Persisting Questions

Especially tricky is the question of whether timing involves a single dedicated mechanism or different mechanisms that can be selected on the basis of duration ranges, current context, or other parameters. Timing in the microsecond range is known to involve delay lines, at least in barn owls, bats, and electric fish (for a review, see Carr, 1993). Coincidence detection mechanisms are considered more suitable to longer-duration ranges (see Mattel et al., this volume; Meck, this volume; Miall, 1992). In fact, various possibilities that account adequately for temporal encoding seem plausible in view of the great diversity of functional phenomena displayed by the central nervous system, whether these phenomena consist of oscillations, multiple convergent inputs, spatio-temporal potentiation, or sustained changes in membrane potentials, among others. The argument of biological plausibility is primordial in selecting models of timing mechanisms. Surprisingly, it is most often evoked in favor of accumulator-free models (e.g., coincidence detector mechanisms fitting with striatal architecture — Matell and Meck, 2000; see Meck, this volume; Meck and Benson, 2002; or delay lines fitting with cerebellar architecture — Moore, 1992), although accumulator-like mechanisms can be traced at various organic levels. For example, increased activity as a function of time is typically observed in neuronal or electroencephalographic recordings during motor preparation (for a review, see Requin et al., 1991). Moreover, in temporal production and discrimination tasks, the level of activity in certain cortical areas is higher when subjects overestimate rather than underestimate the target (Macar and Vidal, 2002; Macar et al., 1999). These data are entirely compatible with the hypothesis that in dedicated timing networks, the level reached in some type of accumulator mechanism determines subjective duration.

Now, how do we reconcile the idea of an accumulator under attentional control and that of implicit timing, pervasive whenever temporal regularities exist in the environment (see Hinton, this volume)? An interesting possibility is that we have a unique timing system that is activated in all cases, but the output of which is labile, and rapidly fades away from memory whenever we process nontemporal parameters (Zakay, 1989), as in many everyday life situations. In these cases, timing judgments would then be based on any information less sensitive to fast decay, such as the amount of work performed during a past interval. Behavioral data showing that in dual-task paradigms the directional error bias typical of prospective judgments (implying preliminary awareness of the timing task) is reversed for retrospective ones (obtained without preliminary warning) strongly suggest that it is not the timer output that is used in the latter case. The idea that the timer is nevertheless activated systematically, even when its output cannot be used, is strengthened by considering the ubiquity of anticipatory behavior. The rule is that a response to a significant event is primed by the earliest index that announces this event (cf. classical conditioning, which reflects basic computational mechanisms displayed by complex as well as very simple organisms). Anticipation supposes the coding of temporal order and leads to accurate timing after the repetition of relatively few trials, even when there is no need for accuracy in the current context (cf. Gallistel and Gibbon, 2000, 2001). Why, then, would timing not be subtended by a nonspecialized population of neurons? Subsets of these neurons might subsequently be selected depending on the prevailing context (see Hopson, this volume).

Certainly, this hypothesis does not reflect the dominant trend in the literature; rather, the timer is thought to be located within key structures of the central nervous system. However, even the dopaminergic effects on timing, which clearly highlight the role of striatal structures, might be viewed as reflecting the pervasive influence of dopaminergic systems on all basic functions, rather than as an argument for specific localization of a central timer in the striatum. Following this line of reasoning, choosing a general mechanism, such as learning, as the principle that guides the development of temporal dynamics within the neural network as a whole is certainly appealing (see Hopson, this volume), though error gradient backpropaga-tion mechanisms seem farther from biological reality than simpler learning algorithms, such as reinforcement (Watkins, 1989) or self-organizing maps (Kohonen, 1987). An even less demanding hypothesis might be to transpose the results achieved in robotics with the automatic synthesis of behaviors to the temporal dimension. Here, no learning algorithm is required, because learning is a border effect, resulting from the mere repetition of behavior (Touzet, 1999).

In sum, although the outcome of research on the neural bases of interval timing, as well as the new propositions it has inspired, is to be counted among the most significant advances of the last decade or so, the debate on whether a central timer can or cannot be identified is likely to persist for the foreseeable future. Perhaps this apparent paradox comes from the fact that the neural structures that are now acknowledged as crucial in temporal processing are involved in multiple functions and subtend elementary to highly complex behaviors. Progress on the functional organization of the brain has likely rendered archaic the idea of localizing the internal clock in a single structure devoted solely to interval timing.

0 0

Post a comment