See discussions, stats, and author profiles for this publication at: [Link]

net/publication/224013703

Implicit Learning

Article in Current Directions in Psychological Science · February 2003

CITATIONS READS
6 749

2 authors:

Peter Frensch Dennis Rünger

Humboldt-Universität zu Berlin University of California, Los Angeles
151 PUBLICATIONS 8,344 CITATIONS 52 PUBLICATIONS 1,935 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Dennis Rünger on 03 June 2021.

The user has requested enhancement of the downloaded file.

 CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 13

Implicit Learning Reber, whose early work in the

1960s rekindled interest in implicit
Peter A. Frensch1 and Dennis Rünger learning, has done most of his em-
pirical work with artificial-gram-
Department of Psychology, Humboldt University, Berlin, Germany
mar-learning tasks. In these tasks,
participants are asked to memorize
a set of letter strings, such as
tion is achieved. Implicit learning— “XXRTRXV” and “QQWMWQP,”
Abstract that are, unbeknownst to the par-
Implicit learning appears to be laxly defined as learning without
awareness—is seemingly ubiqui- ticipants, generated by some rules.
a fundamental and ubiquitous After the memorization phase, par-
process in cognition. Although tous in everyday life.
In this article, we try to provide an ticipants are told that the strings
defining and operationalizing they memorized followed certain
implicit learning remains a overview of the difficulties research
on implicit learning has been facing rules, and are asked to classify new
central theoretical challenge, strings as grammatical (i.e., follow-
scientists’ understanding of im- and of the advances that have been
made in scientists’ understanding of ing the rules) or not. Typically, par-
plicit learning has progressed ticipants can perform this classifi-
significantly. Beyond establish- the concept. More specifically, we
discuss three separate issues. First, cation task with accuracy better
ing the existence of “learning than would be expected by chance,
without awareness,” current re- we address what is meant by im-
plicit learning and how the concept despite remaining unable to ver-
search seeks to identify the cog- bally describe the rules.
nitive processes that support has been empirically approached in
the recent past. Second, we summa- Thus, in a grammar-learning
implicit learning and addresses task, participants learn about per-
the relationship between learn- rize what is currently known with
some certainty about the cognitive missible and nonpermissible com-
ing and awareness of what was binations of letters that are presented
learned. The emerging view of processes underlying implicit learn-
ing and the mental representations simultaneously. By comparison, in
implicit learning emphasizes the another task often used to investi-
role of associative learning that are acquired through it. Third,
we discuss some of the most impor- gate implicit learning, the serial re-
mechanisms that exploit statis- action time task (SRTT), partici-
tical dependencies in the envi- tant current topics of investigation.
pants learn about permissible and
ronment in order to generate nonpermissible combinations of
highly specific knowledge rep- spatial locations that occur over
resentations. time. In the SRTT, participants are
DEFINITION AND asked to select and depress a key
Keywords OPERATIONALIZATION that matches each of the locations
cognitive psychology; learning; at which a stimulus appears on a
consciousness; awareness The one basic theoretical issue that screen. The sequence of locations at
reigns supreme among the difficul- which the stimulus appears is
ties facing researchers concerns the fixed. In general, participants seem
Have you ever wondered why it definition and operationalization of to be able to learn the sequence of
is that you can speak your native implicit learning. Although it seems spatial locations even when they
language so well without making clear that implicit learning needs to are not able to verbally describe it.
any grammatical errors although be viewed in opposition to learning Divergent definitions of implicit
you do not know many of the that is not implicit (often called ex- learning entail divergent opera-
grammatical rules you follow? plicit, hypothesis-driven learning), it tionalizations of the concept, but
Have you ever wondered how it is has so far proven extremely difficult even researchers who agree in their
that you can walk properly al- to provide a satisfactory definition of definitions might use experimen-
though you cannot describe the implicit learning. At least a dozen tal tasks that differ in what exactly
rules of mechanics your body must different definitions have been of- participants might learn. There-
certainly follow? These two exam- fered in the field. fore, it remains an open empirical
ples point to an important human One important consequence of issue to what extent results from a
property, namely, the ability to adapt the heterogeneity of definitions is given task that has been used to
to environmental constraints—to that different researchers have op- probe implicit learning can be gen-
learn—in the absence of any erationalized implicit learning in dif- eralized to other tasks. This point
knowledge about how the adapta- ferent ways. For example, Arthur leads to our first conclusion:

Copyright © 2003 American Psychological Society

14 VOLUME 12, NUMBER 1, FEBRUARY 2003

• Conclusion 1. Implicit learning of port is used to assess awareness of tion scores correlated with their
Task A is not necessarily compa- acquired knowledge, many experi- learning scores for the SRTT. In
rable to implicit learning of Task mental findings appear to support general, many different studies us-
B. Neither the properties of the the conclusion that implicit learn- ing different experimental para-
learning mechanisms involved ing is possible. digms have used forced-choice
nor the acquired mental repre- However, many authors have ar- tests to assess awareness of the ac-
sentations need be the same. It is gued that verbal reports may have quired knowledge, and these stud-
even conceivable that implicit poor validity. First, it has been ar- ies appear not to support the exist-
learning of Task A might be pos- gued that the verbal-report data do ence of implicit learning.
sible, but implicit learning of not pass the information criterion; However, it has been argued
Task B might not. that is, the information assessed by that this particular interpretation
verbal recall tests is not always the rests on the assumption that the
same information that has led to the forced-choice tests are pure assess-
demonstrated learning. Second, ver- ments of awareness (i.e., are pro-
THE KEY ISSUE bal recall tests might not pass the cess pure). This is almost certainly
sensitivity criterion; that is, they may not the case. Indeed, participants
Regardless of how implicit not provide a level of sensitivity that might choose a correct answer on a
learning is defined and operation- is comparable to that of tests demon- forced-choice test not because they
alized, the key empirical issue that strating learning in the first place. are aware of the fact that it is the
research needs to address is whether Many researchers have therefore correct answer but because they
or not learning that is “implicit” is suggested that awareness should be rely on some intuition that they are
possible, and if it is, whether im- assessed by forced-choice tests, such not able to express. The growing
plicit learning is different from as recognition tests, rather than ver- understanding that tests are rarely
learning that is “not implicit.” bal recall. process pure has fostered the use
Many researchers have for practi- In the grammar-learning para- of new methodologies that are not
cal reasons adopted as their defini- digm, participants are sometimes based on this assumption. For ex-
tion of implicit learning “the capac- asked to complete recognition tests ample, Jacoby’s process dissocia-
ity to learn without awareness of after they have categorized letter tion procedure offers a measure of
the products of learning.” Thus, strings as grammatical or non- awareness that is derived from ex-
learning is assumed to be “im- grammatical. For example, in some perimental conditions that are be-
plicit” when participants are un- studies participants were asked to lieved to trigger both implicit and
aware of what they learned. Alter- indicate for each letter string which nonimplicit processes simulta-
natively, learning is assumed not particular letters they thought neously. This consideration of how
to be implicit when participants are made the string grammatical or awareness should be assessed
aware of what they learned. In not. It was found that participants’ leads to our second conclusion:
other words, implicit learning is de- markings correlated with their
fined in terms of its product rather classification performance, sug- • Conclusion 2. Many researchers
than the properties of the learning gesting at least partial awareness of have tried to avoid the difficult
process. the knowledge learned. issue of how to define implicit
Various measures have been Similar findings have been ob- learning and have, often without
proposed to assess awareness of tained in studies that have used stating so explicitly, adopted the
the products of learning. The most other implicit-learning paradigms. stance that implicit learning is
notable measures are verbal re- For example, after participants had the capacity to learn without
ports and forced-choice tests (such completed the SRTT, they were awareness of the products of
as recognition tests). presented sequence patterns of learning. However, it has be-
Participants in implicit-learning varying lengths in numerical form. come clear that the amount of
experiments have consistently Each sequence (e.g., 123432) de- support for implicit learning
been shown to be able to acquire noted a series of locations on the varies considerably with the
knowledge that they cannot ver- computer screen. Participants were specific measure that is selected
bally describe. This appears to be asked to mark patterns that they to assess awareness of what was
true for a wide variety of tasks, in- had encountered during the exper- learned. Thus, by avoiding the
cluding the grammar-learning and iment as true and patterns that issue of how to define implicit
sequence-learning tasks we de- they had not seen as false. It was learning, researchers have intro-
scribed earlier. Thus, if verbal re- found that participants’ recogni- duced the problem of how to de-

Published by Blackwell Publishing Inc.

 CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 15

fine awareness. In the end, the helpful to consider the different ways According to the four remaining
definitional question has not in which, in principle, learning and possibilities, learning and aware-
been resolved, but has merely awareness of the products of learning ness need not be—but might be—
been transferred from one con- might be related. Figure 1 depicts perfectly correlated. According to
cept to another. five of the many distinct possibilities the second possibility, depicted in
that have been proposed. Figure 1b, a single learning mecha-
First, it is, of course, conceivable nism is assumed to create memory
MECHANISMS OF LEARNING that learning and awareness of representations that control behav-
AND AWARENESS what has been learned are perfectly ior. Some of the learned memory
correlated. According to this pro- representations might be open to
Even when implicit learning (in posal, implicit learning does not awareness; some might not be.
the sense of learning that yields exist. As is shown in Figure 1a, The last three possibilities (Figs.
knowledge the learner is not aware learning might be achieved by a 1c–1e) allow for truly implicit
of) is demonstrated conclusively, one single mechanism that generates learning. According to the third pos-
learns little about the mechanisms memory representations a learner sibility, an implicit-learning mech-
underlying implicit learning. It is is always aware of. anism might generate memory rep-

Fig. 1. Possible relations between learning and awareness of what was learned.

Copyright © 2003 American Psychological Society

16 VOLUME 12, NUMBER 1, FEBRUARY 2003

resentations that control behavior. nonimplicit learning), represented support the multiple-systems
The perception of one’s own behav- by the possibilities depicted in Fig- view.
ior, in turn, might lead to nonim- ures 1c through 1e, from the single- Brain-imaging techniques have
plicit (i.e., hypothesis-testing) system hypothesis, represented by increasingly been used to study
learning that might generate aware- the possibilities depicted in Figures implicit learning in the SRTT. Al-
ness of what was learned (Fig. 1c). 1a and 1b. The relative adequacy of though some results suggest that
Under this view, the effects of im- these two hypotheses can be as- partially distinctive brain areas are
plicit learning are an important trig- sessed by exploring the potentially involved in implicit and nonim-
ger for nonimplicit learning. For ex- differing influence of variables plicit forms of learning, it is, at
ample, a tennis player might perceive such as intention to learn, atten- present, not clear whether these
an increased accuracy of her serve. tion, age of participants, individual findings should be interpreted as
She might then conclude that the differences in intelligence, stimulus evidence supporting the multiple-
reason for this improvement is to complexity, and task demands on systems view or as evidence sup-
be found in a slightly higher toss of learning with awareness and on porting a “single-system plus aware-
the ball. learning without awareness. ness” view (depicted in Fig. 1b).
The fourth possibility is that For example, researchers have ex- Consideration of the cognitive
nonimplicit learning might lead to plored the possibility that implicit mechanisms that might be in-
awareness of what was learned, and nonimplicit learning might be volved in implicit learning leads to
and might control behavior. The differentially affected by age. With our third conclusion:
expression of behavior, in turn, the SRTT, it has been found that im-
might provide the input for the op- plicit learning is less affected by age • Conclusion 3. Defining implicit
eration of an implicit-learning than is learning that is based on hy- learning with respect to aware-
mechanism (Fig. 1d). For example, pothesis testing. Indeed, implicit ness of the products of learning
most tennis players know that learning in the SRTT does not appear has drawn attention away from
solid ground strokes require a to begin to decline until relatively old the mechanisms that are respon-
player to move toward the ap- age, and even then, the elderly dis- sible for the generation of differ-
proaching ball instead of away play performance levels that are ent forms of knowledge. Despite
from it. The conscious effort to en- much closer to those of younger a continuously increasing amount
gage in a forward movement may adults for implicit-learning tasks of empirical data, the debate be-
lead to learning within the motor than for nonimplicit-learning tasks tween multiple-systems propo-
systems that lies largely outside of involving, for example, problem nents and single-system propo-
conscious awareness. solving, reasoning, and long-term nents has not been settled yet.
The fifth possibility, shown in Fig- memory. On the whole, this research Furthermore, the question of
ure 1e, is that there exist two distinct therefore lends some credibility to how exactly awareness and
learning mechanisms, with one of the multiple-systems view. learning might be interrelated
the mechanisms generating mem- Also, both neuropsychological has only recently begun to be
ory representations that a learner is studies and neuroimaging stud- addressed empirically.
aware of, and the other mechanism ies have addressed the adequacy
generating representations that a of the multiple-systems and single-
learner is not aware of but that nev- system hypotheses. For example,
ertheless control behavior. early studies suggested that even
Most of the research that has densely amnesic patients can IMPORTANT ADVANCES
been concerned with the differ- show near-normal implicit learn- MADE
ence between implicit and nonim- ing in both the grammar-learning
plicit learning has not addressed and the sequence-learning para- If one agrees with the use of
which possibility in Figure 1 de- digms, although they are specifi- verbal-report measures to assess
scribes the nature of implicit learn- cally impaired on recognition and awareness, then recent research on
ing, but rather has tried to demon- prediction tasks. More recent crit- implicit learning has modified ear-
strate that learning is possible in ical reexaminations have, how- lier theoretical beliefs in important
the absence of learners’ awareness ever, demonstrated that amnesic ways. Earlier work had character-
of the acquired knowledge. How- patients do seem to show a deficit ized implicit learning as a mecha-
ever, several attempts have been in implicit learning compared nism by which abstract knowledge
made to distinguish the two-sys- with normal control participants; of regularities that are present in
tems hypothesis (i.e., that there are it is therefore unclear whether or the environment is acquired auto-
separate systems for implicit and not the findings, on the whole, matically and unintentionally by

Published by Blackwell Publishing Inc.

 CURRENT DIRECTIONS IN PSYCHOLOGICAL SCIENCE 17

mere exposure to relevant instances. matically, without the use of atten- • Conclusion 4. The early proposal
The proposal of a smart uncon- tional resources. By far, most of these of a smart unconscious capable
scious was based, to a large extent, studies have used the SRTT, often of acquiring abstract knowledge
on empirical findings with the manipulating the amount of atten- in an effortless, automatic man-
grammar-learning task that ap- tional resources available to partici- ner has been replaced recently
peared to show that participants pants by asking them to perform the by the assumption of one or
possessed abstract knowledge SRTT either by itself or together with more implicit learning mecha-
about the rules of the grammar that a secondary task (typically a tone- nisms that operate mostly asso-
went beyond the surface character- counting task). ciatively. These mechanisms
istics of the information encountered. In general, it has been found pick up statistical dependencies
This claim seemed further sup- that implicit learning takes place encountered in the environment
ported by findings indicating that both in the presence and in the ab- and generate highly specific
implicitly acquired knowledge sence of a secondary task. What knowledge representations. It is
may transfer across modalities; for remains unclear, at present, is the likely that the mechanisms oper-
example, learning from a task in- extent to which implicit learning is ate only on information that is
volving written letters (visual stim- affected by the attention manipu- attended to and that is relevant
uli) can transfer to performance in lation. Some researchers argue to the response to be made.
a task involving letter sounds (au- that the secondary task interferes
ditory stimuli). with task performance rather than
This view has been challenged, with implicit learning proper (i.e.,
however, by many recent findings. that the secondary task impedes CONCLUSIONS
For example, it has been repeatedly the expression of what has been
shown that implicitly acquired learned). Under this view, implicit Researchers’ understanding of
knowledge might consist of little learning does not depend on the implicit learning has come a long
more than short fragments or chunks availability of attentional resources. way. Today, many believe that im-
of the materials encountered in an Others take the stance that the plicit learning exists, and further-
implicit-learning situation. In the learning process itself is adversely more that it is based on relatively
wake of these findings, neural-net- affected by the presence of a sec- simple learning mechanisms.
work models and fragment-based ondary task and thus requires at- These mechanisms associate envi-
models that are capable of simulating tentional resources. ronmental stimuli that are at-
a great deal of the available experi- On the whole, the experiments tended to and that are relevant for
mental findings have been devel- that have been conducted all suffer behavior. Despite the recent ad-
oped. These models utilize represen- from the problem that attention itself vances, however, the field still suf-
tations of elementary stimuli in the is an ill-defined concept that might fers from a number of unresolved
learning situation (e.g., representa- refer to both mental capacity and se- empirical and theoretical issues.
tions of letters in a grammar-learning lection. In the latter sense, “attention” First, there exist conflicting results
task) and associations between the points to the problem of allocating regarding the role of attention in
representations. Learning consists of cognitive resources to a specific item implicit learning. Second, the exact
a continuous, incremental change in or event. When “attention” is used relation between learning and
the associative pattern that is sensi- synonymously with “mental capac- awareness (see Fig. 1) is very much
tive to the statistical features of the ity,” it instead refers to a limitation of unknown. Third, the key theoreti-
set of items or events encountered. cognitive resources that becomes cal issue of how to define implicit
Thus, a representation of the implicit- apparent when resources have to be learning has still not been resolved.
learning situation that is shaped by shared by concurrent cognitive pro- We strongly believe that progress
statistical constraints gradually cesses. When these two factors are on the former two (empirical) issues
evolves. Although the characteriza- separately and experimentally ma- will be made soon and will be
tion of implicitly acquired knowledge nipulated, it appears that implicit based on improved methodology
is still a matter of debate, the cur- learning occurs only when stimuli and the joint use of computational
rent trend is to assume that ab- are relevant to the task and are at- modeling and functional brain-im-
stract knowledge might not be im- tended to, but that implicit learning aging techniques. Progress on the
plicitly generated. may require no or very little mental key theoretical issue can come, how-
Many recent studies have ex- capacity. ever, only from theoretical advances
plored whether implicit learning, un- Recent advances in researchers’ in understanding of the concepts
like nonimplicit learning (i.e., explicit understanding of implicit learning of “consciousness,” “awareness,” and
hypothesis testing), proceeds auto- lead to our fourth conclusion: “intention.” To achieve this progress

Copyright © 2003 American Psychological Society

18 VOLUME 12, NUMBER 1, FEBRUARY 2003

might require the joint efforts of Cleeremans, A. (1993). Mechanisms of (1998). Handbook of implicit learn-
philosophers, neuroscientists, and implicit learning: Connectionist mod- ing. Thousand Oaks, CA: Sage.
cognitive psychologists. els of sequence processing. Cam-
bridge, MA: MIT Press.
Note
Recommended Reading Reber, A.S. (1993). Implicit learning
and tacit knowledge: An essay on the 1. Address correspondence to Peter
Berry, D.C., & Dienes, Z. (1993). Im- cognitive unconscious. New York: A. Frensch, Department of Psychology,
plicit learning: Theoretical and em- Oxford University Press. Humboldt University, Hausvogteiplatz
pirical issues. Hove, England: 5-7, D-10177 Berlin, Germany; e-mail:
Erlbaum. Stadler, M.A., & Frensch, P.A. (Eds.). [Link]@[Link].

Computer- and Internet-Based Assessment

Psychotherapy Interventions Traditional paper-based self-

report instruments are easily adapted
C. Barr Taylor1 and Kristine H. Luce to the computer format and offer a
Department of Psychiatry, Stanford University Medical Center, Stanford, California number of advantages that include
ensuring data completeness and
standardization. Research has
Abstract time with the therapist, increasing found that computer-administered
Computers and Internet- clients’ participation in therapeu- assessment instruments work as
based programs have great po- tic activities outside the standard well as other kinds of self-report in-
tential to make psychological as- clinical hour, and streamlining in- struments and as well as therapist-
sessment and treatment more put and processing of clients’ data administered ones. Clients may feel
cost-effective. Computer-assisted related to their participation in ther- less embarrassed about reporting
therapy appears to be as effec- apeutic activities. Unfortunately, the sensitive or potentially stigmatiz-
tive as face-to-face treatment for scientific study of these programs ing information (e.g., about sexual
treating anxiety disorders and has seriously lagged behind their behavior or illegal drug use) during
depression. Internet support purported potential, and these in- a computer-assisted assessment than
groups also may be effective and terventions pose important ethical during a face-to-face assessment, al-
have advantages over face-to-face and professional questions. lowing for more accurate estimates
therapy. However, research on of mental health behaviors. Studies
this approach remains meager. show that more symptoms, includ-
ing suicidal thoughts, are reported
COMPUTER-BASED during computer-assisted interviews
Keywords PROGRAMS than face-to-face interviews. Over-
computer applications; Internet all, the evidence suggests that com-
applications; psychotherapy and Information puters can make assessments more
technology efficient, more accurate, and less ex-
A number of studies have demon- pensive. Yet computer-based as-
strated that computers can provide sessment interviews do not allow
In recent years, the increasing information effectively and econom- for clinical intuition and nuance, as-
number of users of computer and ically. An analysis of a large number sessment of behavior, and nonver-
Internet technology has greatly of studies of computer-assisted bal emotional expression, nor do
expanded the potential of com- instruction (CAI) found that CAI is they foster a therapeutic alliance be-
puter- and Internet-based therapy consistently effective in improving tween client and therapist as infor-
programs. Computer- and Internet- knowledge (Fletcher-Flinn & Gravatt, mation is collected.
assisted assessment methods and 1995). Surprisingly, few studies Recently, handheld computers
therapy programs have the potential evaluating the use of CAI for pro- or personal digital assistants (PDAs)
to increase the cost-effectiveness of viding information related to mental have been used to collect real-time,
standardized psychotherapeutic health or psychotherapy have been naturalistic data on a variety of vari-
treatments by reducing contact conducted. ables. For example, clients can record

Published by Blackwell Publishing Inc.

View publication stats