Abstract
Skilled human readers can readily recognize written letters and words. This domain of visual recognition --- known as orthographic processing --- is critical to human reading abilities, but it is unclear how such behaviors are supported by neural populations in the human brain. A dominant theory is that the existing representations known to support general object recognition, i.e. the ventral visual pathway, could be “recycled” to support orthographic processing. Importantly, recent work has shown that non-human primates (baboons) can learn to distinguish written English words from pseudowords (lexical decision task), successfully generalizing to novel strings, and exhibiting behavioral error patterns that are consistent with humans. These observations demonstrate that non-human primates, while not capturing the entirety of human reading abilities, provide a unique opportunity to investigate the neuronal mechanisms underlying orthographic processing. Here, we investigated the representation of letters and letter strings in the ventral visual stream of naive macaque monkeys, and asked to what extent these representations could support word recognition. We recorded the activity of hundreds of neurons in V4 and IT while monkeys passively viewed images of letters, English words and non-word strings. We then used biologically plausible linear decoders to test these neuronal populations on word recognition. We found that decoders trained on the IT population responses achieved relatively high performance on this lexical decision task. Importantly, the pattern of errors obtained from the IT population was highly correlated with the corresponding primate behavioral pattern. This behavioral pattern was also reproduced by leading artificial neural network models of the ventral stream, but not by low-level representations. Taken together, these results suggest that, even in untrained non-human primates, the population of IT neurons forms an explicit representation of written words that could be “recycled” to support orthographic processing behaviors, serving as a precursor of reading acquisition in humans.