Conclusions 439A similar characteristic is found in the perturbation model
of Estes (1972, 1997), an early simulation model that
strongly influenced the development of OSCAR. In the per-
turbation model, items are effectively represented as values
along dimensions, such as temporal or spatial position. The
organization can be hierarchical, meaning that an item might
be represented in terms of its position along an ordered list,
within-list, or within-group dimension (see Lee & Estes,
1981; Nairne, 1991). The crux of the model is the assumption
that the position values are subject to random perturbations
over time: That is, there is a certain probability that each rep-
resented value will drift along its position dimension. The
probability that a perturbation will occur can be specified
mathematically, and the model has been shown to generate
precise predictions about correct and incorrect performance.
For example, the model does an excellent job of explain-
ing the nature of errors in immediate serial recall. As noted
earlier, when people make errors in ordered recall, items tend
to be placed incorrectly in nearby positions, and there is a
regular gradient found as distance increases from the item’s
original position (see Healy, 1974). The perturbation model
not only generates these errors, but it also specifies exactly
how the gradients should change as retention intervals in-
crease. The model also nicely handles empirical dissociations
between item and order memory, particularly the different
forms of the serial position curve that have been reported (see
Healy, 1974). Moreover, as Nairne (1991, 1992) has shown,
essentially the same assumptions that handle data from
immediate serial recall can also fit data across retention
intervals lasting minutes or hours (although see Healy &
McNamara, 1996, for some qualifying arguments). Thus, the
perturbation model can be viewed as a unitary model, ex-
plaining both short- and long-term memory performance, and
neither rehearsal nor fixed decay assumptions are needed to
fit the data (Estes, 1997; Nairne, 1991).
The final model that I discuss is my own feature model
(Nairne, 1988, 1990, 2001; Neath & Nairne, 1995). All for-
getting in the feature model is attributed to interference, ei-
ther from feature overwriting or from incorrect interpretation
of the primary memory trace. Rehearsal can play a role in the
model, as a mechanism for effectively re-presenting list
items, but rehearsal plays no real role in producing standard
immediate memory phenomena such as the word length
effect or even the effects of articulatory suppression on per-
formance (see also Neath, 2000). The model assumes that
residual remnants of perceptual processing remain in primary
memory after list presentation. These primary memory traces
are represented as vectors of features and can be overwritten,
based on similarity, by subsequent list items. At the point of
recall, surviving traces exist in a degraded or blurry form and
must be interpreted prior to recall. Most of the interesting ef-
fects of immediate retention arise out of the interpretation
process.
The feature model has been applied successfully to most
of the standard phenomena of immediate memory, including
the modality and suffix effects. One of its most important as-
sumptions is the idea that the trace interpretation process is
guided by the presence or absence of distinctive features.
Correct performance hinges on the presence of features in the
degraded trace that uniquely specify one of the possible recall
candidates. To the extent that primary memory traces contain
features that are matched in all of the presented items, such as
a common sound or phoneme, performance suffers. It is this
characteristic of the model that explains the phonological
similarity effect, as well as long-standing phenomena such as
the von Restorff effect (see Kelley & Nairne, 2001). More
importantly, the trace interpretation process is assumed to re-
semble the kinds of cue-driven retrieval processes that guide
all forms of remembering, regardless of the time scales in-
volved (see Nairne, 2002). The feature model, like the other
unitary models discussed, assigns no special mnemonic laws
or properties (such as decay) to remembering over the short
term.CONCLUSIONSTransient memories, discussed here in the form of sensory
and short-term memory, clearly serve highly adaptive func-
tions in human cognitive processing. Sensory memories en-
able us to prolong the present, for the briefest of intervals;
short-term memories comprise the ingredients of conscious
awareness and play a vital role, among other things, in the
comprehension and production of spoken language. Com-
pared to the study of long-term retention, studying transient
memories is a relatively recent enterprise, commencing with
full vigor only in the second half of the twentieth century. As
we have seen, many issues remain unresolved, and funda-
mental controversies continue. Does short-term retention fol-
low its own unique operating laws? Is it necessary to propose
processes, such as decay, that apply uniquely to remembering
over the short term? Is sensory persistence truly an evolved
form of remembering, serving its own special function, or is
it simply an artifact of the properties of neural networks?
Despite these controversies, few questions remain about
the data to be explained. In the presence of distractor activity,
we can still remember only a handful of unrelated items for
more than 10 or 20 s; when an array of unrelated letters is
briefly flashed, a partial reporting of the array is still dramat-
ically better than reporting of the whole. Whatever form the