A number of studies
have manipulated the clock component of the model both in animals
(Maricq, Roberts, & Church, 1981; Meck, 1983) and in humans
(Droit-Volet & Wearden, 2002; Penton- Voak, Edwards, Percival,
& Wearden, 1996) and have provided evidence for a
pacemaker–accumulator clock of the type that SET
proposes. Much less attention has been devoted to
memory and decision
components of the model. Some recent studies have explore
timing in the probable absence of reference memory (Allan &
Gerhardt, 2001; Rodriguez- Girones & Kacelnik, 1999; Wearden
& Bray, 2001), but little research (apart from that of Meck,
1983, with rats) has attempted to manipulate the contents of the
reference memory itself.
The reference
memory component of the SET model serves two
functions. The first is to provide a temporal reference
for behavioural stability by storing “important”
times. In animal studies the content of reference memory will
typically be the remembered time of reinforcement; in human
studies the reference memory is assumed to contain memories of
standard durations important for the task in hand. The
second role of reference memory is to generate the scalar
property of timing: the requirement that the standard deviation
of internal “estimates” of time is a constant fraction
of the mean. This scalar property is a form of Weber’s law
and assumes a timing process with constant sensitivity as the
interval timed varies.
A key question in relation to reference
memory is how the temporal representations
develop..
Apriori, successive
temporal memories could be stored in at least two distinct
ways. In one of these, the successive individual examples
could be aggregated together in some way (e.g., averaged into some
mean time value), so increasing numbers of successive presentations
of a standard would tend to make the aggregate value more
representative or accurate than individual examples were. In
this case, the contents of temporal memory would be transformed
versions of the individual standard values provided. In contrast,
another possibility is that the different examples of the
standard are stored completely separately and then sampled. In
this case, the individual identity of each previously presented
standard would be maintained in memory, and the contents of
reference memory might not statistically change with increasing
numbers of standard presentations. Depending on how these
separately stored examples of the temporal reference are used,
increasing numbers of presentations of the standard may not
necessarily change the temporal representation
used.
Although we presented three experiments with
consistent results above, perhaps the main contribution of the
present article is that it argues for a reconceptualization of the
nature of temporal reference memory, in humans and (most
probably) also in animals. This reconceptualization considers the
reference memory not as an extensive store of previous experiences,
but as a much smaller store, either containing a single item or,
as in the perturbation model, as a single item with upper and lower
boundaries, which control whether or not change occurs as a result
of some new experience. Perhaps the most surprising thing about
this view of reference memory is that it is not only compatible
with almost all previous results in both humans and animals, but
also completely consistent with almost all the theoretical
treatment of reference memory provided by Meck and colleagues
(Meck, 1983, 1991; Meck & Angell, 1992; Meck & Church,
1987a, 1987b; Meck et al., 1984), in their development of the
K* concept. The principal mechanism controlling variability
of temporal references might be the K* transformation
itself, rather than any aspect of the storage process of reference
memory.
