Implicit learning of tacit knowledge

Abstract

I examine the phenomenon of implicit learning, the process by which knowledge about the rule-governed complexities of the stimulus environment is acquired independently of conscious attempts to do so. Our research with the two seemingly disparate experimental paradigms of synthetic grammar learning and probability learning, is reviewed and integrated with other approaches to the general problem of unconscious cognition. The conclusions reached are as follows: (a) Implicit learning produces a tacit knowledge base that is abstract and representative of the structure of the environment; (b) such knowledge is optimally acquired independently of conscious efforts to learn; and (c) it can be used implicitly to solve problems and make accurate decisions about novel stimulus circumstances. Various epistemological issues and related problems such as intuition, neuroclinical disorders of learning and memory, and the relationship of evolutionary processes to cognitive science are also discussed. (PsycINFO Database Record (c) 2012 APA, all rights reserved)

Full text

Journal
of
Experimental
Psychology: General
1989,
Vol.
118,
No. 3,
219-235
Copyrighl
1989
by the
American
Psychological
Association,
Inc.
0096-3445/89/500,75
Implicit Learning
and
Tacit
Knowledge
Arthur
S.
Reber
Brooklyn
College
and the
Graduate Center
City
University
of New
York
I
examine
the
phenomenon
of
implicit learning,
the
process
by
which knowledge about
the rale-
governed complexities
of the
stimulus environment
is
acquired independently
of
conscious
attempts
to do so. Our
research
with
the
two, seemingly
disparate
experimental paradigms
of
synthetic grammar learning
and
probability
learning
is reviewed and
integrated
with other
approaches
to the
general problem
of
unconscious cognition.
The
conclusions
reached are as
follows:
(a)
Implicit
learning produces
a
tacit knowledge
base
that
is
abstract
and
representative
of
the
structure
of the
environment;
(b)
such knowledge
is
optimally acquired independently
of
conscious
efforts
to
learn;
and (c) it can be
used implicitly
to
solve problems
and
make
accurate
decisions
about
novel
stimulus
circumstances. Various
epistemological
issues
and related
prob-
1
lems
such
as
intuition,
neuroclinical
disorders
of
learning
and
memory,
and the
relationship
of
evolutionary
processes
to
cognitive science
are
also discussed.
Some
two
decades
ago the
term
implicit
learning
was
first
used
to
characterize
how one
develops intuitive knowledge
about
the
underlying structure
of a
complex stimulus
envi-
ronment (Reber, 1965, 1967).
In
those early writings,
I
argued
that implicit learning
is
characterized
by two
critical features:
(a)
It is an
unconscious process
and (b) it
yields
abstract
knowledge.
Implicit knowledge results
from
the
induction
of
an
abstract
representation
of the
structure that
the
stimulus
environment
displays,
and
this knowledge
is
acquired
in the
absence
of
conscious,
reflective
strategies
to
learn. Since then,
the
evidence
in
support
of
this theory
has
been abundant,
and
many
of the
details
of the
process
have been sharpened. This
article
is an
overview
of
this evidence
and an
attempt
to
extend
the
general
concepts
to
provide some insight into
a
variety
of
related processes
such
as
arriving
at
intuitive judg-
ments, complex decision making, and,
in a
broad sense,
learning about
the
complex
covariations
among events that
characterize
the
environment.
Put
simply, this
is an
article
about
learning.
It
seems curi-
ous,
given
the
pattern
of
psychological investigation
of the
middle decades
of
this century, that
the
topic
of
learning
should
be so
poorly represented
in the
contemporary literature
in
cognitive psychology.
The
energies
of
cognitive scientists
have
been invested largely
in the
analysis
and
modeling
of
existing
knowledge rather than
in
investigations
of how it was
acquired.
For
example,
in an
important
recent
article
on the
general
topic
of
unconscious memorial systems,
Schacter
(1987)
never came
to
grips
with
the
distinction between
implicit
learning
and
implicit memory.
The
latter,
the
focus
of
his
review,
was
dealt
with
historically, characterized, out-
Preparation
of
this
article
was
supported
in
part
by a
grant
from
the
City
University
of New
York
PSC-CUNY
Research
Award
Pro-
gram.
Special
thanks
go to
Rhianon Allen,
Ruth
Hernstadt,
Paul
Lewicki,
and
Robert
McCauley
for
suggestions, insights,
and
gentle criticisms
(which
I
probably should have paid more attention to).
Correspondence concerning this article should
be
addressed
to
Arthur
S.
Reber, Department
of
Psychology,
Brooklyn
College, City
University
of New
York, Brooklyn,
New
York
11210.
lined,
and
analyzed,
but
virtually
no
attention
was
paid
to the
processes
by
which
these implicit memories "got
there."
This
general lack
of
attention
to the
acquisition problem
may be
one of the reasons why
much recent theorizing
has
been
oriented toward
a
nativist
position (e.g., Chomsky, 1980;
Fodor,
1975, 1983; Gleitman
&
Wanner,
1982).
Failure
to
explicate
how
complex knowledge
is
acquired invites
the
supposition
that
"it was
there
all the
time."
What
follows
is an
exploration
of
implicit learning
from
the
point
of
view
that
the
processes discussed
are
general
and
universal. Implicit acquisition
of
complex knowledge
is
taken
as a
foundation
process
for the
development
of
abstract,
tacit
knowledge
of
all
kinds.
The
stepping-off
place
is the
presump-
tion
that there
is, at
this
juncture,
no
reason
to
place
any
priority
on
particular biological determinants
of a
specific
kind.
All
forms
of
implicit
knowledge
are
taken
as
essentially
similar
at
their deepest
levels.
This
position
needs
to be
pushed
as
far
as it can go; it has
considerably more explanatory power
than
has
been generally recognized.
Experimental
Procedures
Research
on
implicit learning
is
properly carried
out
with
arbitrary
stimulus
domains
with
complex, idiosyncratic struc-
tures.
In
order
to
obtain
insight
into
a
process such
as
implicit
learning,
it is
essential
to
work with novel, synthetic systems
and to
focus
on the
capacity
of
one's
subjects
to
induce
knowledge
of a
deep sort
from
such stimulus
fields.
Over
the
years,
a
number
of
different
techniques have been used.
My
colleagues
and I
have chosen,
in our
laboratory,
to
work
with
two
procedures that
we
have
found
to be
extremely
useful:
artificial
grammar learning
and
probability
learning.
The
former
is
well
known
in the
literature
and has
been
used
by
many;
the
latter
is
somewhat obscure but,
as
will
become
clear,
is an
extremely
sensitive
technique that
has
provided
some
intriguing data.
It is
useful
to
provide
a
short
overview
of
each here
and to
outline
the
general procedures
for its
use.
Various
other techniques that have
found
their
way
into
the
laboratory,
such
as the
various procedures developed
by
such
219

220
ARTHUR
S.
REBER
workers
as
Lewicki
(see
1986a)
and
Broadbent
(see
Broadbent,
FitzGerald,
&
Broadbent, 1986),
are
introduced later.
Grammar
Learning
Figure
1
shows
one of the first
synthetic grammars used
along with
the
basic types
of
"sentences"
that
it can
generate.
This grammar
was
first
used
by
Reber
(1965, 1967).
It is a
Markovian
system derived
from
a
simpler system that formed
the
basis
of
George Miller's
Project
Grammarama
(see Miller,
1967)
and has
subsequently been used
to
generate
the
stimuli
for
a
number
of
other studies (Howard
&
Ballas,
1980;
Mill-
ward,
1981;
Reber
&
Lewis, 1977;
Roter,
1985).
It can be
taken
as
representative
of the
grammars used
in a
variety
of
other experiments.
Although
there have been many variations
on a
theme here,
the
basic procedure used
in
these grammar learning
studies
is
to
have
an
acquisition phase, during which subjects acquire
knowledge
of the
rules
of the
grammar,
and a
testing phase,
during
which
some
assessment
is
made
of
what they have
learned. Additional details
are
supplied
as
follows
when
needed.
Several points, however, need
to be
kept clear about these
synthetic
languages
and how
they have been used
to
examine
implicit, unconscious cognitive
processes.
First, they
are
com-
plex
systems,
too
complex
to be
learned
in an
afternoon
in
the
laboratory,
as
Miller
(1967)
noted. Miller
saw
this
as a
liability,
which
it is if one
wishes
to
examine explicit concept
learning. This complexity, however, should
be
regarded
as a
virtue
in the
current context,
for a rich and
complex stimulus
domain
is a
prerequisite
for the
occurrence
of
implicit learn-
ing.
If the
system
in use is too
simple,
or if the
code
can be
broken
by
conscious
effort,
then
one
will
not see
implicit
processes.
Second,
the
grammars given here
are finite-state
systems that generate strings
of
symbols
in a
left-to-right,
nonbierarchical
fashion.
This
fact
should
not be
taken
as
reflecting
any
prejudices
about
the
structural underpinnings
of
natural languages
or
their
acquisition.
We
elected
to use
finite-state
grammars
for
several
reasons
independent
of
the-
OUT
Figure
1.
Schematic
diagram
of a finite-state
grammar.
(Stimuli
are
generated
by
following
any
path
of
arrows
leading
from
the
initial
State
1 to the
terminal
State
6. The
following
are the
five
basic
strings
of
the
grammar
with
the
loops
or
recursions
in
brackets:
1.
T[S]XS;
2.
T[S]XX[[T]VPX]VV;
3.
T[SJXX[[T]VPX]VPS;
4.
P[[T]VPX]W;
5.
P[[TJVPXJVPS.)
oretical
issues
in
linguistics
or
natural language learning: They
are
mathematically
tractable;
they have
an
intrinsic probabi-
listic
structure that
is
well
known; they
can
generate
a
rela-
tively
large number
of
strings
to use as
stimuli; and,
as
mentioned, they
are
sufficiently
complex
so
that
the
under-
lying
formal structure
is not
within
the
conscious memorial
domain
of the
typical
subject
upon
the
subject's entering
the
laboratory. Finally,
as
will
become clear, there
is
nothing
special
about
these stimulus generators
in any
interesting
psychological sense.
The
basic components
of
implicit learn-
ing
emerge
in a
wide variety
of
different
empirical settings
with
a
range
of
different
stimulus environments.
Probability
Learning
(PL)
The
format
that
we
adopted
departs
noticeably
from
the
traditional two-choice procedure
in
which
each trial consists
of
a
"ready"
signal,
a
prediction response,
and an
outcome
event.
The
procedure used
to
explore implicit processes
de-
rives
from
the
proposition that
the
essential nature
of a PL
experiment
has
little
to do
with
the
explicit learning
of
prob-
abilities
of
events. Rather, what passes
in the
literature
for
probability learning
is
actually
a
much more subtle process
in
which
subjects learn implicitly about
the
stochastic structure
of
an
event sequence
to
which they have been exposed.
In the
course
of
making predictions, they mimic
its
structure
and
thereby
generate
a
sequence
of
responses,
one
by-product
of
which
is an
approximate matching
of the
probabilities
of the
events—that
is,
probability
learning.
Accordingly,
the PL
procedure
was
modified
as
follows
(Reber, 1966; Reber
&
Millward,
1968).
The
subject begins
an
experimental session simply
by
observing
the
occurrence
of
a
sequence
of
rapidly presented events. There
is no
ready
signal,
and the
subject makes
no
prediction
responses. In
this
situation,
a
passively observed event
is
functionally
equivalent
to a
traditional
trial,
and a
learning session consisting
of 2 or
3 min of
observing events
at a
rate
of two per
second
is
sufficient
to put a
subject
at an
asymptotic rate
of
responding;
that
is,
subsequent prediction
responses
made
by
subjects
who
have
had
this learning experience show
all of the
characteris-
tics
of
ordinary subjects
who
have
had an
equivalent number
of
traditional trials.
We
dubbed this procedure
the
instant
asymptote
technique.
The
typical experiment with this modified
PL
procedure
consists
of an
acquisition phase, during which subjects observe
event
sequences that
may
have
any of a
variety
of
stochastic
structures,
and a
testing phase, during which subjects make
prediction
responses.
As
with
the
grammar learning studies,
there
are
many variations
on
this
basic theme; they
are
introduced later
as
needed.
Despite
the
many
superficial
differences between
the
prob-
ability
learning paradigm
and the
grammar learning experi-
ment, there
are two
essential commonalities. First,
in
both
cases
the
subject
is
confronted with
a
stimulus environment
about which knowledge must
be
acquired
in
order
to
respond
effectively
during
the
testing
session.
Second, neither
of the
structural systems being used
is
part
of or
even remotely
similar
to the
epistemic
contents
of the
typical subject's long-
term
memory. These points
are
fundamental;
the
whole pur-

IMPLICIT
LEARNING
AND
TACIT KNOWLEDGE
221
pose
of
examining implicit learning
in the
laboratory
is to
develop
understanding
of how
rich
and
complex knowledge
is
initially obtained independently
of
overt, conscious strate-
gies
for its
acquisition. This
process
is
ubiquitous
in
human
experience
and
yet,
as a
focus
of
psychological inquiry,
it is
largely
ignored.
What
follows
is a
brief review
of our
work with these
two
procedures presented
in the
form
of
basic
issues that
the
literature addresses. These issues
are
integrated with
the
grow-
ing
body
of
literature
on
unconscious,
nonreflective,
implicit
processes
and,
finally,
summed
up in a
systematic attempt
to
see how
such cognitive systems could evolve
and how
they
fit
into contemporary struggles
with
a
number
of
classic
prob-
lems
in
pure
and
applied psychology.
The
work
in our
laboratory that
is the
focus
of
this overview
(beginning
with
the
very
first
studies
by
Reber,
1965,
1966)
was
carried
out
with
a
particular
research strategy:
to use a
limited number
of
techniques
to
examine
a
wide variety
of
effects.
The
virtue
of
this approach
is
that
by
developing
a
few
techniques
and
building
a
robust data base,
one can
explore
a
large number
of
issues
and not be
terribly concerned
about
the
vagaries
that
get
introduced with alternative pro-
cedures. Given that
the
problems
of
implicit learning
and
tacit knowledge
can be
explored through these
two
proce-
dures, this heuristic says that they should
be
used
in as
many
circumstances
as
makes scientific sense. Biological
fans
of
E.
coll
will
recognize
this
strategy.
There
is, of
course,
an
alternative strategy:
to
examine these
nonconscious
cognitive
and
perceptual processes
in as
wide
a
variety
of
experimental environments
as
possible.
The
virtue
of
this strategy
is
that
one is not
likely
to be
seduced
by
idiosyncratic properties
of
particular procedures; generaliza-
tions come easier.
If
implicit learning
is
real,
it
should emerge
in
contexts conceptually remote
from
synthetic grammars
and
structured event sequences. Ideally, both strategies should
be
carried out.
As
will become clear, those whose research
programs have taken
the
latter tack, such
as
Lewicki
(1985,
1986a,
1986b; Lewicki,
Czyzewska,
&
Hoffman,
1987),
have
typically
produced data that
are
congruent
and
complemen-
tary.
Empirical Studies
of
Implicit Learning
On the
Exploitation
of
Structure
When
a
stimulus environment
is
structured, people learn
to
exploit
that
structure
in the
sense that they come
to use it
in
order
to
behave
in a
relevant fashion
in its
presence. This
proposition seems
noncontroversial
as a
generalization about
human
cognition;
in
fact
it
lies
at the
core
of
several
ap-
proaches
to
perception
(Gibson,
1966; 1979; Mace, 1974),
decision making,
and
information processing
(Garner,
1974;
Hasher
&
Zacks,
1984)
and can be
seen
as
underlying,
in a
broad
sense,
any of a
number
of
genera]
theoretical
analyses
such
as
Anderson's (1983) production
systems,
Nelson's
(1986)
and
Schlesinger's
(1982)
models
of
natural language
acquisition, Fried
and
Holyoak's
(1984)
model
of
category
induction,
Lewicki's
(1986a)
analysis
of
socialization,
and,
interestingly,
Holland,
Holyoak,
Nisbett,
and
Thagard's
(1986)
model
of
induction
and
Rumelhart
and
McClelland's
(1986)
connectionist
systems.
The
latter
two are of
special
importance because each
can be
seen
as
advances
in the
cognitive
sciences'
hopefully
awakened interest
in
knowledge
acquisition.
In
an
early study
on
this general problem (Reber,
1967),
subjects were shown
to
become interestingly sensitive
to the
constraints
of a
synthetic grammar simply
from
exposure
to
exemplary
strings.
In
that experiment, subjects were
not in-
formed
that they were working
with
rule-governed stimuli.
They
were merely requested
to
memorize strings
of
letters
in
what
was
touted
as a
rote memory experiment. With practice
they
became increasingly adept
at
processing
and
memorizing
strings,
whereas
a
control group working with
nonordered
letter
strings showed
no
such improvement. Furthermore,
after
this neutral learning task, subjects were
able
to use
what
they
had
apprehended
of the
rules
of the
grammar
to
discrim-
inate between
new
letter strings that conformed
to the
gram-
matical constraints
and
letter
strings that violated
one or
more
of
the
rules
of the
grammar.
In
simplest terms, these subjects
can
be
said
to
have been exploiting
the
structure inherent
in
the
stimulus display. This basic
finding is a
robust
one and
has
been
reported by
numerous authors (e.g., Brooks, 1978;
Dulany,
Carlson,
&
Dewey, 1984; Howard
&
Dallas,
1980;
Mathews
et ah, in
press;
Millward,
1981;
Morgan
&
Newport,
1981).
Similar
observations concerning
the
exploitation
of
struc-
ture
have been made
in
somewhat
different
contexts
by
other
researchers.
Broadbent
and his
co-workers
showed that knowl-
edge
of
complex rule systems governing simulated
economic/
production systems
is
also acquired
and
used
in an
implicit
fashion
(Berry
&
Broadbent, 1984; Broadbent
&
Aston,
1978;
Broadbent
et
al.,
1986).
In
those studies, subjects
were
pre-
sented with
an
imaginary manufacturing situation
such
as a
sugar
production plant
and
instructed simply
to
maneuver
variables
such
as
wages, labor peace, worker output,
and the
like
to
yield
a
particular overall production standard.
The
systems,
in
fact,
operate according
to
sophisticated, complex
rule
systems that relate these
factors
to
each other. Achieving
the
required production standards requires that
the
rules
be
"known,"
in
some sense
of
that word. Broadbent
and his
colleagues consistently reported that subjects induced
the
rule
systems implicitly
and
made
appropriate
adjustments
in the
relevant
variables
and did so in the
absence
of
conscious
knowledge
of the
rules themselves.
The
pattern
of
these
find-
ings
strongly parallels those
in the
synthetic grammar learning
studies.
Several
of the PL
studies
have
also yielded analogues
of
this
process. Reber
and
Millward
(1971)
found
evidence that
subjects
can
accurately anticipate
the
changing probabilities
of
events even when
the
anticipatory
response requires
an
integration
of
information across
50
preceding events.
In
this
particular
case,
the
probability
of
each individual event
on
any
Trial
n
was
systematically increased
and
decreased
as
n
moved
through
a
period
of 50
trials. Subjects were
first
given
1,000
instant asymptote trials with this sawtooth event
se-
quence
and
then requested
to
predict successive events. Under
these conditions,
rather
than shadowing
the
changing event

222
ARTHUR
S.
REBER
probabilities,
subjects ultimately learned
to
anticipate
the
shifts
in the
likelihood
of
events
so
that their predictions
of
events rose
and
fell
coincidentally
with
the
actual event
se-
quences. They
had
learned
the
underlying structure
of the
stimulus environment
and
were capable
of
exploiting
it to
direct their
choices.
Millward
and
Reber
(1972),
using event sequences with
short-
and
long-range
contingencies
between events,
reported
an
even more impressive ability
of
subjects
to
exploit sto-
chastic
structures. Subjects were exposed
to
sequences that
contained event-to-event dependencies such
that
the
actual
event
that
appeared
on
any
Trial
n was
stochastically depend-
ent
on
the
event that
had
occurred
on
some previous Trial
n
—
j,
where
j = 1, 3, 5, or 7.
Training consisted
of
several
hundred
trials
of the
instant asymptote procedure with
the
particular
stochastic dependency
for
that
session. During
the
first
session,
j = 1;
during
the
second,
j = 3; and so
forth.
During
testing,
subjects displayed
a
clear sensitivity
to
these
dependencies,
a
sensitivity that reflected
an
ability
to
exploit
structure that required knowledge
of
event dependencies
as
remote
as
seven
trials. What makes this
finding
interesting
is
that this capacity appears
to be
beyond what were
found
in
earlier work (Millward
&
Reber, 1968; Reber
&
Millward,
1965)
to be
limits
on
explicit recall.
In
those experiments,
subjects were asked
to
recall which event
had
occurred
on a
specified
previous trial. Beyond
five
trials, they were virtually
reduced
to
guessing.
Parallel
findings
were recently reported
by
Lewicki
and his
co-workers
(Lewicki
et
al.,
1987;
Lewicki, Hill,
&
Bizot,
1988),
who,
using reaction time measures, showed that subjects
implicitly
knew
the
future
location
of a
stimulus event
even
though,
when
they were asked
to
report explicitly where
the
event
would occur, their performances were
no
better than
chance.
Lewicki
and his
colleagues used
a
complex rule that
specified
which
of
four
quadrants
of a field
would contain
a
target number.
The
actual location
was
based
on the
pattern
of
quadrants
in
which
the
number
had
appeared
on
particular
earlier
trials.
The
situation, like
the PL
studies,
was
based
on
an
arbitrary relationship between events. With extended prac-
tice, subjects showed dramatic decreases
in
reaction time
to
respond
to the
location
of the
target number.
The
improve-
ment
was
clearly
due to
tacit
knowledge
of the
stochastic
relationship
and not
simply
to
increased
facility
with
the
task.
Changing
the
rules produced
an
abrupt increase
in
reaction
time,
but
when,
in a
postexperimental
debriefing session,
subjects
were given
an
opportunity
to
consciously predict
the
critical quadrants, their performances were
no
better than
chance.
An
intriguing element
in
these studies
was
that
all of
Lewicki
et
al.'s
(1988)
subjects were
faculty
members
in a
department
of
psychology,
and all
knew
that
the
research
they
were
involved
in was
oriented
toward the
study
of
noncons-
cious
cognitive processes.
Clearly,
subjects learn
to use the
structural relationships
inherent
in
these various complex stimulus domains.
No
real
surprises
here.
In
many
ways
these various studies
function
basically
as
complicated existence demonstrations; they show
that
it is
possible
to
obtain this kind
of
unconscious, nonre-
flective,
implicit learning
in a
controlled laboratory setting,
that
it can
occur
in a relatively
short time span,
and
that
it
can be
seen
to
emerge when
the
stimulus
is a
structured
domain whose content
is
arbitrary
and
distinctly
remote
from
typical day-to-day experiences with
the real
world.
On
Implicit
Versus
Explicit
Processes
The
experiments reviewed were
all run
under instructional
sets
in
which
subjects were unaware that
the
stimuli were
structured
or
rule defined.
In
these cases,
the
point
was to
maximize
the
emergence
of
implicit learning.
It
is
important
to be
clear about this issue
and the
kinds
of
manipulations that have
been
used.
It is
universally accepted
that
the
college undergraduate whose cognitive
processes
form
(for
better
or
worse)
the
foundations
of our
science
is an
active
and
consciously probing organism, especially
in
regard
to
things with structure
and
'patterns.
The
aforementioned
researchers carried
out
their experiments
by
carefully
circum-
venting
this pattern-searching tendency.
:In
Reber
and
Mill-
ward's
PL
studies
and in
Lewicki's
target location experi-
ments,
they accomplished this
by
"blitzing"
the
subjects with
information
at
rates beyond those
at
which conscious code-
breaking
strategies could
operate.
The
grammar learning stud-
ies
and
Broadbent's
production system experiments were
successful
because
the
structure
of the
stimuli
was
highly
complex
and the
instructions
to the
subjects were calculated
to be
vague.
An
obvious question
is,
What
effect
would
explicit
instructions have? What happens when
subjects
are
informed,
at the
outset, that
the
materials that they
will
be
working
with
reflect
regularities
and
patterns?
The first
manipulation
of the
factor
of
explicitness used
the
PL
technique (Reber, 1966; Reber
&
Millward, 1968).
The
procedure consisted
simply
of
telling some
of the
subjects
exactly
what
was
going
on in the
experiment. Specifically,
by
informing
one
group
of
subjects
of the relative
probabilities
of
the two
events,
the
researchers gave them concrete instruc-
tions about
the
frequency
characteristics
of the
event sequence
that
they
would
be
asked
to
predict. These subjects were then
run
in a
standard
PL
procedure
and
compared with
a
control
group
run
with
the
same event sequences
but
without
the
explicit
information.
The
information about
the
event frequencies
had
virtually
no
effect
on
behavior.
The two
groups
were
statistically indis-
tinguishable
from
each other,
even
on the
first
block
of 25
prediction trials
in
which
the
impact
of the
instructions would
have
been most
likely
to be
felt.
Clearly, probability learning
is
more than
the
learning
of
probabilities. Rather,
what
really
goes
on is the
apprehension
of
deep information about
the
structure
of the
sequence
of
events.
Postexperimental
debriefings
were revealing.
Subjects
were
quite
clear about knowing
which
light would
be the
dominant
one,
and all
said that they believed
the
instructions. But,
in
virtually
every
case,
they claimed that somehow
the
specific
information
lacked meaning that they
felt
they
could use.
It
took
real
experience with
the
event sequence
to
acquire
a
knowledge
base that
was
usable
for
directing choices
on
individual
trials.
This
experiment
used
Bernoulli sequences;
there
was no
"structure"
in the
usual sense
of the
word.
Nevertheless,
subjects
reported achieving
a
sense
of the
nature
of
the
event sequence
from
experience with events that they
did not
derive
from
the
explicit
instructions.
Of
importance

IMPLICIT
LEARNING
AND
TACIT
KNOWLEDGE
223
is
that this occurred despite
the
fact
that,
in
principle, there
is
nothing
to be
extracted
from
the
event sequence other than
the
relative
frequencies
of the two
events.
Several
studies using
the
grammar learning procedure have
also explored
the
boundary between implicit
and
explicit
knowledge.
In
the first
experiment,
Reber
(1976)
used
the
simple device
of
encouraging
one
group
of
subjects
to
search
for
the
structure
in the
stimuli
while
a
comparable
group
was
run
under
a
neutral instructional set. Both groups were given
the
same learning phase, during which they
had to
memorize
exemplars
from
a
synthetic grammar,
and an
identical testing
phase, during which they were asked
to
assess
the
well-
formedness
of
novel
letter
strings.
Well-formedness
is
defined
by
whether
the
grammar could generate
the
string.
Informed
subjects
were told
only
about
the
existence
of
structure; noth-
ing
was
said
about
the
nature
of
that
structure.
The
explicitly instructed subjects
in
this study performed
more poorly
in all
aspects
of the
experiment than
did
those
given
the
neutral instructions. They took longer
to
memorize
the
exemplars, they were poorer
at
determining well-formed-
ness
of
test strings,
and
they showed evidence
of
having
induced rules
that
were
not
representative
of the
grammar
in
use.
The
suggestion
is
that
at
least under these circumstances,
implicit processing
of
complex materials
has an
advantage
over explicit processing.
However,
as
gradually became clear, what this study
ac-
tually
showed
is
that explicit processing
of
complex materials
has a
decided
disadvantage
in
relation
to
implicit processing.
This
is no
mere play
on
words.
The
implicit/explicit distinc-
tion
is
rather more complex
than
it first
appeared.
Analysis
of
the fine
grain
of the
data
from
Reber's
(1976)
article
revealed that
the
explicit instructions seemed
to be
having
a
particular kind
of
interference
effect.
Specifically,
subjects
were
being encouraged
to
search
for
rules that, given
the
nature
of finite-state
grammars with their path-independent,
Markovian
properties
and
given
the
kinds
of
attack strategies
that
the
typical undergraduate
possesses,
they were
not
likely
to find.
Moreover, they tended
to
make improper inductions
that
led
them
to
hold rules about
the
stimuli that were,
in
fact,
wrong.
The
simplest conclusion seems
to be the right
one: Looking
for
rules
will
not
work
if you
cannot
find
them.
In
a
number
of
other
studies,
instructions
of
various
kinds
have
been shown
to
have
any of a
number
of
effects.
Brooks
(1978)
used
finite-state
grammars similar
to the one
used
by
Reber
(1976)
and a
paired-associates
learning procedure
in
which
strings
of
letters
from
grammars were paired with
responses
of
particular kinds (e.g., animal names, cities).
He
found
that
informing
subjects
about
the
existence
of
regular-
ities
in the
letter strings lowered overall performance. Reber,
Kassin,
Lewis,
and
Cantor
(1980)
found
poorer performance
with
explicit instructions when
the
stimuli were presented
in
a
large, simultaneous array
in
which letter strings were posted
on
a
board
in
haphazard fashion. Howard
and
Ballas
(1980)
reported detrimental
effects
of
explicit instructions with struc-
tured
sequences
of
auditory stimuli
when
there
was no
system-
-
atically
interpretable
pattern expressed
by the
stimulus
se~
quences.
In all
these
cases,
the
original
finding
was
basically^
replicated.
However,
Millward
(1981)
found
no
difference
between
explicitly
and
implicitly instructed subjects
in an
experiment
that,
in
principle, looked like
a
replication
of
Reber's
(1976)
with
the
seemingly modest variation that
the
strings used
during learning were
up to
11
letters long, whereas
in
Reber's
study
stimuli were
no
longer than
8
letters.
Abrams
(1987),
using
a
strict replication
of
Reber's procedure with
the
excep-
tion that
the
study
was run on a
computer,
failed
to find the
instructional
effect.
Dulany
et
al.
(1984)
reported
no
signifi-
cant differences between
the two
instructional
sets,
although
in
this case
a
rather
different
testing procedure
may
have
masked
differences.
Mathews
et al. (in
press)
found
a
complex
pattern
of
differences
between instructional
groups,
depending
on
whether
the
letter
set
used
to
instantiate
the
grammar
was
modified
over
the
several days
of the
experiment. They also
used
a
grammar somewhat more complex
than
is
typical.
Danks
and
Gans
(1975)
reported
no
differences
when
they
used
a
synthetic system
that
was
considerably
simpler than
the
Markovian systems used
by
others.
Last, several studies showed
an
advantage
for the
explicitly
instructed subjects. Howard
and
Ballas
(1980)
reported that
the
explicit instructions that
debilitated
performance
when
introduced
under conditions
of
semantic
uninterpretability
could also
function
to
facilitate
performance when
the
stimuli
expressed
semantically
interpretable patterns. Reber
et al.
(1980)
showed that
it was
possible
to
shift
performance about
rather dramatically
by
intermixing instructional
set
with
the
manner
of
presentation
of the
stimulus materials
and
with
the
time during learning when
the
explicit instructions were
introduced.
There
are two
factors here that
help
make these data
somewhat
less haphazard than
they
appear
to be. The first is
psychological salience;
the
second
is the
circumstances under
which
the
instructions
are
given
to the
subjects.
The first of
these
is the
more interesting
and the one
from
which insight
into
process
can be
gained.
In
two of the
instances
in
which explicit instructions
facil-
itated performance,
the
manner
of
presentation
of the
stimuli
was
such that
the
underlying
factors
that represent
the
gram-
mar
were rendered salient.
In
Howard
and
Ballas's
(1980)
study,
the
semantic component focused
the
subjects
on the
relevant
aspects
of the
patterned stimuli.
The
effectiveness
of
such
a
semantic component
has
often
been noted
in
artificial
grammar learning studies (Moeser
&
Bregman,
1972;
Morgan
&
Newport,
1981).
In
Reber
et
al.'s
(1980) study,
the
simple
expedient
of
arranging
the
exemplars
of the
grammar accord-
ing
to
their underlying
form
produced
the
instructional
facil-
itation. Moreover, several other researchers apparently
ar-
ranged
matters, inadvertently,
so
that structural properties
became more salient.
In
Millward's
(1981)
study,
for
example,
the use of
longer strings provided many opportunities
for
subjects
to be
exposed
to the
loops
or
recursions
in the
grammar
(see Figure
1)
and
thereby increased
the
psycholog-
ical
salience
of the
underlying structure.
In
Danks
and
Gans's
(1975)
study,
the relatively
simple nature
of the
stimuli
likely
acted
to
equate
the
mode
of
processing
of the
stimuli
in
both
groups;
that
is,
both groups were likely
using
a
reasonably
explicit
mode independently
of the
instructions. Hence
the
converse
of the
earlier conclusion: Looking
for
rules will
work
if
you can find
them.
Some
cases appear
to be
genuine
failures
to
replicate
the
original
finding:
specifically,
those
of
Dulany
et al.
(1984)
and

224
ARTHUR
S.
REBER
Abrams
(1987).
In
Dulany
et
al.'s
study,
the
procedure used
during
learning should,
in
principle, have yielded
a
difference
during
testing.
These
are
some
interesting
aspects
of
this
experiment,
but
there
are no
obvious reasons
why the
effect
failed
to
emerge.
On the
surface,
Abrams's
study,
which
was
carried
out in our
laboratory, should also have produced
an
instructional
effect.
In
that study, however, there
were
reasons
for
suspecting
the
computer used
to run the
study
as the
culprit.
This
remains
to be
explored,
but
simply presenting
the
explicit instructions
on the
computer screen
may not
have
the
compelling quality that
a
"real"
experimenter
reading
them
has.
As
other work suggests
(Reber
&
Allen,
1978),
implicit learning
is the
default
mode;
therefore,
if the
subjects
do
not
understand
or do not
believe
the
instructions, then
no
differences
would
be
expected. This suggests
a
possible
im-
portant methodological
factor
that
has
gone largely unnoticed:
the
sophisticated equipment that
we use
commonly
in our
laboratories
may
be
having untoward
effects
on our
studies.
In
any
event,
the
literature
on the
implicit/explicit problem
is
clearly complex,
and it
takes
but a
moment's
reflection
to
appreciate
the
fact
that
there
are
still other important issues
lurking
behind these
findings.
First,
it
seems clear that
any
number
of
confounding
factors
may
influence,
either posi-
tively or
negatively,
the
impact
of
explicit instructions (cf.
Lewicki,
1986a). Such instructions
may
introduce
an
element
of
stress
or
anxiety,
may
evoke
a
sense
of
motivation,
may
encourage
one or
another conscious strategy,
and the
like.
To
date,
few
of the
researchers mentioned have taken such
factors
into
account,
and not
much
is
known
about
how
conscious,
explicit
processing systems interact with
the
implicit
and
unconscious.
For
example,
a
recent study suggests that using
instructions that engage
the
explicit system
may
also elicit
anxiety
and
that
anxiety
may be
related
to
poor performance
on
a
standard grammar learning task
(Rathus,
Reber,
&
Kushner,
1988).
Second,
it
seems clear that
we are
still dealing with
a
rather
limited
kind
of
analysis
of
complex learning, particularly
if
one
wishes
to
view
this
research in its
constrained laboratory
setting
as
representing
a
general metaphor
for
real-world
acquisition processes.
In the
real
world
nearly
all
complex
skills
are
acquired with
a
blend
of the
explicit
and the
implicit,
a
balance between
the
conscious/overt
and the
unconscious/
covert.
There
is
surely
a
difference
between simply
informing
a
learner that
the
stimulus materials have structure,
as re-
searchers
in the
aforementioned experiments did,
and
telling
the
learner something
definitive
about that structure.
The
next
section deals with this issue.
Effects
of
Providing
Specific
Information
This issue concerns
the
impact
of
giving
subjects precise
information
about
the
nature
of the
stimulus display that they
will
be
exposed
to.
This
is an
important question
for a
number
of
reasons.
For
one,
this issue broaches
on
some
of the
classic
questions
in
pedagogic theory about
how
best
to
convey
highly
complex
and richly
structured
information
to
students.
It
also
emerges
in
various studies
on the
acquisition
of
expertise
in
such
areas
as
medical diagnosis,
in
which
the relationship
between
specific
knowledge presented
to
medical students
and
their emergent tacit knowledge base
is
turning
out to be
most
complex
(see,
e.g.,
Carmody,
Kundel,
&
Toto, 1984).
Reber
et
al.
(1980)
attempted
to
address
this issue
by
using
the
standard grammar learning procedure.
In
that study,
subjects
were
presented with
the
actual schematic structure
of
the
grammar; that
is,
they
were
presented with Figure
1.
Each
subject
was
handed
the
diagram
and
given
a
7-min
"course"
in
how
such
a
structure
can be
used
to
generate strings
of
symbols. This procedure
was
supported
by an
observation
period during
which
a set of
exemplars
was
shown
to the
subjects.
In
this training
format
a
maximally explicit learning
procedure
was
thus mixed
with
a
maximally implicit one.
Reber
et al.
(1980)
explored
the
manner
of
interaction
between
these
two
modes
of
apprehension
by
introducing
the
explicit
training
at
different
points
in the
observation
period.
One
group
of
subjects received
the
explicit instruction
at the
outset
before
any
exemplars
were
seen;
one
group received
it
part
way
through
the
observation
period;
and for a
third
group,
the
explication
of
structure
was
delayed until
after
they
observed
the
full
set of
exemplars.
As in the
typical grammar
learning
study, knowledge acquired during learning
was as-
sessed
by
means
of a
well-formedness
task.
The key
finding
was
that
the
earlier during
the
observation
training
the
explicit instructions
were
given,
the
more
effective
they
were. From
the
previous discussions
it is
clear that
increasing
the
salience
of the
relationships between symbols
increases
the
effectiveness
of
subjects'
attentional
focus.
It is
also
clear that instructions that encourage
the
subject
to
deal
with
the
stimuli
in
ways
that
are
discoordinate
with
underlying
structure have detrimental
effects
on
acquisition. Thus
the
explication
of the
precise nature
of the
structure underlying
the
stimuli must have
differential
impact
on
learning,
de-
pending
on
which
of
these
two
processes
is
encouraged.
The
most plausible interpretation,
and the one
that
has
interesting implications
for
theories
of
instruction,
is
that
the
function
of
providing explicit instructions
at the
outset
is to
direct
and
focus
the
subjects'
attention.
It
alerts them
to the
kinds
of
structural relations that characterize
the
stimuli that
follow
and
permits appropriate coding schemes
to be
imple-
mented. These instructions
did not
teach
the
grammar
in any
full
or
explicit
fashion;
rather, they oriented
the
subjects
toward
the
relevant
invariances
in the
display that
followed
so
that
the
subjects,
in
effect,
taught themselves.
Accordingly,
when
such explicit instruction
is
introduced
later
in the
observation period,
its
effects
are
different
because
two
sources
of
difficulty
are
introduced.
First,
it
imposes
a
formalization
of
structure that
is, in all
likelihood, discoordi-
nate
with
the
tacit system that
was in the
process
of
being
induced.
Second,
it reduces the
number
of
exemplars that
can
be
used
as a
base
for
extracting
invariance
patterns.
In the
extreme
case
in
which
the
instructions were delayed until
the
completion
of the
observation period, this informational base
is
virtually
eliminated.
These points
can use
some
further
exploration. There
is
every
reason
to
suspect that subjects' tacit representation
of
rules
is
idiosyncratic
in
various characteristics.
The
induction

tMPLICIT
LEARNING
AND
TACIT
KNOWLEDGE
225
routines hypothesized
by
Holland
et
al.
(1986)
predict
this
personalized
aspect
at the
outset.
From
earlier
work,
it is
clear
that
subjects
are
known
to use a
wide variety
of
coding
schemes
in
focusing their
attention
on the
stimuli (see
Allen
&
Reber,
1980;
Reber
&
Allen, 1978;
and
Reber
&
Lewis,
1977,
for
details).
So
long
as
these schemes
do not
entail
inappropriate rule formation, their impact
is
superficial.
In-
dependently
of
individualistic mnemonics,
attentional
focus-
ing
priorities,
or
preferred rehearsal
strategies,
the
implicit
learner will emerge
from
the
training session with
a
tacit,
valid
knowledge base coordinate
with
the
structure
of the
stimulus
environment.
The
degree
to
which
the
explicit instructions introduce
difficulties
will
thus
be
dependent
on the
extent
to
which
the
subject's
tacit
representation
of
structure matches
the
formal-
ization
provided
by the
schematic
of the
grammar
and the
accompanying characterization
of its
generational properties.
In
Dulany
et
al.'s
(1984)
terms, subjects
are
learning
"corre-
lated grammars" whose properties are,
in all
likelihood,
not
commensurate
in any
simple
way
with
the
Markovian
system
in
use. Recent results
from
Mathews
et al. (in
press) strongly
corroborate
this
interpretation.
The
deep
difficulty
here
is
that
there
is a
potentially
infinite
number
of
formalizations
that could account
for the
structure
displayed
in any
given
subset
of
strings
from
one of
these
grammars; present
the
"wrong"
one to a
subject,
and the
instructions
will
not
have
a
salutary
effect.
The
problem
is
apparent:
How
much
can we
expect
a
subject
to
benefit
from
the
specific information that
the set
of
exemplars just observed
and
tacitly coded
as,
say,
bigram
covariation patterns
is "in
reality"
to be
formalized
as a
Markovian process?
To
take
an
obvious
analogy, most
of us
with
our
extensive observation
and
generation
of
utterances
in
English
have
failed
to
derive
any
facilitative
effect
of
explicit instruction
with
transforma-
tional grammar that,
at
least
in
principle,
can be
posed
as a
legitimate
formalization
of our
tacit knowledge. Moreover,
such explicit awareness
of
structure
can
actually
be a
nuisance
when
one
tries
to
fulfill
the
kinds
of
demands placed
on
subjects
in
these experiments,
as in the
discrimination
of
well-
formed,
novel
instances
from
instances
that
contain
some
violation
of the
formal
system.
In
a
study that addressed this point
directly,
Bialystofc
(1981)
found
that subjects learning French
as a
second lan-
guage
could rapidly
and
accurately detect
ungrammatical
sentences
and
could
do so
largely
independently
of the
com-
plexity
of the
grammatical rules violated. However,
when
asked
to
characterize
the
nature
of the
violated rules,
the
complexity factor played
a
significant
role. Complexity,
of
course,
is
defined
in
such cases
by the
kinds
of
grammars that
are
taught
in
"French
as a
second language" courses.
In
summary, although there
are not a lot of
hard empirical
data here, those that
are
available point toward
an
interesting
conclusion.
Specific
instruction concerning
the
materials
to
be
learned
in
complex situations will
be
maximally beneficial
when
it is
representationally
coordinate
with
the
tacit knowl-
edge
derived
from
experience. Because this issue
is
ultimately
critical
for
theories
of
instruction,
it is one
much
in
need
of
close
examination.
On
Deep
and
Surface
Structure
The
issue here
is the
degree
to
which
implicit
learning
can
be
seen
as
acquisition
of
knowledge that
is
based
on the
superficial
physical
form
of the
stimuli
or as
knowledge
of the
deeper, more abstract relations that can,
in
principle,
be
said
to
underlie them.
In
an
early
article,
Reber (1969), reported evidence
for the
proposition that implicit knowledge
is
abstract
and not de-
pendent
in any
important
way on the
particular physical
manifestations
of the
stimuli. This study consisted
of two
sessions during which subjects memorized letter strings
from
a
grammar.
When
the
second session began,
the
stimulus
materials were, without
warning,
modified.
For
some subjects
the
same letters continued
to be
used,
but the
rules
for
letter
order
were
now
those
of a
different
grammar (the
"syntax"
was
altered).
For
other subjects
the
underlying structure
was
not
tampered
with,
but the
letters used
to
represent
the
grammar
were
replaced
with
a new set
(the "vocabulary"
was
changed).
The two
obvious control groups,
one for
which
both aspects were altered
and one for
which neither
was
changed, were
also
run.
The
various manipulations
had
sys-
tematic
effects
on
subjects'
ability
to
memorize stimuli
in the
second session. Modification
of the
rules
for
letter order
produced decrements
in
performance; modifications
of the
physical
form
had
virtually
no
adverse
effects.
So
long
as the
deep rules that characterized
the
stimuli were
left
intact,
their
instantiations
in the form of one or
another
set of
letters
was
a
factor
of
relatively little importance.
The
recent study
by
Mathews
et al. (in
press) supported this
general
finding.
Their experiment
was run
over
a
4-week
period. Subjects
who
received
a new
letter
set
each
week
(which
was
based
on the
same underlying syntax) performed
as
well
as
subjects
who
worked
with
the
same
letter
set
throughout
the
course
of the
experiment.
The
effect
was
quite
striking;
the
transfer
from
letter
set to
letter
set
occurred
smoothly
and
apparently without
the
need
for any
conscious
translation.
Reber
and
Lewis
(1977) reported
an
equally striking
ex-
ample
of the
abstract
nature
of
tacit
knowledge. They
assessed
knowledge
of the
grammar
by
having subjects solve anagram
problems.
After
a
standard training session during which
subjects
memorized exemplars
from
the
language,
they
solved
anagrams
from
the
synthetic language over
a
4-day
period.
For
reasons
to be
discussed,
it is
convenient
to
code letter
strings
in the
form
of
bigrams
and to
note
the
rank order
of
frequency
of
occurrence
of
each possible bigram.
For
example,
the
string
prrrw
contains bigrams
FT,
TV,
and vv
once
each
and
TT
twice. Given
how
this experiment
was
run, three rank
orders
of
frequency
of
occurrence
of
bigrams exist:
(a) one
based
on the
actual solutions
offered
by the
subjects (cor-
rected,
of
course,
for
guessing),
(b) one
based
on the
frequency
of
occurrence
of
each acceptable bigram within
the
artificial
language
itself (within
the
string lengths used),
and (c) one
based
on the
actual bigrams that appeared
in the
learning
stimuli.
Rank-order
correlations among these three
were
revealing.
The
correlation between
(a) and (b) was
.72, whereas that

226
ARTHUR
S.
REBER
between
(a) and (c) was
only .04.
The
interesting point about
these results
is
that
the
comparison between
(a) and (b) is a
comparison between subjects' usable knowledge
and a
deep
representation
of the
frequency
patterns
of the
grammar.
Rank-order
(b) was
formed
on the
basis
of the
full
set of
acceptable strings that
the
grammar could,
in
principle, gen-
erate within
the
specific string lengths. Subjects,
however,
never
saw
this
full
set of
strings; they
were
exposed only
to
the
exemplars chosen
for the
training sessions. These partic-
ular strings were selected
to
ensure that each
subject
saw at
least
one
string
of
each possible length
and,
for
each length
when
it was
possible,
at
least
one
example
of
each
of the
grammar's three loops. This procedure yielded
a set of
strings
that displayed
all of the
deep structural characteristics
of the
finite-state
language but,
in
terms
of
specific
frequency
of
bigrams,
was
distinctly idiosyncratic.
The
comparison between
(a) and (c) is a
reflection
of the
degree
to
which
subjects
are
simply
keying
on the raw
fre-
quency
data
as
displayed
in the
exemplars.
The
failure
for
this correlation
to be
different
from
zero suggests that subjects
were
not
solving
the
anagrams
on the
basis
of
superficial
knowledge
of
frequency
of
bigrams
or on the
basis
of a fixed
set of
memorized instances. They clearly acquired knowledge
that
can be
characterized
as
deep,
abstract,
and
representative
of
the
structure inherent
in the
underlying
invariance
patterns
of
the
stimulus environment.
This
finding is
analogous
to
Posner
and
Keele's
(1968,
1970}
oft-cited abstraction
of
prototype
effect.
The
underlying
prototypes
that
their
subjects
extracted
from
the
exemplary
dot
patterns
are
specifiable only
in
terms
of an
averaging
of
the
spatial relations among
the
various components
of the
patterns. But, psychologically, such
an
averaging
is not
just
a
simple
piling
up of the
features
of the
exemplars.
If
memory
behaved
like
that,
the
resulting representation would
be, not
distinct prototypes that Posner
and
Keele
found,
but a
blob.
The
induction routine that appears
to be
operating
in
situations such
as
these
is
necessarily
one
that results
in an
abstract representation. Moreover,
it is one
that
is
applicable
to
the
classification
of
novel instances
and not
specifically
characterizable
by a raw
compilation
of
experienced instances.
This issue
is one of
considerable complexity.
The
point
of
the
preceding argument
is not
that
all
memorial systems must
be
viewed
as
founded
on
induced
abstractions.
The
evidence
of
Brooks
(1978)
and
others
(cf. Smith
&
Medin,
1981)shows
that memories
are
frequently
based
on
instantiations,
fairly
uninterpreted
representations
of the
stimulus inputs.
The
point
is
that
when
implicit acquisition processes
are
operating,
the resulting
memorial system
is
abstract.
As was
shown
elsewhere
(Allen
&
Reber,
1980;
Reber
&
Allen,
1978),
the
same
subjects working with
the
same grammars
can
emerge
from
a
learning session
with
either
an
instantiated memory
system
or an
abstract one, depending
on the
learning proce-
dures
used.
In
those articles, that
old war
horse
functionalism
was
shown
to
provide
the
best characterization
of
this
issue;
that
is, the
specific
functions
that need
to be
carried
out
invite
the
learner
to
assume
a
cognitive stance that
is
functional,
that will accomplish
the
task
at
hand. Under some circum-
stances (such
as the
paired-associates
learning procedure used
by
Brooks, 1978,
and by
Reber
&
Allen,
1978),
a
rather
concrete, instantiated memorial system
will
be
established;
under
others (the instant asymptote technique
in the PL
studies
by
Reber
&
Millward, 1968,
and the
observation
procedure
in
Reber
&
Allen's
1978
study),
a
distinctly abstract
representation
will
emerge.
In the
pure implicit, unconscious
acquisition mode,
the
default
position
is
abstraction.
On
Mental
Representation
As
Rosch
and
Lloyd (1978) pointed out, sooner
or
later
every
discourse
on
mental process
and
structure must come
to
grips
with
the
problem
of the
form
of the representation of
knowledge
held.
Such
discussions must begin
with
some
presumptions.
The
ones introduced here are,
of
course, open
to
emendation
as
understanding processes. They
are
taken
as
the
starting point simply because
they
have considerable
explanatory
power, more than most contemporary
cognitiv-
ists
have granted.
First,
the
general argument
put
forward
by
such
diverse
theorists
as
Gibson (1966,
1979),
Garner (1974, 1978),
and
Neisser
(1976),
that
the
stimulus
is
more than
the
physical
setting
for the
occurrence
of a
response,
is
taken
as a
given.
This point
is
more than
a
simplistic swipe
at
behaviorism;
it
is
an
argument that stresses
that
the
stimulus domain within
which
we
function
is
extraordinarily
rich and
complex and,
in
all
likelihood, much more
so
than most
cognitivists
have
been willing
to
recognize.
The
underlying causative nature
of
the
stimulus environment
is
rarely explored; most theorists
are
satisfied
with
characterizations that
are
theory driven.
Second,
there
is
general agreement with
the
arguments
put
forward
(in
rather
different
forms,
to be
sure)
by
Palmer
(1978)
and
Anderson (1978, 1979, 1983)
to the
effect
that
most theoretical attempts
to
deal with
the
representation issue
are
misguided. Palmer maintained that
the
confusion derives
from a
failure
to
deal directly
with
metatheoretical
factors
concerning
existing models. Anderson argued that
in
princi-
ple, there
are no
ways
in
which
behavioral data
can be
used
to
identify
uniquely
any one
particular mental
representation.
There
are
some reasons
for
perhaps disputing these claims
(see,
e.g., Hayes-Roth, 1979;
Pylyshyn,
1979,
1980),
but
they
are
not a
concern here.
From
the
point
of
view that
I
take
as
presumptive here,
it
matters
not at all
whether
the
following
interpretations
of
mental
representation are
supported
by a
well-structured con-
sideration
of
metarepresentational
factors
or
whether they
can
be
shown
to be
uniquely specifiable.
The
point
of
view
that
I
take
reflects
that
of.classical
functionalism
as
introduced
in
the
preceding section. Functional theories
are
typically
re-
garded
these days
as
formulations (abstract,
to be
sure)
of
what
is
possible
for a
person
to
process
and
why. This seems
right,
and
as has
been argued elsewhere (Allen
&
Reber, 1980;
Reber
&
Allen,
1978),
the
main consideration should
be
with
characterizing representations,
in
terms
of how the
individual
can
be
seen
as
behaving
in an
adaptive
fashion,
rather than
in
terms
of
pure
representational
theory.
For
example,
as
discussed earlier, there
are
good empirical reasons
for
regard-
ing
the
functional
representation
of the
mental content
of a
finite-state
grammar
as an
ordered
set of
bigrams (and
tri-
grams;
see
Mathews
et
al.,
in
press)
and not as a
formal
Markovian
system.

IMPLICIT LEARNING
AND
TACIT KNOWLEDGE
227
Third,
the
oft-dismissed
position
of
representational realism
is
accepted
as a first
approximation. What
is in the
stimulus
world
is
what ends
up in the
mind
of the
perceiver/cognizer.
The
point
is
that
a
good
way to
start dealing with
the
repre-
sentation problem
is
with
the
physical stimulus itself. Under
various constraints
of
processing
and
various task demands,
enrichment
and/or
elaborative
operations
are
certainly used,
and the
resulting
coded
representation
may
very
well
not be
isomorphic
with
the
stimulus
field.
Nevertheless,
as
Mace
(1974)
put it, a
good initial strategy
is to
"ask
not
what's
inside
your head,
ask
what your
head's
inside of."
Several
findings
from
studies with
artificial
grammars
are
relevant
to the
issue
of
representation. Table
1
gives
the
summary
data
from
14
separate experiments that
reveal
some
interesting
patterns. Some
details
on
procedure
are
needed:
In
all of
these studies, knowledge acquired during learning
was
assessed
through
the
well-formedness
task
in
which
sub-
jects
are
presented with
a
number
of
test strings (typically
100)
that must
be
classified
as
either grammatical
or
nongram-
matical.
In the
typical
experiment,
the 100
trials consist
of 50
unique
items, each
of
which
is
presented twice without
feed-
back
about
the
correctness
of the
response.
This procedure yields
data
that directly address
the repre-
sentational issue.
The
logic
of the
analysis
is
simple. There
are
four
possible outcomes
for
each individual item
for
each
subject:
The
subject
may
classify
it
correctly
on
both presen-
tations (CC),
classify
it
correctly
on
only
one of the two (CE
or
EC),
or
misclassify
it on
both presentations (EE). Assume
that
the
subject
operates
by
using
a
simple decision-making
strategy: When
the
status
of the
item
is
known,
it is
always
classified
correctly;
when
it is not
known,
a
guess
is
made.
This simple model
is
quite
powerful
and
allows
for a
surpris-
ingly
deep analysis
of the
representation problem.
Specifically,
under this model,
(a) the
values
of CE, EC,
and EE
should
be
statistically indistinguishable
from
each
other,
and all
should
be
significantly lower than
the
value
of
CC.
This pattern
is
expected
on the
grounds that
the
items
that
contribute
to CE, EC, and EE are
those about
which
the
subject's
knowledge base
is not relevant, (b) A
value
of EE
significantly
greater than
the
values
of EC and CE is
prima
facie
evidence
for the
elaboration
of
nonrepresentative
rules
on
the
part
of
subjects. Thus
if
subjects emerge
from
the
learning
phase
with
rules (either explicit
or
implicit) that
are
not
accurate
reflections
of the
grammar, this knowledge base
will
consistently lead them
to
misclassify
particular
items,
(c)
The
robustness
of
representative knowledge
can be
assessed
from
the relationship
between
the
values
of EC and CE. If the
value
of CE is
detectably
larger than
EC, we can
reasonably
suspect
that
forgetting
was
occurring during testing;
corre-
spondingly,
if EC is
larger than
CE, we can
infer
that
learning
was
occurring during
testing,
(d) One can
estimate knowledge
of
the
grammar
by
looking
at the
value
of CC,
which
contains
only
those items whose status
was
known
by the
subjects
plus
those guessed correctly
on
both presentations. Last,
(e) one
can
derive
an
overall measure
of
consistency
of
responding
by
taking
the sum of CC and EE.
Of
the
values
from
14
experimental conditions (see Table
1),
the
uninteresting ones
can be
dispensed with
first.
There
are no
cases
in
which
the
values
of CE and EC are
significantly
different
from
each other. Thus there
is no
evidence
of
loss
Table
1
Correct
(C) and
Error
(E)
Response
Patterns
to
Individual
Items
During
Well-Formedness
Tasks
Pattern
Condition/training
procedure
Reber(1967)
1
.
Simple memorization
Reber(I976)
2.
Simple memorization
3.
Memorization/rule search
Reber
and
Allen (1978)
4.
Simple observation
5.
Paired
associates
Reber,
Kassin,
Lewis,
and
Cantor
(1980, Experiment
1)
6.
Random display/implicit
instructions
7.
Random display/explicit
instructions
8.
Structured
display/implicit
instructions
9.
Structured display/explicit
instructions
Reber, Kassin, Lewis,
and
Cantor
(1980, Experiment
2)
10.
Rules
at
beginning
of
observation
1
1
.
Rules
in
middle
of
observation
12.
Rules
at end of
observation
13.
Rules only
14.
Observation only
CC
.69
.66
.53
.73
.65
.51
.48
.52
.68
.67
.58
.57
.54
.48
CE
.07
.10
.12
.08
.12
.16
.12
.16
.10
.11
.12
.13
.11
.15
EC
.12
.11
.12
.09
.07
.14
.14
.16
.10
.12
.14
.15
.16
.13
EE
.12
.13
.23"
.11
.16"
.19
.25"
.16
.11
.11
.16
.16
.18"
.24"
Consistency
.81
.79
.76
.84
.81
.70
.73
.68
.79
.78
.74
.73
.72
.72
*
EE
value
significantly
higher
than
the
mean
of the CE and EC
values.

228
ARTHUR
S.
REBER
of
knowledge during
the
well-formedness
task
and no
evi-
dence
that
any
additional learning
was
taking
place.
The
interesting results
are
those
concerning comparisons
between
the
values
of EE and
those
of EC and
CE.
When
no
difference
is
found
between
EE and the
mean
of CE and EC,
it
is
reasonable
to
conclude that there
was no
evidence
of
nonrepresentative
rules
in
use. Values
of EE
that
are
large
in
relation
to
those
of EC and CE
indicate that subjects were
using
rules that
are not
representative
of the
grammar.
The
footnoted
values
in
Table
1 are the
five
conditions that yielded
evidence that subjects emerged
from
the
learning phase with
notions about structure that
were
not
commensurate with
the
stimulus display.
It is
instructive
to
look closely
at
these
five
cases.
In
Con-
dition
13,
the
subjects were given only
the
schematic diagram
of
the
grammar
but no
opportunity
to
observe exemplars.
It
appears that,
not
surprisingly, such
a
procedure encourages
subjects
to
invent
specific
rules
for
letter order and,
in the
absence
of
complete
learning,
to
elaborate rules about per-
missible letter sequences
that
are not
reflective
of the
gram-
mar.
Conditions
3 and 7
illustrate what happens when subjects
are
under
an
instructional
set
that encourages
the use of
rule
search strategies
but in
which
the
letter strings
are
given
to
them
in a
haphazard order. Such
a set of
demand character-
istics encourages subjects
to
invent
a
sufficient
number
of
inappropriate rules
to
inflate
the EE
values.
In
Condition
5 a
paired-associate
task
was
used
to
impart
knowledge.
As is
discussed elsewhere
(Allen
&
Reber,
1980;
Reber
&
Allen, 1978),
the
very nature
of
such
a
task leads
subjects
to set up an
instantiated memorial system composed
of
parts
of
items
and
some whole items along with their
associated
responses. Hence
the
inflated
EE
value
is not due
to
the
application
of
inappropriate rules; rather,
it is due to
subjects'
tendency
to
misclassify
test strings because inappro-
priate analogies exist
in
instantiated memory.
The fifth
con-
dition with
an
inordinately high
EE
value
was 14.
There
is no
obvious explanation
for
this outcome. This datum
is an
anomaly;
1
such outcome
out of 14,
however,
is not bad at
all.
The
remaining
9
conditions
all
yielded response patterns
that
fit
with
the
proposition that whatever subjects
are ac-
quiring
from
the
training sessions
can be
viewed
as
basically
representative
of the
underlying
structure
of the
stimulus
domains. These consist
of
"neutral set" conditions,
in
which
the
subjects
are led to
approach
the
learning task
as an
experiment
in
memory
or
perception
and no
mention
is
made
of
the
rule-governed nature
of the
stimuli (Conditions
1,
2,4,
6,
and 8), and
"structured set"
conditions,
in
which subjects
are
provided with information concerning rules
for
letter
order
but in a
manner than ensures that conscious rule
searching
will
be
coordinate with
the
kinds
of
rules
in use
(Conditions
9, 10,
11,
and
12).
Taken together, these experiments lend general support
to
the
proposition that implicit learning
functions
by the
induc-
tion
of an
underlying representation
that
mirrors
the
structure
intrinsic
to the
environment. Such
an
induction process takes
place
naturally
when
one is
simply attending
in an
unbiased
manner
to the
patterns
of
variation
in the
environment
or
when
one is
provided with
an
orientation that
is
coordinate
with
these variations.
This characterization
of the
appropriateness
of
mental
rep-
resentation entails nothing about
the
sheer amount
of
knowl-
edge
that
one
takes
out of a
learning session.
In
fact,
it is
relatively
easy
to
show that there
is
little
to
enable
one to
distinguish explicit
from
implicit
processes
here.
The
consist-
ency
values
in
Table
1
reveal surprisingly little variation
from
condition
to
condition, particularly when compared with
the
range
of CC and EE
values. These consistency values
can be
seen
as a raw
estimate
of the
total number
of
rules that
subjects
can be
said
to be
using during decision making,
for
they
are
simply
the
sums
of the CC and EE
values. Taking
a
simple
(and only quasi-legitimate) average across conditions
reveals
that
the
overall mean consistency values
for the
foot-
noted conditions
and the
nonfootnoted
ones
are .75 and
.76,
respectively.
Thus there
is no
evidence
that
either
set of
conditions
produces more rule learning;
the
difference
is
that
explicit learning results
in the
emergence
of a
number
of
inappropriate rules, whereas implicit learning tends
to
yield
representative, veridical rules.
This same model
of
representation
is
supported
by
data
from
other
tasks.
In
Reber
and
Lewis's
(1977)
anagram solu-
tion
task,
subjects worked with
the
same problem sets over
4
days.
In
that
study there
was
improvement over time,
so a
stochastic
model
was fit to the
data
and
used
to
predict
the
pattern
of
error
and
correct responses
to
individual test items
that would
be
expected under
the
assumption that subjects
were
not
using inappropriate rules. (For details
on the
model,
see the
original article.)
The
results
were
in
keeping with
the
general
theme here.
The
EEEE
value (the proportion
of
items
solved incorrectly
on all 4
days
of the
study)
was no
higher
than
would
be
expected under
the
assumption that
subjects
either knew
the
solution
to a
particular anagram
or
made
nonsystematic guesses
for
problems
not
within
the
domain
of
their knowledge base.
Several
experiments
in
which
the PL
procedure
was
used
are
also
of
interest.
The
relevant
data
are the
recency curves.
In
the
standard analysis
of a PL
experiment,
a
recency curve
represents
the
probability
of a
given response plotted against
the
length
of the
immediately preceding
run of
that event.
Recency
curves
may
take
on any of a
number
of
shapes,
depending
on the
conditions
of the
experiment. Negative
recency
is
most common, particularly
early
in an
experiment.
Under
various circumstances,
however,
even positive recency
may
be
observed (see Friedman
et
al.,
1964,
for
details).
The
concern here
is
with
the
recency curves
from
experiments
with
500 or
more trials with
a
Bernoulli event sequence with
probability
of the
more likely
event
set at
.80.
Figure
2
presents
the
pooled recency data
from
five
such
experiments
(see Reber, 1967; Reber
&
Millward,
1968).
All
subjects
were
run
though
a
learning period with either
the
traditional
PL or the
instant asymptote technique.
The
sub-
ject-generated
curve
has
been adjusted
downward
by
exactly
.04 at all
points
to
correct
for a
ubiquitous overshooting
effect
that
is
observed
in all of
these many-trial experiments (see
Reber
&
Millward, 1968,
for a
discussion
of
this
issue).
This
adjustment
in no way
modifies
the
startling aspect
of
these
two
curves.

IMPLICIT
LEARNING
AND
TACIT
KNOWLEDGE
229
With
few
exceptions,
the
curves
sit on top of
each other.
There
is no
evidence whatsoever
for
either
the
positive recency
predicted
by the
early conditioning models
or the
negative
recency
reported
by
many. There
is,
however, overwhelming
evidence
for a
mental representation that
reflects
the
structure
of
the
stimulus environment.
The
simplest characterization
of
this curve, which
is
based
on a
total
of
44,000 responses,
is
that
it
reveals that subjects mimic
the
structure
of the
event
sequence. Subjects' prediction
responses
show
flat
recency
curves because
the
event sequences themselves display
flat
recency
curves—as
they must, being Bernoulli
in
nature.
This
is not a new
point;
it was
made earlier
by
Derks
(1963)
and by
Jones
and
Meyers
(1966),
who
showed that experi-
ments
can
encourage either positive
or
negative recency
by
presenting event sequences
with
either
many
long
or
many
short
runs
of
events.
But the
precision with which subjects'
response patterns
can
reflect
the
event patterns
has
never
really
been appreciated.
To
take
this
point
to a
further
ex-
treme,
data
like
those
in
Figure
2 are so
robust that they
can
actually
be
used
as a
check
on
one's experimental procedure.
In
one PL
study
(Millward
&
Reber,
1972),
the
subjects'
overall
response proportions were .523
and
.476
for the two
events,
a
result that
was
perplexing because each event
had
been
programmed
to
occur
in
exactly half
of the
trials.
The
anomaly
turned
out to be in the
computer program used
to
generate
the
sequences.
A
check revealed that
the two
events
had
actually been presented
to
subjects with proportions
of
.520
and
.480!
Although
the
preceding analyses seem
to
provide support
for
the
representational realist position,
it is
still unclear just
how
far one can
legitimately push such
a
proposition.
In
many
of the
experiments reported here
and in
other related
areas
of
study (see
Schacter,
1987),
subjects respond
in
ways
that indicate that their mental content
may not be
quite
so
neatly
isomorphic
with
that
of the
stimulus
field.
However,
it
also
seems reasonably clear that
when
such transforms
or
constructions
of
representations
are
observed,
"secondary"
processes
are
responsible; that
is,
the
"primary"
process
of
veridical
representation
of
environmental structure becomes
colored either
by
elaborative operations,
as in
experiments
in
which
instructional sets encouraged invention
of
inappro-
priate rules (Howard
&
Ballas,
1980; Reber, 1976),
or by
restrictive
operations,
as in
studies
in
which task demands
led
to the
narrowing
of
attentional
focus
(Brooks, 1978; Cantor,
1980;
Reber
et
al.,
1980).
Also,
careful
scrutiny
of the EE
values
in
Table
1
reveals that
even
in the
nonfootnoted
conditions there
was a
tendency
for
some
nonrepresentational
elaboration
to
take place.
In all
nine
of
these cases,
the EE
value
is
equal
to or
higher than
the EC or
CE
values
(p
< .05
by
a
sign
test).
The
problem
of
mental representation
is
clearly
no
easy
nut to
crack.
The
position
taken here seems
to be a
reasonable
one,
although
it
will
probably
be
shown
to be
wrong
in the
final
analysis.
Tacit
knowledge
is a
reasonably veridical, par-
tial
isomorphism
of the
structural patterns
of
relational
invar-
iances
that
the
environment displays.
It is
reasonably veridical
in
that
it reflects,
with
considerable accuracy,
the
stimulus
invariances
displayed
in the
environment.
It is
partial
in
that
not all
patterns become part
of
tacit
knowledge.
It is
structural
.90
.70
-
S.
O
o
EVENT
SEQUENCES
•—•
POOLED
DATA
(44,000
trials)
3 S 7 9 11
LENGTH
of
E1
RUN
13
15+
Figure
2.
Pooled
recency
data
from
five
separate
probability
learn-
ing
experiments.
(The
open
circles
give
the
average
probability
that
the
more frequent event will
occur
over
all
event
sequences
used;
filled
circles
give
the
average
probability
that
the
subjects
will
make
the
more
frequent
response,
adjusted
for
overshooting.
The
data
are
based
on
44,000
trials
from
88
asymptotic
subjects
tested
with
the
probability
of the
more
frequent event
set at
.80.)
in
that
the
patterns
are
manifestations
of
abstract generative
rules
for
symbol ordering.
On the
Availability
of
Tacit
Knowledge
The
conclusion reached
in the
first
studies
on
implicit
learning (Reber, 1965)
was
that
the
knowledge acquired
was
completely unavailable
to
consciousness.
The
many experi-
ments carried
out
since have
shown
that position
to
have
been
an
oversimplification.
The
picture that
is
emerging,
though
perhaps somewhat less striking,
is
certainly more
interesting. Specifically, knowledge acquired
from
implicit
learning
procedures
is
knowledge that,
in
some
raw
fashion,
is
always ahead
of the
capability
of its
possessor
to
explicate
it.
Hence although
it is
misleading
to
argue that implicitly
acquired
knowledge
is
completely unconscious,
it is not
mis-
leading
to
argue that
the
implicitly acquired
epistemic
con-
tents
of
mind
are
always
richer and
more sophisticated than
what
can be
explicated.
In
Reber
and
Lewis's
(1977)
study, data
were
first
presented
to
support this position. Over
the 4
days
of
that study,
during
which
subjects solved anagram puzzles
on the
basis
of the
syntax
of an
artificial
grammar,
there
was a
general increase
in
the
ability
of
subjects
to
communicate their knowledge
of
the
rule system
in
use. There
was
also
an
increase
in the
ability
to
solve
the
anagrams,
but the
former
never
caught
up
with
the
latter, that
is, as
subjects
improved
in
their ability
to
verbalize
the
rules that they
were
using,
they also developed
richer and
more complex rules. Implicit knowledge remained
ahead
of
explicit knowledge.
In
a
recent study,
Mathews
et al. (in
press) used
a
novel
yoked-control
technique
to
explore this issue. Subjects were
interrupted
at
intervals during
a
well-formedness judgment
task
and
asked
to
explicate
the
rules that they were using.
The
information
was
then
given
to
yoked-control
subjects,
who

230
ARTHUR
S.
REBER
were
then tested
in the
same well-formedness task.
So
equipped,
these control subjects managed
to
perform
at
roughly
half
the
level
of
accuracy
of the
experimental subjects.
Moreover,
as the
experiment progressed
and
each experimen-
tal
subject
improved,
so did
each yoked control,
but the
controls never caught
up
with
the
experimental subjects.
The
most direct attempt
to
deal with
the
issue
of the
degree
to
which implicitly acquired knowledge
is
available
to
con-
sciousness
was
carried
out by
Dulany
et
al.
(1984).
After
a
standard learning procedure, subjects were asked
to
mark
each
well-formedness test item
as
acceptable
or not and to
specify
what
features
of
that item
led
them
to
classify
it as
they
did. Dulany
et al.
argued that
the
features
so
marked
accounted
for the
full
set of
decisions that each subject made,
a
result that,
if
correct,
supports
the
notion that used knowl-
edge
of the
grammar
was
held consciously.
Reber,
Allen,
and
Regan
(1985),
however, argued that
the
nature
of the
task
that Dulany
et al.
used carried
its own
guarantee
of
success;
that
is, the
task forced
the
data
to
appear
as
though they
carried
the
implication
of
consciousness, whereas actually
the
subjects
were
reporting only
vague
guesses about
the
appro-
priateness
or
inappropriateness
of
letter groups.
The
issue
continues
to be
disputed. Dulany, Carlson,
and
Dewey
(1985)
presented reasons
for
doubting Reber
et
al.'s
analysis, whereas
Hayes
(in
press) recently produced evidence
in
support
of the
interpretation
of
Reber
and his
co-workers.
One of the
problems with this line
of
research
is
that
it
fails
to
distinguish between knowledge
that
is
available
to
con-
sciousness
after
attempts
at
retrieval
and
knowledge that
is
present
in
consciousness
at the
time that
the
decisions them-
selves
are
being made.
Carmody
et al.
(1984)
noted
this
problem
in
assessing
the
knowledge base that physicians
are
taught
to use
versus what they actually
use in
diagnosis,
and
Schacter
(1987)
argued
that
conclusions reached about
the
availability
of
implicit information must take account
of a
variety
of
task
constraints
that have their
own
impacts. Never-
theless,
if it is not yet
clear,
the
discussions that
follow
will
emphasize
even further
the
central thesis
of
this line
of re-
search.
To
wit:
A
considerable portion
of
memorial content
is
unconscious, and, even more important,
a
goodly amount
of
knowledge acquisition
takes
place
in the
absence
of
intent
to
learn.
Entailments
and
Implications
The
preceding discussion
is a
reasonably thorough review
of
the
current state
of
affairs
as
regards
the
general issues
of
the
acquisition, usage, representation,
and
availability
of
tacit
knowledge.
As
Schacter
(1987)
pointed
out
recently,
one of
the
intriguing aspects
of the
history
of
work
on
this issue
is
that there
is
such
an
amazing variety
of
implicit processes
that have been observed
and yet
there
is
nothing approaching
a
satisfactory theoretical account
of
them. What
follows
may
or
may not
improve
on
this state
of
affairs.
The
following
is
a
small
flurry
of
speculation concerning
the
possible
entail-
ments
and
implications
of the
research. Each
of the
topics
is
touched upon only
briefly;
the
point
here
is to
provoke
new
avenues
of
study,
not to
draw
any
hard conclusions.
On the
Origins
of
Unconscious
Cognition
Usually
the
header here
is the
"Origins
of
Conscious
Cog-
nition,"
not"
Unconscious."
Traditionally,
the
focus
has
been
on
consciousness
with
the
implication that
defining
and
char-
acterizing
consciousness
will
solve
the
problem; unconscious
processes
will
be
handled
by the
invoking
of
exclusionary
clauses.
The
history
of the
variety
of
ways
in
which
the
unconscious
has
been represented
(Ellenberger,
1970)
shows
this clearly. Consciousness assumes epistemic priority because
it
is so
introspectively
obvious, whereas
the
unconscious must
be
struggled with
in
derivative fashion.
The
point
to be
defended here
is
that this ordering
of
priorities
has
been
an
error.
The
theoretically important
ex-
ercise should
be on the
origins
of
unconscious cognitive
processes. Consciousness,
evolutionarily
speaking,
is a
late
arrival
on the
mental scene. Perhaps
it is not of
such
recent
origin
as
some
have
argued (Jaynes, 1976),
but
surely
it
postdates
a
number
of
fairly
rich and
elaborative
cognitive
processes that
functioned
and
still
function
in our
phyloge-
netic
predecessors
(Griffin,
1981, 1984). There
is,
moreover,
absolutely
no
reason
to
suppose that these presumably adap-
tive
mental
capacities
ought
to
have been lost.
In
fact,
there
are a
number
of
reasons
for
supposing that
they
continue
to
flourish
interpenetrated
by an
emerging executive system,
conscious mentation.
Taking such
a
perspective
gives
unconscious cognition
the
empirical
and
theoretical
priority that
it
deserves
but
ha&
not
enjoyed
since
the era of the
philosophical
emergentista
Un-
conscious cognitive functions should
not
have
to
be
defended
against arguments that deny
their
role
in
action
(see
the
debate
between
Dulany
et
al.,
1984,
1985)
and
Reber
et
al.,
1985).
The
proper stance
is to
assume that unconscious mental
processes
are the
foundations upon which emerging conscious
operations
are
laid.
The
really
difficult
problem, then,
is to
discern
how
these components
of
mind interact.
This perspective
has
some interesting
entailments.
One is
that
it
suggests
a
novel
way to see how the
work
on
implicit
learning
fits in
with
a
good deal
of
other research
on the
cognitive
unconscious. Another
is
that
it
allows
for a new
framework
for:
Parsing
the
Cognitive
Unconscious
A
conspicuously large number
of
processes
and
functions
have
been assigned
to the
unconscious over
the
past century
or
so.
They
have
come
in a
variety
of
forms,
some concerned
with
perceptual processes, some
with
dynamic, some
with
motivational
and
emotional,
and
some with cognitive.
A
number
of
schemata have also been proposed
for
defining
and
classifying
the
subcategories
of
unconscious
functions
and
operations (see Ellenberger, 1970;
Erdelyi,
1985). Herein
is
one
more.
As
a first
approximation, assume
a
relatively high-level
parse that separates unconscious mentation into
two
classes,
one
that most aptly
can be
called
the
primitive
and one
that,
for
reasons
to be
spelled out,
can be
thought
of as the
sophisticated.
The
primitive unconscious encompasses
a va-

IMPLICIT LEARNING
AND
TACIT KNOWLEDGE
231
riety
of
basic
functions,
all of
which
are
carried
out
more
or
less automatically
and are
more
or
less devoid
of
meaning,
affect,
or
interpretation. Included here
is a
range
of
processes
such
Gibson's
(1979)
direct pickup
of
information
in
percep-
tion, Hasher
and
Zacks's
(1984)
automatic encoding
of
infor-
mation
about
the
frequency
of
events
as
they occur,
Lewicki's
(1985,
1986a, 1986b) studies showing unconscious apprehen-
sion
of
feature covariation,
Broadbent
and his
co-workers'
(Berry
&
Broadbent, 1984; Broadbent
&
Aston,
1978;
Broad-
bent
et
al.,
1986) studies
on
simulated economic
and
manu-
facturing
systems, and,
of
course,
the
studies reviewed
on
implicit learning
of
complex covariations displayed
by
syn-
thetic grammars
and
structured event sequences.
The
operations
of
this primitive unconscious seem
to be
about
as
fundamental
for a
species' survival
as any
nonvege-
tative
function
could
be.
Virtually
every
organism must
be
able
to
perform
a
basic
Hasher
and
Zacks's
(1984)
type
of
counting
of the
occurrences
of
ecologically important events.
Ground squirrels presumably count small holes
and
keep
a
kind
of log of
their locations,
and
lions count gnus
and
their
various
properties. Rats, pigeons, dogs,
and
other laboratory
subjects
count covariations between events even
in the
most
basic circumstances.
The
essence
of
Pavlovian conditioning
is
the
apprehension
of a
genuine covariation between
the
conditioned stimulus
and the
unconditioned stimulus
(Res-
corla,
1967,
1988).
The
reason
that
the
work
of
such research-
ers as
Hasher
and
Zacks,
Lewicki,
and
others
is
typically
regarded
as
cognitive
in
nature
and
somehow
different
from
Rescorla's
is
that
it is
typically
carried
out
with mature adult
subjects
and
with
materials that have
a rich
cognitive under-
pinning,
such
as
words, sentences,
and
pictures. Yet, there
is
no
a
priori
reason
to
regard these high-level cognitive
"counts"
as
different
in any
fundamental
way from the
very
simple
countings
of
Rescorla's
subjects. What
is
different
is the
process
by
which
each
organism
comes
to
categorize
the
items
whose
frequency
and
covariation patterns
are
being logged,
not the
mechanism
for
representing
the raw
data.
From this perspective,
the
grammar learning experiments,
the PL
studies,
and the
rest
of the
literature
on
implicit
learning
can be
viewed
as
epistemic
kin of the
most basic
of
the
primitive unconscious functions.
For
example,
Lewicki
(1985)
showed
that
in the
limiting case, only
one
exposure
to
a
target person with
a
salient personality characteristic (e.g.,
kindness, capability)
and
particular physical characteristic
(e.g.,
long
or
short hair)
is
sufficient
to set up a
tacit
knowledge
base that reflects these covariations
and
affects
decisions made
about novel people.
In
cases with
a richer
data
base, such
as
the
structured event sequences
of the
synthetic languages
and
the
probability learning experiments,
the
kinds
of
structural
covariations that
are
apprehended
are
deeper
and
more
ab-
stract. Yet, they
can be
viewed
as
categorical extensions
in
that
the
basic process
is, in
principle, still
one of
counting,
only
what
is
being counted
are
complex interdependent
co-
variations among
events—or,
as
they
are
commonly known
in the
literature,
rides.
These various manifestations
of the
functions
of the
prim-
itive unconscious have
a
number
of
additional
factors
in
common.
First,
and
most
basic,
the
pickup
of
information
takes place independently
of
consciousness
or
awareness
of
what
is
picked
up. Put
another way, adding
the
factor
of
consciousness changes
the
very
nature
of the
process (Reber,
1976;
Reber
&
Allen, 1978; Reber
et
al.,
1980).
As
stated
at
the
outset,
this
may be
taken
as the
defining
feature
of
implicit
learning.
Second, although much
of
what
is
acquired
may
eventually
be
made available
to
conscious expression, what
is
held
or
stored exceeds what
can be
expressed. This
is
displayed
in one
of
two
fashions:
Either predictions
of
performance made
on
the
basis
of
available knowledge
fall
short
of
actual perform-
ance
(Mathews
et
al.,
in
press;
Nisbett
&
Wilson, 1977; Reber
&
Lewis,
1977)
or
differential
effects
of
variables
are
seen
when
the
implicit-explicit instructional
set is
varied (Graf
&
Mandler,
1984; Reber, 1976; Reber
et
al.,
1980;
Schacter
&
Graf,
1986). Schacter (1987) noted that this inequality
be-
tween
implicit knowledge that
is
inaccessible
and
implicit
knowledge
that
can be
articulated explicitly
may
have
to do
with
the
degree
of
elaborative
encoding that
is
allowed.
In-
deed, most
of the
studies
in
which
a
substantial proportion
of
once-tacit knowledge
is
made available
to
consciousness
are
those
in
which considerable overt encoding
is
carried
out
(e.g.,
Dulanyetal.,
1984).
Third,
the
memorial content
of the
primitive unconscious
has
a
causal role
to
play
in
behavior. This proposition goes
almost
without
saying,
given
the
preceding
discussion,
but it
needs
to be
specified;
if
there were
no
causal component
to
unconscious cognition,
we
might
as
well
simply
return
to a
radical behaviorism.
Put
simply,
the
primitive unconscious
processes
are for
learning about
the
world
in
very basic ways.
They
are
automatic
and
ineluctable; they
function
to
pick
up
critical knowledge about categories
and
about covariations
of
aspects
of
categories. They
do
not, however, have
any
func-
tions that involve meaning
or
affect;
these
are the
province
of
the
sophisticated
unconscious.
In
this
latter
class
are
included
such phenomena
as
uncon-
scious perception
of
graphic
and
semantic information (Mar-
cel,
1983), perceptual
vigilance
and
perceptual
defense
(Er-
delyi,
1974),
the
implicit pickup
of
affective
information that
is
based
on
phonological
factors
(Corteen
&
Wood, 1972)
or
geometric
features
(Kunst-Wilson
&
Zajonc,
1980; Seamon,
Brody,
&
Kauff,
1983; Seamon, Marsh,
&
Brody, 1984),
and
repetition priming
effects
with
various linguistic
and
nonlin-
guistic
materials (Jacoby
&
Dallas,
1981; Scarborough,
Cortese,
&
Scarborough, 1977; Scarborough, Gerard,
&
Cortese,
1979, among many others).
In
some
ways
the
evi-
dence
for the
unconscious element
is
stronger here than
it is
with
the
primitive unconscious.
The use of
forced-choice
recognition
tests
as a
measure
of
sensitivity
in the
work
of
Marcel
(1983),
Kunst-Wilson
and
Zajonc
(1980), Seamon
et
al.
(1983),
and
Seamon
et al.
(1984)
supports
the
strong claim
that
these
processes
are
occurring virtually independently
of
awareness.
Although
there
may be
some problems with meth-
odology
(see Holender, 1986), this procedure
is, in
principle,
superior
to
that
used
by
most researchers,
who
mainly
brow-
beat their subjects into telling
what
they know
(Allen
&
Reber,
1980;
Berry
&
Broadbent, 1984; Brooks, 1978; Reber&
Allen,
1978).
What makes these various processes intriguing
and
what
differentiates
these sophisticated processes
from
the
primitive

232
ARTHUR
S.
REBER
is
that
all
share
a
basic operating property: They
all
depend
on
a
rich, abstract knowledge base
that
asserts itself
in a
causal
manner
to
control perception,
affective
choice,
and
decision
making
independently
of
consciousness. This component
of
the
cognitive unconscious depends
on
previously acquired
knowledge,
as
opposed
to the
primitive
component,
which
operated
to
acquire such knowledge.
The
very
epistemic
base
that makes these sophisticated processes functional
can be
seen
as
that derived
from
the
primitive
processes.
These
sophisticated
systems also
differ
from
the
primitive
in
other
ways.
First, they
are
components
of
mind
that
are
generally
available
to
consciousness.
In
other words, there
is
awareness
of the
knowledge base itself;
a
subject
in one of
Marcel's
(1983)
experiments surely
knows
the
target word
and,
moreover,
surely knows that
he or she
knows
it.
What
is
crucial
is
that this overt knowledge
base
has a
higher threshold
for
engagement than
the
covert
one
does. Second, they
are
based
on
knowledge systems
that
have become
highly
auto-
matized. They share this automatic quality with
the
primitive
functions
in the
limit,
but
there
are
good reasons
for
thinking
that much
of
this interpretive
and
semantic knowledge derived
from
explicit processes that became automatic only
after
pained,
conscious
action.
Interestingly, this line
of
argument
parallels
that
taken
by
Dulany
et
al.
(1984, 1985)
in
their
criticism
of the
synthetic grammar learning studies; however,
Dulany
et al.
targeted
the
wrong
level
for
invoking
it.
Last,
these systems
all
function
on a
symbolic level.
All of the
critical components
of the
sophisticated unconscious
involve
semantic
and
affective
properties
of
stimuli. This aspect seems
to be
largely missing
in the
primitive domain.
It
seems that
these sophisticated
processes
are
more uniquely
the
stuff
of
humanity
than
are the
primitive processes, which
are
operat-
ing
systems that
we
share with virtually
all
corticated
species
and are
found
rather
far
down
the
phylogenetic
scale.
The
Robustness
of
Implicit
Processes
There
has
been
a
good deal
of
work
to
suggest that implicit
systems
are
robust
in the
face
of
disorders that
are
known
to
produce serious deficits
in
conscious, overt
processes.
Support
for
this
functional
separation
of
conscious
and
unconscious
cognitive processing
has
come
from the
study
of
various
patient
populations. Classic cases
are
amnesia (see Milner,
Corkin,
&
Teuber,
1968,
for the
early
work
and
Schacter,
1987,
for a recent
overview),
Hindsight
(Weiskrantz,
1986),
prosopagnosia
(Bauer,
1984),
and
alexia
(Shalh'ce
&
Saffran,
1986).
In all
these cases there
is
compelling evidence
of
effective
performance
in the
absence
of
awareness.
The
model
of
mental parsing suggested earlier provides
a
novel
interpretation
of
this work. There
is a
standard heuristic
in
evolutionary biology that older primitive systems
are
more
robust
and
resistant
to
insult than
are
newer, more complex
systems.
The
hypothesis that
the
implicit cognitive processes
are the
functional
components
of the
evolutionarily
older,
primitive
system predicts
that
they should show greater resist-
ance
than should explicit processes.
By
extension,
all of the
various
phenomena that have been cited
as
manifestations
of
primitive
unconscious processes would
be
expected
to
display
similar robustness under conditions
in
which parallel explicit
processes have been diminished
or
even lost entirely.
The
strongest evidence
in
support
of
such
an
interpretation
comes
from
cases
in
which direct comparison between
im-
plicit
and
explicit processes
has
been made
in
clinical settings.
In
an
extended series
of
studies,
Warrington
and
Weiskrantz
(see
1982
for an
overview)
found
no
deficits
in
amnesics when
the
task involved memory
for
words based
on
word-stem
and
word-fragment
cues,
but
performance
was
seriously impaired
when
overt word recognition
and recall
procedures were used.
A
similar pattern emerged with Hasher
and
Zacks's
frequency
encoding task,
in
which performance
was
found
to be
robust
in the
face
of
clinical depression (Hasher
&
Zacks,
1979;
Roy,
1982)
and
even
Korsakoffs
syndrome (Strauss,
Weingartner,
&
Thompson, 1985). Last,
a
recent study
of
Abrams
and
Reber
(1989)
suggested that
even
the
acquisition
of
knowledge
is
undiminished
so
long
as the
task
is a
nonreflective,
uncon-
scious one. They used
an
implicit grammar learning task
and
an
explicit short-term memory task with
a
mixed population
of
institutionalized depressives, schizophrenics,
and
alcoholics
with
organic brain damage.
The
patients performed more
poorly
than
a
normal control group
on the
memory task,
but
the
performances
of the two
groups
were
statistically indistin-
guishable
on the
implicit learning task. This last study
is
particularly important,
for it is one of the
few
that shows that
implicit
learning
is
robust
in the
face
of
serious psychological
and/or
neurological disorders (see Graf
&
Schacter,
1985,
for
another example
involving
a
word-completion task).
On
Intuition
One
of the
gains
of
this
line
of
research
on
implicit processes
is
that
it
provides
the
opportunity
to
reclaim intuition
for
cognitive
psychology. There
is
probably
no
cognitive process
that
suffers
from
such
a gap
between
phenomenological
reality
and
scientific
understanding.
Introspectively,
intuition
is one
of
the
most compelling
and
obvious cognitive processes;
empirically
and
theoretically,
it is one of the
processes least
understood
by
contemporary cognitive
scientists.
The
basic argument
is
simple:
The
kinds
of
operations
identified
under
the
rubric
of
implicit learning represent
the
epistemic core
of
intuition; that
is,
the
introspective qualities
that most
people—from
Bergson
(1913)
and
Croce
(1922)
to
Jung
(1926),
Polanyi
(1958),
and
Westcott
(1968)—identify
when
discussing intuition
are
those processes that have
emerged
in the
studies
of
implicit acquisition
of
complex
knowledge.
Perhaps
the
most compelling aspect
of
intuition,
and the one
most
often
cited
in the
various definitions that
have
been given (see Westcott, 1968),
is
that
the
individual
has a
sense
of
what
is right or
wrong,
a
sense
of
what
is the
appropriate
or
inappropriate response
to
make
in a
given
set
of
circumstances,
but is
largely
ignorant
of the
reasons
for
that mental
state.
This,
of
course,
is how the
typical
subject
has
been characterized
after
a
standard acquisition session
in
an
implicit learning experiment.
The
point
is
that intuition
is a
perfectly
normal
and
com-
mon
mental state/process that
is the end
product
of an
implicit
learning experience.
In
other words, intuition ought
not to be
embedded
in
personality theory
as it was
with Jung

IMPLICIT LEARNING
AND
TACIT KNOWLEDGE
233
(1926),
and
although
it is a
topic
of
some philosophical
interest,
it is
probably best
not
dealt
with
as an a
priori topic
as
it was by
Croce (1922).
It is a
cognitive
state
that emerges
under
specifiable conditions,
and it
operates
to
assist
an
individual
to
make choices
and to
engage
in
particular classes
of
action.
To
have
an
intuitive sense
of
what
is
right
and
proper,
to
have
a
vague
feeh'ng
of the
goal
of an
extended
process
of
thought,
to
"get
the
point"
without really being
able
to
verbalize what
it is
that
one has
gotten,
is to
have
gone
through
an
implicit
learning experience
and
have built
up the
requisite representative knowledge base
to
allow
for
such
judgment.
Summary
This
article
is an
attempt
to
come
to
grips with
an
essential,
although oft-ignored, problem
in
contemporary cognitive psy-
chology:
the
acquisition
of
complex knowledge.
At the
heart
of
the
presented thesis
is the
concept
of
implicit
learning
wherein
abstract, representative knowledge
of the
stimulus
environment
is
acquired, held,
and
used
to
control behavior.
The
operations
of
implicit learning
are
shown
to
take place
independently
of
consciousness; their mental products
have
been
demonstrated
to be
held tacitly; their
functional
con-
trolling
properties have been shown
to
operate
largely
outside
of
awareness.
The
strong argument
is
that implicit learning
represents
a
general, modality-free
Ur-process,
a
fundamental
operation
whereby
critical
covariations
in the
stimulus envi-
ronment
are
picked
up.
The key
problem
in all of
this
is to
specify,
as
clearly
as
possible,
the
boundary conditions
on the
process
of
implicit
learning—that
is, to
outline
the
circumstances under
which
it
emerges
and
those under
which
it is
suppressed
or
over-
whelmed.
A
substantial part
of the
empirical
work
reviewed
here
should
be
seen
in
that light. Last, there
has
been
an
attempt
to
show
how
such
a
process
can be
seen
as
functioning
in
the
context
of
other, complex cognitive operations
and to
speculate
on how it
might
be
viewed
in a
variety
of
other
frameworks,
from
that
of
evolutionary theory
to
those
of
various
clinical syndromes
affecting
cognitive
function
to
those
of
some
novel
considerations
of
intuition.
References
Abrams,
M.
(1987).
Implicit
learning
in the
psychiatrically
impaired.
Unpublished
doctoral
dissertation,
City
University
of New
York.
Abrams,
M.,
&
Reber,
A. S.
(1989). Implicit learning
in
special
populations.
Journal
of
Psycholinguistic
Research,
17,
425-439.
Allen,
R., &
Reber,
A. S.
(1980).
Very
long term memory
for
tacit
knowledge.
Cognition,
S,
175-185.
Anderson,
J. R.
(1978).
Arguments
concerning representations
for
mental imagery.
Psychological
Review,
85,
249-277.
Anderson,
J. R.
(1979). Further arguments
concerning
representa-
tions
for
mental imagery.
Psychology
Review,
86.
395-406.
Anderson,
J. R.
(1983).
The
architecture
of
cognition. Cambridge,
MA:
Harvard
University
Press.
Bauer,
R. M.
(1984).
Autonomic recognition
of
names
and
faces
in
prosopagnosia:
A
neuropsychological
application
of the
guilty
knowledge
test.
Neuropsychologia,
22,
457-469.
Bergson,
H.
(1913).
Introduction
to
metaphysics.
New
York: Liberal
Arts
Press.
Berry,
D. C., &
Broadbent,
D. E.
(1984).
On the relationship
between
task performance
and
associated
verbalizable
knowledge.
Quarterly
Journal
of
Experimental
Psychology,
36,
209-231.
Bialystok,
E.
(1981). Some evidence
for the
integrity
and
interaction
of
two
knowledge
sources.
In R.
Anderson
(Ed.),
New
directions
in
research
on the
acquisition
and use
of
a
second
language.
Rowley,
MA:
Newbury.
Broadbent,
D.
E.,
&
Aston,
B.
(1978). Human control
of a
simulated
economic system. Ergonomics,
21,
1035-1043.
Broadbent,
D.
E.,
FitzGerald,
P., &
Broadbent,
M. H. P.
(1986).
Implicit
and
explicit knowledge
in the
control
of
complex systems.
British
Journal
of
Psychology,
77,
33-50.
Brooks,
L. R.
(1978).
Nonanalytic
concept formation
and
memory
for
instances.
In E.
Rosch
& B. B.
Lloyd
(Eds.),
Cognition
and
categorization
(pp.
169-211).
New
York: Wiley.
Cantor,
G. W.
(1980, April).
On
symbol
string-type
and the
general-
ization
of
the
implicit
learning
paradigm. Paper presented
at the
annual
meeting
of the
Eastern Psychological Association,
New
York.
Carmody,
D. P.,
Kundel,
H.
L.,
&
Toto,
L. C.
(1984).
Comparison
scans
while reading chest images: Taught,
but not
practiced.
Inves-
tigative
Radiology,
19,
462-466.
Chomsky,
N.
(1980). Rules
and
representations.
New
York: Columbia
University
Press.
Corteen,
R.
S.,
&
Wood,
B.
(1972).
Autonomic responses
to
shock-
associated
words
in an
unattended channel.
Journal
of
Experimen-
tal
Psychology,
94,
308-313.
Croce,
B.
(1922).
Aesthetic (2nd
ed.).
New
York: Norwood.
Danks,
J.
H.,
&
Gans,
D. L.
(1975).
Acquisition
and
utilization
of a
rule
structure.
Journal
of
Experimental
Psychology:
Human Learn-
ing
and
Memory,
1,
201-208.
Derks,
P. L.
(1963).
Effect
of run
length
in the
gambler's
fallacy.
Journal
of
Experimental
Psychology,
65,
213-214.
Dulany,
D.
E.,
Carlson,
R.
A.,
&
Dewey,
G. I.
(1984).
A
case
of
syntactical learning
and
judgment:
How
conscious
and how
ab-
stract?
Journal
of
Experimental
Psychology:
General,
113, 541-
555.
Dulany,
D.
E.,
Carlson,
R.
A.,
&
Dewey,
G. I.
(1985).
On
conscious-
ness
in
syntactical learning
and
judgment:
A reply to
Reber, Allen,
and
Regan.
Journal
of
Experimental
Psychology:
General,
114,
25-
32.
Ellenberger,
H. F.
(1970).
The
discovery
of
the
unconscious.
New
York: Basic Books.
Erdelyi,
M. H.
(1974).
A new
look
at the New
Look: Perceptual
defense
and
vigilance.
Psychological
Review,
81,
1-25.
Erdelyi,
M. H.
(1985).
Psychoanalysis:
Freud's
cognitive
psychology.
New
York: Freeman.
Fodor,
J. A.
(1975).
The
language
of
thought. Cambridge,
MA:
Harvard
University
Press.
Fodor,
J. A.
(1983).
Modularity
of
mind:
An
essay
on
faculty
psy-
chology.
Cambridge,
MA: MIT
Press.
Fried,
L.
S.,
&
Holyoak,
K.
J.
(1984). Induction
of
category distri-
butions:
A
framework
for
classification learning.
Journal
of
Exper-
imental
Psychology:
Learning,
Memory,
and
Cognition,
10,
234-
257.
Friedman,
M. P.,
Burke,
C.
J.,
Cole,
M.,
Keller,
L.,
Millward,
R.
B.,
&
Estes,
W. K.
(1964).
Two-choice
behavior under extended
train-
ing
with
shining
probabilities
of
reinforcement.
In R. C.
Atkinson
(Ed.),
Studies
in
mathematical
psychology
(pp.
250-291).
Stanford,
CA:
Stanford University Press.
Garner,
W. R.
(1974).
The
processing
of
information
and
structure.
Hillsdale,
NJ:
Erlbaum.
Garner,
W. R.
(1978).
Aspects
of a
stimulus: Features, dimensions,
and
configurations.
In E.
Rosch
& B. B.
Lloyd
(Eds.),
Cognition
and
categorization
(pp.
99-133).
Hillsdale,
NJ:
Erlbaum.

234
ARTHUR
S.
REBER
Gibson,
J. J.
(1966).
The
senses
considered
as
perceptual
systems.
Boston:
Houghton
Mifflin.
Gibson,
J. J.
(1979).
The
ecological
approach
to
visual
perception.
Boston:
Houghton
Mifflin.
Gleitman,
L.
R-,
&
Wanner,
E.
(1982).
Language acquisition:
The
state
of the
art.
In E.
Wanner
& L. R.
Gleitman
(Eds.),
Language
acquisition:
The
slate
of
the art
(pp.
3—48).
New
York: Cambridge
University
Press.
Graf,
P., &
Mandler,
G.
(1984). Activation makes words more
accessible
but not
necessarily more retrievable.
Journal
of
Verbal
Learning
and
Verbal
Behavior,
23,
553-568.
Graf,
P., &
Schacter,
D. L.
(1985). Implicit
and
explicit memory
for
new
associations
in
normal
and
amnesic subjects. Journal
of
Ex-
perimental
Psychology;
Learning,
Memory,
and
Cognition,
11,
501-518.
Griffin,
D. R.
(1981).
The
question
of
animal
awareness:
Evolutionary
continuity
of
menial
experience.
New
York:
Rockefeller
University
Press.
Griffin,
D. R.
(1984). Animal thinking. Cambridge,
MA:
Harvard
University
Press.
Hasher,
L.,
&
Zacks,
R.
T.
(1979).
Automatic
and
effortful
processes
in
memory.
Journal
of
Experimental
Psychology:
General,
108,
356-388.
Hasher,
L.,
&
Zacks,
R. T.
(1984).
Automatic processing
of
funda-
mental information.
American
Psychologist,
39,
1372-1388.
Hayes,
N. A, (in
press).
Consciousness
cannot
explain
syntactical
judgment:
A
rebuttal
of
Dulany,
Carlson,
and
Dewey.
Memory
&
Cognition.
Hayes-Roth,
F.
(1979). Distinguishing theories
of representation: A
critique
of
Anderson's
"Arguments
Concerning Mental Imagery."
Psychological
Review,
86,
376-392.
Holender,
D.
(1986). Semantic activation
without
conscious identi-
fication
in
dichotic
listening,
parafoveal
vision,
and
visual masking:
A
survey
and
appraisal.
The
Behavioral
and
Brain
Sciences,
9, 1-
66.
Holland,
J. H.,
Holyoak,
K.
J.,
Nisbett,
R.
E.,
&
Thagard,
P. R.
(1986).
Induction:
Processes
of
inference,
learning
and
discovery.
Cambridge,
MA: MIT
Press.
Howard,
J.
H.,
&
Dallas,
J. A.
(1980). Syntactic
and
semantic factors
in
the
classification
of
nonspeech
transient patterns.
Perception
and
Psychophysics,
28,
431-439.
Jacoby,
L.
L.,
&
Dallas,
M.
(1981).
On the
relationship between
autobiographical memory
and
perceptual learning.
Journal
of
Ex-
perimental
Psychology:
General,
110,
306-340.
Jaynes,
J.
(1976).
The
origins
of
consciousness
in the
breakdown
of
the
bicameral
mind. Boston: Houghton
Mifflin.
Jones,
M.
R.,
&
Myers,
i. L.
(1966).
A
comparison
of two
methods
of
event randomization
in
probability learning.
Journal
of
Experi-
mental
Psychology,
72,
909-911.
Jung,
C.
(1926).
Psychological
types
(H. G.
Baynes,
Trans.).
London:
Routledge
&
Regan
Paul.
Kunst-Wilson,
W.
R.,
&
Zajonc,
R. B.
(1980).
Affective
discrimina-
tion
of
stimuli
that cannot
be
recognized.
Science,
207,
557-558.
Lewicki,
P.
(1985).
Nonconscious
biasing
effects
of
>single
instances
on
subsequent judgments.
Journal
of
Personality
and
Social
Psy-
chology,
48,
563-574.
Lewicki,
P.
(1986a),
Nonconscious
social
information
processing.
New
York: Academic Press.
Lewicki,
P.
(1986b).
Processing information about covariations that
cannot
be
articulated.
Journal
of
Experimental
Psychology:
Learn-
ing,
Memory,
and
Cognition,
12,
135-146.
Lewicki,
P.,
Czyzewska,
M.,
&
Hoffman,
H.
(1987). Unconscious
acquisition
of
complex procedural knowledge.
Journal
of
Experi-
mental
Psychology:
Learning,.
Memory,
and
Cognition,
13,
523-
530.
Lewicki,
P.,
Hill,
T.,
&
Bizot,
E.
(1988). Acquisition
of
procedural
knowledge
About
a
pattern
of
stimuli that cannot
be
articulated.
Cognitive
Psychology,
20,
24-37.
Mace,
W. M.
(1974).
Ecologically
stimulating cognitive psychology:
Gibsonian
perspectives.
In W.
Weimer
& D.
Palermo
(Eds.),
Cog-
nition
and the
symbolic
processes.
Ilillsdale,
NJ:
Erlbaum.
Marcel,
A. J.
(1983).
Conscious
and
unconscious perception: Exper-
iments
on
visual masking
and
word recognition.
Cognitive
Psy-
chology.
15,
197-237.
Mathews,
R.
C.,
Buss,
R.
R.,
Stanley,
W.
B.,
Blanchard-Fields,
F.,
Cho,
J.-R.,
&
Druhan,
B. (in
press).
The
role
of
implicit
and
explicit
processes
in
learning
from
examples:
A
synergistic
effect.
Journal
of
Experimental
Psychology:
Learning,
Memory,
and
Cognition.
Miller,
G. A.
(1967).
The
psychology
of
communication.
New
York:
Basic Books.
Millward,
R. B.
(1981).
Models
of
concept formation.
In R. E.
Snow,
P.A.
Frederico,
& W. E.
Montague
(Eds.),
Aptitude,
learning,
and
instruction:
Cognitive
process
analysis.
Hillsdale,
NJ:
Erlbaum.
Millward,
R.
B.,
&
Reber,
A. S.
(1968). Event-recall
in
probability
learning.
Journal
of
Verbal
Learning
and
Verbal
Behavior,
7,
980-
989.
Millward,
R.
B.,
&
Reber,
A. S.
(1972).
Probability
learning:
Contin-
gent-event
sequences
with
lags. American
Journal
of
Psychology,
85,
81-98.
Milner,
B.,
Corkin,
S.,
&
Teuber,
H. L.
(1968).
Further analysis
of
me
hippocampal amnesic syndrome:
14-year
follow-up
study
of
H.M.
Neuropsychologia,
6,
215-234.
Moeser,
S.
D.,
&
Bregman,
A. S.
(1972).
The
role
of
reference
in the
acquisition
of a
miniature
artificial
language.
Journal
of
Verbal
Learning
and
Verbal
Behavior,
11.
759-769.
Morgan,
J.
L.,
&
Newport,
E. L.
(1981).
The
role
of
constituent
structure
in the
induction
of an
artificial
language.
Journal
of
Verbal
Learning
and
Verbal
Behavior,
20,
67-85.
Neisser,
U.
(1976).
Cognition
and
reality.
San
Francisco: Freeman.
Nelson,
K.
(1986).
Making
sense:
The
acquisition
of
shared
meaning.
New
York: Academic Press.
Nisbett,
R.
E.,
&
Wilson,
T. D.
(1977).
Telling more than
we
know:
Verbal
reports
on
mental
processes.
Psychological
Review,
84,
231-
259.
Palmer,
S. E.
(1978).
Fundamental aspects
of
cognitive representa-
tion.
In
E.
Rosen
& B. B.
Lloyd
(Eds.),
Cognition
and
categorization
(pp.
259-303). Hillsdale,
NJ:
Erlbaum.
Polanyi,
M.
(1958).
Personal
knowledge:
Toward
a
post-critical
phi-
losophy.
Chicago: University
of
Chicago Press.
Posner,
M.
I.,
&
Keele,
S. W.
(1968).
On the
genesis
of
abstract ideas.
Journal
of
Experimental
Psychology,
77,
353-363.
Posner,
M.
I.,
&
Keele,
S. W.
(1970).
Retention
of
abstract ideas.
Journal
of
Experimental
Psychology,
83,
304-308.
Pylyshyn,
Z.
(1979).
Validating computational models:
A
critique
of
Anderson's
indeterminacy
of
representation claim.
Psychological
Review,
86,
383-394.
Pylyshyn,
Z.
(1980). Computation
and
cognition: Issues
in the
foun-
dation
of
cognitive science.
The
Behavioral
and
Brain
Sciences,
3,
111-169.
Rathus,
J.,
Reber,
A.
S.,
&
Kushner,
M.
(1988).
Implicit
learningand
anxiety
levels.
Manuscript
in
preparation.
Reber,
A. S.
(1965).
Implicit
learning
of
artificial
grammars.
Unpub-
lished
master's thesis,
Brown
University.
Reber,
A. S.
(1966).
A
perceptual
learning
analysis
of
probability
learning.
Unpublished doctoral dissertation,
Brown
University.
Reber,
A. S.
(1967).
Implicit learning
of
artificial
grammars.
Journal
of
Verbal
Learning
and
Verbal
Behavior,
77,
317-327.
Reber,
A. S.
(1969). Transfer
of
syntactic structure
in
synthetic
languages.
Journal
of
Experimental
Psychology,
81,
115-119.
Reber,
A. S.
(1976).
Implicit learning
of
synthetic languages:
The
role

IMPLICIT
LEARNING
AND
TACIT KNOWLEDGE
235
of
instructional set.
Journal
of
Experimental
Psychology:
Human
Learning
and
Memory,
2,
88-94.
Reber,
A. S., &
Allen,
R.
(1978). Analogy
and
abstraction strategies
in
synthetic grammar learning:
A
functionalist
interpretation.
Cog-
nition,
6,
189-221.
Reber,
A.
S.,
Allen,
R.,
&
Regan,
S.
(1985).
Syntactical learning
and
judgment,
still unconscious
and
still
abstract:
Comment
on
Dulany,
Carlson,
and
Dewey.
Journal
of
Experimental
Psychology:
General,
114,
17-24.
Reber,
As
S,,
Kassin,
S. M.,
Lewis,
S.,
&
Cantor,
G. W.
(1980).
On
the
relationship
between
implicit
and
explicit
modes
in the
learning
of
a
complex
rule
structure.
Journal
of
Experimental
Psychology:
Human Learning
and
Memory,
6,
492-502.
Reber,
A.
S.,
&
Lewis,
S.
(1977).
Toward
a
theory
of
implicit learning:
The
analysis
of the
form
and
structure
of
a
body
of
tacit
knowledge.
Cognition,
5,
333-361.
Reber,
A.
S.,
&
Millward,
R. B.
(1965). Probability learning
and
memory
for
event sequences.
Psychonomic
Science,
3,
431-432.
Reber,
A.
S.,
&
Millward,
R. B.
(1968).
Event
observation
in
proba-
bility
learning.
Journal
of
Experimental
Psychology,
77,
317-327.
Reber,
A.
S.,
&
Millward,
R. B.
(1971).
Event tracking
in
probability
learning.
American
Journal
of
Psychology,
84,
85-99.
Rescorla,
R. A.
(1967).
Pavlovian
conditioning
and its
proper control
procedures.
Psychological
Review,
74,
71-80.
Rescorla,
R. A.
(1988). Pavlovian
conditioning:
It's
not
what
you
think
it is.
American
Psychologist,
43,
151-160.
Rosen,
E., &
Lloyd,
B. B.
(1978).
Representations.
In E.
Rosen
&
B.
B.
Lloyd
(Eds.),
Cognition
and
categorization
(pp. 213-215).
Hillsdale,
NJ:
Erlbaum.
Roter,
A.
(1985).
Implicit
processing:
A
developmental
study.
Unpub-
lished
doctoral dissertation, City University
of New
York.
Roy,
F. A.
(1982). Action
and
performance.
In A. W.
Ellis
(Ed.),
Normality
and
pathology
in
cognitive
functions.
New
York: Aca-
demic Press.
Rumelhart,
D.
E.,
&
McClelland,
J. L.
(Eds.) (1986).
Parallel
distrib-
uted
processing.
Cambridge,
MA: MIT
Press.
Scarborough,
D.
L.,
Cortese,
C., &
Scarborough,
H. S.
(1977).
Fre-
quency
and
repetition
effects
in
lexical memory.
Journal
of
Exper-
imental
Psychology:
Human
Perception
and
Performance,
3,
1-17.
Scarborough,
D.
L.,
Gerard,
L.,
&
Cortese,
C.
(1,979).
Accessing
lexical
memory:
The
transfer
of
word
Repetition
effects
across task
and
modality.
Memory
&
Cognition,
7,
3-12.
Schacter,
D. L.
(1987}.
Implicit
memory:
History
and
current
status.
Journal
of
Experimental
Psychology:
Learning,
Memory,
and
Cog-
nition,
13,
501-518.
Schacter,
D.
L.,
&
Graf,
P.
(1986),
Effects
of
elaborative
processing
on
implicit
and
explicit
memory
for new
associations.
Journal
of
Experimental
Psychology:
Learning,
Memory,
and
Cognition,
12,
432-444.
Schlesinger,
I. M.
(1982).
Steps
to
language-
Toward
a
theory
of
native
language
acquisition.
Hillsdale,
NJ:
Erlbaum.
Seamon,
J.
G.,
Brody,
N., &
Kauff,
D. M.
(1983).
Affective
discrim:
ination
of
stimuli
that
are not
recognized:
Effects
of
shadowing,
masking,
and
cerebral
laterality.
Journal
of
Experimental
Psychol-
ogy:
Learning,
Memory,
and
Cognition,
9,
544-555.
Seamon,
J.
G.,
Marsh,
R.
L.,
&
Brody,
N.
(1984). Critical importance
of
exposure duration
for
affective
discrimination
of
stimuli that
are
not
recognized.
Journal
of
Experimental
Psychology:
Learning,
Memory,
and
Cognition,
10,
465-469.
Shallice,
T., &
Saffian,
E.
(1986). Lexical processing
in the
absence
of
explicit word identification: Evidence
from a
letter-by-letter
reader.
Cognitive
Neuropsyehology,
3,
429-458.
Smith, E.E.,
&
Medin,
D. L.
(1981).
Categories
and
concepts.
Caj%,.
bridge,
MA:
Harvard University Press.
Strauss,
M.
E.,
Weingartner,
H.,
&
Thompson,
K.
(1985). Remem-
bering
words
and how
often
they
occurred
in
memory-impaired
patients.
Memory
&
Cognition,
13,
1507-1510.
Warrington,
E.
K.,
&
Weiskrantz,
L.
(1982).
Amnesia:
A
disconnec-
tion syndrome?
Newopsychologia,
20,
233-248.
Weiskrantz,
L.
(1986).
Blindsight.
New
York:
Oxford
University
Press.
Westcott,
M.
R.
(1968).
Toward
a
contemporary
psychology
of
intui-
tion.
New
York: Holt,
Rinehart
&
Winston.
Received
January
11,
1988
Revision
received
May 10,
1988
Accepted
May
11,
1988
1990
APA
Convention
"Call
for
Programs"
The
"Call
for
Programs"
for the
1990
APA
annual
convention
will
be
included
in the
October
issue
of the APA
Monitor.
The
1990
convention
will
be
held
in
Boston,
Massa-
chusetts),
from
August
10
through
August
14
Deadline
for
submission
of
program
and
presentation:proposals
is
December
15,
1989.
This
earlier
deadline
is
required
because
many
university
and
college
campuses
will
close
for the
holidays
in
mid-December
and
because
the
convention
is in
mid-August.
Additional
copies
of the
"Call"
will
be
available
from
the
APA
Convention
Office
in
October.