438 Sensory and Working Memory
possible to generate precise and sometimes novel predictions.
For example, the primacy model predicts a unique and inter-
esting kind of output error called afill-in error. It turns out that
when people miss an item in the recall sequence, as when they
fail to recall the letter B in the second serial position, there is a
significant tendency to report the missed item in the next serial
position (i.e., as occurring in Position 3). Page and Norris
(1998) argue that the assumptions of the primacy model
correctly predict this tendency, whereas other models do not
(although see Neath, Kelley, & Surprenant, 2001).
The connectionist model of Burgess and Hitch (1999) also
attempts to formalize the operation of the phonological loop.
Serial recall is more clearly cue-driven in this model, rather
than based on activation levels, and the cues in this case
are elements of a moving or drifting context signal. List items
are associated with a snapshot of the context, which is set at
the moment an item is presented. Adjacent items, because the
context signal is slow-moving, tend to be associated to simi-
lar contextual cues, which helps to explain, in part, why peo-
ple tend to transpose adjacent items in recall output. At the
point of recall, the context signal is reset, and a competitive
selection process proceeds based partly on the activations
that items receive via their connections with the context.
The Burgess and Hitch (1999) model also assumes that
connections exist between phonemic information and to-be-
recalled items. This phonemic layer, when activated, serves
essentially as the analogue for the phonological store in the
working memory model. Rehearsal strengthens otherwise de-
caying connections between the phonemic layer and items,
thereby helping to account for phenomena such as the word
length effect. Again, the details are beyond the scope of this
chapter, but the model nicely handles a variety of phenomena
in immediate retention, including the occurrence of sound-
based errors, the effects of articulatory suppression, and even
temporal grouping effects (see also Hitch et al., 1996).
Another recently developed hybrid model is the Start-End
model of Henson (1998). Once again, the major assumptions
of the standard juggler model are reproduced in the form of
decaying representations that are refreshed through internal
rehearsal. The locus of the word length effect, and the general
relationship between articulation rate and span, is placed in
the trade-off between rehearsal and decay. The unique aspect
of the Start-End model is its machinery for handling the
recovery of serial order. Henson (1998) assumes that, during
presentation, list items are coded relative to the start and
end of the list. Items near the beginning of the list are associ-
ated more strongly with a beginning-of-the-list start marker,
and recency items more strongly with an end marker.
These position codes are then reinstated at test and used to
activate associated items, and an item is selected for recall.
Because of the nature of the position codes, adjacent items
tend to be associated with overlapping cues, leading to sys-
tematic error gradients in recall. The Henson (1998) model,
like the other hybrid models, can be shown to mimic the
major phenomena of immediate retention.Unitary ModelsThe second class of simulation model, the unitary models,
typically rejects decay and rehearsal as the major determi-
nants of immediate memory performance. As noted earlier
in the chapter, there are both empirical and theoretical rea-
sons to question whether decay and rehearsal are viable ex-
planatory constructs in immediate retention. Although people
certainly do rehearse, and rehearsal can play a role in unitary
models, its role in unitary models is not to refresh otherwise
decaying representations. Instead, rehearsal is typically
viewed as another kind of stimulus presentation, which can,
depending on the circumstance, either facilitate or interfere
with subsequent retention (see G. D. A. Brown et al., 2000;
Tan & Ward, 2000).
In the OSCAR model (which stands for OSCillator-based
Associative Recall), forgetting is caused entirely by various
forms of interference (G. D. A. Brown et al., 2000). By rely-
ing on interference and rejecting decay, OSCAR shares an
important property with most conceptions of how forgetting
occurs in long-term memory (see Crowder, 1976; Neath,
1998). In a fashion similar to the Burgess and Hitch (1999)
model, associations are formed between to-be-recalled items
and snapshots of a moving context signal (instantiated
through sets of slow and fast temporal oscillators). The con-
text is reset for recall, and there is cue-driven competition for
output. Interference in the model occurs because of response
competition during the selection process, output interference,
and inherent capacity limitations in the storage mechanism
that is employed (see Brown et al. for details).
Regarding rehearsal, the OSCAR model includes no active
mechanism for the rehearsal of items during list presentation.
Although its authors did consider the possibility of strategic
rehearsal, the concept was rejected because, essentially, “we
have found no need for it in accounting for the phenomena
under consideration” (G. D. A. Brown et al., 2000, p. 172).
With regard to decay, the concept is rejected for many of the
same reasons discussed earlier in the chapter—that is, certain
data seem antagonistic to the proposal of decay—but also be-
cause OSCAR is designed to explain data from both short- and
long-term retention environments. Indeed, OSCAR fits data
from retention intervals lasting seconds as well as it fits data
from intervals lasting hours, without changing any of its main
assumptions.