The perception of sensory input to the brain surely depends on "what actually comes in", i.e. measurable physical properties of the stimulus like intensity.
Input is received by the receptors (photoreceptors in the eye, hair cells in the cochlear, cells being sensitive for touch, pain, temperature etc. on the skin), passes the brainstem, feeds into the thalamus and eventually reaches the primary sensory areas in the brain. From there the signals are analyzed by secondary systems and eventually converge in polymodal areas of the brain forming the actual percept. This process is commonly referred to as
"bottom-up processing" as the signal comes in at a low stage of processing and is further rened and analyzed so the signal at higher stages is dependend on input from lower stages.
Top-down processing refers to the opposite principle: Information from a higher stage of processing is used to actually modulate new input at lower stages. This principle is found quite often in the brain. One example is the attenuation or amplication of sounds in the cochlea made by the outer hair cells, which receive input from the perioliviar nucleus.
While bottom-up processing e.g. enables the individual to rapidly shift atten-tion to a salient feature of potential importance, top-down processing enables cognition to alter these processes because the individual may be hungry and thus looking for a colored spot which may be food (Connor et al. 2004 [7]).
The common synonyms, "stimulus driven" for bottom-up and "goal directed"
for top-down, are good descriptions for these two dierent ways of
modulat-ing processmodulat-ing in the brain. Nonetheless care has to be taken not to see them as independent but overlapping principles that optimize e.g. attentional per-formance (Sarter et al. 2001 [44]).
This section will give a summary of the research with a focus on top-down processes and the auditory domain.
1.2.1 Attention
The role of bottom-up versus top-down processes in attention has been the topic of many recent studies (e.g. Debener et al. 2002 [9], Debener et al.
2003 [10], Hamker 2006 (in press) [18], Ogawa & Hidehiko 2006 [28]) and reviews (e.g. Deco & Tolls 2005 [11], Egeth & Yantis 1997 [14], Sarter et al.
2001 [44]).
Debener et al. (2002) [9] conducted a study where they tried to distinguish top-down and bottom-up processing by using two components of the auditory event-related potentials, namely the "novelty P3", detectable at about 300ms post stimulus for bottom-up and the "target P3", detectable at about 400-600ms post-stimulus for top-down. They used a paradigm called "Novelty Oddball". In this paradigm three stimuli are used: one frequent tone, one infrequent tone and several "novel" sounds. While the EEG is recorded subjects hear the frequent tone 80%, the infrequent tone 10% and one of the "novel" sounds 10% of the time. The task for the subjects is to count the occurence of the infrequent tone. The infrequent tone should therefore elicite the "target P3" while the novel sounds should elicite the "novelty P3". Therefore, the infrequent tone served as target-stimulus for the subject requiring him to actively pay attention to it. As attention is directed to the stimulus by cognitive processes from higher levels (the motivation to detect the infrequent tone) this reects top-down modulation. The opposite is true for the "novel" sounds. The subject does not actively pay attention to these sounds and does not expect them. As they are also very dierent from the tones (in the experiment the tones were pure sine-tones and the sounds were environmental sounds belonging to categories as "animals" or "machine") the subject will nonetheless detect them as odd and automatically draw attention
towards them. As attention is automatically directed to the stimulus by the stimulus itself (or better: by low processes) and not by higher level cognitive processes attention this reects bottom-up modulation. Debener et al. (2002) [9] conducted two recording sessions each consisting of two blocks, separated by one week. They found that the "novelty P3" showed habituation within one session while the "target P3" did not. The "target P3" however showed a signicant decrease in the second session compared to the rst while this was not shown for the "novelty P3". Although they admit several shortcomings of their design they conclude that in a novelty oddball design it is possible to distinguish between top-down and bottom-up processes by analyzing the
"novelty P3" versus the "target P3".
One year later, Debener et al. (2003) [10] used the same paradigm to show that top-down attentional processing enhances gamma-band activity at about 40Hz. They showed a signicant enhancement of gamma-band ac-tivity for the target-stimulus compared to the standard whereas the acac-tivity remained the same for the novel stimulus.
1.2.2 Expectancy
Although attention is by far the most studied aspect of top-down modulation of behavior and neural correlates, there has at least been some research about expecting a positive or negative reward (e.g. Breiter et al. 2001 [3]) and expectancy that should lead directed attention (e.g. Kastner et al. 1999 [20]).
In 1985 Perruchet [31] studied the role of expectancy on conditioning. In his experiment subjects heard a tone and were told in advance that in 50% of the trials this tone would be followed by a short pu of nitrogen to the eye that elicits an eye-blink response. In terms of Pavlovian conditioning, 50% of the trials had an US (air-pu) - CS (the tone) contingency while the other 50%
did not. The subject's task was to rate the likelihood of the US to appear.
Without the subjects knowing the contingency was organized in runs. One run consisted of 1 to 4 consecutive CS-US pairings or CS alone. Referring to the phenomenon called "gambler's-fallacy" (Ayton & Fischer 2004 [2]) he
made two speculations what could happen to the propability of the CR (the eye-blink response) in dependence of how often a pairing or no pairing would be presented in a row.
1. Propability of the CR increases with many consecutive pairings and decreases with many consecutive non-pairings.
2. Propability of the CR decreases with many consecutive pairings and increases with many consecutive non-pairings.
While the rst hypothesis is derived from strength-theory, i.e. the better the contingency the more likely the CR, the second is derived from expectancy-theory, i.e. the expectation of the subjects that after a certain number of consecutive trials it should nally change because overall probability is 50%.
In cognitive psychology this logical fallacy - i.e. the belief that events in the past aect random events - is called gambler's fallacy (for a recent discussion see Ayton & Fischer 2004 [2]).
For example: You toss a fair coin three times and it happens to come out three times "tail". When you now ask someone what would be the most likely result the next time you toss, he will most likely say "heads" because if the coin was fair the chances of getting "tail" four times in a row a minimal (Ayton & Fischer 2004 [2]). Of course, this is the wrong answer as every round is independent from the previous.
For the expectancy-rating only one hypothesis was made stating that accord-ing to expectancy-theory the rataccord-ing should decrease with more consecutive paired trials and increase with more consecutive non-paired ones. The results conrmed the strength-theory-hypothesis for the probability of CR and the expectancy-theory-hypothesis for the expectancy-rating.
Although Perruchet does not discuss his ndings with the view on top-down versus bottom-up modulation it seems worth doing so. The ndings of Per-ruchet seem ambiguous at rst view as we nd a decrease of one behavioral parameter and an increase of the other. Considering the nature of these responses, it becomes clear that there is a distinction to be drawn.
While the conditioned eye-blink response is automatic and not aected by higher cognitive processes as intention, the opposite is true for the expectancy-rating, which is an intentional response that is actually made by cognitive processes as expectancy, intention and experience. With reference to the in-troduction of top-down and bottom-up processing using the synonyms "goal-directed" and "stimulus-driven" further gives a distinction between the two processes of the experiment. A conditioned response is without doubt driven only by the stimulus while the expectancy-rating is directed by the goal of delivering an adequate forecast of the next event. Therefore, we can separate Perruchet's results to one modelling a bottom-up processing and one mod-elling a top-down processing, which can and should be interpreted for their own.