Binding binding: Departure points for a different version of the perceptual retouch theory

In the perceptual retouch theory, masking and related microgenetic phenomena were explained as a result of interaction between specific cortical representational systems and the non-specific sub-cortical modulation system. Masking appears as deprivation of sufficient modulation of the consciousness mechanism suffered by the target-specific signals because of the temporal delay of non-specific modulation (necessary for conscious representation), which explicates the later-coming mask information instead of the already decayed target information. The core of the model envisaged relative magnitudes of EPSPs of single cortical cells driven by target and mask signals at the moment when the nonspecific, presynaptic, excitatory input arrives from the thalamus. In the light of the current evidence about the importance of synchronised activity of specific and non-specific systems in generating consciousness, the retouch theory requires perhaps a different view. This article presents some premises for modification of the retouch theory, where instead of the cumulative presynaptic spike activities and EPSPs of single cells, the oscillatory activity in the gamma range of the participating systems is considered and shown to be consistent with the basic ideas of the retouch theory. In this conceptualisation, O-binding refers to specific encoding which is based on gamma-band synchronised oscillations in the activity of specific cortical sensory modules that represent features and objects; C-binding refers to the gamma-band oscillations in the activity of the non-specific thalamic systems, which is necessary for the O-binding based data to become consciously experienced.


INTRODUCTION
When visual cognition is studied from an interdisciplinary perspective, researchers typically try to understand how the specific data-processing modules in the cortex mediate perception of and attention to features, objects, and events. It was only in the eighties when researchers of cognitive processes began to pay attention also to the contribution of the so-called non-specific systems of modulation to the perceptual and attentional processes (Baars, 1988;Bachmann, 1984;Crick, 1984). As one particular instance of such an approach, the theory of masking named perceptual retouch theory was introduced (Bachmann, 1984(Bachmann, , 1994(Bachmann, , 1999. interactive effects of processing sub-systems within a larger set of brain systems, which are considered the very mechanism of conscious experience. Basically, masking was explained as the result of relative deprivation for specific data processing (that of the target) of the service by the processes that typically perform the function of generating conscious experience for actual sensory information. In normal perception which is accompanied by conscious experience of the perceptual object, specific data (features) about that object, as represented by the driver-neurons' cortical activity, has to be modulated by presynaptic facilitatory input from the non-specific sub-cortical systems. Without this kind of non-specific modulation, the data represented in the specific cortical modules remains pre-conscious (Bachmann, 1984(Bachmann, , 1994Bogen, 1995;Crick & Koch, 2003;Llinás, 2001;Magoun, 1958;Rees, Kreiman, & Koch, 2002;Schiff & Purpura, 2002). The operation of causing pre-conscious specific perceptual information to become explicit in conscious representation was termed perceptual retouch by Bachmann (1984Bachmann ( , 1994. The spatio-temporal properties of the functioning of the specific representational systems and non-specific modulation systems enabled to be put forward a masking theory which was surprisingly well consistent with quite many empirical facts from masking experiments (Bachmann, 1984(Bachmann, , 1994. The most important of these properties are as follows: 1. Sensory stimulation evokes both specific data coding in the cortical sensory areas (SP) and a non-specific arousal-like process in the sub-cortical (especially reticular and thalamic) centers (NSP). The delay with which evoked activity reaches cortical parts of SP is substantially shorter (e.g., a few dozen ms) than the delay with which the NSP activity or a dynamic change in NSP activity, evoked through collaterals, arrives at the designated driver neurons in the same cortical SP locations. The boost of NSP-impulses that is necessary for creating an explicit representation of sufficient saliency arrives at the cortex when the SP-processes are already more or less stabilised and their activity is about to decay.
2. While receptive fields of SP neurons are small and allow detailed representation, with specific contents varying from driver to driver (detector to detector), receptive fields of NSP neurons are large and unspecific regarding detailed contents (Brooks & Jung, 1973;Churchland & Sejnowski, 1992;Crick & Koch, 2003;Purpura, 1970). This property enables stimuli that are separated in space and represent different specific contents to evoke activity and interact through the activity of the same NSP unit. For instance, an initially presented stimulus (S1) evokes NSP-activity that can presynaptically modulate both the SP-units representative of S1 itself and SP-units representative of S2. These interacting stimuli need not be spatially superimposed, although they may be. (Figure 1 illustrates the functional architecture of the dual-process approach that lays the grounds for the retouch theory.) Backward masking (including metacontrast) was explained in the following way. S1 leads to (1) fast coding within cortical SP and (2) a slower NSP-process. When S2 is presented very soon after S1 (e.g., with stimulus onset asynchrony, SOA, equal to 15 ms), a more or less simultaneous process of feature-coding and object formation is going on in SP for S1-and S2 features, and a common ("blended") pre-conscious representation of a pseudo-object is formed. When the delayed modulation from NSP arrives presynaptically onto S1 and S2 related SP-units in the cortex, the result of retouch for consciousness will be that a blended pseudo-object is perceived. Whether both S1 and S2 can be distinctly perceived depends (a) on the intensity relations between S1 and S2 (a more intense stimulus' features and surfaces dominating), and (b)

Figure 1.
A schematic of the functional architecture of the two interacting systems for sensory data processing. Specific pathways (SP) send sensory signals upstream to the specific cortical modules that encode stimuli features and integrate objects in terms of their specific contents. This fast system builds perceptual representations also pre-consciously. A slower, non-specific system (NSP), which is located in feature-wise non-specialised thalamic and reticular centers (e.g., intralaminar nuclei, reticular nucleus, globus  SOAs between S1 and S2, S1 can be perceived well or not so well, depending on the peculiarities of interstimulus interaction within SP. When S2 is presented after S1 with an intermediate delay (e.g., SOA = 50-80 ms), the NSP-modulation boost evoked by S1 arrives at the cortical SP at the moment when the S2 specific process is at its maximum (e.g., EPSP level is maximised), but the S1 specific process has begun to decay (e.g., EPSP level has somewhat subsided already). As a result, in the retouched perceptual image, S2 saliency is higher than S1 saliency and S2 dominates S1, as is the case in mutual masking (e.g., Bachmann & Allik, 1976;Michaels & Turvey, 1979) or in metacontrast (Breitmeyer, 1984). Subjects attend to S2 and it will replace S1 in subjective perceptual representation. With long SOAs above 150-200 ms, subjects perceive distinct successive objects -S1 and S2; both objects have had their own retouch cycles and they are entered into and held in short-term memory.
In this conceptualisation, the activity of single units was postulated to represent the activity of the whole pool of responsible neurons. Perceptual retouch theory, besides what was described above, was also able to predict perceptual latency priming (PLP, Bachmann, 1989;Neumann & Scharlau, in press; press), backward masking with common-onset, asynchronous offset displays (Cohene & Bechtoldt, 1974;Di Lollo et al., 2000), a variety of psychophysiological effects where experimental facilitation of the NSP leads to unusually efficient perception of S1 (e.g., Bachmann, 1994), and some more effects. Despite this, several controversial aspects of the retouch theory became evident. While Breitmeyer and Öğmen (2000) suggested testing a unique retouch-theory prediction that there could be an illusory temporal order reversal between S1 and S2, the properties of this illusion  did not fit with retouch explanation. With PLP, the time properties of the maximum priming effect predicted by the retouch theory (at about 50-100 ms) did not conform easily to several instances of much higher PLP values found in recent experiments (e.g., Scharlau, in press;Scharlau et al., 2005).
In the retouch theory, the effects of increased visibility and saliency that ensue due to NSP-modulation were not differentially related to the contour system and surface representation system responses. However, manifold evidence shows that time-course functions of masking can substantially differ for those two perceptual properties of objects in masking Ishikawa et al., 2006). Moreover, retouch theory is undeveloped to account for the intriguing differences between backward (metacontrast) masking, where the same local vernier targets and masks allow either strong masking or unmasking depending on whether the so-called shinethrough test-and-mask combinations are used or not (e.g., Herzog, 2006). All this enforces thinking about the revision or additional development of the retouch theory.
But this is not all. In the retouch theory, the core mechanism was the mechanism for generating consciousness as it was understood until 1984. Since then, important developments have also changed the understanding of the mechanisms of conscious experience.
Although the basic principle -SP has to be modulated by NSP in order to be able to explicitly communicate SP contents -has remained the same, many new characteristics of how SP and NSP interact so as to produce consciousness have become clearer (Bogen, 1995;Edelman & Tononi, 2000;Llinás & Ribary, 2001;Rees, Kreiman, & Koch, 2002;Sherman & Guillery, 1998;Singer, 1998;Steriade, 1996a, b;Steriade, Jones, & Llinás, 1990;Steriade, Jones, & McCormick, 1997;Ward, 2003). This also necessitates some revision of the perceptual retouch theory. The remaining part of the present article is devoted to outlining the premises for such a revision (or rather -development).

ATTENTION ENHANCES GAMMA-REsPONsEs
Although gamma-synchronicity is a response given also to unattended stimuli, attention and awareness-related status tend to enhance gamma-oscillations.
Thus, Summerfield et al. (2002) showed that awareness of backward-masked stimuli correlated with gamma-activity in occipital and temporal cortices.
High-contrast, small, periodic stimuli elicit gain and synchrony of gamma responses in visual areas when the stimuli are attended (Womelsdorf et al., 2006).
Yet, unattended stimuli also evoke a burst of gamma activity, although the spike-field coherence is smaller than in attended conditions. The onset-related firing rate was maximal at about 150 ms, post-stimulus. In a shape-tracking task, successful allocation of attention enhanced gamma-response (Taylor et al., 2005). But unattended changes in visual shapes also were accompanied by gamma boosts. Thus attention necessarily boosts gamma responses, but cannot be regarded as a sufficient mechanism for consciousness. In binocular rivalry, transient bursts of increased global phase synchrony in the gamma band were associated with visibility (Doesburg et al., 2005). As in rivaly no strong input transients are involved and because the gamma-band activity begun to peak 400-250 ms before subjects responded to the change, all this may point to the possibility that we deal here with endogeneous gamma-enhancement (an equivalent of retouch activity?) that predicts recruitment of SP-representations for consciousness. One way or another, gamma-synchrony appears to be associated with coherent conscious percepts. But again, it seems necessary, but we do not know on what conditions it also becomes sufficient.
It is known that lateral occipital and temporal areas display gamma oscillations to attended stimuli (Tallon-Baudry et al., 2005). The latency of the response equals about 100 ms. Gamma-oscillations in the calcarine gyrus are characterised by a fastemerging, high-frequency pattern (even more than 70 Hz). In a visual discrimination task that involves feature binding, gamma-response to an attended object emerges within only 50-150 ms (Herrmann & Mecklinger, 2001).
In the author's present thinking, both attention and the consciousness-related property of perception are strongly associated with gamma-frequency brain activities, but the double dissociation for (1) attentionrelated gamma activity and (2) consciousness-related gamma activity is yet to be demonstrated in numerous replication studies. The arguments why I prefer not to put an equation mark between attention and consciousness can be found in Bachmann (2006). Most importantly, fully focused and intense attending to a stimulus or location (e.g., in metacontrast masking, http://www.ac-psych.org binocular rivalry or motion-induced blindness) that also brings about a gamma burst in the brain does not automatically guarantee consciousness for the attended to or expected stimulus. And vice versa: for information processing that is biased and facilitated by selective attention, and that should produce gamma enhancement, there is no guarantee that the corresponding stimulus-information becomes consciously apprehended (e.g., Jaśkowski et al., 2002;Kentridge et al., 2004). Indirectly, this supports the idea that we need to have not only one variety or mechanism of gamma-activity as related to attention/consciousness, but it may be better to look for at least two brain systems prone to gamma-range dynamics when selectively processing information, but at the same time possessing relative functional autonomy. This is what fits with the agenda of the following part of this article.  (Kotchoubey, 2005). On the other hand, relatively small injuries or narrowly localised anaesthetic targeting can render subjects totally unconscious (Baars, 1997;Bogen, 1995;Newman, 1995;Steriade & McCarley, 2005).
One of the best models so far to describe SP/NSP oscillatory interaction in generating conscious representation has been offered by Rodolfo Llinás (e.g., Llinás, 2001;Llinás et al., 2002Llinás et al., , 2005.

VISUAL BACKWARD MASKING AND RELATED PHENOMENA IN THE LIGHT OF "BINDING BINDING"
Let me explain backward masking by the interaction of O-binding and C-binding. After having been presented, S1 evokes and sets the SP-and NSP oscillatory activity in motion. The part of modulating oscillatory activity which is caused by S1 transient becomes effective at the cortical level later than the cortical burst of SPsystem gamma-oscillations for S1 had emerged. At the same later time, the gamma-burst of S2-evoked oscillatory activity is generated. C-binding has to deal with two competing oscillatory neuronal active ensembles -that for S1 (already decaying) and that for S2 (showing the most-vigorous, "fresh" pattern of oscillations with higher amplitude and perhaps with slightly better coherence characteristics). Moreover, reentrant signals within the cortical SP meet more driving input from S2 than from S1, which has been switched off Why is it that in metacontrast the first-coming target is often totally suppressed, although an interpretation of the retouch theory considered by Breitmeyer andÖğmen (2000, 2006) would predict some diminished, pyramidal neuron in the sensory cortex specific pathway for sensory information transmission non-specific thalamocortical pathway for modulation of the activity of neurons that carry specific information channels for lateral cortical interactions

) The specific pathway activates pyramidal neurons and inhibitory interneurons (upper red), producing cortical oscillations by direct activation and feedforward inhibition. Collaterals from this pathway produce thalamic feedback inhibition through the reticular nucleus (lower red). The return corticothalamic pathway (curved green arrow) from pyramidal cells returns this oscillatory loop to specific and reticular thalamic nuclei (yellow and red lower circles). The non-specific thalamocortical pathway projects to the cortex and gives collaterals to the reticular nucleus. Pyramidal neurons return the oscillation to the non-specific and reticular thalamic nuclei (green and red lower circles). This forms the second resonant loop (curved green arrow on the right). The conjunction of the specific and non-specific loops is hypothesised to generate functional binding by temporal coincidence.
http://www.ac-psych.org masking is diminished or eliminated when spatial attention is directed to the target location before its presentation (Enns, 2004). In terms of the revised retouch theory, the pre-cue evokes C-binding processes ahead in time and when the target appears, SP-oscillations are quickly integrated into the syn-chronised NSP+SP, oscillatory ensemble. The target becomes visible at once.
According to the results of our recent study (Luiga & Bachmann, in press), release from substitution masking is obtainable with local spatial pre-cues, but not with central pre-cues that direct spatial attention in an abstract, encoded format (and this holds even for very long SOAs between pre-cue and target-plusmask stimulus, where there is plenty of time for the pre-cue to be processed and interpreted). My explanation is that it is difficult to engage a sufficiently ef- The temporal dissociation of different aspects of masking, such as between contour-and brightnessprocessing mechanisms Ishikawa et al., 2006), as well as absence of metacontrast with opposite-polarity luminous targets and masks (Becker & Anstis, 2004), are a valuable recent addition to the masking literature. In , meta-and paracontrast was studied, and subjects had to judge the surface brightness of target discs or else discriminate the contours of target discs (with a small edge segment cut off at different locations). Targets were masked by surrounding rings as in the many earlier classic studies. It appeared that optimum SOAs for the contour task were much shorter than those for the brightness task. In paracontrast, where the mask precedes the target in time, target contrast facilitation was found (consistent with even the earlier version of the retouch theory). Ishikawa et al. (2006) varied grating-orientation and -spatial frequency of the surface of targets and masks, and they also applied a metacontrast task requiring detection of targets. They found that at short SOAs, metacontrast magnitude strongly depended on stimulus feature specificity, whereas at longer SOAs (above 40 ms), masking demonstrated strong contrast sensitivity and low stimulus feature specificity. In the earlier retouch theory version (Bachmann, 1994) it was claimed that metacontrast is unspecific to spatial-frequency properties of the stimuli. Now this remains to be revised.
The above described effects are both accountable by assuming variations in the oscillatory activity within the O-binding system. This variation can be a function of temporal properties of the brightness, surface and contour encoding sensory systems. In some instances, http://www.ac-psych.org parallel oscillatory activity between target-related and mask-related object binding may be possible when the channels (e.g., on-system and off-system) can involve oscillatory activity in parallel, with the result emerging that C-binding explicates both the target and mask. In some other instances, as is the case with inter-contour conflict, C-binding explicates severe metacontrast with one range of timing; in the case of brightness-processing mechanisms being involved, the timing characteristics may differ.
The earlier version of the retouch theory predicted U-shaped metacontrast functions without any further oscillatory shape of the masking function as dependent on SOA (Bachmann, 1994). Besides masking, retouch theory was used to explain several other phenomena such as flash-lag effect, Fröhlich effect, PLP and some others as well (Bachmann, 1999(Bachmann, , 2006. In the experiments demonstrating the flash-lag effect, two types of stimulation are juxtaposed: an object that continuously changes its feature value is presented for some time, and another object that carries an invariant feature value is briefly flashed alongside the changing object (e.g., the spatial location of a moving bar is changing or the colour of a stationary disc gradually changes from yellow to red while another bar is flashed at a stationary location as aligned with the moving bar or another disc is flashed nearby and has the same colour as the changing disc precisely at the moment of flash presentation). Flash-lag effect means an illusion where the feature value of the flashed object (e.g., location, colour) lags behind the perceived feature value of the changing object. In the Fröhlich efect (Fröhlich, 1923), the perceived first position of a moving object that comes from behind an occluder is located not at the position it actually became exposed ( Actually, as seen in Figure 3, PLP values tend to deviate from the theoretically expected y = x, function. (Instead, y = kx seems to happen, with k equal to about 0.5.) The revised retouch theory can be specified so as to be able to explain this puzzle. We can assume that it is not the latency with which the first discharges in the cortex, caused by subcortical presy-

WELL-KNOWN MASKING THEORIES AND "BINDING BINDING"
As a dual-process theory, the revised retouch theory 1. The RECOD model of masking , which outsprung from the earlier very An illustration of the functional relationship relating SOA (set between prime and target) with perceptual asynchrony between targets presented in control conditions without prime and main experimental conditions where prime precedes target. The slope of the function is about 0.5. (Adapted from Aschersleben and Bachmann, 2004, unpublished.) http://www.ac-psych.org ing reentrant signalling and partial decay of S1 at the pattern level in favour of S2 representation) have been already carried out. When, due to distractors, attention is dispersed, NSP-resources cannot be rigorously and rapidly invoked and mask information becomes the dominating data for retouch because C-binding becomes effective only at the moment when the O-binding process emphasises mask-object representation.
When C-binding has been set on in advance, substitution masking obviously disappears, but the pre-cue has to be sensory in nature and spatially localised close to the target (Luiga & Bachmann, in press).

ENDCOMMENTS
To end the acquaintance-tour of this sketch of the modified perceptual retouch theory, a few general remarks are necessary. Due to its emphasis on the temporally extended process of SP/NSP interaction, retouch theory naturally fits with the notions about minimum excitatory duration, which is necessary for a conscious percept to emerge (e.g., Libet's or Koch's works -see Koch, 2004), and about the importance of considering the object updating operations in addi-