What Can We Learn From Humans About Orthographic Processing in Monkeys? A Reply to Frost and Keuleers (2013)

In their commentary on our two recent articles (Grainger, Dufau, Montant, Ziegler, & Fagot, 2012; Ziegler et al., 2013), Frost and Keuleers (2013; this issue) state that the conclusions drawn on the basis of our orthographic-learning studies with baboons are “logically fallacious and do not withstand empirical scrutiny” (p. 1868). Here, we clarify our conclusions, and we show that they follow the logic of scientific inference and are supported by the empirical data. We then show that Frost and Keuleers’s conclusion that monkeys simply learn probabilistic conjunctions of unordered individual shapes is wrong.

Frost and Keuleers take issue with what they claim to be our main conclusion: “Because baboons do not have a linguistic system but nevertheless perform like humans do, the neural mechanisms underlying orthographic processing in the two species must be similar and therefore nonlinguistic” (p. 1868). Paraphrased and out of context, this statement provides a somewhat distorted version of the reasoning developed in our two articles. Some clarification is therefore in order.

Following the standard logic of scientific enquiry, our conclusions were drawn on the basis of the hypotheses under test, the predictions derived from these hypotheses, the means employed to test these predictions, and the results that were obtained. The starting point of this whole enterprise was the simple fact that printed words are both visual objects and linguistic entities. From this we derived the main hypotheses that human orthographic processing might share basic mechanisms involved in visual object identification and that prior linguistic knowledge might not be a necessary requisite for acquiring elementary orthographic-processing skills. This led to the prediction that baboons, who presumably identify objects much like humans do, should exhibit some form of orthographic processing, such that they would not process printed words as holistic entities but as a set of position-coded letter identities. We then sought evidence for such elementary orthographic processing in baboons by training them to classify strings of letters as being real English words or nonwords and by examining the information they used to accurately perform this orthographic classification task.

We reported five major findings. First, word stimuli seen for the first time were classified as words significantly more often than were nonword stimuli seen for the first time (generalization). Second, we observed a significant correlation between bigram frequency and accuracy of responses to words. Third, baboons systematically made more errors on nonwords that were orthographically more similar to known words (based on the orthographic Levenshtein distance 20 metric; Yarkoni, Balota, & Yap, 2008). Fourth, baboons made significantly more errors when presented with nonwords formed by transposing two letters of a known word than when presented with nonwords formed by substituting two letters, and, finally, no more errors were made to nonwords formed by substituting a single letter of a known word with a visually similar letter than to stimuli in the same category with a visually dissimilar letter.

From these findings, we drew four conclusions. First, we inferred that monkeys use a truly orthographic rather than a visual code, that is, they decompose words into their component letters rather than process them as a whole visual shape. Second, it appears that prior linguistic knowledge is therefore not a necessary prerequisite in order to achieve humanlike orthographic processing. Third, orthographic processing is, at least partly, constrained by general principles of visual object processing shared by monkeys and humans, and, finally, the front end of reading is supported by neural mechanisms that are much older than the behavior itself and are not purely linguistic in nature.

Frost and Keuleers come to a different conclusion, namely that “the results simply show that baboons can learn probabilistic conjunctions of three to four individual shapes and that this learning does not extend to the order of the shapes” (p. 1869). However, both points are wrong. First, this is not about being able to discriminate “probabilistic conjunctions of three to four individual shapes.” Words and nonwords were both made up of the same 26 letters. Thus, there were potentially thousands of conjunctions that had to be learned. Second, shape information does not seem to play a role, because nonwords with visually similar letter shapes did not produce more false-positive errors than nonwords with visually dissimilar letter shapes (Ziegler et al., 2013). Third, finding a transposed-letter effect does not mean that order is irrelevant. On the contrary, order does matter for humans and baboons, and it is precisely the mechanism used to code for order that gives rise to transposed-letter effects (Grainger & Whitney, 2004).

We agree with Frost and Keuleers that there are important differences between humans and baboons that will have to be worked out in the future. This is the very foundation of our comparative approach to understanding reading. We also agree that false-positive errors to transposed-letter nonwords are much higher in baboons than in skilled human readers. However, this does not invalidate our results. Indeed, errors to transposed-letter nonwords are also much higher in developing readers (Grainger, Lété, Bertrand, Dufau, & Ziegler, 2012), and yet transposed-letter effects are used as a classic marker for orthographic processing in human children (Acha & Perea, 2008; Castles, Davis, Cavalot, & Forster, 2007). As concerns the final issue of how one should perform comparative psychology, the rat example is nice, but Frost and Keuleers ignore that the strength of our comparison lies in the fact that, contrary to rats, humans and baboons have highly similar genetic makeups and comparable visual systems.