Conscious, preconscious, and subliminal processing: a testable taxonomy Stanislas Dehaene1,2, Jean-Pierre Changeux2,3, Lionel Naccache1, Jerome �� �� Sackur1 and Claire Sergent1 1INSERM-CEA Cognitive Neuroimaging Unit, Service Hospitalier Fre ��de ��ric Joliot, Orsay, France 2Colle `ge de France, Paris, France 3CNRS Unit, Receptors and Cognition, Institut Pasteur, Paris, France Of the many brain events evoked by a visual stimulus, which are specifically associated with conscious percep- tion,andwhichmerelyreflectnon-consciousprocessing? Several recent neuroimaging studies have contrasted conscious and non-conscious visual processing, but their results appear inconsistent. Some support a correlation of conscious perception with early occipital events, others with late parieto-frontal activity. Here we attempt to make sense of these dissenting results. On the basis of the global neuronal workspace hypothesis, we propose a taxonomy that distinguishes between vigilance and access to conscious report, as well as between sub- liminal, preconscious and conscious processing. We suggest that these distinctions map onto different neural mechanisms, and that conscious perception is system- atically associated with surges of parieto-frontal activity causing top-down amplification. Introduction Understanding the neuronal mechanisms of conscious- ness is a major challenge for cognitive neuroscience. Recently, great progress has been achieved by contrasting brain activation images obtained during minimally different experimental conditions, one of which leads to conscious perception while the other does not. Surpris- ingly, however no coherent picture has emerged from those experiments. On the contrary, a controversy has arisen, as some studies suggest that consciousness depends mostly on the thalamus and brain stem [1], others on early visual areas [2,3], and yet others on higher prefrontal and parietal association areas [4���9]. Here, we propose that those apparent contradictions can be resolved by a relevant theorizing of the physiologi- cal conditions for conscious processing of sensory stimuli. Based on the recent proposal of a large-scale thalamo- cortical formal network and its simulations [4,5], we tentatively propose a plausible and testable taxonomy of brain activity states associated with conscious and non- conscious processing. In particular, within non-conscious processing, we distinguish a transient ���preconscious��� state of activity in which information is potentially accessible, yet not accessed. An enabling condition: vigilance The term ���consciousness��� has multiple meanings, one of them intransitive (e.g. ���the patient regained conscious- ness���), and the other transitive (e.g. ���consciousness of color���). To avoid further confusion, we abandon the term and use ���states of vigilance��� to refer to the non-transitive meaning, i.e. a continuum of states which encompasses wakefulness, sleep, coma, anesthesia, etc. Being in an appropriate state of vigilance (e.g. awake rather than asleep) is an obvious enabling condition for conscious processing of sensory stimuli. Empirically, awakening into the vigilant state correlates with a progressive increase in regional cerebral blood flow, first in the brainstem and thalamus, then in the cortex with a particularly important increase in prefrontal-cingulate activation and functional connectivity [10]. Anesthesia, sleep, vegetative state and coma [1,11] are all associated with modulations of the activity of this large-scale thalamocortical network which also shows high baseline activity during vigilant rest [12] and encompasses prefrontal, cingulate and inferior parietal nodes. These observations may be captured by a recent implementation of the neural workspace model [4] in which ascending brain stem nuclei (e.g. cholinergic among others) send globally depolarizing neuromodulatory sig- nals to a thalamic and cortical hierarchy. Simulations show a progressive increase in spontaneous firing as a function of neuromodulator release, which evolves into what is known in dynamical systems theory as a Hopf bifurcation: spontaneous firing increases continuously in intensity, but high-frequency oscillations appear suddenly in the gamma band (20���80 Hz). By increasing spon- taneous activity, and thus bringing a broad thalamo- cortical network closer to firing threshold, vigilance lowers the threshold for external sensory inputs. In summary, vigilance is a graded variable, and a minimum level is essential for placing thalamo-cortical systems into a receptive state. Corresponding author: Dehaene, S. (dehaene@shfj.cea.fr). Opinion TRENDS in Cognitive Sciences Vol.10 No.5 May 2006 www.sciencedirect.com 1364-6613/$ - see front matter Q 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.tics.2006.03.007

Early visual activation is not sufficient for conscious report We now consider the neural bases of the second, transitive meaning of consciousness, which we term ���access to conscious report���. How do we consciously perceive a visual stimulus? Many neuroimaging experiments have demon- strated a tight correlation between the conscious visual perception and the activation of striate and extrastriate visual areas [13���18]. For instance, unmasking of a visual stimulus increases activity in extrastriate areas in tight correlation with subjective reports of stimulus visibility [18]. Furthermore, extrastriate regions clearly play a causal role in conscious visual perception, because their selective lesioning eliminates the corresponding contents from experience ��� for instance a lesion of area V4 can destroy color perception in the contralateral hemifield [19]. On the basis of such data, Zeki [2] has proposed that the conscious perception of a given visual attribute resides in the extrastriate area specialized for that attribute (e.g. area MT/V5 for motion, or area V4 for color). A ���micro- consciousness��� would be involved whenever that area receives a sufficient amount of activation. We argue, however, that early sensory activation is necessary but not sufficient for conscious access, because activity in extrastriate visual areas is frequently observed while participants deny having seen any stimulus [14,20��� 23]. When invisibility is caused by masking [20] or by dichoptic stimulation [14] this stimulus-evoked activity remains weak, and one might argue that its small amplitude alone could explain the absence of conscious perception [2,14]. However, the visual activation evoked by invisible stimuli can also be very strong, for instance when invisibility is caused by neglect [21] or inattention [22,23]. In a recent study of the attentional blink, we observed that up to about 180 ms after stimulus presen- tation, the occipito-temporal event-related potentials evoked by a invisible word were large and essentially indistinguishable from those evoked by a visible word [23]. Yet on invisible trials, the participants��� visibility ratings did not deviate from the lowest value, used when no word was physically present. Thus, intense occipito-temporal activation can be accompanied by a complete lack of conscious report. Top-down amplification, long-distance reverberation, and reportability We [4���6] and others [7,8,24] have suggested that, in addition to vigilance and bottom-up activation, a third factor underlying conscious access is the extension of brain activation to higher association cortices intercon- nected by long-distance connections and forming a reverberating neuronal assembly with distant perceptual areas. Why would this brain state correspond to conscious access? Neurocomputational simulations show that once stimulus-evoked activation has reached highly intercon- nected associative areas, two important changes occur: (1) The activation can reverberate, thus holding information on-line for a long duration essentially unrelated to the initial stimulus duration (2) Stimulus information can be rapidly propagated to many brain systems. We argue that both properties are characteristic of conscious information processing which in our view is associated with a distinct internal space, buffered from fast fluctuations in sensory inputs, where information can be shared across a broad variety of processes including evaluation, verbal report, planning and long-term memory [25]. Empirically, access of sensory stimuli to conscious report correlates with the activation of higher associative cortices, particularly parietal, prefrontal and anterior cingulate areas. In fMRI, the activation of those regions systematically separates masked versus unmasked pre- sentations of words [20] or images [26] undetected versus detected changes during change blindness [27,28] extin- guished versus seen visual stimuli in neglect patients [29] or missed versus reported stimuli during the attentional blink [9,22,23,30���32]. In many of these paradigms, anterior activation is accompanied by an amplification and an increase in functional correlation with posterior stimulus-specific areas [20,26,30]. Sudden parieto-frontal activation and top-down amplification are two frequent signatures of conscious perception. Is attention a confound or a necessity for conscious access? Some have argued that many of the above neuroimaging paradigms are inappropriately controlled because con- scious perception is confounded with increased attention and more extended stimulus processing. For instance, a conscious word can be attended, repeated or memorized while a non-conscious word cannot. Such confounds would suffice to explain the greater parieto-prefrontal activity to unmasked words [20]. For this reason, Tse et al. [18] have argued that one should prefer experimental designs in which attention is drawn away from the stimulus. They show that, in such a situation, correlates of stimulus visibility are found solely in occipital areas, not in higher associative regions, and therefore argue that the mech- anisms of conscious visual perception lie in extrastriate cortex. We obviously agree on one point: it is important to design paradigms in which conscious perception is not confounded with massive changes in overt or covert behaviour. However, this goal has been achieved in several studies. In our recent study of the attentional blink [23], for instance, subjects viewed a fixed stimulus and made similar motor gestures on seen and not-seen trials, yet those were still distinguished by strong parieto- frontal activation. We question, however, the proposal that inattention is an appropriate control. Under conditions of diverted attention, such as those studied by Tse et al. [18], even an unmasked stimulus is not guaranteed to be consciously perceived. On the contrary, considerable evidence indi- cates that without attention, conscious perception cannot occur. In the inattentional blindness paradigm, even a 700-ms stimulus presented in the fovea, when unat- tended, might fail to be seen [33]. During the attentional blink, a mildly masked stimulus, normally quite visible, becomes invisible when attention is diverted to another task [23,34]. Opinion TRENDS in Cognitive Sciences Vol.10 No.5 May 2006 205 www.sciencedirect.com