and in particular mirror reversals) and motion are less important. Only later, and only in the expert reader’s brain when viewing words in the reader’s language, is a “symmetry-breaking” observed, where only the left letterbox region responds.
There seem to be TWO visual pathways that are quite distinct operating in the visual cortex: the ventral one that is concerned with object recognition and exhibits symmetric brain activation, and the dorsal one that is concerned with motion and spatial orientation, and does not exhibit this symmetry. Here’s where the symmetry can be broken: and it’s aided by motor learning (since that’s the provenance of the dorsal cortical system attuned to motion in the environment), hence the effectiveness of children tracing sandpaper letters to aid their reading acquisition in the Montessori method. Neat!
Dehaene calls for a “neuro-anthropology” that he realizes is beyond our reach, but that he believes to be a viable research orientation and program. How might we look at cultures neuroscientifically? Structural anthropologists a la Levi-Strauss already do it to an extent, positing the existence of a kind of “deep grammar” of culture that gives rise, unconsciously, to universal cultural forms, among them religion, mathematics, music and arts, facial expressions, games, legal systems, etc. Dan Sperber frames it as, “the image that emerges from the cumulated ethnographic record is not at all one of indefinite variability, but rather one of extremely elaborate variations within a seemingly arbitrary restricted set.” There’s more scientific evidence (vs. anthropological) than you might think: consider the research of Paul Ekman, who painstakingly developed a system that can encode all facial “micro-expressions,” and demonstrated that people across cultures assign the same meanings to most of these facial expressions. In music, we see that infants across cultures are sensitive to octaves and fifths, though other intervals and scales vary.
Sperber and Ekman’s data, though, leads them to posit too much constraint on culture, in Dehaene’s view. Sperber thinks of the brain in terms of different modules that developed for certain evolutionarily helpful reasons and then can be stretched a little to apply in somewhat different domains: universal structures that allow a “fringe” of cultural variability. Dehaene thinks there’s more variation possible: art forms aren’t just a “super-stimulus” that work at the limits of cortical stimulation within each of these regions, but they arise out of associations between a wide selection of mental representations that elicit complex emotions. To the extent that there are cortical modules that artists exploit, those modules have nebulous boundaries, and education and cultural acquisition can rewire their connections and boundaries substantially (perhaps the best argument for early, intensive arts education I’ve ever heard).
Moreover, like Changeux thinks, masterpieces come from novel stimulation of many regions–but not just that novel stimulation: the interactions between those regions, and their synchronous and “harmonious” (reinforcing) activation, are key. So too are amplifications of reality: as Ramachandran puts it, in a way many aesthetic theorists would like, “the purpose of art…is not merely to depict or represent reality…but to enhance, transcend, or indeed even to distort it…to amplify it in order to more powerfully activate the same neural mechanisms that would be activated by the original object.” (310) These connections are uniquely amplified in the human brain, which developed neuronal connections between distant cortical areas in the course of tool learning, has an abnormally large frontal cortex that can synthesize novel ideas and associations like metaphors, and is remarkably sensitive–even in infancy–to the communicative intentions of others. I think there’s an interesting parallel to common descriptions of modernism here: art arising out of representing common forms in the most novel, non-traditional ways possible, art taken to the extremes of what has gone before along each possible dimension we can use to describe artistic forms.
Granted–as Dehane grants it–we’re a long way from any rigorous neuroscientific theory of aesthetics or complex cultural productions. But it’s fun and enlightening to speculate, and in the meantime, Dehaene gives us a core of very rigorous explication and a case study in how to exhibit a brain- and mind-based theory of a complex learned phenomenon without excluding its historical, social and cultural context.
As a final note in that vein–the New York Times review of this book criticized Dehaene for seeming to disagree with himself: at the same time saying that the mind is tightly constrained in its learning by the brain’s structure, and that it is remarkably plastic. I don’t think these things need to be in opposition. For one, Dehaene makes it clear that he’s positing constraints partly as a corrective to the “default social science model” of cultural acquisition via general learning mechanisms starting from a blank slate. More substantially, though, evo-devo and several other scientific developments have clearly shown us how tight constraints can produce enormous variety of mechanism and ultimate form. This is a sophisticated account of reading in culture that carefully avoids both reductionism and relativism, while incorporating both biology and historical and cultural context.