In this issue | PNAS

Neural code for word recognition

Learning to read leads to the emergence of a region of the brain called the Visual Word Form Zone (VWFA). How this region specializes in processing written words compared to other visual stimuli is unclear. Thomas Hannagan, Aakash Agrawal et al. used a biologically plausible computer model to assess the potential developmental origins of VWFA. The results support the neural recycling hypothesis, which proposes that VWFA results from the reassignment of other brain regions originally involved in the recognition of faces and objects. The authors found that an artificial neural network initially trained to categorize objects could later learn to recognize 1,000 written words with 80% or better accuracy, despite large variations in physical characteristics such as case, font and size. The model replicated other known VWFA properties, such as higher activation of words relative to faces and objects, and loss of reading skills after a simulated injury that eliminated 20% of selective word units. Additionally, the model predicts that the neural code of written words is sparse, so that only a small set of neurons that select the most responsive words are needed to identify a given word. According to the authors, the model’s predictions could become testable in the future, depending on advances in functional magnetic resonance imaging or intracranial recordings in the human brain. – JW

Read online


Highlight recent and topical articles selected by member laboratories of the Academy