One word, one module?

In a discussion of (Juola and Plunkett, 2000), (Thomas and de Wet, 1998) argue that the apparent double dissociations found in that study simply show that the backpropagation network used did not have enough hidden units and that the network did (or could) not develop “fully distributed internal representations.” In other words, the dissociations found were, in their opinion, probably not just pathological cases, but represented hidden modularity that had somehow sneaked into the system despite the original researchers’ best efforts at prevention. The logic of inferring functional modularity from dissociations remains strong and compelling, even if (as will be argued) wrong.