Modeling as Part of Perception: a Hypothesis on the Function of Neural Oscillations

We argue that the effectiveness of synchronization of oscillatory neural activities coding simple features, as it relates to perceptual organization, may originate in the temporal characteristics of resonance that develops in a two-stroke architecture of neural information processing – a cycling between bottom-up and top-down mechanisms. We provide empirical evidence to support the idea that resonance involves the generation and evaluation of ‘models’ of spatial and temporal stimulus attributes. By virtue of temporal modeling, temporally assisted spatial segmentation comes to be very precisely determined by the combination of both global and local stimulus phase. Neural oscillations in electrophysiological investigations are found throughout the cortex, at a wide range of frequencies and at every level of resolution. The extent to which the brain appears to employ oscillations and the complexity of the physiologically observed phenomena challenge any simple and somewhat more general explanation of their functional role. This situation is partly caused by the necessity to restrict electrophysiological investigation to certain segments of the cortex, certain frequency ranges and/or limited levels of resolution. In this contribution we aim to show that the frequency locking of fast oscillatory activity between successive stages of processing promotes the segmentation of figure and ground. In addition we suggest a very general functional interpretation of the oscillations involved: They serve to: (a) generate interpretations of sensory attributes in bottom-up processing which are evaluated by (b) modeling prototypical sensory patterns to anticipate and match the sensory activities by means of top-down processing. It is important to note that, in contrast to hypotheses derived from electrophysiological investigations, we are able to reach this conclusion by examining the possible functions of neural oscillations from an information processing perspective only: i.e. by employing the psychophysical research agenda. In the historical context this is perhaps not remarkable: Fechner, who was the first to speculate about the role of oscillations in conscious perception, also demonstrated the first, but still puzzling empirical evidence for an oscillatory sensory function through demonstrations of color perception via the rapid and periodic alternation of black and white stimuli (i.e. Fechner’s Colors, see Fechner, 1838 and for a modern approach see Herrmann & Elliott, this issue). Fechner (1860) also suggested that the most likely neurophysiological process by which psychological events might be generated were neuronal oscillations. Recent, empirical studies have revealed more specific evidence to suggest that perceptual mechanisms operate with very well defined temporal relations: Kristofferson, (1980, 1990) found that breaks in a stepwise-increasing discrimination function correspond to a successive doubling of the Weber fraction. Local invariances in absolute thresholds of very low sinusoidal sounds (v. Békésy 1936) and of apparent motion (Geissler, Schebera and Kompass, 1999; Kompass & Geissler 2001) exhibit similar systems of preferred temporal intervals with a rich structure of simple integer size relations. It has been suggested that these phenomena can result from phase locking between coupled oscillators across a range of different, but necessarily related, frequencies (see e.g. Geissler, 1987; Geissler et al., 1999), although the timing of the apparent motion stimuli seems to indicate that temporal intervals may also play a fundamental role as units of perceptual processing. These results are complemented by a rapidly growing body of physiological data demonstrating a correspondence between particular perceptual experiences and zero-phase lag synchronisation of the neural activities coding elementary stimulus attributes (see Gray, 1999 for review). One piece of psychophysical evidence in which these ideas converge is the synchronypriming effects reported by Elliott and Müller (1998, 1999, 2000, 2001). In Elliott and Müller’s experiments, a local, figurally-relevant priming stimulus was presented within the context of a flickering premask matrix of four frames, each presented at 10 Hz and, by virtue of frame-frame asynchronies of 25 ms with a global matrix frequency of 40 Hz (see Figure 1). The local 10-Hz priming frame was found to generate a prime with a 40-Hz structure, which, by Elliott and Müller’s (2000) account may be explained if one considers local prime activity to have inherited the frequency of global frame presentations. Given that priming stimuli were not detected by observers and given that they did not promote the deployment of attention to the priming stimulus location in the flickering display, Elliott and Müller considered the priming effects to be generated very early in visual processing. Early visual mechanisms do not possess receptive fields with sufficient dimensions to be capable of responding to the entire premask matrix, although later cortical mechanisms do. Consequently, Elliott and Müller (2000) proposed that early and local neural activity might become primed at 40 Hz, by virtue of this temporal code being fed back from later mechanisms, responding at 40 Hz to the premask matrix as a whole. Successive phase shifts across this network of recurrency would serve to shift the phase of local neural activity coding the prime, such that it became temporally segmented from other activity coding distractor frames. Given that the priming frame is also located in a different place in the premask matrix relative to the distractor frames, temporal segmentation may be considered to spatially segment the priming frame, thereby leading to spatially-specific priming effects as reported by Elliott and Müller (1998).