CP 122
1050 Bruxelles BELGIUM
axcleer@ulb.ac.be
Abstract
How does implicit learning interact with the availability of
explicit information? In a recent series of experiments,
Curran & Keele (1992) demonstrated that sequence learning
in a choice reaction setting involves at least two different
processes, that result in differing availability of the acquired
knowledge to conscious inspection, and that are differentially
affected by the availability of attentional resources. In this
paper, I propose a new information-processing model of
sequence learning and explore how well it can account for
these data. The model is based on the Simple Recurrent
Network (Elman, 1990; Cleeremans & McClelland, 1991;
Cleeremans, 1993), which it extends by allowing additional
information to modulate processing. The model implements
the notion that awareness of sequence structure changes the
task from one of anticipating the next event based on
temporal context to one of retrieving the next event from
short-term memory. This latter process is sensitive to the
availability of attentional resources. When the latter are
available, performance is enhanced. However, reliance on
representations that depend on attentional resources also
results in serious performance degradation when these
representations become less reliable, as when a secondary
task is performed concurrently with the sequence learning
task.
Introduction
In recent years, sequence learning in choice reaction
settings has elicited considerable interest as a vehicle to
study implicit information processing (e.g., Cleeremans
& McClelland, 1991; Lewicki, Hill, & Bizot, 1988;
Nissen and Bullemer, 1987; Perruchet, & Amorim,
1992). In such tasks, subjects are presented with a
visuo-spatial choice reaction task, but, unknown to
them, the sequence of successive stimuli is structured,
so that the uncertainty about the next event may be
reduced based on the constraints set by previous
events. Typically, subjects exhibit detailed sensitivity
about these sequential constraints, yet their explicit
knowledge of the sequence remains very limited. This
kind of outcome, where detectable performance
improvements are not accompanied by correlated

The author is a Senior Research Assistant of the National
Fund for Scientific Research (Belgium).
improvements in explicit, reportable knowledge, is
referred to as implicit learning (Reber, 1989). Implicit
learning contrasts with explicit learning (exhibited for
instance by subjects engaged in problem-solving
behavior), in which processing is usually goal-directed
and fully available to conscious inspection. This notion
of two modes of learning has led many to formulate
dichotomous theories of cognition in which implicit
and explicit processing are generally thought to be
complementary (in the sense of one mode being most
efficient in the exact conditions where the other is least
efficient) and independent (see Hayes and Broadbent,
1988, Reber, 1989, for examples).
However, it seems reasonable to assume that
learning in general is never purely implicit or purely
explicit. On the contrary, it is likely that most tasks that
have been dubbed implicit do in fact involveto
various degreesexplicit strategies and knowledge.
Goal-directed, intentional processing cannot simply be
turned off. Many recent studies (Curran & Keele,
1993; Perruchet & Amorim, 1992, Howard, Mutter, &
Howard, 1992) have begun to explore the effects of
various factors relevant to the implicit/explicit
distinction on performance in implicit learning tasks.
These factors are maybe best described as
characteristics of explicit learning, that is, (1)
awareness of the material, (2) intentionality, and (3)
sensitivity to the availability of attentional resources.
The picture that emerges from these studies is far too
complex to be discussed in detail here, but in a
nutshell, all three factors may facilitate or interfere
with performance in implicit learning tasks, depending
on other factors such as stimulus salience or material
complexity. Thus, there seems to be reasonable
empirical grounds for distinguishing between learning
processes that are differentially affected by the
variables listed above.
Taking such an implicit/explicit dichotomy for
granted, if only in a purely functional sense, one may
have different theories about the nature of the
representations and mechanisms that produce this
dichotomy. Three positions have been expressed in the
implicit learning literature. First, some authors (e.g.,
Perruchet & Amorim, 1992) argue that performance in
implicit learning tasks does not necessarily reflect the
operation of an independent implicit learning system.
Rather, performance would be mostly based on explicit
processing, but the resulting knowledge is fragmented enough that verbal reports probing for general
information are unlikely to reveal the extent of
subjects knowledge. Other authors (e.g., Knowlton,
Ramus & Squire, 1992) assume that implicit and
explicit learning are supported by different memory
systems, and that these systems are completely
independent from each other. Implicit and explicit
learning would thus proceed in parallel, but without
interacting. They produce different kinds of
knowledge, and are most likely to operate efficiently in
contrasted settings. Finally, there may be an
intermediate position where one assumes that implicit
and explicit processing indeed rely on distinct memory
systems, but in which some interactions between the
two systems are allowed, and in which some
processing resources are shared.
In this paper, I would like to explore how one may
start thinking about these issues by proposing a new
information-processing model of learning of sequential
material in choice reaction settings. The model is based
on the simple recurrent network (SRN) connectionist
architecture first proposed by Elman (1990), and
subsequently applied to implicit learning phenomena
by Cleeremans and McClelland (1991). By contrast
with the SRN and other models of sequence
processing, this model uses different sources of
knowledge to produce its responses. Thus, it
instantiates the third theoretical position described
above. To start exploring how well this kind of model
is able to account for relevant sequence-learning data, I
compared its performance with that of human subjects
in three experiments conducted by Curran and Keele
(1993). In the next section, I describe these
experiments and provide an empirical context for the
simulation work described in the rest of this article.
The Curran and Keele Studies
Curran and Keele (1993) conducted four experiments
that explore how implicit and explicit learning interact
in a sequence-learning task. For lack of space, and
because Experiment 4 is somewhat different from the
others, I will not discuss it in this paper. In the first
three experiments, subjects were exposed to a four-
choice reaction time task divided in blocks of 120 trials
each. Curran and Keele manipulated three factors.
First, the material could either be random or sequential.
When sequential, the targets movement followed a
repeating sequence of length six (e.g., 1-2-3-2-4-3).
Positive differences between RTs elicited by random
blocks and RTs elicited by sequential blocks would
indicate that subjects are learning about the sequence.
Second, an attention-demanding secondary task could
either be present or absent. When present, either a low-
pitched or a high-pitched tone appeared between any
two RT trials. Subjects were to count the number of
high-pitched tones and report their count at the end of
the block. Third, subjects could either receive typical
implicit learning instructions (incidental subjects), or
could be told that the material would sometimes follow
a sequence, and that knowing the sequence would be
helpful in carrying out the main RT task (intentional
subjects). These latter subjects were also given a
minute to study the actual sequence.
All three experiments started with 2 blocks of
practice on random material in dual-task conditions. In
Experiment 1 (see Figure 2, top panel), a group of
intentional subjects and a group of incidental subjects
were first exposed to 4 single-task, sequential blocks.
Next, they received one block of random material
followed by another sequential block, again in single-
task conditions. Learning was assessed by averaging
performance on the last two sequential blocks and by
subtracting this average from performance on the
intermediate random block. In a second, dual-task,
phase of the experiment, subjects were exposed to 2
blocks of random material, followed by one block of
sequential material and a final block of random
material. Learning in this second phase was again
assessed by computing the RT