Home | People | Papers | Projects | Methods | Links | Graduate Programs | Subjects

The neural development and organization of abstract word recognition

The common contexts hypothesis. As with the first line of research, establishing the existence of an abstract word form area in extrastriate cortex raises the question of how it arose. And given the success of the co-occurrence hypothesis in explaining the localization of letter recognition, one might naturally wonder whether a correlation-based hypothesis could explain the development of abstract word forms. We have indeed proposed just such a hypothesis based on the realization that different visual forms of the same letter tend to occur in common contexts. The idea is that the visual contexts in which different forms of the same letter appear interact with correlation-based Hebbian learning in the visual system of the brain to produce ALIs and that, as a result, the representation of words is abstract. Specifically, because many letters look similar regardless of case, font, etc., different visual forms of the same letter share similar distributions of visual contexts. We proposed that this correlation leads the visual system to produce representations corresponding to ALIs.

Here is the idea. People are exposed to any given word written in a variety of different ways: all capitals, lowercase, in different fonts, in different sizes, etc. Because the shapes of some letters are relatively invariant over these transformations, they tend to make the contexts in which one visual form of a letter appear similar to the contexts in which other visual forms of that same letter appear. For example, if the visual form "a" occurs in certain contexts (e.g., between "c" and "p" in "cap", before "s" in "as"), then the visual form "A" will also occur in visually similar contexts (between "C" and "P" in "CAP", before "S" in "AS"). Of course, the different visual forms of some letters are fairly different (e.g., "D" vs. "d") and so there will be some contexts that are unique to one visual form of a letter (e.g., "a" but not "A" occurs in the context "d-d"). But given that a number of letters are similar in their uppercase and lowercase forms (Cc, Kk, Oo, Pp, Ss, Uu, Vv, Ww, Xx, Zz, and to some degree Bb, Ff, Hh, Ii, Jj, Ll, Mm, Nn, Tt, and Yy; the obvious exceptions are Aa, Dd, Ee, Gg, Qq, and Rr) and that letters in different fonts and sizes tend to look similar, there will be a significant proportion of common contexts for dissimilar-looking uppercase and lowercase letter forms. We now describe a neural network model that demonstrates how this common contexts hypothesis could work as well as a behavioral study that confirms one of its predictions.

Neuroimaging | Theory | Computational Model | Behavioral Studies | Discussion | Papers