Implicit Learning 503Figure 18.2 Implicit learning in a serial response time task by amnesics
and normals. Source:Nissen and Bullemer (1987).
Other researchers using different measures of explicit
knowledge have shown high correlations between explicit
knowledge and performance (e.g., Perruchet & Amorim,
1992), as well as evidence of explicit knowledge of the se-
quence at the point where performance differences between
random and sequential presentation first become evident,
casting some doubt on the Willingham et al. (1989) finding of
relative independence of the two measures of knowledge.
However, given the evidence of sequence learning in
Korsakoff amnesics, we see that learning without awareness
certainly can occur.
Implicit Learning and Attention
It is interesting to ask whether the learning seen in the SRT
task is dependent on the availability of attention to process
relations between elements in the task. One way to assess this
is to compare learning in a single-task condition, in which
only the task to be learned is performed, with a dual-task
condition, in which a secondary task is performed. The re-
quirement to perform a secondary task should take attention
away from the SRT task. Such an experiment was carried out
by A. Cohen, Ivry, and Keele (1990) using a sequence task in
combination with distractor tasks of different difficulties.
They specially constructed sequences to contain either
unique associations, in which each stimulus uniquely speci-
fied the following (e.g., A always followed by C), ambiguous
associations, such that A might be followed by C in one case
and by D in another, or both. They found that ambiguous
sequences were not learned under the dual-task conditions
but that unique associations were. On the basis of these
results, they suggested that sequence learning depends on
two processes. The first process is automatic, in that it can
occur without attention’s being directed to it. This process
forms associations between adjacent items. The second
process, which requires attention to operate, is proposed to
build hierarchical codes based on parsing the sequence at a
higher level (i.e., into bigger subsequences) than associations
between only two items.
A. Cohen et al. (1990) showed that simple associations
could be learned under conditions of distraction, but the
amount of learning may nonetheless be affected by atten-
tional load. For example, Frensch, Buchner, and Lin (1994)
showed that whereas learning could take place under both
single- and dual-task conditions, and for both simple and
ambiguous sequences, the amount of learning that took place
was reduced when a distractor task was present, suggesting
that sequence learning is modulated by attention.
Jiménez and Méndez (1999) attempted to resolve the issue
of whether general attentional demands modulate sequence
learning by using sequences that were unlikely to be explicitly
learned and a secondary task that should produce little disrup-
tion. They examined the roles of both selective processing
requirements (attending to to-be-associated elements) and
generalized mental load (taxing attentional resources by
adding a secondary task to the sequence-learning task). An
SRT task was used in which the sequences of stimulus loca-
tions were generated following a noisy finite-state grammar
and response keys were pressed corresponding to the position
of the stimuli. In addition, the identity of a stimulus on a given
trial gave probabilistic information about where the next stim-
ulus would appear. A single-task condition was contrasted
with a dual-task condition in which two of four possible target
shapes had to be counted and the total reported at the end of the
block. Participants in both single- and dual-task conditions ex-
hibited sequence learning, showing faster reaction times and a
lower error rate for practice with grammatical sequences than
for random ones. Learning of the predictive relation between a
stimulus on one trial and the location of the stimulus on the
next trial was assessed by examining the difference between
valid (in which the predictive relation held) and invalid (in
which the position was not predicted) trials. Only in the dual-
task conditions, in which participants had to attend to target
identity in order to perform the counting task, were these rela-
tions learned. Thus, selective attention does seem to be neces-
sary for such learning to occur, and this learning occurs even
though (or precisely because) a secondary task must be per-
formed. In other words, paying attention to a predictive di-
mension seems to be necessary for this dimension to enter into
a predictive relationship.
In summary, sequence learning can occur implicitly, and
this learning is at least partly the result of automatic associa-
tive processes. Associative processes can be carried out