The present experiments focused on how orthographic processing develops during reading acquisition. Specifically, a large, cross-sectional sample of children from grade 2 to grade 4 was exposed to pairs of words, pseudowords, digit strings, and pseudo-letter (Armenian) strings while their sensitivity to transpositions (T) and substitutions (S) of internal characters was investigated in a perceptual matching task. The results showed that the development of identity and position decoding diverged between the four stimulus categories. Most importantly, sensitivity improved with longer exposure to formal education and higher reading level to "both" S and T pairs for digit strings, but "only" to S pairs for words and pseudowords. The results were successfully reproduced in two small independent samples. We propose a general framework, the Adaptive Specialization Hypothesis, to accommodate the results. According to this hypothesis, the transposed-letter effect is not a hard-wired feature of the orthographic processing system but an adaptive response of the developing orthographic system to the constraints of lexical access in several orthographies.