Serial processing of two words becomes parallel when they combine to form a known compound word

According to one model of reading, words are recognized one at a time with serial shifts of focused attention. This serial strategy would be required if, as some prior research suggests, there is a bottleneck in the brain that cannot process two words simultaneously. Consistent with this serial model, we first show that participants can judge the lexical status of only one of two unrelated words that are flashed briefly above and below the point of gaze fixation and then masked. We then investigate whether two words that together compose an existing compound word (e.g., bottle + neck) can nonetheless be processed in parallel. The results demonstrate that indeed, under the same conditions in which two unrelated words cannot be recognized simultaneously, accuracy for recognizing either or both of two words that form a compound exceeds the prediction of the serial model. This result complicates theories of a serial bottleneck in word recognition, especially in the context of natural reading. We propose a model that begins with parallel orthographic processing, initially serial lexical activation, and then interactive activations that can amplify the representations of two words that form a known compound.