When bottom-up meets top-down.

Studying the intriguing question of how words or pictures are encoded, processed and stored has been a long-standing problem in cognition. Both the written word frog, and a picture of the animal, can lead to an observer to say /frog/, or otherwise to the association that frogs leap, croak, need water, etc. Two different types of account have been formulated to explain word and picture processing: multiple-semantic-system approaches, [e.g. different systems for word ('logogen') and picture ('imagen') processing], and common-semantic-system approaches (e.g. single-code models). A recent study reported two experiments that looked at the effects of semantic and contextual factors, as well as perceptual familiarity, on word and picture recognition 1. The aim was to obtain, using ERPs, a detailed electrophysiological picture using an elegant and complex design of how the brain responds to words and pictures as a function of meaning, context and novelty. At a general level, the study seemed to lend support to the single-code model, as both words and pictures showed a similar time course and gave rise to an enhanced negativity about 300–500 ms after stimulus presentation (N400). However, at a more detailed level, remarkable differences between word and picture processing appeared. In the frontal region, as early as 100 ms (N1) and 200 ms (P1) after stimulus presentation, picture processing but not word processing started to be affected by semantic factors, but only when the participants were led to expect one particular picture. The second experiment suggested that this difference could be attributed to a perceptual effect only, namely stimulus uncertainty or 'perceptual predictability', to use the authors' term – this semantic effect on picture processing disappeared after visual familiarization with the items. The very elaborate data set used in this study make it unlikely that a single, amodal semantic system model can account for the electrophysiological findings. Instead, a specificity model based on spatially distinct regions of the brain and different patterns of activation seems more convincing at this stage. Refreshingly, the authors' conclusion that top-down and bottom-up processes are integrated at around 100 ms should be taken as encouragement to study similar problems again, rather than as being the final answer. (2001) Meaning and modality: influences of context, semantic memory organization, and perceptual predictability on picture processing. Two new face illusions The perception of human faces is different from that of other shapes: it has long been known that specialized mechanisms exist …