ArticlePDF Available

Sequence learning under dual-task conditions: Alternatives to a resource-based account

Authors:

Abstract and Figures

In two experiments with the serial reaction-time task, participants were presented with deterministic or probabilistic sequences under single- or dual-task conditions. Experiment 1 showed that learning of a probabilistic structure was not impaired over a first session by performing a counting task, but that such an interference arose over a second session, when the knowledge was tested under single-task conditions. In contrast, the effects of the secondary task arose earlier for participants exposed to deterministic sequences. This difference between deterministic and probabilistic sequences disappeared in Experiment 2, where the counting task was performed on tones associated to the locations. Comparisons between sessions indicated that the secondary task affected not only the expression but also the acquisition of sequence learning, and that greater interference was observed in those conditions that yielded more explicit knowledge. These results suggest that the effects of a dual task on the measures of implicit sequence learning may be partly due to the intrusion of explicit knowledge and partly due to the disruption of the sequence produced by the inclusion of random events.
Content may be subject to copyright.
ORIGINAL ARTICLE
Luis Jime
´
nez Æ Gustavo A. Va
´
zquez
Sequence learning under dual-task conditions:
alternatives to a resource-based account
Received: 18 October 2004 / Accepted: 26 November 2004 / Published online: 26 April 2005
Springer-Verlag 2005
Abstract In two experiments with the serial reaction-
time task, participants were presented with deterministic
or probabilistic sequences under single- or dual-task
conditions. Experiment 1 showed that lea rning of a
probabilistic structure was not impaired over a first
session by performing a counting task, but that such an
interference arose over a second session, when the
knowledge was tested under single-task conditions. In
contrast, the effects of the secondary task arose earlier
for participants exposed to deterministic sequences. This
difference between deterministic and probabilistic se-
quences disappeared in Experiment 2, where the count-
ing task was performed on tones associated to the
locations. Comparisons between sessions indicated that
the secondary task affected not only the expression but
also the acquisition of sequence learning, and that
greater interferen ce was observed in those conditions
that yielded more explicit knowledge. These results
suggest that the effects of a dual task on the measures of
implicit sequence learning may be partly due to the
intrusion of explicit knowledge and partly due to the
disruption of the sequence produced by the inclusion of
random events.
Introduction
Learning sequential skills such as those necessary to
typewrite a text, to use a joystick or to play a musical
instrument require learners to perform the same task
repeatedly, while paying selective attention to the indi-
vidual components of the sequence. In addition to
coping with these task demands, it is not clear whether
this kind of learning can be aided by performing any
other resource-demanding operations, such as trying to
commit a relevant set of movements to memory, or
testing different hypotheses about which responses are
likely to be required after a given series. Such strategies
can surely benefit the generation of explicit knowledg e
(Curran & Keele, 1993; Reed & Johnson, 1994), but the
question remains as to whether some analogous re-
source-consuming operations can be considered neces-
sary to produce implicit learning as well. In the present
study we aim to analyze whether dual-task procedures,
which are believed to reduce these central attentional
resources, impair not only explicit but also implicit
learning.
In the laboratory, this learning has been modeled
through differ ent implicit learning paradigms, in which
participants are typically not instructed to learn, but are
simply presented with an orienting task that requires
them to process and to respond to each relevant stimulus
(see Cleeremans, Destrebecqz, & Boyer, 1998, for a re-
view of these tasks). Under these circumstances, learners
appear to acquire more knowledge about the underlying
structures than they can consciously convey, and they
have been observed to bring forth this knowledge even
when they are not instructed to use it, or when the task
demands explicitly require them to avoid relying on it
(e.g., Destrebecqz & Cleeremans, 2001; Jime
´
nez, Me
´
n-
dez, & Cleeremans, 1996; but see Shanks, Wilkinson, &
Channon, 2003; Wilkinson & Shanks, 2004). This kind
of incidental knowledge, which is difficult to express
(Dienes & Berry, 1997) and difficult to control (Cleere-
mans & Jime
´
nez, 2002), has been considered by some
authors as a side effect of information processing that
would arise immediately from any interaction with a
structured environment (Barsalou, 1995; Cleeremans &
Jime
´
nez, 2002; Jime
´
nez, 2003; see also Frensch, 1998;
Keele, Ivry, Mayr, Hazeltine, & Heuer, 2003). However,
some evidence appears to contradict this view by
showing that the addition of a secondary task that
makes extensive demands on central attentiona l
L. Jime
´
nez (&) Æ G. A. Va
´
zquez
Universidad de Santiago,
Facultad de Psicologı
´
a,
Santiago, 15782, Spain
E-mail: jimenez@usc.es
Psychological Research (2005) 69: 352–368
DOI 10.1007/s00426-004-0210-9
resources may have an adverse effect on this learning
(e.g., Nissen & Bullemer, 1987; Shanks & Channon,
2002; see also Shanks, 2003). The main goal of this paper
is to review the evidence, to discuss the alternatives
proposed so far to account for these results, and to re-
port on two experiments that may shed some new light
on this controversial issue.
The paradigm of sequence learning has become
dominant within the last few years in the area of implicit
learning, in part because it is particularly well-suited to
explore these issues. In this paradigm, participants are
presented with a serial reaction-time (SRT) task, in
which they are told to respond as fast and as accurately
as possible to the location of a stimulus that appears in
each trial in one of several possible locations on a
computer screen. The series of locations follows a reg-
ularity that is often repeated over many cycles. Partici-
pants are usually not informed about the existence of
this pattern, but learning is inferred indirectly, by com-
paring responses to structured and control trials. The
question as to whether this learning can be achieved and
expressed without awareness has proved to be very dif-
ficult to settle, but the conclusion that this can be the
case is supported by results showing that at least part of
this knowledge escapes deliberate control, can be more
clearly expressed through indirect measures (Jime
´
nez
et al., 1996), and remai ns effective even under conditions
in which learners are instructed to try to avoid its
influence (Destrebecqz & Cleeremans, 2001, 2003; but
see Wilkinson & Shanks, 2004).
In addition to the questio n as to whether sequence
learning can proceed without consciousness, a second
source of dispute has been concerned with whether this
sequence learning depends on the availability of a cer-
tain amount of attentional resources beyo nd those re-
quired to perform the SRT task. This topic was first
addressed by Nissen and Bullemer (1987), who con-
cluded that this learning could be taken as implicit, but
still required the allocation of central attentional re-
sources, as it was prevented when participants were
trained under dual-task conditions.
Since this first report, dozens of different studies have
turned to this issue, either presenting new evidence or
arguing for new interpretations of the same basic pattern
of results (see Hsiao & Reber, 1998, for a review). Some
studies reported that sequence learning can actually be
obtained under dual-task conditions, but that learning is
typically reduced under distraction, either in the scope of
the associations (Cohen, Ivry, & Keele, 1990), or in the
strength of the effects (Frensch, Buchner, & Lin, 1994;
Reed & Johnson, 1994). Other authors have approached
the issue from a more conceptual standpoint, arguing for
several alternative accounts that avoid the implication
that this learning, despite its seemingly automatic nat-
ure, makes extensive demands on central attentional
resources.
These alternative accounts may be grouped into three
main categories. First, some authors have suggested that
distraction does selectively suppress the expression of
learning, rather than its acquisit ion during the SRT task
(Frensch, Lin, & Buchner, 1998; Frensch, Wenke, &
Ru
¨
nger, 1999). A second group of researchers has pro-
posed that the differences in learning between single-task
and dual-task conditions may be attributed to the in-
trusion of explicit effects on the measures of perfor-
mance that are usually believed to reflect exclusively
implicit knowledge (Cleeremans & Jime
´
nez, 1998; Cur-
ran & Keele, 1993; Jime
´
nez & Me
´
ndez, 1999). Finally,
a third alternative sustains that an interference may arise
because of the disruption of the original sequence that is
brought about by the processing of random stimuli in
the conditions of the secondary task (Rah, Reber, &
Hsiao, 2000; Schmidtke & Heuer, 1997; Stadler, 1995,
see also Frensch & Miner, 1994; Willingham, Greenberg,
& Thomas, 1997). In the following paragraphs we con-
sider the evidence that sustains each of these three al-
ternatives in more detail.
As for the suppression hypothesis, it received support
from a number of studies by Frensch and colleagues
(Frensch, 1998; Frensch et al., 1998, 1999, but see also
Curran & Keele, 1993; Shanks & Channon, 2002). In
these studies, participants were trained under different
conditions that might or might not include a secondary
counting task, but learning was independently tested
both in the presence and in the absence of this secondary
task. The overall pattern of results indicated that the
differences between training conditions arose specifically
when the secondary task was extended over testing, and
not when all participan ts were tested uniformly without
distraction. In accordance with these results, Frensch
and his co-workers claimed that dual-task conditions
might be suppressing the expression of implicit learning,
rather than its acquisition in the course of the SRT task.
A number of studies have cast some doubts over the
sufficiency of this hypothesis to account for all the dual-
task effects observed in this paradigm. On the one hand,
the results of the study on which Frensch et al. (1999)
based their own design (Curran & Keele, 1993, Experi-
ment 3) indicated that learning was smaller under dual-
task conditions than under single-task conditions, even
when all participants were tested under homogeneous,
single-task conditions. Moreover, the results reported in
Frensch et al. (1998) have been challenged by Shanks
and Channon (2002), who pointed out three methodo-
logical shortcomings that might have produced an arti-
factual decrease in the differences observed between
single-task and dual-task performance. When these po-
tential problems were dealt with in a series of new
experiments, the authors observed that the measures of
learning were affected by the training conditions, and
that it was so regardless of whether testing took place
under single-task or dual-task tests. Overall, these results
indicate that, in addition to any possible suppression
effect, the secondary task can also affect learning to
some extent. What remains to be tested is whether such a
decrease in learning should be due to its implicit com-
ponent, and whether it needs to be attributed to the
scarcity of attentional resources.
353
The explicit intrusion hypothesis maintains that some
explicit knowledge may arise during the SRT task, and
that these explicit effects could be larger when partici-
pants perform without distraction, thus being poten-
tially responsible for the difference observed between
single-task and dual-task conditions. This hypothesis
has been indirectly sustained in studies that show that,
when the sequential structure is made more complex by
including probabilistic noise, the interferen ce provoked
by a secondary task is usually smaller than that obtained
with simpler, deterministic sequences. Indeed, the effect
of distraction has been shown to completely disappear
under certain circumstances by using probabilistic
structures (e.g., Cleeremans & Jime
´
nez, 1998; Schvane-
veldt & Gomez, 1998; Jime
´
nez & Me
´
ndez, 1999, 2001).
Cleeremans and Jime
´
nez (1998), for instance, showed
that the effect of a dual task was greater when the
structures were deterministic, and that the transfer from
single-task to dual-task conditions was specifically
harmful for participants trained with deterministic se-
quences. This is exactly what could be expected if par-
ticipants trained with deterministic sequences were
relying on some explicit knowledge, and if this knowl-
edge were more easily acquired and expressed under
single-task conditions.
Schvaneveldt and Gomez (1998) have found evidence
consistent with this hypothesis by using only probabil-
istic structures. In their Experiments 2 and 3, they gen-
erated a sequence of locations by choosing 90% of the
trials according to a second-order conditional (SOC)
sequence (see Reed & Johnson, 1994), and by using a
different SOC sequence to generate the remaining 10%
of the trials. Differential improvements in RTs to
probable and improbable transitions showed that all
participants learned about the sequence, and the absence
of a significant interaction between learning and dis-
traction suggested that learning was not affected by the
dual task. The an alysis of errors, however, was not en-
tirely consistent with this view, as it showed that par-
ticipants under single-task conditions made more
anticipation errors than did those presented with the
secondary task. This result may be open to different
interpretations. However, it can be accommodated by
the explicit intrusion hypothesis by assuming that the
use of probabilistic sequences may reduce the acquisi-
tion of explicit knowledge, but that awareness of some
chunks may still selectively arise under single-task con-
ditions, and that it may be responsible for the increase in
the number of anticipation errors.
If this reasoning is correct, then more complex
probabilistic structures should be arranged as a way to
further control for the absence of these expl icit effects.
Jime
´
nez and M e
´
ndez (1999, 2001) used a finite-state
grammar to generate a complex structure in which the
average conditional probability for probable successors
was as low as .49, whereas the corresponding average for
the improbable successors amounted to .15. To allow
participants to learn about this complex structure, the
SRT task was extended over several sessions. The task
required participants to respond to the location of dif-
ferent stimuli that appeared in each trial in one of four
possible locations and, in addition to the sequence of
locations, the shapes of these stimuli also carried some
predictive information about the next location. Dual-
task participants were required to keep a running count
of the number of trials in which some target shapes
appeared within each block, whereas single-task partic-
ipants were not told to do anything about the shape s.
The res ults indicated that performing both tasks simul-
taneously did not produce a reduction in sequence
learning relative to that found under single-task condi-
tions. Moreover, participants presented with the sec-
ondary task also learned about the predictive
contingencies established between shapes and locations,
whereas those who performed the SRT task alone did
not develop any sensitivity to this covariation. Overa ll,
these results were interpreted as showing the resilience of
implicit sequence learning not only with regard to task
load, but also to the presence of a potentially competing
source of predicti ve learning.
According to the intrusion hypothesis, the absence of
dual-task effects observed in these studies can be inter-
preted by assuming that explicit learning might have
been thoroughly controlled in this case and that, in the
absence of such explicit learning, the mechanisms of
implicit acquisition are seen to be independent of task
load. Alternatively, it can be argued that shape counting
may have become automated after being performed over
several thousand trials, and that it may have reached a
point at which it would no longer impose significant
attentional demands (cf. Shanks, 2003; Shanks, Row-
land, & Ranger, this issue). We will discuss this argu-
ment more extensively later in the article, but at this
point it is necessary to describe another possible account
of these results, which has to do with the disruption
hypothesis briefly presented above as a potential account
of dual-task effects.
According to the disruption hypothesis, dual-task ef-
fects can be attributed to the inclusion of random stimuli
that cannot be successfully ignored by those participants
assigned to a dual-task condition and that introduce
noise in an otherwise structured sequence. This account
would predict that interference arises selectively when
participants need to pay attention to a set of random
stimuli, but not when the secondary-task stimuli are
systematically associated with the primary events. In the
experiments reported by Jime
´
nez and Me
´
ndez (1999,
2001), given that each shape was related to a different
location, these stimuli would not be expected to interfere
with the learning of the sequence, but they would con-
stitute further predictive events.
Schmidtke and Heuer (1997) raised basically the same
argument to account for the results of some experiments
in which they found that learning about a sequence of
locations was not hindered by the inclusion of a sec-
ondary task when this task was performed on a series of
tones that followed a sequence that could be easily
integrated with the sequence of locations. Participants in
354
this experiment were not required to perform a counting
task, but instead to give punctual responses to some
tones. The authors interpreted thei r results by arguing
that the learners would automa tically tend to integrate
the two streams of stimuli into a single sequence, and
that the effect of interference would arise selectively
when the secondary-task stimuli could not be integrated
with the main sequence. According to this ‘‘task inte-
gration hypothesi s,’’ Schmidtke and Heuer observed
that the removal of the secondary task actually pro-
duced a decrease, instead of an increase, in the observed
effects of sequence learning, when participants had been
trained with two easily integrable sequences.
Although task integration may occur under certain
circumstances, and it may contribute to alleviating dual-
task interference, this process does not appear to ac-
count for all the cases in which sequence learning is
observed to be free from interference in a dual-task
setting. For instance, in Jime
´
nez and Me
´
ndez (1999), the
use of structured shapes as secondary-t ask stimuli did
not prod uce any interference with learning, nor did the
removal of the secondary task produce a disruption in
the observed effect, as would be predicted if both se-
quences had become integrated. Moreover, Schmidtke
and Heuer (1997) also found that the use of structured
tones decreased the interference from dual-task condi-
tions even when this structure was designed to be diffi-
cult to integrate with the sequence of locations (see their
Fig. 1). Therefore, it appears that the use of systematic
stimuli for the secondary task may alleviate the effects of
distraction, even when these two structures do not be-
come integrated into a single sequence.
Other possible accounts for the disruption effects
have pointed to a temporal disorganization of the se-
quence. The addition of the secondary task can affect the
temporal parameters of the sequence either because
updating the counter in some trials may introduce
inconsistency in the response-to-stimulus intervals
(RSIs; Stadler, 1995; Willingham et al., 1997) or because
it increases the overall inter-stimulus intervals, thereby
making it less likely that the represent ations of succes-
sive trials could remain activated in working memory
(Frensch & Miner, 1994). However, neither of these
accounts can easily explain why the effects of distraction
disappear when the secondary-task stimuli are struc-
tured, and thus the question remains as to whether some
combination of these hypotheses accounts for the whole
set of available data or whether, on the contrary, it could
still be reasonable to sustain that implicit sequence
learning depends on the resources subtracted by per-
forming these secondary tasks. The following experi-
ments were designed to investigate these issues further.
Overview of the experiments
This series of experiments initially focused on the ex-
plicit intrusion hypothesis. We sought to compare the
effects of dual task on the learni ng of two otherwise
similar deterministic and probabilistic structures, as well
as to show whether these two types of structures pro-
duced different amounts of explicit knowledge.
To minimize the likelihood that the secondary task
could become automated, as suggested by Shanks (2003)
and Shanks et al. (this issue), we decided to provide far
less training than that provided in Jime
´
nez and Me
´
ndez
(1999). Hence, probabilistic structures were simplified
with respect to those used in previous experiments. They
were generated on the basis of two SOC sequences, by
selecting 80% of the trials according to the training se-
quence, and generating the remaining 20% according to
a control sequence. The substitution process was per-
formed on a serial basis, by randomly interspersing two
repetitions of the control sequence with eight repetitions
of the training sequence.
We assessed conscious knowledge through a cued
generation task similar to that recently used by Wil-
kinson and Shanks (2004, Experiment 3). In using this
task, we do not endorse the assumption that all knowl-
edge expressed through this measure is necessarily con-
scious, but only that such a measure gets closer than any
of its common alternatives (namely, those taken from
recognition and free generation tasks) to fulfil the
information and sensitivity criteria put forward by
Shanks and St. John (1994). In this cued generation task,
participants were presented with a number of standard
SRT trials and, after those trials, they were prompted to
generate the most likely successor of the series. Thus, the
measure was directly assessing the same discrimination
between predictable and unpredictable successors that
could be held responsible for the indirect effect observed
through the SRT task. We assume that responding to
this generation task may be influenced by both conscious
and unconscious knowledge, but also that any infor-
mation that is expressed through the SRT task and that
is not manifested through this similar direct measure,
could most likely be taken as relying on non conscious
knowledge (for a complete articulation of this assump-
tion, see Jime
´
nez et al., 1996; Reingold & Merikle, 1988).
In addition to the explicit intrusion hypothesi s, we
were also interested in testing both the suppression and
the disruption hypotheses. The suppression hypothesis
was tested in both experiments, by arranging a second
training session containing a test block in which all
participants were tested under comparable, single-task
conditions. Finally, the disruption hypothesis was tested
over Experiment 2, by replacing the use of random tones
with tones that were consistent with the series of loca-
tions, and that were presented simultaneously with the
visual stimuli.
Experiment 1
The aim of this experiment is to try to reconcile the
results obtained by Jime
´
nez and Me
´
ndez (1999)with
those reported by Shanks and Channon (2002). In the
former study, the authors observed that performing
355
a secondary task produced no interference with learning
about a complex probabilistic structure . At variance
with these results, Shanks and Channon reported that
performing a similar counting task interfered with
learning about a much simpler, deterministic sequence.
Although there are many other differences between these
two designs, the most obvious of them is concerned with
their sequential structure. Hence, our first purpose was
to replicate Shanks and Channon’s results, and to test
whether a similar pattern of interference may arise when
a probabilistic, but otherwise similar, structure was ar-
ranged in this procedure. We designed four experimental
conditions by crossing the factors of task load (single-
task versus dual-task training) and sequential structure
(probabilistic versus deterministic).
Methods
Participants
Eighty-two students from the University of Santiago
were recruited to participate in the experiment. Partici-
pants were randomly assigned to one of the four con-
ditions formed by crossing the factors of task load and
sequential structure. After dis carding 10 participants
who produced counting errors exceeding the criterion of
10% (Shanks & Channon, 2002), 72 participants re-
mained, 18 of them assigned to each condition. Partici-
pants were paid a minimum of 6 Euros for participating,
and they received additional incentives in terms of their
performance (they earned an average of 10.40 Euros).
Apparatus
The experiment was run on PCs with 33-cm color
monitors and standard QWERTY keyboards, and was
programmed and run on INQUISIT 1.31 software. Four
white boxes (3.1 cm wide · 2.1 cm high) were arranged
horizontally at the bottom of the computer screen
against a green background. In each trial, a black dot
(2 mm in diameter) appeared in the center of one of
these boxes, and it disappeared when the participant
pressed the appropriate key. The response keys were
those corresponding to the letters V, B, N, and M, lo-
cated across the bottom of the keyboard. These keys
were assigned to each location by following a consistent
spatial mapping (i.e., the leftmost location required
pressing the V key, and so on). Participants pressed these
keys by using the middle and index fingers of both
hands. Two tones of 1,000 and 2,000 Hz were generated
by the computer, and were administered through LAB-
TEC headphones when prescribed by the design.
Materials
Two SOC sequences of 12 locations were used to gen-
erate both training and control trials. If we refer to
target locations from left to right as 1–4, the sequences
were as follows: SOCa: 1-2-1-4 -3-2-4-1-3-4-2-3, and
SOCb: 3-2-3-4-1-2-4-3-1-4-2-1. As can be observed, in
both sequences each location appear s with the same
likelihood, and each first-order transition (with the
exception of repetitions, which are forbidden) is also
equally likely. In addition, the second-order conditionals
include a minimum of reversals (i.e., one per sequence,
as in 1-2-1), and they are maximally discriminative be-
tween sequences, so that the successor of any given
context is always different between SOCa and SOCb.
Finally, the sequences were designed to avoid highly
salient runs, such as 1-2-3-4. For this reason we decided
to use SOCs different from those used by Shanks and
Channon (2002). As in their SOCs, however, our se-
quences were also structurally identical to each other,
and were related by a simple transformation 1 M 3.
Procedure
The experiment was divided in two sessions, performed
on two consecutive days. In session 1, participants were
randomly assigned to one of the four experimental
conditions, and were provided with instructions appro-
priate to their condition. They were told that the goal of
the experiment was to analyze the effect of practice on
relatively simple tasks, and were informed about the
structure of the experiment as well as about how to place
their fingers on the keyboard. Participants were also
informed about the presence of a tone that would appear
during the RSI intervals, and were required to have the
headphones on throughout the experiment. Participants
in single-task conditions were told that they didn’ t need
to do anything concerning the tones, whereas those as-
signed to dual-task conditions were instructed to keep a
count of the number of high-pitched tones that appeared
during each block, and to report on it at the end of the
block. After these instructions, and before proceeding to
the experimental phase, they were presented with a pre-
training block of the SRT task, composed of 40 random
and unrecorded trials. Then, they performed 15 con-
secutive blocks of 120 trials of the SRT task. Session 2
started by repeating the instructions, and it included 11
additional blocks of the SRT task, which was followed
by a generation task, to be described below.
SRT task During each trial of the SRT task, a dot
appeared at the cen ter of one of the four boxes, and it
remained present up to the moment in which the par-
ticipant pressed the key corresponding to its location. If
the answer was not correct, an error was registered, and
the participant had to try it again. One hundred milli-
seconds after the correct response, a random tone of
1,000 or 2,000 Hz was presented through the head-
phones for another 100 ms, and then the next visual
stimulus immediately followed. At the end of each block,
participants in dual-task conditions were prompted to
report on the number of target tones present ed over this
block, and were given fee dback about the correct num-
356
ber. They were also instructed to copy these two values
into a register sheet, and were made aware that an
average error of more than five occurrences per block
would force the experimenter to invalidate their data. All
participants were informed at the end of each block
about the accuracy and mean latency of their responding
over this block, and were told to keep track of these
measures by copying them into the register sheet. They
were informed that their earnings would depend on these
measures of performance. After registering their scores
for each block, and before starting a new one, they were
allowed a discretionary rest period.
SOCa and SOCb were used either as training or as
control sequences for half of the participants assigned to
each condition. Each training block began by randomly
selecting two non-repeated locations for the first two
trials, and it continued by following the rules corre-
sponding to their training SOC. For participants as-
signed to a deterministic condition, each block started at
a random point in the sequence, and continued by
repeating the training sequence 10 times, over the fol-
lowing 120 trials. For participants assigned to a proba-
bilistic condition, two series of 12 trials generated
according to the control sequenc e were interspersed with
eight series of 12 trials generated according to the
training sequence. The first series of each block was
forced to follow the training sequence. From here on,
the next series could follow the control SOC with a
probability of .20. If no control series had been ran-
domly chosen by the program over the first four series,
the fifth one was forced to present the control sequence.
Likewise, if the control series had appeared only once
after nine series, the tenth ser ies was forced to follow the
control SOC. In all cases, a new series started with the
location that was consistent with the tw o final items of
the previous series.
These constraints were maintained over the whole
SRT task, with the exception of two test blocks
(block 13 from session 1 and block 8 from session 2). In
these two blocks all participants were presented with a
probabilistic structure, but it was inverted with respect
to that shown during training for the probabilistic con-
dition. Hence, 80% of the trials were generated
according to the control sequence, and only 20% were
generated according to the training sequence. These test
blocks allowed us to obtain a conventional, inter-block
measure of learning, by comparing responding to con-
trol trials within these test blocks with responding to the
training trials from its neighboring blocks. This
arrangement also allowed us to obtain within-block
measures of learning by comparing responding to
training and control trials within the same block.
However, in this report we will restrict the analyses to
the inter-block comparisons, since these measures have
been more widely used in the field, and they have been
most frequently associated with the dual-task effects
(e.g., Shanks & Channon, 2002).
The main purpose of including a second training
session was to arrange a second test block that may al-
low us to assess learning under equivalent conditions for
all groups. However, we also wanted to present the cued
generation task in a context in which participants would
have had recent practice with the SRT task, but not so
much continuous practice that fatigue could prevent
them coping with the task demands. We arranged a re-
freshing period of six training blocks, followed by three
more blocks (7– 9) in which the secondary task was re-
moved for all particip ants, and two more training blocks
(10 and 11) in which the original conditions were rein-
stated. Block 8 was arranged as a test block and, after
block 11, participants were presented with the cued
generation task.
Cued generation task The design of this task was sim-
ilar to that devised by Wilkinson an d Shanks (2004,
Experiment 3). Participants were told to respond to
series of six stimuli. The first five trials of each series
were regular SRT trials, with the exception that tones
were no longer presented. The sixth trial was a genera-
tion trial, in which participants had to generate the most
likely location according to the sequence observed dur-
ing the previous trials. Unlike W ilkinson and Shanks, we
avoided the risk of providing information about the
target sequence during these trials by presenting not only
the 12 contexts that can be generated from the training
sequence, but also the 12 contexts that can be built out
of the control sequence. Moreover, we randomly pre-
sented each series three times, thus obtaining a measure
of consistence for each generation response. Finally, in
order to have a clear random baseline, and given that we
wanted to ascertain whether participants were able to
discriminate between training and control successors of
each series, we restricted the possible generation re-
sponses to just two candidates, corresponding to the
successors that followed the two final trials of the series
according to either the training or the control sequence.
Thus, the generation task was performed over 72
series, each composed of five regular SRT trials plus one
generation trial. These 72 series corresponded to three
repetitions of each of the 24 five-trial contexts that can
be generated by following either the training or the
control sequence. Each one of these 24 contexts ap-
peared randomly in each round. In the sixth trial of each
series, participants were required to generate the most
frequent successor of the series by choosing between two
candidates marked with question marks that corre-
sponded to the possible successors of the two latter tri-
als, according to the training or the control sequences.
No feedback was provided regarding the accuracy of
these generati on responses, and no speed pressure was
exerted in these trials.
Results
We will present the results obtained during the SRT and
generation tasks separately. The analysis of generation
performance was conducted exclusively on the responses
357
given to the contexts corresponding to the training se-
quence. The remaining trials had been included to avoid
this task to provi de information biased toward the
training sequence, and so they were disregarded. If par-
ticipants learned consciously about their training se-
quence, they would be expected to be able to discr iminate
on the appropriate successors of these training contexts.
As for the SRT task, the first two trials from each
block were discarded as buffer trials. Hit rates and mean
RTs for correct responses were computed over each
block, and separately for training and control trials
when this distinction was applicable. Given that the re-
sults obtained with hit rates showed analogous trends to
those observed through RTs, we will only report on the
analyses corresponding to RTs. Unless explicitly noted,
a significance level of .05 was used for all analyses.
We computed the mean RT in those trials that con-
formed to the most common sequence in each block,
disregarding those trials generated according to the
alternative structure. In most blocks, this common se-
quence was the training sequence, but in the test blocks
(i.e., block 13 from session 1 and block 8 from session 2)
it corresponded to the control sequence. Figure 1 shows
the mean RTs for these trials across blocks and over the
two sessions, plotted separately for probabilistic and
deterministic structures, and for single-task and dual-
task conditions. With these data, we conducted pre-
liminary analyses of performance over the training
blocks, and analyzed learning by comparing responding
to the control blocks with responding to their neigh-
boring training blocks. These analyses were conducted
both for session 1, where participants were tested under
their own training conditions, and for session 2, where all
participants were transferred to single-task conditions.
Session 1
Inspection of Fig. 1 shows that performance was slower
under the dual-task condition, that RTs decreased sig-
nificantly with practice, and that this decrease was larger
for dual-task conditions than for single-task groups. All
these observations were confirmed through a three-way
analysis of variance (ANOVA) with task load (dual vs.
single) and structure (probabilistic vs. deterministic) as
between-participants variables, and with practice
(blocks 1–12) as a within-participants variable. It re-
vealed significant effect s of task load, F(1, 68) = 100.31,
MSE = 14,747,569.31, and practice, F(11, 748) =
82.86, MSE = 137,016.89, as well as an interaction
practice · task load, F(11, 748) = 13.25, MSE =
21,906.311. There was no significant effect or interaction
involving structure.
To analyze sequence learning within this session, we
compared the mean RTs obtained in control trials over
the test block 13 with the mean RTs averaged over the
training trials from its neighboring blocks 12 and 14. A
three-way ANOVA with task load and structure as be-
tween-participants variables, and sequence (training vs.
control) as a within-participants variable, showed effects
of task load, F(1, 68) = 81.16, MSE = 1,601,552.27,
and sequence, F(1, 68) = 171.55, MSE = 90,925.95,
but not an effect of structure. The interactions
sequence · task load, F (1, 68) = 28.92, MSE
= 1,5326.66, sequence · structure, F(1, 68) = 22.86,
MSE = 12,115.15, and sequence · task load · struc-
ture, F (1, 68) = 14.92, MSE = 7,905.78 were all
significant. These interactions indicated that the effects
of sequence had been larger for single-task conditions
than for dual-task conditions (71 vs. 30 ms), and for
deterministic structures compared with probabilistic
structures (68 vs. 32 ms). The three-way interaction was
explored through follow-up analyses conducted sepa-
rately on both probabilistic and deterministic groups.
These analyses showed effects of seque nce in both con-
ditions, F(1, 34) = 40.68 MSE = 18,330.47, and
F(1, 34) = 138.98, MSE = 84,710.63, but, crucially,
the interaction sequence · task load was significant only
in the deterministic group, F(1, 34) = 37.12, MSE
= 22,623.91. Thus, performing the secondary task
significantly reduced learning when the sequence was
deterministic (124 to 33 ms), but not when it was
probabilistic (38 to 26 ms). The effect of 26 ms obtained
under probabilistic and dual-task conditions was
significant too, F(1, 17) = 7.83, MSE = 6,129.64, thus
showing that a manipulation that dramatically
decreased the effect of learning under deterministic
conditions did not show any influence on the weaker
effect obtained under probabilistic conditions.
Session 2
A three-way ANOVA conducted on the six training
blocks over session 2 with task load and structure as
between-participants variables, and with practice as a
within-participants variable, revealed effects of task
load, F(1, 68) = 87.72, MSE = 3,865,189.64, and
practice, F(5, 340) = 5.46, MSE = 2,826.90, but not
an effect of structure. There was a significant three-way
Fig. 1 Mean RTs across training for sessions 1 and 2 from
Experiment 1, plotted separately for each condition
358
interaction practice · task load · structure,
F(5, 340) = 2.68, MSE = 1,390 .71, which could be
interpreted as showing a different effect of practice over
the four conditions. Follow-up analyses conducted for
each combination of structure and task load showed
that only those participants presented with the easiest
condition (i.e., a deterministic sequence under single-
task conditions) showed persistent effects of practice
over this second session, F(5, 85) = 15.95, MSE
= 3,056.30.
As for the sequence learning manifested in this ses-
sion, it was analyzed by comparing responding to con-
trol trials within the test block 8 with the averaged
responding to training tri als over its neighboring
blocks 7 and 9. Throughout these three blocks, all par-
ticipants performed under single-task conditions,
regardless of their training condition. The corresponding
ANOVA showed significant effects of task load,
F(1, 68) = 18.86, MSE = 82,964.30, and of sequence,
F(1, 68) = 502.85, MSE = 159,511.83, but no effect of
structure. The interaction sequence · task load,
F(1, 68) = 88.47, MSE = 28,065.54, showed that the
effects of sequence were larger for those participants
who were trained under a single-task condition com-
pared with those trained with a secondary task (94 vs.
39 ms). The interaction sequence · structure ,
F(1, 68) = 81.90, MSE = 25,980.67, also showed that
the effects of the sequence were larger for deterministic
than for probabilistic structures (93 vs. 40 ms). Finally,
the three-way interaction sequence · task load · struc-
ture, F(1, 68) = 24.11, MSE = 7,649.54 was significant
too.
The meaning of this three-way interaction was ex-
plored through follow-up analyses conducted separately
on probabilistic and deterministic groups. These analy-
ses produced significant effects of sequence in both
conditions, F(1, 34) = 141.73, MSE = 28,370.60, and
F(1, 34) = 361 .82, MSE = 157,121.90. However, unlike
what occurred in the first session, the interaction se-
quence · task load was significant for both determinis-
tic, F(1, 34) = 74.83, MSE = 3,2509.79 and
probabilistic structures, F(1, 34) = 16.01, MSE =
3,205.29. Thus, although the impact of the secondary
task on sequence learning was larger for deterministic
structures, these analyses showed that the task load
conditions during training had affected learning of both
deterministic (136 to 51 ms) and probabilistic sequences
(53 to 26 ms).
Comparison of sessions
A comparison of conditions with regard to the trends
produced over these two tests allowed us to assess
whether the effects of distraction could be construed
either as an expression deficit or as an interference with
the acquisition of sequence knowledge. In general, the
differences between sessions in the effects of learning can
be attributed both to the changes produced over the test
conditions and to the general increase in practice.
According to the suppression hypothesis, if learning is
equivalent between conditions, and if transferring par-
ticipants from dual- to single-task tests can only con-
tribute to increasing the expression of learning, then a
larger improvement in the effect of learning would be
expected to arise when participants were transferred
from dual- to sin gle-task conditions than in the group in
which both measures were obtained under single-t ask
conditions.
To analyze this issue, we computed differential
learning scores by taking the mean RTs obtained in
those trials that conformed to the control sequence over
the test blocks, and subtracting from them the mean
RTs obtained over the training trials from their neigh-
boring blocks. A three-way ANOVA on these scores
with task load and structure as between-participants
variables, and with session as a within-part icipants
variable, showed the expected effects of task load, F(1,
68) = 73.46, MSE = 84,872.36, and structure, F(1, 68)
= 63.69, MSE = 73,578.76, as well as a significant
interaction between task load and structure, F(1, 68) =
26.93, MSE = 31,108.53. More to the point, there was
also a significant effect of session, F(1, 68) = 17.76,
MSE = 9,574.62, and a significant interaction of ses-
sion · structure, F(1, 68) = 4.85, MSE = 2,612.88. Th is
pattern indicated that learning scores did generally in-
crease between sessions (50–67 ms), and that the in-
crease was larger for deterministic sequences (69–93 ms)
rather than for probabilistic structures (32–40 ms).
Importantly, there was also a marginally significant
interaction session · task load, F(1, 68) = 3.55, MSE =
1,912.04, p=.06, but it did not support the prediction of
the suppression hypothesis. Instead, it indicated a
greater increase in learning between sessions for those
participants who had been uniformly trained and tested
under single-task conditions (71–94 ms), relative to that
found for learners who were trained with a secondary
task (30– 39 ms). Furthermore, the effect of session was
not even significant in this latter group, thus suggesting
that transferring participants from dual- to single-task
tests did not relieve any hidden influence of learning, nor
did it show any further increase over the learning effect
already observed over the first session.
Figure 2 represents the probability of hits produced
in the generation task over the 36 trials featuring train-
ing contexts , averaged over each experimental condition.
As can be readily observed by looking at the standard
error bars, all the scores were significantly larger than
those expected by chance (i.e., .50), thus showing that
the measure was sensitive to some learning about the
training structures. The effect of task load was larger for
those participants exposed to deterministic sequences,
but otherwise there appeared to be no overall effect of
structure. A two-way ANOVA conducted on these
scores with task load and structure as between-partici-
pants variables confirmed these impressions, revealing a
significant effect of task load, F(1, 68) = 14.55, MSE =
.15, and a significant interaction task load · structure,
359
F(1, 68) = 7.08, MSE = .07, but no significant effect of
structure. Two follow-up analyses conducted for each
structure showed that the effect of task load was sig-
nificant for deterministic sequences, F(1, 34) = 13.84,
MSE = .22, but not for probabilistic structures.
Discussion
The results obtained over this generation task are con-
sistent with the claims that:
1. All participants may have acquired some explicit
knowledge about the training sequence
2. The amount of explicit knowledge is smaller for those
participants who perform the SRT under distraction
3. The effect of distraction on the acquisition of explicit
knowledge appears to be larger when the sequence is
deterministic
As for the indirect measures of learning obtained
through the SRT task, they indicate that these measures
can be affected by distraction under some conditions,
and that the effects of distraction are larger when the
sequences are deterministic. Thus, the pattern of results
is generally consistent with the explicit intrusion
hypothesis, in showing that the effects of distraction are
specifically larger in those conditions in which the
resulting knowledge appears to be more explicit.
Alternatively, given that all the groups showed
above-chance generation scores, we may try to interpret
these results as showing that all the knowledge mani-
fested through the generation task is explicit, and thus
that it generally depend s on the availability of atten-
tional resources. However, this claim is difficult to rec-
oncile with the interaction observed in the generation
results between structure and task load. Plainly, the
evidence does not show that the dual task produced a
uniform interference over learning, but instea d it sug-
gests that participants trained with deterministic se-
quences were more affected by distraction, whereas
participants trained with probabilistic structures ex-
pressed roughly the same knowledge over the generation
task regardless of whether they had been trained in the
presence or absence of distraction.
The results obtained in the SRT task are also gener-
ally consistent with the claim that distraction affects
learning in a different way depending on the complexity
of the sequence. During the first session of training,
learning about a deterministic sequence was found to
depend heavily on the integrity of attentional resources,
whereas learning of a probabilistic structure remained
largely unaffected by dual-task conditions. Over the
second session, however, the results showed an effect of
dual task on the learning of both deterministic and
probabilistic sequenc es, but yet the interference was
smaller for the more complex, probabilistic structures
compared with that shown for the learning of deter-
ministic sequences.
This qualitative difference between the learning of
probabilistic and deterministic structures has been re-
ported before (Cleeremans & Jime
´
nez, 1998; see also
Jime
´
nez & Me
´
ndez, 1999, 2001) and was interpreted as
evidence in favor of the claim that distraction does se-
lectively affect the explicit components of learning.
However, the defende rs of a resource-based account of
sequence learning have proposed at least two different
lines of reasoning to account for such a difference (see
Shanks, 2003; Shanks et al., this issue).
First, they have noticed that prior evidence showing
that probabilistic sequence learning can proceed without
interference from a secondary task came from experi-
ments that provided too much training in the dual-task
conditions, thus allowing ‘‘that the secondary task may
have become automatic, therefore allowing resources to
be allocated to sequence learning’’ (Shanks et al., this
issue). Against this argument, the pattern of results ob-
served in this experiment shows that the absence of
interference can be obtained with in a single session, and
when the secondary task still requires resources, as at-
tested by the interference produced by the same dis-
tractor task over the learning of an otherwise similar,
but deterministic, sequence. Moreover, the results of this
experiment also show that a pattern of interference that
was absent during a first session can arise later on, over a
second session. A possible account for these later effects
may assume that those learners presented with a prob-
abilistic sequence may start by acquiring some implicit
knowledge over the first session and then, upon noticing
the beha vioral changes brought about by their implicit
learning, they may develop an intentional search for
contingencies that could be especially useful if they are
not engaged in performing a secondary task (cf.,
Frensch et al., 2003).
A second argument raised by the defenders of a re-
source-based view to account for the different effects of
distraction on the learning of deterministic and proba-
bilistic structures suggests that learning may result from
a single pool of resource-demanding operations, but that
these operations would require more resources as
learning grows stronger (Shanks, personal
Fig. 2 Proportion of correct responses to the generation task, and
standard error of each measure, represented separately for each
condition and for Experiments 1 and 2
360
communication, March, 2004). According to this view,
participants can learn deterministic sequences very
thoroughly if they are able to allocate enough resources,
whereas learning of probabilistic structures would be
hard to achieve even with full resources. A corollary of
this view is that, if the learning of a probabilis tic se-
quence is conceived as depending on precisely the same
mechanisms, then any manipulation that affects learning
of a deterministic sequence should also affect learning of
a probabilistic structure, even though the net effect may
be larger for the deterministic case. From such a unitary
framework, however, it would be difficult to account for
a pattern of interference that arises exclusively when the
learning is stronger, but that does not affect learning
when it becomes weaker. The results obtained over the
first session of this experiment, in which learning of a
deterministic sequence was significantly hindered by
distraction, whereas learning about a probabilistic
structure was produced in the same way regardless of the
presence versus absence of distraction, would strongly
indicate that two different learning mechanisms are at
work in this case, one depending on attentional re-
sources, and the other proceeding automatically as an
obligatory effect of responding to each trial in the con-
text of the SRT task.
The results obtained over the first session of this
experiment are difficult to reconcile with a resource-
based account of the effects of distraction, but those
obtained over the second session provide new challenges
to any simple account of these results. First, a compar-
ison of the learning effects observed over these two
sessions for each condition is difficult to reconcile with
the suppression hypothesis, in that it shows that par-
ticipants uniformly trained and tested under single-task
conditions improved more over sessions than did those
who were transferred from dual-task to single-task
conditions over the second test. Finally, although we
have been arguing that the general pattern of results is
most compatible with the explicit intrusion hypothesis,
there is also some evidence that can be taken as prob-
lematic to this view. For instance, during this second
session the measures of learning taken from RTs sug-
gested that the learning obtained in single-task and dual-
task conditions was different even for those groups ex-
posed to probabilistic structures. However, according to
the generation scores, the explicit knowledge gained at
the end of this session did not differ significantly between
these two groups. Thus, either the generation score was
not being sensitive to all the explicit knowledge acquired
by these two groups, or the explicit intrusion hypothesis
was not able to account for all the dual-task effects. To
further analyze this issue, we conducted a new analysis
of the indirect measures of learning over this second
session by selectively considering those parts of the
training sequence that had not been generated by each
individual participant above the level of chance.
As described above, our generation task contained
three repetitions of each of the 12 possible five-trial
contexts that could be made out of the training se-
quence, and each time participants were asked to gen-
erate the most likely successor of the series. Thus, we
had a consistency measure that could tell us whether
each learner was able to reliably discriminate on the
appropriate successor of each series. With this measure
in hand, we reasoned that if a learner failed to generate
the appropriate successor of a given context any more
often than its alternative location (if, in fact, he or she
generated the alternative successor more often than the
appropriate one), then we could be confident that this
learner was not conscious of this part of the sequence.
For each participant, we selected those trials corre-
sponding to series that were generated worse than pre-
dicted by chance (two participants from the single-task
deterministic group were excluded from the analysis
because they generated better than chance for all the
series), and re-evaluated the effect of learning as ex-
pressed for those specific trials over the second session of
the SRT task. Perhaps surprisingly, an ANOVA con-
ducted on those trials closely replicated the pattern ob-
tained with the whole data, thus showing significant
effects of task load, F(1, 66) = 12.31, MSE
= 58,431.77, and sequence, F(1, 66) = 199.43, MSE
= 105,298.34, as well as significant interactions se-
quence · task load, F(1, 66) = 35.47, MSE
= 18,726.31, sequence · structure, F (1, 66) = 35.33,
MSE = 18,657.33, and sequence · task load · struc-
ture, F(1, 66) = 6.28, MSE = 3,316.76. The analyses
conducted separately on probabilistic and deterministic
structures also showed that the effect of sequence was
observed in both cases, F(1, 34) = 75.57, MSE =
18,205.88, and F(1, 32) = 123.85, MSE = 103,174.97,
and that the interaction sequence · task load was also
significant for both types of structures, F(1, 34) =
13.44, MSE = 3,238.64, and F(1, 32) = 22.02, MSE
= 18,346.62.
These results have two very different implications.
First, they constitute a strong indication that some
learning was implicit in this paradigm, since they show
that sequenc e learning can be observed through an
indirect measure of performance, even for those frag-
ments that, in a comparable direct measure, received
scores that reflected null sensitivity to this sequential
structure. Second, the results also showed that the effects
of dual task could not be completely accounted for by
the intrus ion of explicit knowledge manifested through
the generation task, since dual-task effects are still evi-
dent after removing the fragments generated above
chance performance by each individual participant.
Up to this point, we have discussed the resul ts of this
experiment against a resource- based account of the ef-
fects of dual task, and against two of its main alterna-
tives, na mely those correspo nding to the suppression
and the explicit intrusion hypotheses. In the following
experiment, we will consider the third alternative, which
accounts for these effects in terms of the disruption
produced in the sequence by the inclusion of random
stimuli that need to be processed and responded to in
coping with the secondary task demands.
361
One of the main differences between Experiment 1
and some previous experiments that have found no effect
of distraction for probabilistic structures, even after
much training (Jime
´
nez & Me
´
ndez, 1999, 2001), has to
do with the random versus structured nature of the
secondary-task stimuli. In those previous experiments,
the secondary-task stimuli were systematically asso-
ciated with the sequence of loca tions, whereas in Ex-
periment 1 those stimuli were random. So, in
Experiment 2, we arranged a systematic relation be-
tween the secondary-task stimuli and the sequence of
location, to ascertain whether this relation may alleviate
the effects of interference found in Experiment 1.
Experiment 2
The goal of this experiment was to replicate Experi-
ment 1, but arranging the tones so that any disruption
caused by these secondary-task stimuli was minimized.
We did so by systematically associating each tone with a
pair of location s, and by presenting the tones simulta-
neously with the visual stimuli, instead of during the
RSI. In doing so, we came closer to the procedure de-
vised by Jime
´
nez and Me
´
ndez (1999, 2001), in which
participants performed the secondary counting task on a
dimension of the stimuli associated with that on which
the SRT task was being performed.
Methods
Participants
Seventy-nine students from the University of Santiago
participated in the experiment, and were assigned to one
of four conditions analogous to those described in
Experiment 1. After discarding seven participants who
made more than 10% tone counting errors, 72 partici-
pants remained, who were evenly assigned to each con-
dition. Participants were paid a minimum of 6 Euros for
participating, and received additional incentives
depending on their performance (they earned an average
of 10.47 Euros).
Procedure
Both the SRT task and the generation task were ar-
ranged to follow the procedure described for Experi-
ment 1, with the exceptions mentioned below. Instead of
presenting random tones during the RSI, the tones ap-
peared simultaneously with the visual stimuli, and each
tone was systematically associated with a pair of loca-
tions. The pairing was randomly selected for each par-
ticipant, and it was maintained throughout the training
so that, for instance, one participant could hear the
high-pitched tone associated with locations 1 and 3, and
the low-pitched tone associated with locations 2 or 4.
During the generation task the tones were no longer
presented. Given that such a pairing between tones and
locations brought about the undesired consequence that
the high-pitched tones appeared in exactly half of the
trials, we randomly varied the size of each block, so that
the number of tones could vary around a similar range
as it did over the previous experiments. Thus, each block
contained a random number of trials between 110 and
130.
It should be noted that the fact that each tone was
associated with a pair of locations did not entail that the
tones could bear any further predictive information
about the following locations, beyond that provided by
the locations with which they covariated. Indeed, given
that the identity of the tone was completely determined
by the location, participants in dual-task conditions
could as well have decided to stop paying attention to
the tones, and just count the number of trials in which
the corresponding locations occurred. Our intention was
to minimize the disruption provoked by the tones
without removing their presence nor the task load im-
posed by the counting task. Hence, participants und er
dual-task conditions were explicitly informed about the
fact that the high-pitched tones always appeared in trials
corresponding to the same pair of locations. If the
attentional resources invested in performing the count-
ing task were necessary to produce sequence learning, we
could expect to obtain the same pattern of results ob-
served in Experiment 1. On the contrary, if the inter-
ference was due to the disruption brought about in the
sequence by responding to a set of random stimuli
interposed between successive locations, then this
manipulation wou ld minimize the effects of the second-
ary task.
Results
Figure 3 represents the mean RTs across blocks and
over the two sessions, plotted separately for prob abilistic
and deterministic structures, and for single-task and
dual-task conditions. In the following paragraphs we
Fig. 3 Mean RTs across training for sessions 1 and 2 from
Experiment 2, plotted separately for each condition
362
highlight the main differences and similarities observed
between this and the previous experiment.
As in Experiment 1, the three-way ANOVA con-
ducted over the 12 practice blocks with task load and
structure as between-participants variables revealed
significant effects of task load, F(1, 68) = 18.11, MSE
= 3,603,050.23, and practice, F(11, 748) = 110.17,
MSE = 144,197.25, as well as an interaction practi-
ce · task load, F (11, 748) = 11.89, MSE = 15,561.77,
which indicated that RTs improved more over practice
under dual-task conditions. No effect or interaction
involving structure approached significance.
Session 1
To analyze sequence learning within session 1, we pro-
ceeded as in Experiment 1, comparing RTs in response
to control trials over the test block 13, with RTs aver-
aged over the training trials from blocks 12 and 14. A
three-way ANOVA with task load and structure as be-
tween-participants variables, and sequence (training
versus control) as a within-participants variable, showed
effects of task load, F(1, 68) = 15.02, MSE
=333,371.32, and sequence, F (1, 68) = 120.82, MSE
= 140,789.81, but not an effect of structure. The inter-
action sequence · structure was significant too,
F (1, 68) = 14.54, MSE = 16,949.39. Import antly,
however, no significant interaction involving task load
approached significance. Thus, these results showed that
participants responded faster in single-task conditions
(374 vs. 470 ms ), that they learned to respond faster to
training than to control trials (391 vs. 453 ms), and that
the effect of learning was larger for participants pre-
sented with deterministic sequences than for those
trained with probabilistic structures (the effects
amounted to 84 and 41 ms respectively). However, they
showed that the effects of task load did not affect
learning (63 vs. 62 ms), and this pattern did not change
by considering separately either de terministic or proba-
bilistic structures.
Session 2
A three-way ANOVA conducted on the six training
blocks included in session 2, with task load and struc-
ture as between-participants variables, and with practice
as a within-participants variable, revealed effects of task
load, F(1, 68) = 29.91, MSE = 1,078,542.23, and
practice, F(5, 340) = 12.50, MSE = 5,181.57, but no
effect of structure. As in Experiment 1, there was also a
significant triple interaction practice · task load ·
structure, F(5, 340) = 2.79, MSE = 1,158.85, indicat-
ing that practice produced larger improvemen ts in per-
formance for those participants presented with
deterministic sequences under single-task conditions.
As for the sequence learning manifested in this ses-
sion, it was analyzed just as in Experiment 1, by com-
paring responding to control trials over block 8 with
responding to training trials over blocks 7 and 9, in all
of which participants performed under single-task
conditions. The ANOVA conducted on the ave rage
RTs using sequence (control versus training) as a within-
participants variable, and structure and task load as
between-participants variables, showed significant
effects of task load, F(1, 68) = 7.30, MSE = 20,752.30,
as well as effects of sequence, F(1, 68) = 338.64, MSE
= 166,488.73, but no effect of structure. The interaction
sequence · task load, F(1, 68) = 34.92, MSE
= 17,165.12, showed that the effects of learning were
larger for participants trained under single-task condi-
tions than for those trained with a secondary task (90 vs.
46 ms). The interaction sequence · structure ,
F(1, 68) = 52.26, MSE = 28,641.81, also showed that
the effects of the sequence were larger for deterministic
than for probabilistic structures (97 vs. 40 ms). Finally,
the three-way interaction sequence · task load · struc-
ture did not reach significance (p > .10), thus indicating
that the impact of imposing a secondary task over the
learning of probabilistic structure (24 vs. 56 ms) was not
different from that produced over the learning of a
deterministic sequence (61 vs. 124 ms).
Comparison of sessions
A comparison of sessions was conducted by computing
differential learning scores, as described in Experiment 1,
and subjecting these scores to a three-way ANOV A with
task load and structure as between-participants vari-
ables, and with session as a within-participants variable.
This analysis showed significant main effects of task load,
F(1, 68) = 8.11, MSE = 18,433.71, and structure,
F(1, 68) = 39.45, MSE = 89,657.58, but no significant
effect of session. Importantly, the interaction between
structure and task load did not approach significance in
this analysis, and the only other significant interaction
was that between task load and session,
F (1, 68) = 15.31, MSE = 15,941.75. This interaction
showed that the effects of learning increased between
sessions for participants trained under single-task
conditions (63–90 ms, F(1, 34) = 17.84, MSE
= 12,651.83), whereas they actually decreased in abso-
lute terms (although nonsignificantly so) for participants
trained in the presence of a secondary task (62 vs. 46 ms).
Cued generation task
The average percentage of hits obtained for each con-
dition during the generation performance is represented
in Fig. 2, together with those obtained for the previous
experiment. As in Experiment 1, these results showed
that participants performed more accurately in the
generation task when they were trained with determin-
istic sequences and under single-task conditions. The
corresponding two-way ANOVA conducted on struc-
ture and task load confirmed this impression, by
showing significant main effects of task load,
363
F(1, 68) = 6.85, MSE = .087, and structure,
F (1, 68) = 19.03, MSE = .241. The interaction task
load · structure did not reach significance (p = .08),
but independent analyses conducted for each type of
structure showed that the effect of task load was sig-
nificant only for deterministic sequences (.75 vs. .63),
F(1, 34) = 9.89, MSE = .12, and not for those par-
ticipants who learned under probabilistic conditions (.58
vs. .56, F <1).
Discussion
As in Experiment 1, the results obtained over the gen-
eration task are consistent with the claims that all par-
ticipants may have acquired some explicit knowledge
about the training sequence, that the amount of explicit
knowledge was smaller for participants trained under
dual-task conditions, and that the effect of distraction
on the acquisition of this explicit knowledge was larger
when the sequences were deterministic. As for the indi-
rect measures expressed through the SRT task, they
showed different patterns between sessions. In session 1,
the results showed that any interference caused by
interposing a tone-counting task on the learning of the
sequence was completely overcome by using tones cor-
related with the successive locations. Importantly, the
absence of distraction effects was generalized for both
probabilistic and deterministic sequences. However, as
occurred in Experiment 1, the effect of distraction arose
latter, during the second session, and it was significant
for both types of structures in a test in which learning
was consistently measured under single-task conditions.
The results observed over the first session are com-
patible with the disruption hypothesi s, in that they show
that learning can proceed unaffected by performing a
secondary counting task, if the secondary-task stimuli
are associated with those on which the SRT task is being
carried out. However, the disruption hypothesis cannot
account for the results observed over the second session.
On the other hand, the comparison of sessions pro-
duced results that were again inconsistent with the
suppression hypothesis, by showing that participants
trained with dual-task conditions did not express any
more knowledge when the distractor task was removed
over the second test than they did over the first session
when they were tested under dual-task conditions. In
contrast, participants who had been consistently trained
and tested under single-task conditions produced a sig-
nificant increase in learning between sessions, due to the
additional blocks presented between these two tests.
As for the resource-based account of sequence
learning, the results are also difficult to interpret within
this view, since they show that sequence learning can
proceed independently from distraction precisely during
the first learning stages. The defenders of this view
(Shanks, 2003; Shanks et al., this issue) have suggested
that sequence learning could only be expected to be free
from dual-task interference after much training at
combining the two tasks, when performing this task may
become automated. However, and against this predic-
tion, the results of this experiment suggested that the
interference can work the other way roun d, so that
learning may proceed independently of attentional load
during the first stages, and distraction can start making a
difference later on.
This pattern of interference is difficult to interpret
according to any single view, but we surmise that it
could be understood by appealing to a blend of the ex-
plicit intrusion and the unexpected-event hypotheses
(Frensch et al., 2003). Thus, implicit learning may be
taken to proceed regardless of distraction at an early
stage, but then, as a consequence of this implicit learn-
ing, some learners may be able to notice that their re-
sponses are be coming smoother over training, thus
adopting an intentional search for contingencies that
could be especially useful when the sequence is deter-
ministic, and when training takes place under single-task
conditions. According to this account, the effect of dis-
traction can be seen as delaying the process by which
implicit knowledge is transformed into reportable
knowledge, and thus it would be expected that explicit
effects would tend to arise precisely during the later
phases of learning (cf. Cleeremans & Jime
´
nez, 2002).
General discussion
In this series of experiments, we set out to compare the
effects of dual-task conditions on the learning of two,
otherwise similar, deterministic and probabilistic se-
quences. Our empirical aim was to reconcile previous
evidence showing that learning about a deterministic
sequence can be hindered by the inclusion of a tone-
counting task (e.g., Shanks & Channon, 2002)with
results indicating that learning of a more complex,
probabilistic structure can proceed unaffected by a
largely comparable shape-counting task (Jime
´
nez &
Me
´
ndez, 1999, 2001). At a co nceptual level, our main
purpose was to analyze whether the observed patterns of
interference were consistent with a resource-ba sed
account of implicit sequence learning or whether, on the
contrary, they coul d be accounted for by a number of
alternative views, cast in terms of the suppression of
some acqu ired knowledge, the intrusion of explicit
knowledge in the measures of performance, or the
disruption produced by interspersing random stimuli on
an otherwise structured sequence.
In Experiment 1, participants were assigned to an
SRT task in which the series of locations followed either
a deterministic SOC sequence, or a probabilistic struc-
ture generated for eac h block by randomly combining
eight repetitions of the training sequence with two rep-
etitions of a control sequence. Half of the participants
performed the SRT task alone, whereas the other half
were told to perform a secondary counting task on the
number of high-pitched tones presented during the RSI
intervals. The results indicated that:
364
1. Sequence learning as assessed over the first session
was significant for both probabilistic and determin-
istic structures
2. Task load modulated the effect of learning selectively
for those participants exposed to deterministic se-
quences
3. Sequence learning as assessed over the second session
through a single-task test was also significant for all
conditions
4. The effect of task load modulated the effect of
learning for both types of structures, although its
impact was still larger for deterministic structures
5. The comparison of the measures taken over each of
these sessions indicates that transferring participants
from dual-task to single-task conditions did not re-
lease any learning effect that might have been sup-
pressed when it was assessed under dual-task
conditions
6. A cued generation task performed at the end of
training showed that the effect of task load was sig-
nificant only for those participants exposed for
deterministic sequences, whereas those trained with
probabilistic structures predicted the successor of
each series with about the same accuracy regardless
of their task load during training
Experiment 2 was a conceptual replication of the first
one, replacing the random tones employed in Experi-
ment 1 with tones that were presented simultaneously to
the primary stimuli, and that were systematically asso-
ciated with them. In this case, we also found that:
1. Sequence learning as tested over the first session was
significant for both structures
2. Task load did not modulate learning in any of these
conditions
3. When learning was tested over the second session
under single-task conditions the results indicate that
learning was affected by task load for both structures,
and that its effects were stronger for deterministic
sequences
4. Again, the comparison of sessions indicates that
transferring participants from dual-task to single-task
conditions did not provide any evidence in favor of
the suppression hypothesis
5. The measures of generation performance again
showed an advantage for single-task conditions only
for those participants who were trained with a
deterministic sequence
Overall, the comparison of Experiments 1 and 2 is
supportive for the disrupti on hypothesis, and thus it
adds to the previous studies that reported a lack of
interference from the dual task when this task was per-
formed on stimuli that were systematically related with
the primary task dimension (e.g., Jime
´
nez & Me
´
ndez,
1999, 2001; Schmidtke & Heuer, 1997 ). In the first ses-
sion of our Experiment 2, we found essentially the same
lack of interference for both deterministic and proba-
bilistic stru ctures by using ton e-counting as the sec-
ondary task, and by presenting the tones simu ltaneously
and associated with the primary stimuli. From this study
alone it is not possible to ascertain whether the lack of
interference observe d early in training for de terministic
sequences may be due either to the association between
tones and locations, or to the fact that they were
simultaneous. However, when considered together with
results such as those by Schmidtke and Heuer, it appears
that the integration hypothesis is the more plausible
account of this preserved learning.
The disruption hypot hesis, however, is qualified by
two complementary results obtained in this study. First,
even though participants in Experiment 1 were forced to
process random tones, this secondary task did not affect
sequence learning over the first session in the probabi-
listic condition, whereas this early interference was
found for participants trained with deterministic se-
quences. Hence, it appears that the disruption can pro-
duce less deleterious effects on the learning of complex
sequences. Second, in Experiment 2, even though the
tones were systematically associated with the primary
stimuli, and thus no disrupti on effect could be expected,
a significant effect of distraction arose later in training.
An account of these seemingly contradictory results
can be articulated on the cognitive and neural architec-
ture of sequence representation recently proposed by
Keele et al. (2003). According to this model, implicit
sequence learning could be based simultaneously on two
different learning modules. First, a dorsal module com-
prising parietal and supplementary motor cortex could
be devoted to learning unidimensional relations in an
implicit way, without being subject to any kind of
attentional constraint. Second, a ventral module
involving more occipital, temporal, and prefrontal areas
would be in charge of associating eve ry attended input
regardless of whether they belong to a single or multiple
dimensions. According to this view, this latter ventral
module will only be subject to selective attentional
constraints, but it would be responsible for providing the
basis on which, under certain circumstances, the repre-
sented knowledge could be upgraded to yield an expli cit
status. In this model, the process by which some of the
contents of the multidimensional module may end up
generating awareness has been left undefined, but it can
be described along the lines proposed by Frensch et al.
(2003) in their unex pected-event hypothesis.
From this general framework, the learning of a
deterministic series under single-task conditions can be
assumed to involve both modules simultaneo usly, thus
providing the best possible context for the emergence of
awareness. On the other hand, the inclusion of a sec-
ondary task performed on random stimuli would be
expected to hinder the activity of the ventral module
and, therefore, to prevent the acquisition of explicit
knowledge, thus producing a large effect of interference
on sequence learning. As for the learning of probabilistic
structures, this may also involve both the dorsal and
ventral modules, but the inclusion of some random
material within the sequential structure could make it
365
much more difficult for the representation developed in
the ventral module to sustain the generation of aware-
ness. Without these explicit effects, the contribution
provided by this multidimension al module may become
negligible, at least early in training, and thus the impact
of the secondary task may be importantly reduced.
In Experiment 2, when the secondary task was per-
formed on stimuli associated with the primary events,
the multidimensional module would be able to learn
about an informative series, but the secondary task
could still be assumed to hinder the emergence of explicit
knowledge, according to the frame work delineated by
the unexpected-event hypothesis. Under these circum-
stances, the disruption of the sequence would no longer
be an issue, and thus we may expect to obtain no effect
of the dual task early in training. However, if the
structures are deterministic, the emergence of some ex-
plicit knowledge may still be expected to arise later in
training, specifically in single-task conditions, and then
some effects of distraction may be observed at this point.
Alternatively, learning in the multidimensional module
may produce lower benefits when the sequence com-
prises an alternating series of locations and tones rather
than when it is composed exclusively of a set of loca-
tions, and such a difference may account for the effect
detected late in training for the probabilistic structures,
even in the absence of differences in generation scores.
As a whole, the integration suggested between the
model of implicit learning by Keele et al. (2003) and
the framework proposed by Frensch et al. (2003) on the
generation of awareness provide a principled description
of how the disruption and the explicit intrusion
hypotheses can be combined together to account for the
complex pattern of results observed in this study. This
integrated model provides a useful tool to understand
the results of many other studies conducted with several
variations of the dual-task procedure. For instance, in
this context it is possible to understand that Jime
´
nez and
Me
´
ndez (199 9) did not find any effect of interference,
even after several sessions of practice, when they used
secondary-task stimuli that were associated with the
primary stimuli, and when they arranged probabilistic
structures that were much more complex than those used
in the present study. Under these circumstances, no
disruption effect was expected to arise in dual-task
conditions, but neither was it likely that the learners
exposed to a single task would be able to obtain more
conscious knowledge than those presented with the
secondary task. Thus, both groups may be expected to
yield roughly equivalent levels of sequence learning.
From this standpoint, it is also possible to account
for all the studies that found that the secondary task
impaired sequence learning when the knowledge is
consciously accessible (e.g., Shanks et al., this issue). In
short, this findin g just confirms that both the acquisition
and the use of explicit learning depend on attentional
resources, but this claim has no bearing on the issue of
whether this dependence can be extended to implicit
sequence learning as well.
More specifically, our results and those reported by
Shanks et al. in this special issue show a number of
commonalities that are worth mentioning, especially
since our interpretations are so different. First , in both
series of experiments there are indicators suggesting that
the effects of distraction under probabilistic conditions
are only mild, at most, during the first training session.
For instance, in their Experiments 1 and 2 the effects of
learning did not differ between single-task and dual-task
conditions when they were assessed through a continu-
ous comparison of responding to probable and
responding to improbable events, and they only became
significantly different (or marginally so) when learning
was assessed through a transfer block in which learning
conditions were changed for those participan ts trained
with the secondary task. Thus, the evidence available to
sustain that the secondary task impaired probabilistic
sequence learning over the first learning stages cannot be
considered to be strong.
Moreover, in both studies the generation tasks ar-
ranged at the end of training were sensitive to some
learning, but each of us interpreted these results in a
rather different way. Shank s et al. (this issue) extracted
the general conclusion that sequence knowledge was
consciously accessible, and thus that there was no evi-
dence for this learning to be implicit. In contrast, we
only assumed that the measure provide d by the gener-
ation task was probably more sensitive to conscious
information than that provided by the RTs. Thus, other
things being equal, if some knowledge was revealed
exclusively through the SRT task, we proposed that we
could be reasonably confident that this knowledge was
implicit. By relying on this set of assumpt ions, we ana-
lyzed performance over the second session from Exper-
iment 1, and obtained evidence showing that the indirect
measures remained sensitive to sequence information
even on those fragmen ts that were not generated better
than chance. By considering for each learner all the trials
that had been generated below the level of chance, rather
than just those trials generated at chance, we made it
highly unlikely that a difference in RTs between
responding to probable and improbable trials could be
discarded as reflecting an artifact from an independent
distribution of random errors (cf. Shanks & Perruchet,
2002). Therefore, we consider the results of this analysis
to be strongly indicative of the contribution of implicit
learning to the measures of performance in the SRT
task.
Overall, the results of these experiments are relevant
to the aim of this special issue on the generation of
awareness, and they call for a very specific view on the
relation between the contents of learning and the con-
tents of consciousness. In contrast to the identity as-
sumption espou sed by Shanks et al. (this issue), we
assume that they may be engaged in a complex dynamic
relation where the contents of learning are ultimately
determined by the contents of consciousness (e.g., by the
specific stimuli that participants are selectively attending
to, or by whether or not they are intending to extract a
366
regularity) and where the contents of consciousness also
evolve as a result of previous learning episodes (cf.
Cleeremans & Jime
´
nez, 2002).
More specifically, we have shown that the differences
observed with practice in the way in which participants
learn either deterministic or probabilistic structures, and
under single- or dual-task conditions, are not easy to
account for in terms of a single learning system. The
pattern appears to be much easier to understand by
relying on the existence of different implicit and explicit
learning mechanisms. According to this view, implicit
learning would be taken to arise as an automatic result
of paying selective attention to each stimulus, and it
would not require additional resources beyond those
needed to process and respond to each relevant stimu-
lus, although it would only produce a slow and pro-
gressive accrual of statistical information. On the other
hand, the results of this statistical learning may promote
not only a change in performance, but also a change in
the contents of consciousness that, under specific cir-
cumstances (i.e., when the sequences are simple and
participants have enough attentional resources avail-
able) may call upon the action of further explicit
mechanisms that would actively learn by testing specific
hypotheses.
Acknowledgements This research was supported by grants PB97–
0525 and BSO2003–05095 from the Ministerio de Educacio
´
n
(Spain). The authors wish to thank George Mandler and David R.
Shanks for their insightful comments on a previous version of this
manuscript.
References
Barsalou, L. (1995). Storage side-effects: Studying processing to
understand learning. In A. Ram & D. B. Leake (Eds.), Goal-
driven learning (pp. 407–419). Cambridge, MA: MIT Press.
Cleeremans, A., & Jime
´
nez, L. (1998). Implicit sequence learning:
The truth is in the details. In M. A. Stadler & P. A. Frensch
(Eds.), Handbook of implicit learning (pp. 323–364). Thousand
Oaks, CA: Sage.
Cleeremans, A., & Jime
´
nez, L. (2002). Implicit learning and con-
sciousness: A graded, dynamic perspective. In R. M. French &
A. Cleeremans (Eds.), Implicit learning and consciousness: An
empirical, computational and philosophical consensus in the
making? (pp. 1–40). Hove, UK: Psychology Press.
Cleeremans, A., Destrebecqz, A., & Boyer, C. (1998). Implicit
learning: News from the front. Trends in Cognitive Sciences, 2,
406–416.
Cohen, A., Ivry, R. I., & Keele, S. W. (1990). Attention and
structure in sequence learning. Journal of Experimental Psy-
chology: Learning, Memory, and Cognition, 16, 17–30.
Curran, T., & Keele, S. W. (1993). Attentional and nonattentional
forms of sequence learning. Journal of Experimental Psychol-
ogy: Learning, Memory, and Cognition, 19, 189–202.
Destrebecqz, A., & Cleeremans, A. (2001). Can sequence learning
be implicit? New evidence with the process dissociation proce-
dure. Psychonomic Bulletin & Review, 8, 343–350.
Destrebecqz, A., & Cleeremans, A. (2003). Temporal effects in se-
quence learning. In L. Jime
´
nez (Ed.), Attention and implicit
learning (pp. 181–213). Amsterdam: Benjamins.
Dienes, Z., & Berry, D. (1997). Implicit learning: Below the
subjective threshold. Psychonomic Bulletin and Review, 4,3
33.
Frensch, P. A. (1998). One concept, multiple meanings: On how to
define implicit learning. In M. A. Stadler & P. A. Frensch
(Eds.), Handbook of implicit learning (pp. 47–104). Thousand
Oaks, CA: Sage.
Frensch, P. A., Buchner, A., & Lin J. (1994). Implicit learning of
unique and ambiguous serial transitions in the presence and
absence of a distractor task. Journal of Experimental Psychol-
ogy: Learning, Memory, and Cognition, 20, 567–584.
Frensch, P. A., Haider, H., Ru
¨
nger, D., Neugebauer, U., Voigt, S.,
Werg, J. (2003). The route from implicit learning to verbal
expression of what has been learned: Verbal report of inciden-
tally experienced environmental regularity. In L. Jime
´
nez (Ed.),
Attention and implicit learning (pp. 335–366). Amsterdam:
Benjamins.
Frensch, P. A., Lin, J., & Buchner, A. (1998). Learning versus
behavioral expression of the learned: The effects of a secondary
tone-counting task on implicit learning in the serial reaction
time task. Psychological Research, 61, 83–98.
Frensch, P. A., & Miner, C. S. (1994). Effects of presentation rate
and individual differences in short-term memory capacity on an
indirect measure of serial memory. Memory & Cognition, 22,
95–110.
Frensch, P. A., Wenke, D., & Ru
¨
nger, D. (1999). A secondary
tone-counting task suppresses expression of knowledge in the
serial reaction task. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 25, 260–274.
Hsiao, A. T., & Reber, A. S. (1998). The role of attention in im-
plicit sequence learning: Exploring the limits of the cognitive
unconscious. In M. A. Stadler & P. A. Frensch (Eds.), Hand-
book of implicit learning (pp. 471–494). Thousand Oaks, CA:
Sage.
Jime
´
nez, L. (2003). Intention, attention, and consciousness in
probabilistic sequence learning. In L. Jime
´
nez (Ed.), Attention
and implicit learning (pp. 43–70). Amsterdam: Benjamins.
Jime
´
nez, L., & Me
´
ndez, C. (1999). Which attention is needed for
implicit sequence learning? Journal of Experimental Psychology:
Learning, Memory, and Cognition, 25, 236–259.
Jime
´
nez, L., & Me
´
ndez, C. (2001). Implicit sequence learning with
competing explicit cues. Quarterly Journal of Experimental
Psychology (A), 54, 345–369.
Jime
´
nez, L., Me
´
ndez, C., & Cleeremans, A. (1996). Direct and
indirect measures of sequence learning. Journal of Experi-
mental Psychology: Learning, Memory, and Cognition, 22, 948–
969.
Keele, S. W., Ivry, R., Mayr, U., Hazeltine, E., & Heuer, H. (2003).
The cognitive and neural architecture of sequence representa-
tion. Psychological Review, 110, 316–339.
Nissen, M. J., & Bullemer, P. (1987). Attentional requirements of
learning: Evidence from performance measures. Cognitive
Psychology, 19, 1–32.
Rah, S., Reber, A. S., & Hsiao, A. T. (2000) Another wrinkle on
the dual-task SRT experiment: It’s probably not dual-task.
Psychonomic Bulletin and Review, 7, 309–313.
Reed, J., & Johnson, P. (1994). Assessing implicit learning with
indirect tests: Determining what is learnt about sequence
structure. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 20, 585–594.
Reingold, E. M., & Merikle, P. (1988). Using direct and indirect
measures to study perception without awareness. Perception
and Psychophysics, 44, 563–575.
Schmidtke, V., & Heuer, H. (1997). Task integration as a factor in
secondary-task effects on sequence learning. Psychological Re-
search, 60, 53–71.
Schvaneveldt, R. W., & Gomez, R. (1998). Attention and
probabilistic sequence learning. Psychological Research, 61,
175–190.
Shanks, D. R. (2003). Attention and awareness in ‘‘implicit’’ se-
quence learning. In L. Jime
´
nez (Ed.), Attention and implicit
learning (pp. 11–42). Amsterdam: Benjamins.
Shanks, D. R., & Channon, S. (2002). Effects of a secondary task
on ‘‘implicit’’ sequence learning: Learning or performance?
Psychological Research, 66, 99–109.
367
Shanks, D. R., & Perruchet, P. (2002). Dissociation between
priming and recognition in the expression of sequential
knowledge. Psychonomic Bulletin and Review,9, 362–367.
Shanks, D. R., & St. John, M. (1994). Characteristics of dissociable
human learning systems. Behavioral and Brain Sciences, 17,
367–447.
Shanks, D. R., Wilkinson, L., & Channon, S. (2003). Relationship
between priming and recognition in deterministic and prob-
abilistic sequence learning. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 29, 248–261.
Stadler, M. (1995). The role of attention in implicit learning.
Journal of Experimental Psychology: Learning, Memory, and
Cognition, 21 , 674–685.
Wilkinson, L., & Shanks, D. R. (2004). Intentional control and
implicit sequence learning. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 30, 354–369.
Willingham, D. B., Greenberg, A. R., & Thomas, R. C. (1997).
Response-to-stimulus interval does not affect implicit motor
sequence learning, but does affect performance. Memory and
Cognition, 25 , 534–542.
368
... As with the AGL paradigm, concerns arose over the content of the knowledge representations (Reed & Johnson, 1994) and subsequent alternations were made to the task. Jimenez and Vazquez (2005) argue that the probabilistic version of the SRT task is as close to a 'pure' measure of implicit learning that has been developed so far. The purity of the results stems from the fact that the probabilistic SRT does not contain any first-order information that could account for learning, as after each location, any other location of the stimuli is possible. ...
... As discussed in Chapter 2, pp. 32-34, there are reasons to believe that the probabilistic version of the SRT task is a purer measure of implicit learning, compared to other versions (Jimenez & Vazquez, 2005). In the SRT task, participants were required to focus on a computer monitor where they saw a stimulus appear at one of four locations. ...
... Thus, the final iteration of the factor analysis was focused on the three implicit learning outcomes -AGL, VSL, and SRT. While the SRT still maintained the strongest loading in this model, which was suspected to be in part due to the level of purity by which this task can measure implicit ability (Jimenez & Vazquez, 2005;Jimenez, 2003), the other two tasks also had increased loading strength in this final iteration which explained 27% of the common variance between these three tasks. ...
Thesis
Full-text available
Is there is a common implicit ability analogous to I.Q. (i.e., a latent factor determined by strong inter correlations of psychometric measures)? Based on my Ph.D. research, there is just enough evidence to argue both ways.
... Previous studies investigating the effect of a secondary task on statistical learning yielded inconsistent findings: Some conclude that statistical learning is resistant to a dual task manipulation 14,[18][19][20]73 , while in other cases degraded performance is observed 16,19,[21][22][23][24] . Importantly, statistical learning in the language domain seems to be more affected by a secondary task 16,22, cf. ...
... Importantly, statistical learning in the language domain seems to be more affected by a secondary task 16,22, cf. 23 , compared with statistical learning in the visuomotor domain 14,[18][19][20] , in line with our results. However, as already mentioned in the Introduction, some of these studies used a secondary task related to a second stimulus stream resulting in a selective attention manipulation where good performance on both tasks can be achieved if attention is switched between the two tasks, potentially affecting the stimulus processing, as well 17,74,75 . ...
... In the present experiment, we chose the cuing of the repeating sequence embedded in the same stimulus stream as the probability-based associations, while all stimuli of that stream are similarly processed, and attention is divided between the cued and the uncued stimuli. Based on our and the previous results 14,18,20 , we conclude that visual statistical learning is not affected by divided attention. ...
Article
Full-text available
Statistical learning facilitates the efficient processing and prediction of environmental events and contributes to the acquisition of automatic behaviors. Whereas a minimal level of attention seems to be required for learning to occur, it is still unclear how acquisition and consolidation of statistical knowledge are affected when attention is divided during learning. To test the effect of divided attention on statistical learning and consolidation, ninety-six healthy young adults performed the Alternating Serial Reaction Time task in which they incidentally acquired second-order transitional probabilities. Half of the participants completed the task with a concurrent secondary intentional sequence learning task that was applied to the same stimulus stream. The other half of the participants performed the task without any attention manipulation. Performance was retested after a 12-h post-learning offline period. Half of each group slept during the delay, while the other half had normal daily activity, enabling us to test the effect of delay activity (sleep vs. wake) on the consolidation of statistical knowledge. Divided attention had no effect on statistical learning: The acquisition of second-order transitional probabilities was comparable with and without the secondary task. Consolidation was neither affected by divided attention: Statistical knowledge was similarly retained over the 12-h delay, irrespective of the delay activity. Our findings can contribute to a better understanding of the role of attentional processes in and the robustness of visuomotor statistical learning and consolidation.
... The effect of a secondary task on implicit sequence learning has been studied extensively in the last few decades. Evidence for impaired [5][6][7][8][9][10][11], intact [4,7,9,12,13], or even improved performance [6] was found during the acquisition of implicit sequence knowledge. Despite the vast literature on the practice, followed by a 24-hour offline period (Fig 1). ...
... The effect of a secondary task on implicit sequence learning has been studied extensively in the last few decades. Evidence for impaired [5][6][7][8][9][10][11], intact [4,7,9,12,13], or even improved performance [6] was found during the acquisition of implicit sequence knowledge. Despite the vast literature on the practice, followed by a 24-hour offline period (Fig 1). ...
... Based on those results, three outcomes were conceivable for our study. First, it was possible that dual-tasking would impair the retrieval of probabilistic sequence knowledge, as the majority of previous studies reported detrimental effects of a secondary task during sequence learning [5][6][7][9][10][11]. Various explanations were offered to account for these results: during dual-tasking, we might disadvantageously integrate the sequenced and non-sequenced information [36], the secondary task might disrupt the organization of the incoming information about regularities [37], or that parallel response selection slows down the learning process [38]. ...
Article
Full-text available
The characteristics of acquiring new sequence information under dual-task situations have been extensively studied. A concurrent task has often been found to affect performance. In real life, however, we mostly perform a secondary task when the primary task is already well acquired. The effect of a secondary task on the ability to retrieve well-established sequence representations remains elusive. The present study investigates whether accessing well-acquired probabilistic sequence knowledge is affected by a concurrent task. Participants acquired non-adjacent regularities in an implicit probabilistic sequence learning task. After a 24-hour offline period, participants were tested on the same probabilistic sequence learning task under dual-task or single-task conditions. Here, we show that although the secondary task significantly prolonged the overall reaction times in the primary (sequence learning) task, access to the previously learned probabilistic representations remained intact. Our results highlight the importance of studying the dual-task effect not only in the learning phase but also during memory access to reveal the robustness of the acquired skill.
... This leads to various behavioral differences between implicit and explicit knowledge. Implicit, as opposed to explicit learning, can be rather independent from attentional capacities, distractions or working memory load (Jiménez & Vázquez, 2005;Won & Jiang, 2015). Both types of learning can be of advantage in different situations. ...
... Typically, dual-task paradigms are used to determine whether implicit learning is impaired under attentional load. Although many studies have demonstrated unimpaired learning under attentional load (e.g., Cohen et al., 1990;Frensch & Miner, 1995;Reed & Johnson, 1994;Shanks & Johnstone, 1999), others have shown impaired learning, slowed reaction times, and diminished transfer effects (e.g., Jimenez & Vazquez, 2005;Shanks & Channon, 2002;Shanks, Rowland, & Ranger, 2005). Here, however, we find selfreports of inattention (i.e., mind wandering) were not associated with performance in the implicit group, but that they were negatively associated with performance in the explicit group. ...
Article
According to the attentional resources account, mind wandering (or “task-unrelated thought”) is thought to compete with a focal task for attentional resources. Here, we tested two key predictions of this account: First, that mind wandering should not interfere with performance on a task that does not require attentional resources; second, that as task requirements become automatized, performance should improve and depth of mind wandering should increase. Here, we used a serial reaction time task with implicit- and explicit-learning groups to test these predictions. Providing novel evidence for the attentional resource account's first prediction, results indicated that depth of mind wandering was negatively associated with learning in the explicit, but not the implicit, group, indicating that mind wandering is associated with impaired explicit, but not implicit, learning. Corroborating the attention resource account's second prediction, we also found that, overall, performance improved while at the same time depth of mind wandering increased. From an implicit-learning perspective, these results are consistent with the claim that explicit learning is impaired under attentional load, but implicit learning is not. Data, analysis code, manuscript preparation code, and pre-print available at osf.io/qzry7/.
... Finally, even though we did not find a direct correlation between specific questions and lowered performance in testing, it is possible that the insertion of questions led to an overall decrement in learning because it forced participants to do the training procedure under dual-task conditions, which has been shown to be detrimental for implicit learning (Hendricks et al., 2013;Jiménez and A Vázquez, 2005;van Schagen, 2017). In an extension of Radulescu et al. (2019), van Schagen (2017 tried overloading subjects' working memory with a dual task instead of increasing entropy: not only did the experimental manipulation not improve learning, but participants in the dual-task condition actually performed worse than those in a single-task control condition (van Schagen, 2017). ...
Thesis
Full-text available
Language proficiency largely relies on implicit knowledge, which is unconscious and operates independently of voluntary control. Implicit learning is a process of incidental learning which results in the acquisition of implicit knowledge. We know that adult learners can acquire knowledge of novel L2 linguistic rules through implicit learning, as evidenced by their performance on receptive tasks. However, it is unclear whether implicit learning processes can also support the development of L2 production skills. The central question of this dissertation was whether it would be possible for learners to acquire implicit knowledge of a new rule through implicit learning and use it directly in spoken production. Our second question concerns the relationship between production and comprehension: we asked whether implicit knowledge acquired through a production task would also lead to improved performance in comprehension. To address these questions, we trained participants on a semiartificial language based on a rule naturally found in Czech: specifically, the usage rule for a pair of spatial prepositions (v and na) which alternate depending on the distinction between open and enclosed spaces. Training was carried out using a novel methodology based on elicited oral imitation, which was also used to test productive knowledge. Participants were also tested on comprehension, using both reaction time and recognition memory paradigms. Our findings suggest that it is possible to acquire implicit productive knowledge through a production-based task, and to generalise it to new instances in spoken production. The results of our experiments also show that learning outcomes were sensitive to the specific procedure used to train participants, which appeared to interact with individual differences in working memory. Finally, we found limited evidence that implicit knowledge acquired through production could be transferred to comprehension, supporting a skill-specific account of implicit knowledge.
Thesis
Full-text available
Predictability is increasingly recognized as an important principle in perception and motor learning. The pursuit of increased predictability seems to one of the main goals that the human system pursues. Therefore, providing predictability in one of the most challenging situations that humans face, namely multitasking, a promising line of research. In this thesis the impact of predictability was systematically investigated in five experiments. In the first four experiments predictability was achieved by implementing a repeating pattern in one task, or both tasks. Participants acquired knowledge of these patterns either explicitly or implicitly in several training sessions, under single-task or dual-task conditions. We tested whether this increased predictability helped dual-task performance after the training sessions. The results suggest that predictability is helpful for dual-task performance, although the benefits are confined to the predictable task itself. In a fifth experiment we focused on providing between task predictability, which led to a large performance improvement in both tasks, prompting the discussion about what constitutes a task, in the sense of when can two tasks be perceived as a single task comprising both, a theoretical problem we tried to tackle in one of the articles. Explanations for the findings, theoretical implications, methodological issues and suggestions for future research are given in the general discussion
Article
To investigate whether working memory and visual processing have the same role or different roles in A/B and A/not A prototype category learning, the present study adopted an A/B or A/not A category learning task in control and dual conditions. The results of Experiment 1 showed that an additional dual visual working memory task rather than a dual verbal working memory task reduced accuracy of the A/B task, whereas no dual tasks influenced accuracy of the A/not A task. The results of Experiment 2 revealed that an additional dual visual processing task impaired accuracy of the A/B task, whereas the dual visual processing task did not influence accuracy of the A/not A task. These results indicate that visual working memory and visual processing play different roles in A/B and A/not A prototype category learning, and support that these two types of prototype category learning are mediated by different memory systems.
Article
Full-text available
Knowing when the brain learns is crucial for both the comprehension of memory formation and consolidation and for developing new training and neurorehabilitation strategies in healthy and patient populations. Recently, a rapid form of offline learning developing during short rest periods has been shown to account for most of procedural learning, leading to the hypothesis that the brain mainly learns during rest between practice periods. Nonetheless, procedural learning has several subcomponents not disentangled in previous studies investigating learning dynamics, such as acquiring the statistical regularities of the task, or else the high-order rules that regulate its organization. Here we analyzed 506 behavioral sessions of implicit visuomotor deterministic and probabilistic sequence learning tasks, allowing the distinction between general skill learning, statistical learning, and high-order rule learning. Our results show that the temporal dynamics of apparently simultaneous learning processes differ. While high-order rule learning is acquired offline, statistical learning is evidenced online. These findings open new avenues on the short-scale temporal dynamics of learning and memory consolidation and reveal a fundamental distinction between statistical and high-order rule learning, the former benefiting from online evidence accumulation and the latter requiring short rest periods for rapid consolidation.
Article
Full-text available
Implicit learning refers to the incidental acquisition and expression of knowledge that is not accompanied by full awareness of its contents. Implicit sequence learning (ISL) represents one of the most useful paradigms to investigate these processes. In this paradigm, participants are usually instructed to respond to the location of a target that moves regularly through a set of possible locations. Although participants are not informed about the existence of a sequence, they eventually learn it implicitly, as attested by the costs observed when this sequence is violated in a reduced set of control trials. Interestingly, the expression of this learning decreases immediately after a control trial, in a way that resembles the adjustments triggered in response to incongruent trials in interference tasks. These effects have been attributed to a control network involving dorsolateral prefrontal cortex (DLPFC) and cingulate (ACC) structures. In the present work, we reviewed a group of recent studies which had inhibited DLPFC top-down control by means of non-invasive brain stimulation to increase the acquisition of ISL. In addition, as no previous study has investigated the effect of inhibiting top-down control on releasing the automatic expression of ISL, we present a pre-registered – yet exploratory – study in which an inhibitory continuous theta burst stimulation protocol was applied over an anterior-ventral portion of the dorsolateral prefrontal cortex (DLPFC) highly interconnected with the ACC, and whose activity has been specifically linked to motor control (i.e., Right DLPFC, n=10 or the Left DLPFC, n=10), compared to active Vertex stimulation (n=10). Contrary to our hypotheses, the results did not show evidence for the involvement of such region in the expression of ISL. We discussed the results in the context of the set of contradictory findings reported in the systematic review.