Visual presentations of words or digits can be made invisible by masking. Yet when a threshold is crossed, the same stimuli suddenly become visible. This masking paradigm raises two central questions : What is the depth of subliminal processing of masked stimuli? And what major changes occur at the c onsciousness threshold?

In my talk, I will review the behavioral and neuroimaging evidence that my colleagues and I have obtained on those two issues. Multiple cortical areas can be activated by subliminal stimuli, in direct correspondence with behavioral manifestations of priming at the lexical, semantic, and motor level. I will describe two recent experiments that replicated our subliminal repetition priming effect in the inferior temporal lobe and used it to study the details of word identification processes in alphabetic, Kanji and Kana
scripts.

Our results suggest, however, that subliminal words fail to elicit the intense and distributed pattern of frontal, parietal and cingulate activity which is seen when words are unmasked and become consciously identifiable. In particular, the anterior cingulate cortex seems to be uniquely associated with the conscious processing of conflicts, and not to be involved in unconscious conflict resolution. A model of a conscious neuronal workspace may account for these data. The model postulates that a visual stimulus becomes reportable when its neural representation in posterior areas is amplified and made available, via long-distance neural connections, to multiple distant sites. Neuronal simulations of the workspace model, using realistic spiking neurons and thalamo-cortical connectivity, provide insight into the neural correlates of conscious access. In particular, the simulations suggest that a phase transition, corresponding to the sudden, all-or-none "ignition of a large-scale reverberating and self-sustained assembly, underlies the consciousness threshold. New empirical tests of the critical predictions of this model will be presented.

Altogether, theory and experiments suggest a simple and testable basis for conscious access: a macroscopic state change in brain-scale distributed neuronal networks, with local amplification and long-distance correlation, leading to information availability and therefore to reportability. This simple phase transition or "ignition should be easily identified in both behavior and neuroimaging.