Visualizing Language

Evidence for highly selective neuronal tuning to whole words...
http://www.ncbi.nlm.nih.gov/pubmed/19409265

Written language is one of humanity's most important developments. It has allowed the expansion and management of societies and nations on a large scale, enabled rapid and competitive global commerce, and informed further success through the recording of history. The ability of the human brain to learn and interpret written symbols and words is a complex and somewhat poorly understood process. As scientists continue to explore vision in mammals, some basic tenets of how the brain 'sees' have become clear. Mammalian vision is hierarchical - this means that our brain interprets images by first breaking them down into simple components, and then adding these components together to create a representation of what our eyes observe. The most basic component of an image that nerve cells (neurons) in our brain respond to is the line. There are many different neurons that respond to lines of different orientations (horizontal, vertical, and all angles in between). These 'first-order' neurons send signals about the orientation of lines within an observed scene to the next level of neurons in the hierarchy, which begin to assemble these lines into more complex figures - squares, circles, and so on. Think of it as though the brain is drawing a comic from the Sunday funnies, starting with some basic lines, adding shapes, curves, and finally color. Eventually there are neurons very high up in the hierarchy than can specifically recognize very complex images - faces, for example. Somewhere in your brain there is a very small number of neurons that become active when you see a picture of your mother, but that remain silent in response to any other face.

How does this system apply to written words from languages that utilize alphabets (Latin, Arabic, etc.)? During recent years, functional non-invasive brain imaging in awake human patients has demonstrated that a similar system exists for letters, and even for combinations of letters. Again, using the same hierarchy of neurons, letters are built up starting with lines, then curves and simple shapes, and on to letters. However, the next step in the process, the brain's interpretation of entire words, remains a topic of debate. Previous studies have failed to find a distinction in the way that the higher levels of the visual neuron hierarchy respond to real words or fake words, such as 'fast' versus 'tast'. This has led to the hypothesis that word recognition neurons respond not to entire words but only to two- or three-letter combinations, as opposed to face recognition neurons which specifically respond to entire faces.

Today's experiment addressed this issue using a twist on the standard brain imaging technique (known as fMRI, or functional magnetic resonance imaging). The researchers argue that a system for whole word recognition does exist in the brain, and that previous studies have missed a nuance of its function. Instead of responding only to a specific word, it may be that high-level neurons in the word recognition area respond strongly to a specific word, but weakly to similar fake words. This is distinct from face recognition neurons, which respond either strongly or not at all. In this model of word interpretation, the word 'fast' would strongly activate 'fast'-specific neurons, but the fake word 'tast' might weakly activate 'fast', 'cast', 'east', 'last', 'past', 'mast', 'vast', 'test', and 'taste' neurons. When looking at the aggregate response of a brain region as is commonly done with fMRI, these two types of responses would look the same when added together. For example, if a strong response is a 10, and a weak response is a 1, 'fast' would produce a value of 10, while 'tast' would produce nine values of 1, adding up to 9 - not a significant difference. To separate these two types of responses, the researchers paired real words with fake words while performing their fMRI experiment, and essentially subtracted fake responses from real responses. This allowed them to demonstrate that the word recognition region of the brain does indeed contain neurons that respond strongly to specific real words, and weakly to similar fake words, but have no response to similar real words.

It is gratifying to the neuroscience community to find another example of hierarchical organization in the brain. It suggests that our brain does indeed function in a 'simple to complex' order, breaking incoming information down into simple parts and re-assembling it. This study in particular suggests that alphabetic languages are not superior to pictographic languages, for in the end each word does activate a specific, small number of neurons. An interesting direction in which to take this research might be the study of dyslexic patients, to understand at what level of the hierarchy their difficulties with language occur.