Thoughts Beyond Words: When Language Overshadows Insight

Abstract

Four experiments examined whether verbalization can interfere with insight problem solving. In Exp 1, Ss were interrupted during problem solving and asked either to verbalize their strategies (retrospective verbalization) or engage in an unrelated activity (control). Ss in the retrospective verbalization condition were significantly less successful than control Ss at solving the problems. Exp 2 replicated the finding of Exp 1 and demonstrated that the control Ss' advantage was not due to any beneficial effect of the interruption. In Exp 3, concurrent, nondirective verbalization impaired the solving of insight problems but had no effect on noninsight problems. In Exp 4, the effect of concurrent verbalization on insight was maintained even when Ss were encouraged to consider alternative approaches. Together, these findings are consistent with the hypothesis that verbalization can result in the disruption of nonreportable processes that are critical to achieving insight solutions. (PsycINFO Database Record (c) 2012 APA, all rights reserved)

Full text

Journal
of
Experimental Psychology: General
1993,
Vol.
122.
No. 2,
166-183
Copyright
1993
by
the
American
Psychological
Association
Inc
0096-3445/9_V$3.fX)
Thoughts Beyond Words: When Language Overshadows Insight
Jonathan
W.
Schooler, Stellan Ohlsson,
and
Kevin Brooks
Four
experiments examined whether verbalization
can
interfere with insight problem
solving.
In
Experiment
1, Ss
were interrupted during problem solving
and
asked either
to
verbalize their
strategies (retrospective verbalization)
or
engage
in an
unrelated activity (control).
Ss in the
retrospective verbalization condition were significantly
less
successful than control subjects
at
solving
the
problems. Experiment
2
replicated
the
finding
of
Experiment
1 and
demonstrated
that
the
control
Ss'
advantage
was not due to any
beneficial
effect
of the
interruption.
In
Experiment
3,
concurrent,
nondirective verbalization impaired
the
solving
of
insight problems
but had no
effect
on
noninsight
problems.
In
Experiment
4, the
effect
of
concurrent verbalization
on
insight
was
maintained
even when
Ss
were encouraged
to
consider alternative approaches. Together, these
findings
are
consistent
with
the
hypothesis
that
verbalization
can
result
in the
disruption
of
nonreportable
processes
that
are
critical
to
achieving
insight
solutions.
Although thought
processes
often closely correspond
to
the
contents
of
inner
speech
(e.g.,
Ericsson
&
Simon,
1980, 1984; Sokolov, 1972; Vygotsky,
1934/1989),
certain
thoughts
have
a
distinctly nonverbal character.
A
long tra-
dition
of
scholars have suggested that creative thoughts,
and
in
particular
"insights"
(problem solutions that occur
unexpectedly following
an
impasse),
are
distinct from lan-
guage processes (e.g., Bergson, 1902;
Bruner,
1966; Ghis-
elin,
1952;
Hadamard,
1954; Koestler, 1964; Maier, 1931;
Poincare,
1952; Polanyi, 1967; Wallas, 1926;
Wertheimer,
1959).
One of the
most eloquent spokespersons
for the
nonverbal characteristics
of
certain thoughts
is
William
James
(1890),
who
noted that many important insights
are
reported
to
have occurred
in the
absence
of
words:
"Great
thinkers have vast premonitory glimpses
of
schemes
of re-
lations between terms, which hardly even
as
verbal images
enter
the
mind,
so
rapid
is the
whole
process"
(p.
255).
Al-
bert Einstein (cited
in
Schlipp, 1949) later provided elo-
quent
support
for
James's claim that creative insights
often
precede their translation into language, noting that
"These
thoughts
did not
come
in any
verbal formulation.
I
very
rarely think
in
words
at
all.
A
thought comes,
and I may
try
to
express
it in
words afterwards"
(p.
228).
In
addition
to
the
anecdotal
reports
of
great thinkers, empirical
re-
search,
to be
discussed shortly,
has
also provided evidence
for
nonreportable insight
processes
(e.g., Bowers, 1991;
Bowers,
Regehr,
Balthazard,
&
Parker, 1990;
Durkin,
1937; Metcalfe, 1986a, 1986b; Metcalfe
&
Wiebe, 1987).
Jonathan
W.
Schooler,
Stellan
Ohlsson,
and
Kevin
Brooks,
De-
partment
of
Psychology, University
of
Pittsburgh.
This
research
was
supported
by a
grant
to
Jonathan
W.
Schooler
from
the
National Institute
of
Mental Health.
We are
grateful
for
the
comments
of
Anders
Ericsson,
Marte
Fallshore,
Kenneth
Koedinger,
Joseph
Melcher,
Mitchell
Nathan,
William
Oliver,
Robert
Ryan,
Carmi
Schooler,
Lael
Schooler,
and
Timothy
Wilson.
Correspondence
concerning
this
article
should
be
addressed
to
Jonathan
W.
Schooler,
635
Learning
Research
and
Development
Center,
3939
O'Hara
Street, University
of
Pittsburgh, Pittsburgh,
Pennsylvania
15260.
If
insights
are in
some sense distinct
from
language, then
this
raises
the
question, What
is the
relationship between
language
and
insight?
One way to
empirically
address
this
question
is to
examine
the
effect
of
attempting
to put
insight
processes
into words.
If
insight processes
are
either com-
pletely dependent
or
independent
of the
language system,
then
trying
to
articulate these
processes
should
be of
little
consequence.
If,
however, insight
processes
are
both distinct
from
language
and at the
same time influenced
by it,
then
verbalization might well interact with insight. Specifically,
attempts
to
translate insight problem solving into words
might promote
a
reliance
on
verbalizable
processes
while
disrupting
the use of the
hypothetical nonreportable com-
ponents
of
insight.
In
this article
we
explore this possibility
by
examining
the
potential interference that
may
result
from
articulating insight processes.
The
Effects
of
Verbalization
on
Cognition
A
number
of
recent studies have indicated that verbaliza-
tion
of
nonverbal tasks
can
interfere with successful perfor-
mance (e.g., Fallshore
&
Schooler, 1993; Schooler
&
Engstler-Schooler,
1990;
Schooler,
Ryan,
&
Reder,
1991;
Wilson
et
al.,
in
press;
Wilson
&
Schooler,
1991).
For ex-
ample, Schooler
and
Engstler-Schooler
(1990)
examined
the
effects
of
verbalization
on the
primarily nonreportable pro-
cess
of
face recognition.
Face
recognition requires consid-
erable information that cannot
be
adequately verbalized; that
is,
people
are
remarkable
in
their ability
to
recognize faces
yet
they have great
difficulty
describing
the
basis
for
their
recognition judgments (Ellis, 1984; Polanyi, 1967; Schooler,
1989).
Schooler
and
Engstler-Schooler
observed
that when
subjects
attempted
to
describe
a
previously seen face, their
subsequent
recognition
was
substantially reduced compared
with subjects
who did not
describe
the
face. Schooler
and
Engstler-Schooler
called this interference verbal overshad-
owing
because
of the
evidence that verbalization focuses sub-
jects
on the
verbally relevant information
and
thereby over-
shadows information that
is not
readily verbalized.
For
166

THOUGHTS BEYOND WORDS
167
example,
the
effects
of
verbalization
were
found
to
depend
on
the
verbalizability
of the
stimulus; verbalization impaired
memory
for
both faces
and
colors
while marginally improv-
ing
memory
for a
verbal stimulus
(a
spoken statement). Fur-
ther
support
for the
overshadowing nature
of the
verbaliza-
tion came
from
the
observation that when subjects'
recognition time
was
limited,
the
negative
effects
of
condi-
tion
were attenuated. Presumably, limiting
the
recognition
time
reduced subjects' opportunity
to
retrieve
the
verbaliz-
able
information
and
thereby prevented that information
from
overshadowing
the
original visual memory trace.
A
recent study
by
Wilson
and
Schooler (1991) provided
further
evidence that verbalization
can
increase
the
salience
of
the
verbalizable attributes
of a
stimulus
and
thereby over-
shadow
the
nonverbalizable attributes. Wilson
and
Schooler
examined
the
effects
of
verbalizing
affective
judgments,
which
is
another domain
for
which subjects
are
unable
to
fully
articulate their thought
processes
(Nisbett
&
Wilson,
1977).
For
example,
in one
study, subjects were asked
to
evaluate
different
brands
of
strawberry jams. Control sub-
jects tasted
the
jams
and
then rated them,
and
verbalization
subjects
tasted
the
jams
and
then wrote
the
reasons
for
their
preferences prior
to
rating them. When subjects' ratings were
compared with those
of
taste experts
from
Consumer
Re-
ports,
it was
found
that
the
control subjects agreed quite well
with
the
experts, whereas verbalization subjects showed little
agreement. Wilson
and
Schooler concluded that verbaliza-
tion
causes subjects
to
base their evaluations
on the
report-
able
aspects
of
their taste experience
and to
ignore those
aspects
that
cannot
be
adequately
expressed
in
words.
In
support
of
this hypothesis,
it was
observed that
the
attitudes
implied
by
subjects'
written
evaluations
of the
jams strongly
correlated with their ultimate ratings.
Although
potentially relevant
to the
many discussions
of
when
language
does
and
does
not
mediate thought (for
a
recent
review
see
Hunt
&
Agnoli, 1991), most studies that
have
examined
the
effects
of
articulating thought processes
have
been primarily motivated
by
methodological concerns
regarding
the use of
"think-aloud" protocols.
For
example,
Ericsson
and
Simon
(1980,1984)
supported their conclusion
that
think-aloud protocols
can be
used
as a
valid source
of
data
with
an
extensive review indicating that under most
circumstances verbalizing cognitive processes
has
little
or no
effect
on
performance. Although Ericsson
and
Simon sug-
gested
that verbalization
is
typically benign, they
did
identify
one
situation
in
which
it may be
reactive, namely, when sub-
jects
attempted
to
articulate information that would
not
oth-
erwise
have been heeded. Although this constraint
on the
benign
properties
of
verbalization
is
consistent
with
the
ver-
bal
overshadowing approach outlined
above,
Ericsson
and
Simon
concluded that
receding
of
nonverbal processes into
verbal
form
does
not
cause subjects
to
alter
the
information
that
they consider. Rather,
in
their view,
receding
introduces
an
extra step
and
thereby slows down,
but
does
not
quali-
tatively
alter, cognitive processing (see Ericsson
&
Simon,
1984,
p.
79). According
to
Ericsson
and
Simon, qualitative
effects
on
cognitive processing
do not
result
from
verbalizing
particular
types
of
processes,
but
rather
are
created
by
par-
ticular types
of
verbalization instructions. Ericsson
and
Crutcher
(1991)
gave
the
following summary
of
Ericsson
and
Simon's
(1980,
1984) review:
They [Ericsson
&
Simon]
found
that think-aloud
and
retro-
spective reports yield valid data
on
cognitive
processes
with-
out
interfering
with
the
normal course
of
thought sequences.
They also
found
that reactive
effects
and
poor validity
of
verbal
reports were inevitably
due to
other types
of
verbal
report procedures that attempted
to
obtain more information
than
subjects spontaneously
report:
for
example, instructing
subjects
to
give reasons
for
generated
thoughts,
(p. 66)
There
are a
number
of
possible reasons
for the
differences
between
the
conclusions
of the
verbal overshadowing
ap-
proach
as
outlined
by
Schooler
and
Engstler-Schooler
(1990)
and
Ericsson
and
Simon's
(1980,
1984) suggestion that ver-
bal
receding
of
nonverbal
processes
should
not
qualitatively
alter
those
processes.
This
disparity
may be due to
differ-
ences
in the
type
of
cognitive
processes
that were evaluated.
For
example, Schooler
and
Engstler-Schooler considered
recognition
of
visual stimuli
and
Wilson
and
Schooler
(1991)
considered affective judgments.
In
contrast, Ericsson
and Si-
mon
focused more
on
higher order cognitive processes such
as
reasoning
and
problem solving.
It may be
that
the hy-
pothesized verbal overshadowing resulting
from
verbaliza-
tion only occurs
for
very basic processes
and
that
as
Ericsson
and
Simon suggest,
receding
does
not
substantially interfere
with
nonreportable
processes
associated with higher order
tasks such
as
problem solving.
Differences
between
the
conclusions
of the two ap-
proaches
may
also
be due to the
fact
that
the
verbalization
used
by
Schooler
and
colleagues
differs
in
significant ways
from
the
type
of
verbalization that
Ericsson
and
Simon
(1980, 1984) suggested
is
nonreactive. Ericsson
and
Simon
argued
that verbalization should
not
affect
performance
when
it is
nondirected
and
done during
the
cognitive activity
(concurrent). However, Schooler
and
Engstler-Schooler
(1990)
asked subjects
to
describe
the
appearance
of a
face
that
was
previously seen (retrospective verbalization).
In ad-
dition, they specifically directed subjects
to
describe
the ap-
pearance
of the
face, which
may not be
equivalent
to ex-
pressing naturally occurring thoughts. Wilson
and
Schooler
(1991)
asked subjects
to
verbally
justify
the
basis
for
their
decisions and, thus, their instructions
not
only elicited ver-
balization
but
also
focused subjects
on
particular
types
of
thoughts, that
is,
reasons
for and
inferences about their
de-
cision
process.
In
short,
the
question remains open whether
the
verbal
overshadowing that
has
been hypothesized
to
result
from
verbalization
of
basic cognitive processes such
as
perception,
recognition,
and
affective
responses
can
generalize
to
higher
order
processes such
as
problem solving. Moreover,
if
verbal
overshadowing
does
generalize
to
higher order domains,
it
is not
clear whether
it
would occur
if
subject's directions
were
nondirective
and
their verbalization concurrent.
To ad-
dress this issue,
it is first
necessary
to
identify
a
higher order
cognitive activity that involves considerable nonreportable
processing
and
then examine
the
effects
of
verbalization
on
this
activity.

168
J.
SCHOOLER,
S.
OHLSSON,
AND K.
BROOKS
Evidence That Insight Problem Solving Involves
Nonreportable
Processes
One
higher order cognitive activity that
has
frequently
been hypothesized
to
involve
nonreportable
processes
is in-
sight
problem solving. Although
it is
difficult
to
study
the
nonreportable
processes
hypothesized
to be
associated
with
the
thinking
of
creative geniuses, examination
of
insight
puz-
zle
problem solving
in the
laboratory
may
serve
as a
model
for
more important creative insight.
By an
insight problem
we
mean
a
problem that
(a) is
well within
the
competence
of
the
average subject;
(b) has a
high probability
of
leading
to
an
impasse, that
is, a
state
in
which
the
subject does
not
know
what
to do
next;
and (c) has a
high probability
of
rewarding
sustained
effort
with
an
"Aha"
experience
in
which
the im-
passe
is
suddenly broken
and
insight into
the
solution
is
rap-
idly
attained.
The
impasse-insight
sequence
was first ob-
served
by
Gestalt
psychologists
(Duncker,
1945; Kohler,
1921;
Wertheimer,
1959)
who
invented
and
studied
a
number
of
such problems (see
Ohlsson,
1984a,
for a
review
of the
Gestalt work).
Many
discussions
of
insight processes have proposed
the
existence
of
nonreportable
or
unconscious
processes
(e.g.,
Bergson, 1902;
Burner,
1966; Ghiselin, 1952;
Hadamard,
1954;
Koestler,
1964; Maier,
1931;
Poincare,
1952; Polanyi,
1967;
Wallas, 1926; Wertheimer, 1959). Evidence
of
the
non-
reportable qualities
of
insight processes comes
from
a
variety
of
sources.
As
mentioned earlier,
one
type
of
source
is the
anecdotal
reports
of
scientists
and
other
creative
individuals
who
report experiencing their discoveries
as
occurring
in
wordless thoughts. Similar types
of
experiences have also
been observed with
the
more mundane discoveries
of
solu-
tions
to
laboratory insight problems.
For
example, Durkin
(1937)
observed that problem solvers typically grow silent
immediately
before
an
insight.
She
asked subjects
to
ver-
balize their thought processes while attempting
to
solve
in-
sight
problems. Durkin observed that
the
solutions
to
these
problems involved "sudden reorganization"
(p.
80). Durkin
further
noted that these sudden reorganizations were
typi-
cally
preceded
by an
inability
to
verbalize
thoughts:
"Usually
a
rather short very quiet pause occurs just preceding
the
sud-
den
reorganization"
(p.
80). More recent discussion
of
pro-
tocols associated with insight problem solving have also
noted
the
degree
to
which
the
critical steps
in the
insight
solution
are not
reported (e.g., Ericsson
&
Simon, 1984;
C.
A.
Kaplan
&
Simon, 1990).
Further evidence
for the
nonreportable processes associ-
ated with insight comes
from
analyses
of
subjects' ability
to
anticipate
their nearness
to
insight solutions.
For
example,
Metcalfe
and
Wiebe
(1987)
found
that subjects know when
they
are on the
verge
of
solving analytic
problems
(such
as
those found
in
standardized tests),
as
evidenced
by
accurate
ratings
of
closeness
to
solution. However, subjects
had
little
ability
to
rate their closeness
to
solutions
to
insight problems,
supporting
the
hypothesis that
the
processes that prepare
for
insights
are
inaccessible. Finally, Maier (1931)
asked
sub-
jects
to
retrospectively report their solutions
to an
insight
problem
for
which they
had
been given
a
(seemingly acci-
dental)
hint
by the
experimenter. Subjects
who
could report
the
stepwise construction
of the
solution also reported
the
hint
and its
effect
on
their problem solution,
but
subjects
who
reported
the
solution
as
having arrived
in a
flash
of
insight
gave
no
evidence
of
being aware
of the
hint.
Early
accounts
of the
mechanisms underlying unconscious
production
of
insight solutions suggested that elaborate
un-
conscious inference
processes
produced insight solutions.
More
recent explanations
of
insight have focused instead
on
nonreportable memory
and
perceptual processes
and in
par-
ticular
on
spreading activation (Bowers,
1991;
Bowers
et
al.,
1990; Langley
&
Jones, 1988; Ohlsson, 1984b, 1992).
For
example,
Ohlsson (1992) describes
an
information process-
ing
model
of
insight problem solving
in
which inappropriate
perception
of a
problem causes subjects
to
focus
on
solution
approaches that
do not
directly activate
the
memory entries
that
are
needed
to
solve
the
problem.
At the
same time, how-
ever,
subawareness
activation
may
accumulate
on
knowl-
edge
structures that
are
critical
to the
correct
solution.
For
example,
consider
the
following
problem:
A man and his son
are in a
serious care accident.
The
father
is
killed
and the son
is
rushed
to the
emergency room. Upon arrival,
the
attending
doctor looks
at the
child
and
gasps, "This child
is my
son!"
Who is the
doctor?
Of
course
the
answer
is
"The doctor
is
the
mother." However,
failure
to
solve this problem
may
result
if
subjects
fail
to
retrieve into consciousness
the
critical
memory
element that "Doctors
can be
mothers." Although
unable
to
retrieve
the
"mother" memory
entry,
activation
to
that
element
may
accumulate
at a
subawareness level.
Over
time,
a
variety
of
processes
may
contribute
to the
subawareness
activation
of
these critical entries,
including
elaboration, reencoding,
and
constraint relaxation. Elabora-
tion
involves adding
information
to a
problem representa-
tion,
perhaps
by
closer
scrutiny
of the
problem situation.
Reencoding involves changing
the
perceptual interpretation
of the
situation,
for
example,
by
changing figure-ground
re-
lations.
Constraint relaxation involves rejecting features
of
the
solution that were previously thought necessary. When
the
mental representation
of the
problem
is
changed
in any
of
these ways, activation
will
spread down alternative path-
ways
in
memory, raising
the
level
of
activation
of
relevant
memory
entries.
As
soon
as the
activation
to the
critical mem-
ory
causes
it to
enter into awareness,
the
crucial element
necessary
for the
solution
to the
problem
is
retrieved, thereby
resulting
in the
suddenness
of the
insight experience.
Ac-
cordingly,
in the
previous example, when
the
activation
to the
construct
of
"mother"
or the
fact
that
"Doctors
can be
moth-
ers" reaches
the
level
of
awareness,
a
sudden solution
is
reached.
Further empirical evidence
for
subawareness activation
of
elements critical
to
insight puzzle problem solving
has
come
from
research
examining subawareness
activation
in
tasks
that
are
similar
to
insight problems. Bowers
et al.
(1990)
and
Bowers
(1991)
looked
at the
attempted responses provided
by
subjects
who
were unable
to
generate
the
correct solution
to
"insightlike"
problems such
as
recognizing distant seman-
tic
associates (e.g., What word relates
to
both arsenic
and
shoe?)
and
solving anagrams. Bowers observed that
incorrect
guesses often have some semantic relationship
to the
correct
solution,
indicating that solution-relevant information
was

THOUGHTS BEYOND WORDS
169
being activated
and
implying that this
subawareness
activa-
tion
may
ultimately contribute
to the
solving
of the
problem.
Bowers
et
al.
concluded that "The
suddeness
with which
insights
sometimes occur thus
represents
an
abrupt aware-
ness
of a
mental product
or end
stage generated
by
more
continuous,
sub
rosa cognitive
processes"
(p.
95).
Another result that
has
been suggested
to
imply
the
pos-
sibility
of
subawareness activation
of
insight solutions
is
priming
effects
resulting
from
tip-of-the-tongue states.
Yaniv
and
Meyer
(1987)
observed
that
even
when subjects were
unable
to
retrieve answers
to
factual questions
(e.g.,
What
is
a
16th-century navigational
instrument?),
those answers
were nevertheless more quickly responded
to in a
subsequent
lexical decision task. Yaniv
and
Meyer concluded that
the
solutions, though
not
retrieved, were still activated
at a
sub-
awareness threshold,
and
they suggested that analogous pro-
cesses
may
occur during insight problem solving.
In
sum, there seems
to be
considerable evidence that pro-
cesses associated with insight problem solving
are not
avail-
able
to
verbal
report.
The
nature
of
these
nonreportable
in-
sight
processes
is
somewhat
less
clear.
There
is,
however,
accumulating
evidence
for the key
role
of
spreading activa-
tion
to
knowledge structures that were
not
originally per-
ceived
to be
relevant
to the
problem solution.
Could
Verbalization Impair
Insight
Problem Solving?
The
characterization
of
insight problem solving
as
involv-
ing
nonreportable information processing components,
in
conjunction
with
the
suggestion that verbalization
can
over-
shadow
difficult-to-report
perceptual
and
memory
processes,
suggests that
requests
to
verbalize
attempted
solutions
to in-
sight
problems
may
interfere with successful problem solv-
ing.
Problem solving requires both
the
search
for
relevant
information
and
also
the
manipulation
of
that information.
As
Ericsson
and
Simon
(1984)
noted,
people
cannot verbalize
the
memory retrieval
process
itself
but
only
the
product
of
a
memory search.
At the
same time,
people
are
quite adept
at
articulating
the
manner
in
which they manipulate
infor-
mation
that
is
already
in
working memory.
If
verbalization
causes subjects
to
neglect those
processes
that
are not
readily
verbalized,
then verbalization during problem solving
may
cause
subjects
to
favor working memory manipulation
over
long-term memory retrieval.
If
insight problem solving
re-
quires
the
retrieval
of
nonobvious memory elements, then
a
reduced emphasis
on
long-term memory retrieval processes
is
likely
to
lower
the
probability
of
attaining insight. Met-
aphorically
speaking, verbalization
may
cause such
a
ruckus
in
the
"front"
of
one's
mind that
one is
unable
to
attend
to
the
new
approaches that
may be
emerging
in the
"back"
of
one's
mind.
Although
there
are
theoretical
reasons
to
believe that ver-
balization might disrupt insight problem solving, relatively
little
research
has
directly
addressed
this
issue.
In
their
re-
view
of the
effects
of
verbalization, Ericsson
and
Simon
(1984) cited
two
studies (Bullbrook, 1932; Weisberg
&
Suls,
1973) that they viewed
as
evidence that verbalization does
not
affect
insight problem solving. Careful examination
of
these studies, however, indicates that they
are not
really
in-
consistent with
the
prediction that verbalization
can
impair
performance
on
insight problems. Bullbrook
(1932)
com-
pared
the
solution strategies
of
individual subjects
who
thought
aloud while solving insight problems with
the
ret-
rospective reports
of
subjects
who
solved insight problems
in
groups. Bullbrook noted that subjects reported similar
strategies
in the two
conditions. However, Bullbrook
did not
provide
the
frequency
of
successful solutions
in the two
con-
ditions, making
it
impossible
to
determine
the
effects
of
ver-
balization
on
performance.
In
a
more recent study, Weisberg
and
Suls (1973) reported
performance levels
of
subjects
who
solved insight problems
under
both think-aloud
and
silent
conditions. Although Weis-
berg
and
Suls,
as
well
as
Ericsson
and
Simon (1984), viewed
the
results
of
these experiments
as
indicating that verbal-
ization
had no
effect
on
insight problem solving,
in
fact,
the
frequency
of
successful insight solutions
was
numerically
lower
in the
think-aloud experiment than
in the
silent
ex-
periment (24%
vs.
44%). Given that these results were
ob-
served
in
different
experiments that varied
in
other respects
besides
verbalization,
comparison
of
these
means
is not
strictly
appropriate. Nevertheless,
if
anything, they
hint
at the
possibility that verbalization
may
impair
the
successful
so-
lution
of
insight problems. Thus,
the
effects
of
verbalization
on
insight problem solving
are
still
an
open issue.
In
this study,
we
assessed
the
applicability
of the
verbal
overshadowing approach
to
higher
order
cognitive processes
by
conducting
a
systematic examination
of the
possible dis-
ruptive
effects
of
verbalization
on the
solving
of
insight prob-
lems.
In
Experiments
1 and 2, we
examined whether ver-
balization would disrupt insight problem solving under
conditions
as
comparable
as
possible
to
those
in
previous
verbal overshadowing experiments (e.g., Schooler
&
Engstler-Schooler, 1990). Specifically,
we
examined
the ef-
fects
of
directed retrospective verbalization, using
a
para-
digm
in
which subjects were interrupted while solving
in-
sight
problems
and
then probed regarding
the
strategies that
they
were using.
In
Experiments
3 and 4 we
examined
whether verbalization would
affect
insight problem solving
under conditions suggested
to be the
least likely
to
produce
reactive
effects,
that
is,
nondirected concurrent verbalization
(Ericsson
&
Simon, 1984).
If
verbalization reliably impairs
insight
solutions
under
conditions
comparable
to
both
the
verbal overshadowing paradigm
and the
standard think-
aloud paradigm, such
a
finding
would provide
at
least pre-
liminary
evidence that some component
of
insight problem
solving
may be
overshadowed
by
language.
Experiment
1
Experiment
1
examined whether retrospective verbaliza-
tion
would disrupt insight problem solving under conditions
as
comparable
as
possible
to
those
in
previous verbal over-
shadowing
experiments
(e.g.,
Fallshore
&
Schooler,
1993;
Schooler
&
Engstler-Schooler,
1990).
In the
standard
retro-
spective verbalization paradigm,
the
verbalization occurs
af-
ter the
completion
of the
primary task (e.g., Russo, Johnson,
&
Stephens,
1989);
thus,
by the
time subjects verbalize about
a
problem they have already completed
it.
However,
in
pre-

170
J.
SCHOOLER,
S.
OHLSSON,
AND
K.
BROOKS
vious
verbal overshadowing experiments, subjects verbalize
about stimuli that they have encountered
but for
which they
have
not yet
been tested.
To
maximize
the
comparability
of
the
present study
to
previous verbal overshadowing exper-
iments
we
modified
the
standard retrospective verbalization
procedure
to
enable subjects
to
reflect
on a
problem before
they
had
finished
working
on it.
Specifically,
in
Experiment
1,
subjects were interrupted
one
third
of the way
through their
allotted time
for
solving each problem
and
were asked
to
articulate
what strategies they
had
been using.
As a
control
for
the
effects
of
interruption,
the
verbalization condition
was
compared with
a
condition
in
which subjects were inter-
rupted
and
engaged
in an
unrelated activity.
If
subjects
in the
retrospective verbalization condition performed significantly
worse than control subjects, this would provide evidence that
the
verbal disruption observed
in
previous verbal overshad-
owing studies
may
generalize
to
insight problem solving.
Method
Subjects
Subjects
were
86
undergraduates
from
the
University
of
Pitts-
burgh
who
received course credit
for
their participation. Subjects
were
run in
groups
of up to 4 at a
time.
Four
subjects
were
elim-
inated
from
the
analysis because they were
familiar
with
one or
more
of the
stimulus
problems.
Materials
Pilot
research
was
conducted using
a
pool
of
insight problems
obtained
from
previous research
on
insight (e.g., Davidson
&
Stern-
berg,
1984;
Metcalfe
&
Wiebe,
1987).
Out of 10
problems,
6
were
chosen
for
which mean pilot
performance
was
closest
to
50%.
In
the
midst
of the
study,
one of the
problems
had to be
removed from
the
problem
set
because
of its use in a
lecture attended
by the
subject
pool. This problem
was
replaced
with
one for
which pilot perfor-
mance
was
comparable.
All
seven problems used
are
reproduced
in
Appendix
A.
Procedure
Each group
of
subjects
was
randomly
assigned
to one of two
conditions:
verbalization
or
unrelated interruption.
At the
beginning
of
each session, subjects were
fully
informed
of the
procedure
of
the
experiment
and the
instructions they would
be
expected
to
fol-
low.
In
each session, there were
six
trials involving
six
different
insight
problems.
The
problems were presented
in
written form,
one
at
a
time. Problem order
was
counterbalanced
to
control
for
order
and
position
effects.
Subjects were instructed
to
show
all
solutions
to
the
experimenter immediately. When subjects believed they
had
a
correct solution, they informed
the
experimenter,
who
then
checked
it. If
correct, subjects were given
an
unrelated task
to
com-
plete
for the
remainder
of the
time allotted
to the
problem.
If in-
correct,
subjects were encouraged
to
continue with
the
problem.
In
both conditions, subjects were interrupted after
2
min
of
work-
ing
on
each problem.
The
interruption lasted
1.5
min. During that
time,
those
in the
verbalization condition were instructed
to
write
out
all
relevant thoughts that
had
occurred during
the first 2 min of
solving
the
problem.
The
exact instructions were:
"Please
stop
working
on the
problem
now and
write down,
in as
much detail
as
possible, everything
you can
remember about
how you
have been
trying
to
solve
the
problem. Give information about your approach,
strategies,
any
solutions
you
tried,
and so
on." Subjects
in the un-
related
interruption condition spent
the
interruption time working
on a
standard crossword puzzle.
At the end of the
1.5
min, subjects
in
both conditions were instructed
to
resume working
on the
prob-
lem.
All
subjects
had 7.5 min to
solve each problem
(6 min of
problem-solving
time
and 1.5 min on the
control
or
experimental
activity).
Subjects worked
on
each problem
until
they arrived
at a
correct solution
or
until
7.5 min had
passed.
Results
Accuracy
Overall.
The
dependent measure
of
accuracy
was the
total
number
of
problems solved correctly
in the
allotted time
for
each subject. Subjects
in the
verbalization condition
solved significantly fewer problems
in the
allotted time than
subjects
in the
unrelated interruption condition,
r(74)
=
2.126,
p <
.05. Mean percentages
correct
were 35.6%
for
verbalization
and
45.8%
for
unrelated interruption.
Relative
to the
interruption. Overall, there
was no
sig-
nificant
difference between performance before
and
after
the
interruption,
F(l,
74) =
1.179,
p >
.05. There
was
also
no
interaction between
the
effects
of
retrospective verbalization
and
the
number
of
problems solved before
and
after
inter-
ruption,
F(l,
74) <
1.
This latter finding indicates that
the
magnitude
of the
difference between
the
percentage
of
prob-
lems solved
by
subjects
in the
verbalization
and
unrelated
interruption
conditions prior
to the
interruption (15.3%
vs.
21.7%)
was
comparable
to the
difference between these
two
conditions after
the
interruption
(20.3%
vs.
24.1
%). It
should
be
noted, however, that neither
of
these differences
was
sig-
nificant
alone
(p >
.05,
simple
effects
test).
Item
and
serial
position
effects.
Some items were more
difficult
than others. F(5, 370)
=
25.805,
p <
.0001.
How-
ever,
there
was no
interaction between item
and
condition,
F(10,
370)
< 1. An
analysis
of
variance (ANOVA) yielded
no
evidence
of a
serial position effect, F(5, 370)
=
1.362,
p
>
.05,
and no
interaction between serial position
and
ver-
balization,
F(5, 370)
=
1.604,p
>
.05.
Time
There
was no
significant difference
in the
amount
of
time
taken
to
correctly solve
the
problems
in the
unrelated inter-
ruption
and
verbalization conditions, with means
of
168.46
s
and
191.21
s,
respectively,
?(70)
=
1.274,
p >
.05.
Discussion
In
Experiment
1,
subjects
who
attempted
to
verbalize
how
they
had
been
trying
to
solve insight problems solved sig-
nificantly fewer
problems
than control subjects
who
engaged
in
an
unrelated activity
for an
equivalent period
of
time.
Thus, Experiment
1
provides support
for the
hypothesis that
verbalization interferes with
the
processes
associated
with
insight
in a
manner
comparable
to
that
observed
for
other
difficult-to-report
stimuli (e.g.,
Schooler
&
Engstler-
Schooler,
1990; Wilson
&
Schooler,
1991).

THOUGHTS
BEYOND
WORDS
171
Previous discussions
of the
effects
of
verbalization
have
made
a
distinction between
effects
that result
in
qualitative
differences
in the
course
and
structure
of a
problem solving
effort
and
effects,
such
as
time
to
solution, that cause changes
in
quantitative
aspects
(Ericsson
&
Simon, 1984; Russo
et
al.,
1989).
A
possible
explanation
of the
result
of
Experiment
1
is
that verbalization caused subjects
to
take more time
to
solve
the
experimental problems, perhaps
because
they were
aware that they would have
to
report their
strategies.
Because
the
time allotted
for
each problem
was
restricted,
a
manip-
ulation
that increased
the
time needed
to
solve
a
problem
would
cause fewer problems
to be
solved. However,
if the
verbalization request slowed down problem solving, then
it
would
be
expected that correctly
solved
problems should
have
taken longer
in the
verbalization condition. However,
there
was no
significant
difference
between
the
times
re-
quired
by
subjects
in
each condition
to
solve
the
problem.
Moreover, when problems were solved,
the
mean time
re-
quired
was
well under
the
time limit (2.5
min
less than
45%
of
the
allotted
time).
Even
if
subjects were slowed down
by
verbalization,
they still would have
had
time
to
solve
the
problem. Thus,
the
slowing down hypothesis does
not ac-
count
for our
results.
One
possible concern raised
by the
results
of
Experiment
1
is
that
the
effects
of the
instructions
to
verbalize
were
comparable
in
magnitude before
and
after
the
explicit
act of
verbalization.
Although
the
effects
of the
verbalization
in-
structions prior
to the
interruption were
not
statistically
significant,
the
comparable
difference
between
the
unrelated
interruption
and
verbalization
conditions
before
and
after
the
interruption suggests
the
possibility that subjects were
affected
by the
knowledge that they would have
to
verbalize
their
strategies.
In
fact,
a
similar
effect
of
retrospective
verbalization
was
recently reported
by
Russo
et al.
(1989),
who
found
that
in
many
cases
the
effects
of
instructing
subjects
to
verbalize their strategies
after
they
had
com-
pleted solving
a
problem (retrospective verbalization) were
comparable
to the
effects
of
verbalization during problem
solving
(concurrent verbalization). Russo
et al.
similarly
suggested that subjects
in the
retrospective condition were
presumably
influenced
by the
knowledge that they would
have
to
verbalize.
It
seems likely,
for
example, that
the
expectation
of
having
to
verbalize
one's
problem-solving
strategies
may
elicit internalized verbalizations during prob-
lem
solving.
We
will return
to the
issue
of
possible
effects
of
concurrent verbalization
in
Experiment
3.
A
second possible concern about
the
results
of
Experiment
1
relates
to the
rather moderate size
of the
effect.
Although
the
difference
between subjects' performance
in the
verbal-
ization
and
unrelated interruption conditions
was
statistically
significant,
the
absolute difference
in
percentage
of
problems
solved
in the two
conditions
was
relatively small (approx-
imately
10%)
and
might
not be
considered conceptually sig-
nificant.
However, there
are two
reasons
to
consider this dif-
ference
important. First, because
of the
overall
low
levels
of
performance,
relatively speaking, unrelated interruption sub-
jects
solved approximately
30%
more problems than ver-
balization subjects. Second, verbalization subjects per-
formed
less well than unrelated interruption subjects despite
the
fact that they
were
actually given
more
time
to
think about
the
experimental problems.
The
unrelated interruption sub-
jects spent
the
interruption period engaged
in an
unrelated
activity,
but the
verbalization subjects spent
the
correspond-
ing
amount
of
time thinking about
the
problem. Verbalization
subjects
spent
approximately
1.5 min
more
on
each
problem
than
did
unrelated interruption subjects, biasing
the
exper-
iment against
the
observed
outcome.
Although
the
additional time allotted
to
verbalization sub-
jects increases
the
conceptual significance
of
their poorer
performance,
it
also introduces
a
possible
alternative expla-
nation.
The
difference between
the two
conditions might
be
due
not to
impairment
in the
verbalization condition,
but
rather
to
facilitation
in the
unrelated interruption condition
resulting
from
incubation (Wallas, 1926). Experiment
2 ad-
dressed this issue.
Experiment
2
The
first
goal
of
Experiment
2 was to
replicate
the
findings
of
Experiment
1.
The
second goal
was to
determine more
precisely
the
cause
of the
differences between
the two
con-
ditions
in
Experiment
1
(assuming that
the
difference
was
replicated).
The
results
of
Experiment
1
were consistent with
the
hypothesis
that
verbalization
impairs
performance
on in-
sight
problems,
but the
experiment
is
open
to the
alternative
interpretation
that
the
control activity improved perfor-
mance.
The
interruption might have served
as an
incubation
period
and
thus increased
the
probability
of
insight solutions
in
the
control
group.
Incubation
can be
defined
as an
increase
in the
probability
of
finding
the
solution
to a
problem
after
a
pause,
as
com-
pared with
after
continued problem solving. Wallas (1926)
proposed that
all
creative problem solving follows
a se-
quence
of
four
stages
that
he
called
preparation, incubation,
illumination,
and
verification.
In a
recent review
of the
lab-
oratory evidence regarding incubation,
C. A.
Kaplan (1989)
discussed
18
studies. Twelve
of
these studies
found
evidence
for
incubation,
two
failed
to
find
an
incubation
effect,
and the
remaining
four
studies were
difficult
to
interpret because
of
their
nonstandard
definition
of
incubation.
In
addition,
Ka-
plan
reported three
new
experiments that strongly supported
the
reality
of
incubation
effects.
The
plausibility
of the
incubation phenomenon
has
also
increased with
the
formulation
of
precise theories
of the
mechanisms
that might
be
responsible
for the
beneficial
ef-
fects.
The
concept
of
memory decay
is
central
to
such
ex-
planations. Although
the
possible role
of
memory decay
was
suggested
by
Woodworth
(1938), more sophisticated decay
theories
of
incubation awaited
the
information processing
approach
to
problem solving.
For
example, Simon (1986)
suggested that
different
kinds
of
information
about
a
problem
decay
at
different
rates. During
a
pause
in
problem solving,
information
about previous approaches
to the
problem
2
decay faster than information about
the
problem itself.
After
a
pause,
the
information about previous solution
at-
tempts
will
have decayed,
increasing
the
probability
that
some alternative approach
will
come
to
mind. Similar decay-
based
mechanisms
of
incubation have been proposed
by

172
J.
SCHOOLER,
S.
OHLSSON,
AND
K.
BROOKS
Anderson (1981)
and
recently supported
by
Smith
and
Vela
(1991).
In
Experiment
2 we
sought
to
distinguish between
the in-
cubation
and
verbal interference explanations
of
Experiment
1
by
introducing
a
third condition
in
which subjects worked
on the
problem
for the
full
duration without interruption.
If
the
results
from
Experiment
1 are due to
incubation, then
the
subjects
in the no
interruption condition should perform sim-
ilarly
to the
subjects
in the
verbalization condition.
If, on the
other hand, Experiment
1
reflected
a
disruptive
effect
of
ver-
balization
on
insight, then
the
subjects
in the no
interruption
condition
should perform
as
well
as the
subjects
in the un-
related
interruption
condition.
Method
Subjects
Subjects
were
123
undergraduates
from
the
University
of
Pitts-
burgh
who
received
course
credit
for
their participation. Subjects
were
run in
groups
of up to 4 at a
time.
Materials
Materials included
the
first
six
problems used
in
Experiment
1,
reproduced
in
Appendix
A.
Procedure
The
procedure
for
Experiment
2 was
replicated
with
the
addition
of
a
third condition
in
which subjects were
not
given
an
interruption
activity.
Subjects
in
this
no
interruption condition worked
on
each
problem without interruption
for a
maximum
of 6
min.
As in Ex-
periment
1,
subjects
in the
verbalization
and
unrelated interruption
conditions
also spent
up to 6 min on
each problem with
an
additional
1.5
min
devoted
to the
interruption activity
(i.e.,
verbalization
or
crossword puzzle) after
the
2-min
mark.
Results
Accuracy
Overall.
An
analysis
of
variance (ANOVA) revealed
a
main
effect
for
condition, F(2, 120)
=
3.958,
p <
.05. Mean
percentages
correct
for the no
interruption, unrelated inter-
ruption,
and
verbalization conditions were
51.2%,
47.5%,
and
37.3%, respectively.
There
was a
replication
of the
effect
of
verbalization observed
in
Experiment
1,
with subjects
in
the
unrelated interruption condition performing significantly
better than subjects
in the
verbalization
condition
(p <
.05,
simple
effects
test).
Subjects
in the
unrelated
interruption
condition similarly outperformed subjects
in the
verbaliza-
tion condition
(p <
.05, simple effects test).
There
was, how-
ever,
no
significant difference between performance
in the
unrelated interruption
and no
interruption conditions
(p >
.05, simple effects test).
Relative
to the
2-min mark.
Overall,
significantly
more
problems were solved
after
the first 2 min of
problem solv-
ing,
F(l, 120)
=
7.58,
p <
.01.
However, there
was no in-
teraction between
the
number
of
problems solved before
and
after
the
2-min mark
and
condition,
F(2,
120)
<
1.
Again,
this
latter
finding
indicates that
the
magnitude
of the
differ-
ence between
the
percentage
of
problems solved
by
subjects
in
the
verbalization, unrelated interruption,
and no
interrup-
tion
conditions prior
to the
interruption (15.85%, 19.52%,
and
21.95%, respectively)
was
comparable with
the
differ-
ence between these
two
conditions
after
the
interruption
(21.55%, 28.05%,
and
29.26%, respectively).
Item
and
serial
position
effects.
As in
Experiment
1,
there
was a
strong item
effect
indicating variation
in
item
difficulty,
F(5,
600)
=
43.29,
p<
.0001.
Item analysis also
revealed
a
significant
interaction
between
problem
and
con-
dition,
F(10, 600)
=
2.263,
p <
.05. This interaction
oc-
curred because
of
performance
on one
particular problem
(the
pyramid problem,
see
Appendix
A), for
which subjects
appeared
not to
have recognized
the
solution even when
they
reached
it. On
this problem,
it was
discovered
after
the
completion
of the
experiment that
a few
subjects
in the
verbalization
condition
had
written
the
solutions
in
their
verbalizations
without
recognizing that they
had the
correct
answer.
It
seems
likely
that
subjects
in the
other
two
con-
ditions
may
have
similarly
considered
the
correct solution,
but
because
they
did not
have
the
opportunity
to
report
the
dismissed solution, they simply
did not
state
it.
When
the
correct solutions
that
were only mentioned
in the
verbaliza-
tions
are not
included,
the
item interaction disappears,
F(10,
600)
=
1.5,
p >
.05. Moreover, when this problem
is not
included
in the
analysis,
the
interaction between condition
and
problem
is
also
not
observed, F(8, 480)
=
1.43,
p >
.05.
An
ANOVA revealed
no
evidence
of a
position
effect,
F(5, 600)
< 1, and no
interaction between position
and
condition,
F(10,
600)
< 1.
Time
For
those problems that were correctly solved, there
was
no
significant difference between conditions
in
time required
to
reach
the
solution, F(2, 120)
<
1.
The
mean times spent
on
correctly
solved
problems
in the no
interruption, unrelated
interruption,
and
verbalization conditions were 181.3, 179.9,
and
183.4
s,
respectively.
Discussion
Experiment
2
successfully replicated
the
findings
of Ex-
periment
1:
Subjects were less successful when interrupted
to
verbalize their
strategies
than when interrupted
to
perform
an
unrelated activity. Experiment
2
extended
the
findings
of
Experiment
1 by
demonstrating that
the
difference between
the
verbalization
and
unrelated interruption conditions
was
not
due to
benefits
of the
unrelated interruption.
If
partici-
pation
in the
unrelated interruption condition
had
actually
improved performance, then subjects
in
this condition would
have been
expected
to
perform
better
than
the no
interruption
subjects.
The
similar
performance
of
subjects
in
these
two
conditions
indicates
that
the
difference
between
the
verbal-
ization condition
and the
unrelated interruption condition

THOUGHTS BEYOND WORDS
173
cannot
be
attributed
to an
improvement resulting
from
in-
cubation,
but
rather must reflect impairment caused
by the
effort
to
verbalize insight
processes.
Although
the
lack
of a
beneficial
effect
of the
unrelated
interruption
provides
no
support
for the
reality
of
incubation
effects,
this result should
not be
taken
as
strong evidence
against
the
existence
of
incubation effects under
other
cir-
cumstances. Studies that have observed benefits
of
unrelated
interruption
have typically been associated with problem-
solving
attempts
of
longer duration
and
longer incubation
periods.
The
present study
may not
have provided
sufficient
time
for
incubation
effects
to
appear.
The
outcome
of Ex-
periment
2
only shows that
the
incubation hypothesis cannot
account
for the
observed difference
in
performance between
the
verbalization condition
and the
unrelated interruption
conditions
in
these
two
experiments.
Another
possible explanation
for the
differential perfor-
mance
of
subjects
in the
verbalization
and
unrelated inter-
ruption
conditions
of
Experiments
1 and 2
involves
the
spe-
cific
verbalization procedure used. Both Experiments
1 and
2
used
a
modified retrospective verbalization technique
de-
signed
to be as
comparable
as
possible
to
previous verbal
overshadowing
experiments (e.g., Schooler Engstler-
Schooler,
1990).
This technique requires that subjects access
a
memory trace
of
what they were doing while they were
solving
the
problem. Also,
our
instructions explicitly asked
the
subjects
to
write down their strategies, thus encouraging
them
to
theorize about
the
reasons
and
causes
of
their
own
problem-solving
behavior. Ericsson
and
Simon
(1980,1984)
pointed
out
that directed retrospective verbalization requests
are
very
different
from
requests
for
concurrent verbalization,
which
does
not
specify
a
particular category
of
information
to
be
verbalized. Directed retrospective verbalizations
may
be
biased both because they rely
on
memory
and
because
subjects'
theories about their
own
behavior
may
influence
what
they
say.
If
directed retrospective reports
are
more
likely
than nondirected concurrent reports
to be
biased, then
it
seems reasonable,
as
Ericsson
and
Simon (1984) sug-
gested, that
they
would also
be
more likely
to
have reactive
effects
on
performance. This analysis thus raises
the
question
of
whether
the
effects
observed
in
Experiments
1 and 2
were
due
to the
directed
retrospective
nature
of the
verbalization
or
whether they were
a
more general consequence
of any
effort
to
verbalize insight problem solving, retrospectively
or
otherwise.
In
short,
the
question remains whether
the im-
pairment
observed
in the
previous Experiments
was
caused
by
the act of
retrieving strategies
and
problem-solving steps
from
memory
or by the act of
clothing
the
problem solving
in
words. Experiments
3 and 4
explored this issue.
Experiment
3
In
their review
of the
effects
of
verbalization, Ericsson
and
Simon
(1980,1984)
concluded that concurrent verbalization,
which
does
not
require
the
subject
to
retrieve past memory
traces
and
which does
not
request that subjects explain their
strategies, should
not
cause
a
disruption
of
primary task per-
formance.
However, Ericsson
and
Simon based this conclu-
sions
on two
assumptions that
may not
apply
to
insight
problem-solving situations. First, they assumed that
efforts
to
solve problems typically yield
reportable
products that
are
heeded
in the
normal course
of
problem solving. Although
this
claim
may
often
be
true, current theories
of
insight claim
than
insight solutions require
a
number
of
processes, such
as
spread
of
activation, that
are not
easily verbalized (Ohlsson,
1992).
Second,
Ericsson
and
Simon assumed that when
no
reportable products
are
produced subjects simply
fall
silent.
However,
although subjects solving insight problems
often
fall
silent prior
to
solutions
(Durkin,
1937),
it is
quite possible
that
many
of the
subjects
who
fail
to
solve insight problems
under
talk-aloud conditions
may
have performed subopti-
mally
because
they continued talking. Specifically, this
on-
line verbalization
may
highlight reportable aspects
of the
task
and
overshadow critical information
and
processes
that
are
not
readily verbalized.
If so,
then requests
to
verbalize might
interfere
with solutions
to
insight problems, even
if the
ver-
balization
is
concurrent rather than
retrospective,
and
even
if
the
subject
is not
requested
to
theorize about
his or her own
behavior
or to
verbalize some particular category
of
infor-
mation.
To
explore this possibility, Experiment
3
examined
the
effects
of
nondirective concurrent verbalization
on in-
sight
processes.
A
second
issue
addressed
by
Experiment
3 was
whether
verbalization
differentially
affects
performance
on
insight
and
noninsight problems.
If, as
suggested, verbalization
im-
pairs
performance
on
insight problems because
of the
non-
reportable
processes associated with insight, then
it
should
be
expected that verbalization should
not
affect
problem
solving
that
is
hypothesized
to
have little
reliance
on
non-
reportable
processes.
Most characterizations
of
solving stan-
dard analytic noninsight problems have indicated little
re-
liance
on
nonreportable
processes
(e.g.,
Metcalf
&
Wiebe,
1987). Rather, these problems appear
to be
solved
by an
incremental
series
of
steps,
each producing
a
reportable prod-
uct
(Ericsson
&
Simon, 1984). Thus,
if the
negative
effects
of
verbalization
are
associated specifically with nonreport-
able
processes,
then
it
would
be
expected that verbalization
should
have little
effect
on
noninsight analytic problems.
Alternatively,
if the
effects
of
verbalization
are due to
some
other
mechanism, then
negative
effects
on
matched
nonin-
sight
problems might also
be
expected.
For
example, Russo
et
al.
(1989) suggested that when subjects
are
engaged
in
difficult
problems, attention devoted
to
verbalization could
drain
the
resources available
for
solving
the
problem. Com-
petition
for
scarce processing resources would presumably
impair
the
solution
of any
type
of
problem,
but the
effect
would
be
stronger
the
more
difficult
the
problem
(Kahne-
man,
1973). Hence,
if
scarce
resources were
the
cause
of the
effects
of
verbalization, then verbalization should also
affect
noninsight
problems that
are
comparable
in
difficulty
to the
insight
problems.
The
same prediction follows
if the
effect
of
verbalization
is due to
such factors
as
additional stress
(Zajonc,
1965), motivational
shifts
(Wilder
&
Harvey, 1971),
or
increased sense
of
accountability (Tetlock
&
Kim, 1987).
To
address these issues, Experiment
3
examined
the
effects
of
concurrent verbalization
on
insight
and
noninsight prob-
lems
that were matched
for
difficulty.

174
J.
SCHOOLER,
S.
OHLSSON,
AND K.
BROOKS
Method
90
-,
Subjects
Subjects
were
41
undergraduates
from
the
University
of
Pitts-
burgh
who
received course credits
for
their participation.
All
sub-
jects were
run
individually.
The
data
of one
subject were eliminated
because
of
familiarity
with
one of the
problems.
Materials
Seven problems were used; three were insight problems
and
four
were
noninsight problems (including
one
practice problem).
The
insight
problems were those problems
from
Experiments
1 and 2
that
showed
the
strongest verbalization
effect.
The
noninsight prob-
lems
were comparable
to the
incremental problems used
by
Met-
calfe
and
Wiebe
(1987)
and
were drawn from
a
variety
of
sources
including
Grosswirth
and
Salny
(1983)
and P.
Kaplan
(1964).
These
problems
can be
solved
in a
logical, incremental fashion
and do not
require
any
nonobvious approaches
to
reach
the
correct solution.
From
a
pilot study involving
10 of
these noninsight problems,
3
problems
were chosen,
for
which mean performance
was
similar
to
that
of the
insight problems (approximately
50%
accuracy).
A
fourth
problem (for which performance
was
considerably
higher)
was
used
as
a
practice problem
(see
Appendix
B). In
problem presentation,
the
practice problem always appeared
first.
The
insight
and
non-
insight
problems were always blocked. Position
of
problems within
each
group
was
randomized,
and the
order
in
which
the two
groups
occurred
was
counterbalanced across subjects.
Procedure
Each
subject
was
randomly assigned
to
either
the no
verbalization
or
verbalization conditions.
All
subjects were
run
individually
and
were given
a
maximum
of 6
min
to
solve each
of the
seven prob-
lems.
When subjects believed they
had a
correct solution, they
in-
formed
the
experimenter,
who
then checked
it. If
correct, subjects
were
given
the
next problem.
If
incorrect, subjects were encouraged
to
continue with
the
problem.
In the
verbalization condition,
sub-
jects were additionally instructed
to
think aloud while solving each
problem.
The
exact instructions given were
as
follows:
While
you
work
on
each problem,
I
want
you to
think
aloud.
That
is,
verbalize
any
thoughts
you
have while trying
to
solve
the
problem; that
means
any
information
you are
thinking
about,
anything
you
read,
questions
you ask
yourself
and so
forth.
Whatever
crosses
your mind
as you
work
on the
prob-
lems
should
be
said aloud.
Try not to
plan what
to say or
come
up
with ideas that sound good. Just allow your thoughts
to
come
out in
words
as
naturally
as
possible.
As
dictated
by
standard think-aloud procedures (Ericsson
&
Simon,
1984),
verbalization subjects were also prompted
to
continue
ver-
balizing
whenever there
was a
period
of
silence exceeding
15s.
Verbal
protocols were audiotaped
with
each
subjects'
full
consent.
Subjects
in the no
verbalization group were instructed
to
work
on
each problem
until
it was
solved
or
until
the
allotted time
was up.
Results
Accuracy
As
can be
seen
in
Figure
1, the
performance
of
verbal-
ization subjects
was
substantially lower than that
of the no
verbalization
subjects
for
insight problems; however,
for the
noninsight
problems,
the
performance
of the two
groups
was
o
O
70
-
o>
£L
60
-
50
-
Insight Problems
Non-insight Problems
Figure
1.
Mean percentage
of
correctly solved insight
and
non-
insight
problems
in
Experiment
3.
(Open symbols indicate
no
verbalization;
solid symbols indicate verbalization.)
approximately
equal.
An
ANOVA revealed
a
significant
in-
teraction between verbalization
and
type
of
problem,
F(
1,
38)
=
5.221,
p <
.05. This interaction reflects
the
observa-
tion
that verbalization impaired performance
on the
insight
problems
but had no
effect
on
performance
on
noninsight
problems.
hem
Effects
Separate item analyses were conducted
for the
insight
and
noninsight
problems.
For the
insight problems there
was a
significant
item
effect
indicating that some items were more
difficult
than others, F(2,
76) =
7.053,
p <
.01. However,
there
was no
interaction between item
and
condition, F(2,
76)
=
1.0,
p >
.05,
indicating
that
verbalization
did not
differentially affect
the
three insight problems.
For the
non-
insight
problems, there
was no
significant
effect
of
item, F(2,
76)
=
1.72,
p >
.05,
nor any
interaction between items
and
verbalization, F(2,
76) =
1.96,
p >
.05. Because
of the
ran-
domization
of
problem order,
it was not
possible
to
determine
order
effects.
Time
The
mean durations required
to
solve
the
insight
and
non-
insight
problems
are
shown
in
Figure
2. An
analysis
of
time
spent
on
those problems that were correctly solved revealed
no
main
effect
for
condition,
F(l,
34) <
1,
and no
interaction
between condition
and
type
of
problem (insight
or
nonin-
sight),
F(l,
34) <
1.
There
was, however,
a
main
effect
for
type
of
problem, F(l,
34) =
47.55,
p <
.001. Overall,
the
noninsight problems required more time than
the
insight
problems.
Discussion
In
Experiment
3
concurrent verbalization impaired
the so-
lution
of
insight problems while having virtually
no
effect
on
the
solution
of
noninsight
problems.
This
finding suggests
that
the
disruptive
effects
of
verbalization observed
in Ex-

THOUGHTS
BEYOND WORDS
175
300
n
200
-
a 100 -
Insight
Non-insight
Figure
2.
Mean
time
taken
to
correctly solve
insight
and
nonin-
sight
problems
in
Experiment
3.
(Open symbols
indicate
no
ver-
balization;
solid
symbols
indicate
verbalization.)
periments
1 and 2
were
not
artifacts
of the
retrospective par-
adigm; that
is,
they were
not
simply
the
result
of
informal
theorizing
on the
part
of the
subjects about their
own be-
havior.
Rather,
it
appears that
the
disruption
observed
in the
previous experiments
was
directly associated with
the
efforts
to
verbalize insight
processes.
The
results
of
Experiment
3
further
support
our
hypothesis
that
verbalization disrupts critical
nonreportable
processes
by
providing evidence against some alternative explanations
for
the
disruptive
effects
of
verbalization
on
insight problem
solving.
The
hypothesis that verbalization reduces
the re-
sources available
for the
primary task predicts greater
im-
pairment
for
more
difficult
problems (Kahneman, 1973).
However,
the
performance
of the no
verbalization subjects
shows that
the
noninsight problems were,
if
anything, more
difficult
than
the
insight
problems
and
therefore should
have
shown
at
least
as
great
an
effect
of
divided resources.
The
lack
of an
effect
on the
matched noninsight problems also
argues against other explanations that
do not
differentiate
between insight
and
noninsight problems.
For
example, ver-
balization
in the
presence
of an
experimenter might increase
arousal, which could impair performance. However, this
hy-
pothesis, too, would predict that equally
difficult
tasks would
show
similar impairment (Zajonc,
1965),
which
is not
what
we
observed.
The
aforementioned comparisons suggest that verbaliza-
tion
disrupts
some
process
or
processes
that
are
unique
to
insight
problems.
We
have suggested that these unique pro-
cesses
may
involve nonreportable components, such
as
spreading activation, that ultimately allow subjects
to re-
trieve critical memory
elements
necessary
for
insight prob-
lem
solving.
An
alternative reason
why
subjects
may be un-
able
to
retrieve
the
critical elements necessary
for
insight
solutions
is
that
the act of
verbalization
may
produce
a
"de-
mand"
to
stay
in the
originally defined problem
space.
Ex-
periment
4
explored this issue.
Experiment
4
It
is
well established that experimental situations
can in-
troduce demands
on
subjects that
may
profoundly influence
their
performance (Orne,
1962).
Problem solving
is not im-
mune
from influences
of
experimenter demands.
For
exam-
ple, Wilder
and
Harvey
(1971)
observed
that having
an ex-
perimenter present reduced
the
number
of
moves necessary
to
solve
a
move problem. Russo
et
al.
(1989)
suggested that
verbalization improved subjects' performance
in
deciding
between
two
gambles
by
increasing their motivation
to use
a
more
effortful
strategy (mental
multiplication).
In a
per-
sonality
prediction task, Tetlock
and Kim
(1987) observed
that
subjects
who
believed they were publicly accountable
for
their
decisions
used more complex
processes
and
showed greater predictive accuracy than subjects
who did
not
believe.
Verbalizing
one's
strategies
in the
presence
of an
exper-
imenter
clearly presents
the
possibility
for
various perceived
demands.
In
Experiment
3, all
subjects were tested individ-
ually
with
the
experimenter attentively nearby
to
minimize
differences
between
the
demands
in the no
verbalization
and
verbalization
conditions. However, there
is one
type
of de-
mand
that might
be
particularly disruptive
for
insight prob-
lem
solving. Specifically, verbalization
may
produce
an ex-
perimental demand
to
continue with
the
approach that
has
been verbalized, because changing
one's
approach
is to
admit
failure,
which might
be
socially embarrassing
for
some sub-
jects.
Insight
problems
are
characterized
precisely
by the ne-
cessity
to
reject
the
initial, obvious approach
in
order
to
find
the
solution, whereas noninsight problems
can be
solved
by
pursuing
the
obvious approach. Hence,
a
reluctance
to
admit
failure
and
search
for a
different
approach would
affect
in-
sight
problems more than noninsight problems.
To
counteract this hypothesized demand
to
avoid consid-
ering alternative approaches,
we
replicated
the
procedure
of
Experiment
3 and
introduced
a new
condition aimed
at
pro-
viding
the
opposite
demand,
that
is, a
demand
to
consider
new
problem approaches. Specifically,
we
used
a
set-
breaking hint
(Olton
&
Johnson, 1976)
that
described what
a
mental
set is,
that provided
an
example
of an
insight prob-
lem
with
a set and how it was
broken,
and
that encouraged
subjects
to
search
for
alternative approaches
if
they believed
that
they were stuck
in an
inappropriate approach.
It is
rea-
sonable
to
assume that this procedure would counteract
any
perceived demand
to
stick with
a
single approach even when
stuck,
so if the
effect
of
verbalization results
from
an
implicit
demand
to
that
effect,
then such
a
hint should attenuate
the
impairment.
If,
however, verbalization prevents subjects
from
using
the
nonreportable perception
and
memory pro-
cesses
necessary
to find the
appropriate approach, then
the
set-breaking hint should have little
effect
on
subjects' per-
formance
in
solving
the
insight problems.
With
respect
to
solving noninsight problems,
a
priori
it is
difficult
to
anticipate
the
effect
of
providing
a
hint
to
consider
alternative approaches.
On the one
hand, because
the
issue
of
"mind-set"
is
less relevant
to
noninsight problems, sub-
jects
in
both conditions might simply ignore
the
hint while
solving
the
noninsight problems.
On the
other hand,
if the
mind-set hint makes subjects
process
noninsight problems
more like insight problems (i.e.,
by
using memory retrieval
processes
to
search
for
alternative approaches), then
it is
possible that verbalization might also disrupt noninsight
problems when they
are
accompanied with
the
mind-set
hint.

176
J.
SCHOOLER,
S.
OHLSSON,
AND
K.
BROOKS
Method
Subjects
Subjects were
83
undergraduates from
the
University
of
Pitts-
burgh
who
received
$5 for
their
participation.
All
subjects
were
run
individually.
Three
subjects were eliminated because
of
familiarity
with
one of the
problems.
Materials
The
problems used
in
Experiment
4
were identical
to
those used
in
Experiment
3.
Procedure
The
procedure
for
Experiment
4
replicated Experiment
3,
with
the
addition
of a
hint condition
in
which subjects were told that they
would
be
working
on
both insight
and
noninsight problems
and
were given information about
the
general nature
of
these
two
types
of
problems. Noninsight problems were described
as
involving con-
ventional, step-by-step, logical thinking.
In
contrast, insight prob-
lems were described
as
typically leading
one to get
stuck
on a
problem-solving
approach
that
does
not
work. Solving this type
of
problem
was
suggested
to
require considering
it
from
a new
per-
spective.
To
illustrate this point, subjects were shown
a
sample
insight
problem
and
were given
an
explanation
of the
process
of
insight
as it
related
to the
problem. Hint subjects were advised that
if
they
felt
they were stuck
on a
problem that they should consider
the
possibility that they were
in a
mind-set
and
should
try an al-
ternative
approach. Subjects were
further
reminded that
not all of
the
problems were insight problems
and
that some
of
them could
be
solved
in a
conventional, logical way. After
hint
subjects worked
on a
problem
for
approximately
2
min,
they were again reminded
that
if
they
felt
stuck
on a
problem, then
it
might
be an
insight
problem
and
they might
be
caught
in a
mind-set.
Results
Accuracy
The
mean number
of
problems solved
in the
various con-
ditions
of
Experiment
of 4 are
presented
in
Figure
3.
o
O
o
<D
Q-
100
H
80-
60-
40-,
/
f
Insight
Non-insight
No
Hint
Insight
Non-insight
Hint
Figure
3.
Mean percentage
of
correctly solved insight
and
non-
insight problems
in
Experiment
4.
(Solid symbols indicate
no
verbalization;
open symbols indicate verbalization.)
Effects
of
verbalization. Overall, subjects
in the
verbal-
ization conditions solved significantly fewer problems than
those
in the no
verbalization (control) conditions, F(l,
76)
=
12.482,
p <
.001.
There
was
also
a
trend toward
an in-
teraction between verbalization
and
type
of
problem,
in
which verbalization caused
greater
impairment
for
insight
problems than
for
noninsight problems,
F(l,
76) =
3.017,
p
<
.09. This trend
is
better understood
in
light
of the
signif-
icant three-way interaction between verbalization, hint,
and
type
of
problem,
F(l,
76)
=
6.789,
p
<
.05. This interaction
reflects
differences
in the
interaction between verbalization
and
type
of
problem
in the
hint
and no
hint condition.
In the
no
hint condition, which involved
the
identical procedure
as
used
in
Experiment
3,
there
was an
interaction between ver-
balization
and
type
of
problem,
F(l,
76) =
9.434,
p
<
.01.
This interaction represents
a
replication
of the
observation
in
Experiment
3
that verbalization subjects performed mark-
edly
less well than
no
verbalization subjects
on
insight prob-
lems
(a 35%
difference), whereas there
was
virtually
no
dif-
ference between
the
performance
of
these
two
conditions
on
noninsight
problems
(less
than 2%).
In the
hint condition,
there
was a
main
effect
of
verbalization, F(l,
76) =
8.309,
p
<
.01,
but no
interaction between verbalization
and
type
of
problems,
F(l,
76) <
1,
indicating that verbalization sub-
jects
performed less well than
no
verbalization subjects
on
both insight problems
(a 22%
difference)
and
noninsight
problems
(a 29%
difference).
Effects
of the
hint. Overall, subjects
who
received hints
solved fewer problems than those
who did not
receive hints,
though this difference only approached significance,
F(
1,76)
=
3.619,
p <
.06.
The
hint
and
type
of
problem variables
yielded
a
significant interaction, F(l,
76) =
4.715,
p <
.05.
This interaction reflects
the
finding that
the
hint manipulation
caused impairment
on
noninsight problems
but had no
effect
on
insight problems.
Item
effects.
Separate
item analyses were conducted
for
the
insight
and
noninsight problems.
For the
insight problems
there
was a
significant item
effect
indicating that
some
items
were more
difficult
than others, F(2, 152)
=
4.62,
p <
.05.
However, there
was no
significant interaction between items
and
verbalization, F(2, 152)
=
2.7,
p >
.05; between items
and
hint, F(2,
152)
=
1.157,
p >
.05;
nor
among items, hint,
and
verbalization,
F(2, 152)
<
1.
For the
noninsight
prob-
lems, there
was no
significant
effect
of
item, F(2, 152)
=
2.21,
p >
.05,
and no
significant interactions between items
and
verbalization, F(2, 152)
<
1;
between items
and
hint,
F(2,152)
<
1;
nor
among items, hint,
and
verbalization,
F(2,
152)
<
1.
Because
of the
randomization
of
problem order,
it
was not
possible
to
determine
order
effects.
Time
The
mean time
to
correctly solve
the
insight
and
noninsight
problems
is
presented
in
Figure
4. As in the
earlier
experi-
ments, subjects took significantly longer
to
solve
the
non-
insight
problems
than
the
insight problems,
F(l,
60) =
101.
48, p <
.001.
There
was no
main
effect
on
solution time
of
either
verbalization
or
hint,
F(
1,60)
< 1,
in
both
cases.
There
was
also
no
significant interaction between verbalization
and

THOUGHTS
BEYOND
WORDS
177
§ 300
n
1
o
f
200
8
8
100
-
c
as
a>
Insight
Non-insight
No
Hint
•
r
Insight
Non-insight
Hint
Figure
4.
Mean
time
taken
to
correctly
solve
insight
and
nonin-
sight
problems
in
Experiment
4.
(Solid
symbols
indicate
no
ver-
balization;
open
symbols
indicate
verbalization.)
hint,
F(l,
60) <
1;
between verbalization
and
type
of
prob-
lem,
F(l,
60) =
2.02,
p >
.05; between type
of
problem
and
hints,
F(l,
60) =
2.25,
p >
.05;
nor
among type
of
problem,
hint,
and
verbalization, F(l,
60) =
1.05,
p >
.05.
Discussion
The
results
of
Experiment
4
replicate
the
findings
from
Experiment
3. In the
absence
of the
set-breaking hint, con-
current verbalization impaired performance
on
insight prob-
lems
but had
little
effect
on
noninsight problems.
The
per-
formance
of
subjects
who
received
the
set-breaking hint
offers
further
evidence against
the
suggestion that
the
verbal
impairment
is due to an
implicit demand
to
continue
with
one's
initial
approach. Even when
the
experimenter gave
ex-
plicit
instructions
to
consider alternative approaches, sub-
jects'
verbalization performance
on the
insight problems
was
impaired. Thus,
the
effect
observed
in
Experiments
3 and 4
cannot
be
explained
as a
result
of
implicit demands
to
con-
tinue
with
one's
initial approach.
One
might have expected
the
hint
to
improve performance.
C. A.
Kaplan
and
Simon
(1990)
found
that
one
important step
in
the
solution
of
insight problems
is the
realization that
the
present approach
is not
working,
and
therefore, that
one
should
search
for
alternative approaches.
A
set-breaking hint
might
have increased subjects' awareness that they needed
to
search
for
alternative approaches. However,
Olton
and
Johnson
(1976)
also
failed
to find an
effect
of a
set-breaking
hint.
The
failure
of the
hint
to
improve performance
may be
indirectly
related
to the
reasons
why
verbalization impairs
performance:
The
critical
processes
are
outside
of
awareness
and
therefore
are not
only
difficult
to
verbalize,
but
difficult
to
control voluntarily. Without volitional control over
the
critical
processes,
the
subjects cannot
act on the
advice
in the
hint.
Although
the
hint
did not
improve performance,
it
nev-
ertheless proved
to be a
powerful manipulation. Although
having
virtually
no
effect
on
subjects' performance
in the no
verbalization
condition
nor on
verbalization subjects' per-
formance
on
insight problems,
the
hint markedly impaired
verbalization subjects' ability
to
solve noninsight problems.
Although
we did not
have
a
strong
a
priori
prediction
of
this
particular finding,
one
reasonable interpretation
is
that
the
hint
caused subjects
to
process noninsight
problems
as if
they
were insight problems.
It
seems plausible that when subjects
work
on
noninsight problems that
offer
a
straightforward
solution
approach,
they
do not
give much
consideration
to
whether they
are on the right
track. However, when given
the
hint,
subjects
who are
solving noninsight problems
may be-
gin
to
treat
the
problems
as if
they
may be
insight
problems.
For
example, they
may
attempt
to use
memory retrieval pro-
cesses
to
determine whether there
is
"something
they
are
missing."
Because this type
of
memory retrieval
is
likely
to
be
difficult
to
verbalize,
it
follows that noninsight problems,
when
treated
as if
they
may be
insight
problems,
might also
be
expected
to be
impaired
by
verbalization.
In
short,
al-
though
noninsight problems typically
do not
appear
to
elicit
processes
that
are
vulnerable
to
verbalization,
it
seems
that
the
critical factor
in
whether verbalization disrupts problem
solving
is not
simply
the
nature
of the
problem
per se.
Rather,
what
seems
to be
most important
is the
nature
of the
pro-
cesses
that subjects bring
to
bear
on the
problem.
General
Discussion
Summary
The
hypothesis underlying this
series
of
studies
is
that
a
request
for
verbalization interferes with
the
successful
so-
lution
of
insight problems because
it
interferes with
the
suc-
cessful
application
of the
nonreportable
processes
hypothe-
sized
to be
associated
with such problems.
The
present
series
of
studies documents
the
predicted
effect
of
verbaliza-
tion
on
insight
and
argues against
a
number
of
alternative
interpretations.
In
Experiment
1,
subjects
who
were interrupted while solv-
ing
insight problems
and
asked
to
retrospectively verbalize
their strategies performed less well than
no
verbalization sub-
jects
who
were interrupted
and
engaged
in an
activity
un-
related
to
their problem-solving
effort.
Although this result
is
consistent
with
the
hypothesized disruptive
effects
of
ver-
balization
on
insight,
it
might have been
due to the
beneficial
effects
of
taking
a
pause
and
engaging
in an
unrelated
ac-
tivity.
However, Experiment
2
ruled
out
this latter possibility
by
demonstrating that interruption alone
has no
effect
on
performance
in the
present paradigm.
We
conclude that paus-
ing
and
verbalizing
one's
problem-solving strategies impairs
performance
on
insight problems.
In
Experiments
3 and 4 we
studied concurrent verbaliza-
tion
in
which subjects were given standard think-aloud
in-
structions.
The
instructions encouraged
the
subjects
to
ver-
balize their thoughts
as
they occurred
and did not
direct them
toward
a
particular category
of
information. Experiment
3
compared
concurrent verbalization with silent problem solv-
ing
and
found
higher performance
in the
silent condition.
Significantly,
this difference
was
observed
for
insight prob-
lems
but not for
noninsight problems.
In
Experiment
4, the
negative
effects
of
verbalization
on
insight problem solving

178
J.
SCHOOLER,
S.
OHLSSON,
AND K.
BROOKS
were shown
to
hold
even when
the
subjects were given
a
hint
that
they might
be in a
mind-set
and
might need
to
change
their
view
of the
problem.
Alternative Explanations
for the
Current Findings
We
believe that
a
likely explanation
for the
present results
is
that
verbalization disrupts
the
nonreportable
processes
as-
sociated
with
insight
problem
solving.
Before further expli-
cating this hypothesized mechanism,
we
briefly consider
a
variety
of
alternative explanations.
One
possible
interpretation
of the
present
findings
is
that
verbalization
can
consume general processing
resources
that
otherwise would have
been
available
for the
problem-solving
effort
(e.g.,
Russo
et
al.,
1989).
Although verbalization
may
influence
the
manner
in
which resources
are
allocated (i.e.,
emphasizing
reportable
processes
and
information
at the ex-
pense
of
nonreportable ones),
the
present findings argue
against
the
suggestion that verbalization reduces
the
general
resources available
for
problem solving. Specifically,
the
general
resource consumption explanation predicts that ver-
balization
should interfere with noninsight problems
of
com-
parable
difficulty,
which
was not the
case
in
Experiment
3 nor
in
the no
hint condition
of
Experiment
4. The
differential
effect
of
verbalization
on
insight
and
noninsight problems
similarly
argues against
a
variety
of
other explanations, such
as
task arousal (Zajonc, 1965), which assumes that verbal-
ization
adds
an
element
of
difficulty
to the
problem-solving
effort.
A
related alternative explanation
for the
effect
of
concur-
rent
verbalization
is
that
it
slows down problem solving with-
out
qualitatively altering
it.
Ericsson
and
Simon
(1980,1984)
proposed
that this
is to be
expected when
the
verbalized
in-
formation
is
heeded,
but not in
verbal form,
and
thus
in
need
of
receding.
Because
our
subjects were given
a
limited time
for
each problem, increasing
the
time
to
solution would result
in
a
reduced number
of
solved problems. However,
we
found
no
evidence that verbalization slowed down performance
on
insight
problem solving.
If
verbalization slowed down per-
formance,
then
it
would
be
expected
that
the
peak
frequency
of
insight solutions would occur later
in the
verbalization
condition
compared with
the
nonverbalization condition.
However, there
was no
evidence
in any of the
four experi-
ments that verbalization increased
the
time taken
to
solve
the
insight
problems.
It
appears that
in the
case
of
insight prob-
lems,
you
either
get
them
or you do
not,
and if you are
ver-
balizing
you are
simply
less
likely
to get
them.
It
was
also possible that
the
observed effects
of
verbal-
ization
were
due to an
interaction between
the
demand char-
acteristics
of the
verbalization situation
and
problem type.
If
subjects
are
reluctant
to
reveal
to the
experimenter that they
are on the
wrong track, then
the
think-aloud instruction might
create
a
strong
tendency
to
continue
with
the
current
ap-
proach. This tendency would
be
expected
to
affect
insight
problems,
in
which
finding
the
right view
of the
problem
is
the
crucial step, more than noninsight problems,
in
which
the
main
difficulty
is to
execute
or
carry
out the
obvious
ap-
proach. However,
in
Experiment
4
subjects were given
a
hint
by
the
experimenter that they might
be in a
mind-set
and
that
they
might need
to
consider alternative views
of the
problem.
This
hint should have created
a
demand characteristic
in fa-
vor
of
considering
alternative
approaches.
Nevertheless,
the
results
from
Experiment
3
were replicated: Verbalization
still
impaired
insight problems more
than
noninsight problems
even when subjects were encouraged
to
consider alternative
approaches. Thus,
an
explanation
in
terms
of
demand char-
acteristics
is
also
insufficient
to
explain
our
results.
The
Mechanism
of
Interference
The
mechanism underlying
the
disruptive
effects
of
ver-
balizing insight problem solving seems likely
to be
related
to the
negative
effects
of
verbalizing other activities
that
in-
volve
critical
difficult-to-report
components.
As
previously
mentioned,
recently there
has
been growing evidence that
verbalization
may
cause nonreportable aspects
of a
task
to
become overshadowed
by
those that
are
more readily
re-
ported
(e.g.,
Brandimonte, Hitch,
&
Bishop, 1992; Fallshore
&
Schooler, 1993; Schooler
&
Engstler-Schooler, 1990; Wil-
son &
Schooler,
1991).
Our
explanation
for the
effect
of
verbalization
on
insight
is
that
a
request
for
verbalization
may
similarly cause
the
verbalizable task components
to
overshadow those that
are
less readily verbalized.
In the
case
of
noninsight problems, verbalization would
not be
expected
to
produce much disruption because these problems involve
a
series
of
incremental steps, each
of
which
is
separately
reportable
(e.g.,
Newell
&
Simon,
1972).
In
contrast,
the
solutions
to
insight problems occur suddenly (Durkin,
1937;
Metcalfe
&
Wiebe, 1987), thus suggesting that
the
critical
steps leading
to the
solution
are
unavailable
for
conscious
inspection.
Despite
the
opaqueness
of
insight
processes,
sub-
jects
are
nevertheless quite willing
to
think aloud while
at-
tempting
to
solve insight problems.
It
seems that
the
report-
able
processes
are
simply
not the
ones that
will
lead
to a
correct solution. Consistent with this suggestion
is
Met-
calfe's
(1986b) observation that subjects' belief
that
they
are
close
to a
solution
is
actually predictive
of
a
failure
to
solve
the
problem. Metcalfe suggested that subjects
who
believe
that
a
solution
is
imminent
are
engaging
in a
"gradual
ra-
tionalization
process"
(p.
623) that focuses them
on an in-
accurate
yet
reportable approach.
It
seems likely
that
a
sim-
ilar
process
may
occur
in the
case
of
verbalization, namely
that
the
gradual reportable
processes
elicited
by
verbalization
may
overshadow
the
critical nonreportable
processes
nec-
essary
for
successful solutions.
Although
the
present
findings
are
consistent
with
the
sug-
gestion that verbalization overshadows critical
difficult-to-
report insight
processes,
the
nature
of
these processes
re-
mains
to be
specified.
One
likely nonreportable component
of
insight
is
memory retrieval,
and in
particular spreading
activation
processes
(e.g., Bowers
et
al.,
1990; Langley
&
Jones,
1988; Ohlsson,
1992).
Inspection
of the
insight
and
noninsight problems
in the
Appendixes indicates that
the
noninsight problems
are
relatively self-contained, that
is,
subjects
do not
need
to
draw
on any
additional world knowl-
edge
to
solve them. Rather,
the
difficulty
of
noninsight prob-
lems
typically lies
in
correctly sequencing
and
executing
the
operators necessary
to
solve
the
problem.
In
contrast, insight

THOUGHTS BEYOND WORDS
179
problems
require identifying
an
alternative
approach
to the
problem, which
often
requires retrieving world knowledge.
For
example,
in the
case
of the
problem
of
identifying
how
a
dealer knows that
a
bronze coin marked
544
B.C.
is a
fraud
(see Appendix
A),
subjects must
retrieve
the
fact
that B.C.
corresponds
to
"before
Christ"
and the
implications
of
that
fact.
A
related insight process that
may be
disrupted
by
ver-
balization
is
constraint relaxation (Ohlsson,
1992).
In
many
insight
problems,
the
initial representation
is
likely
to be
unnecessarily constrained
by
false
assumptions,
often
re-
ferred
to as an
incorrect mind-set
or
Einstellung
(Luchins,
1942).
For
example,
in the
previously discussed problem
of
determining
the
parental identity
of a
surgeon,
it may be
falsely
assumed that
the
solution
can
only
be
fulfilled
by a
man.
Verbalization
may
interfere with
the
ability
to
over-
come such unnecessary constraints both
by
strengthening
the
activation
of the
incorrect assumption(s)
and by
interfering
with
the
retrieval
of the
necessary
counterinformation
(e.g.,
that women
can be
surgeons).
Another
nonverbalizable insight
process
that
may
also
be
disrupted
by
verbalization
is
perceptual reorganization
(Ohlsson, 1992).
In
some insight problems, critical
shifts
in
defining
the
problem space
may
arise from perceptual remap-
ping
of the
objects
involved
(e.g.,
the
Necker
cube
phenom-
enon).
Given
the
nonverbal characteristic
of
such perceptual
processes,
it
seems quite possible that verbalization
may
dis-
rupt
these
processes
as
well. Although
a
disruption
in
per-
ceptual
reorganization might account
for
reduced perfor-
mance
on
some
of the
insight
problems
(e.g.,
the
triangle coin
problem
in
Appendix
A), it is
important
to
note that verbal-
ization
equally
affected
the
insight problems that
did not
include perceptual components.
Future research will
be
necessary
to
isolate
the
mechanism
by
which
verbalization
focuses subjects
on
reportable
pro-
cesses
at the
expense
of
nonreportable ones.
It
seems likely,
however,
that because
the
distinction between reportable
and
nonreportable cognition
is
often
mapped onto
the
distinction
between automatic
and
controlled processes (e.g., Jacoby,
Ste-Marie,
&
Toth,
in
press;
Schneider,
Dumais,
&
Shiffrin,
1984), verbal overshadowing
of
nonreportable processes
may
be
comparable
to
situations
in
which "the attentive sys-
tem
overrides
the
automatic"
(Eriksen,
Webb,
&
Founder,
1990,
p.
486). Specifically,
a
number
of
researchers have
suggested that automatic
processes
may be
disrupted when
attention
is
directed
toward automatized tasks (Eriksen
et
al.,
1990; Kimble
&
Perlmuter, 1970; Langer
&
Imber,
1979;
Norman
&
Shallice, 1986).
In the
present study,
the
nonre-
portable insight processes, such
as
spreading activation, that
we
suggest
may be
disrupted
by
verbalization
are
typically
characterized
as
being automatic (e.g., Bowers
et
al.,
1990).
Moreover, subjects' inability
to
benefit
from
instructions
to
consider alternative approaches
in
Experiment
4
suggests
that
this critical step
to
successful insight solutions
is not
under
volitional control
and
therefore involves automatic
processes.
It
thus remains
an
intriguing possibility that ver-
balization
may,
in
some sense, increase
the
brightness
of the
attentional
spotlight
and
thereby overshadow
the
automatic
processes necessary
for
insight problem solving.
Implications
for the
Study
of
Insight
Some researchers have suggested that insight
processes
reduce
to the
same conscious
processes
that
are
associated
with
other types
of
problem solving (Perkins,
1981;
Weis-
berg, 1986; Weisberg
&
Alba,
1982).
The
present
series
of
experiments presents some
difficulty
for
such
a
view.
The
differential
effect
of the
verbalization request
on
subjects'
performance
on
insight
and
noninsight problems
is
strong
evidence that
the
underlying
processes
are not
identical.
The
present results thus extend
Metcalfe's
(1986a)
and
Metcalfe
and
Wiebe's
(1987) observation that
people
can
accurately
report their distance
from
the
solutions
to
noninsight prob-
lems
but are
unable
to
report their nearness
to
insight solu-
tions.
These
results
are
consistent with
the
present view that
processes
leading
to
insight solutions
differ
from
those
lead-
ing
to
noninsight solutions
in
that they
are not
available
for
conscious inspection
and
therefore leave
few
hints
as to
their
progress. Thus,
a
strong argument
for the
existence
of
distinct
insight
processes
is
that they readily account
for
qualitative
differences
between insight
and
noninsight problems
ob-
served
in two
different
paradigms.
Methodological
Implications
and the
Generality
of
Verbalization
Effects
The use of
verbal protocols
is now a
widely used method
for
studying cognitive
processes
in
large part because
of
Eric-
sson
and
Simon's
(1980,1984)
theory suggesting that verbal
protocols
when
properly
collected
are
both
valid
and
non-
reactive.
Our
research
is in
agreement with Ericsson
and
Simon's general theoretical assumption that verbalization
will
primarily interfere with performance when subjects
at-
tempt
to
verbalize information that
is not
normally heeded.
However,
the
present
research
suggests
that
Ericsson
and
Simon
may
have underestimated
the
situations under which
verbalization
may
cause normally unheeded information
to
be
considered. Specifically, Ericsson
and
Simon suggested
that
verbalization only qualitatively alters cognitive pro-
cesses when subjects
are
explicitly
directed
to
verbalize
in-
formation
that they would
not
otherwise consider (e.g., when
asked
to
engage
in
retrospective verbalization about
the
strat-
egies that they used). However,
the
present findings suggest
that tasks that involve
a
considerable amount
of
nonreport-
able processing
may be
disrupted
by
verbalization, even
if
subjects
are
given concurrent, non-directive, think-aloud
instructions.
The
disruptive
effects
observed
in the
present study
are
particularly notable because they were incurred with problem
solving, which represents
the
most common application
of
think-aloud
procedures.
In the
past, although there were
many
examinations
of the
effects
of
verbalization
on
higher
order
processes
such
as
reasoning
and
problem solving,
as
Ericsson
and
Simon
(1980,
1984)
noted, there
was
little
ev-
idence that nondirective think-aloud procedures could dis-
rupt
such performance
in a
manner other than simply slowing
performance
or
increasing arithmetic slips (Russo
et
al.,
1989).
The
present research indicates that
it is not
only per-
ceptionlike nonreportable
processes
that
can be
disrupted
by

180
J.
SCHOOLER,
S.
OHLSSON,
AND K.
BROOKS
verbalization. Rather, more generally,
any
cognitive activity
that
relies
primarily
on
nonreportable
processes
and
infor-
mation
may be
vulnerable
to
verbalization, including cre-
ativity
(Finke, 1990; Koestler, 1964), implicit concept learn-
ing
(Reber, 1989), implicit memory (Schacter, 1987),
and
automated complex motor skills (Norman
&
Shallice,
1986).
The
potential breadth
of
cognitive tasks that
may be
vul-
nerable
to
verbalization strongly suggests that
all
researchers
using think-aloud procedures should,
at a
minimum, consider
the
degree
to
which
the
cognitive
processes
they wish
to
examine
are
likely
to
involve nonreportable components.
However, because
we do not at
present have
an
adequate
theory
for
identifying
whether
a
task
involves
nonreportable
components (although this
is
clearly
an
important area
for
future
theory
and
research), researchers using think-aloud
techniques should seriously consider including silent control
groups
to
determine whether verbalization
is
influencing
per-
formance (c.f. Russo
et
al.,
1989).
Language
and
Thought
The
finding that verbalization disrupts insight
processes
suggests
two
implications
for the
age-old issue
of the
rela-
tionship between language
and
thought (for
a
recent review,
see
Hunt
&
Agnoli,
1991).
First,
the
finding that subjects
are
less
effective
at
solving insight problems when they
are
com-
pelled
to put
their thoughts into words provides additional
support
for the
claim that insight involves processes that
are
distinct
from language. Second,
the
observation that verbal-
ization qualitatively alters performance suggests
one
situa-
tion
in
which insight problem solving
may
become increas-
ingly
influenced
by
language, namely when subjects attempt
to
articulate their thoughts.
In
short,
the
present research
suggests that
the
relationship between language
and
thought
is
not
always symbiotic. Rather,
in
some situations, language
may
interfere with thought.
The
claim that verbalization impairs thinking contradicts
the
common wisdom
of the
benefits
of
"talking through"
a
problem. This notion
has
received some support from
a
num-
ber of
studies showing that verbalization
can
improve prob-
lem
solving (e.g.,
Ahlum-Heath
&
Di-Vesta, 1986; Davis,
Carey, Foxman,
&
Tarr,
1968; Gagne
&
Smith, 1962; Wilder
&
Harvey,
1971).
However,
all of
these studies used non-
insight
step-by-step problems with
a
limited number
of op-
tions
at
each step,
and
they
did not
require
the
retrieval
of
nonobvious operators.
As
long
as a
problem
is
reasonably
straightforward,
it
appears that verbalization should
be, at a
minimum,
benign
and
that verbalization sometimes
may
even help
by
highlighting
useful
information. However,
when
the
basic
difficulty
is
that
the
obvious approach
is the
wrong
approach
and the
crucial step
is to
retrieve information
that
will help
to
change
one's
view about
the
problem,
then
talking through
the
problem appears
not to be an
effective
tactic.
At
least
in the
case
of
insight problems
it may be
better
to
"think before
you
speak."
References
Ahlum-Heath,
M.
E., &
Di-Vesta,
F. J.
(1986).
The
effect
of
con-
scious controlled verbalization
of a
cognitive strategy
on
transfer
in
problem solving. Memory
&
Cognition,
14,
281-285.
Anderson,
J. R.
(Ed.).
(1981).
Cognitive skills
and
their acquisition.
Hillsdale,
NJ:
Erlbaum.
Bergson,
H.
(1902).
An
introduction
to
metaphysics.
New
York:
Putnam.
Bowers,
K. S.
(1991, June).
How
sudden
is
sudden insight
and how
can
we
tell? Paper presented
at the
Annual
Meeting
of the
Amer-
ican
Psychological Society, Washington,
DC.
Bowers,
K.
S.,
Regehr,
G.,
Balthazard,
C., &
Parker,
K.
(1990).
Intuition
in the
context
of
discovery. Cognitive Psychology,
22,
72-110.
Brandimonte,
M.
A.,
Hitch,
G.
J.,
&
Bishop,
D. V. M.
(1992).
In-
fluence
of
short-term memory
codes
on
visual image processing:
Evidence
from
image transformation tasks. Journal
of
Experi-
mental Psychology: Learning, Memory,
&
Cognition,
18,
157-
165.
Bruner,
J.
(1966).
On
knowing:
Essays
for the
left
hand. Cambridge,
MA:
Harvard
University
Press.
Bullbrook,
M. E.
(1932).
An
experimental inquiry
into
the
existence
and
nature
of
"insight." American Journal
of
Psychology,
44,
409-453.
Davidson,
J.
E.,
&
Sternberg,
R. J.
(1984).
The
role
of
insight
in
intellectual
giftedness.
Gifted
Child Quarterly,
28,
58-64.
Davis,
J.
H.,
Carey,
M.
H.,
Foxman,
P. N., &
Tarr,
D. B.
(1968).
Verbalization,
experimenter
presence,
and
problem
solving.
Jour-
nal
of
Personality
and
Social Psychology,
8,
299-302.
Duncker,
K.
(1945).
On
problem solving. Psychological Mono-
graphs, 58,(5, Whole
No.
270).
Durkin,
H. E.
(1937).
Trial-and-error,
gradual analysis
and
sudden
reorganization:
An
experimental study
of
problem solving.
Ar-
chives
of
Psychology, 210, 1-85.
Ellis,
H. D.
(1984).
Practical aspects
of
face memory.
In G. L.
Wells
& E. F.
Loftus
(Eds.),
Eyewitness testimony: Psychological per-
spectives (pp.
12-37).
Cambridge, England: Cambridge
Univer-
sity
Press.
Ericsson,
K.
A.,
&
Crutcher,
R. J.
(1991). Introspection
and
verbal
reports
on
cognitive
processes—Two
approaches
to the
study
of
thinking:
A
response
to
Howe.
New
Ideas
in
Psychology,
9,
57-71.
Ericsson,
K.
A.,
&
Simon,
H. A.
(1980). Verbal reports
as
data.
Psychological Review,
87,
215-251.
Ericsson,
K.
A.,
&
Simon,
H. A.
(1984). Protocol analysis. Cam-
bridge,
MA: MIT
Press.
Eriksen,
C. W.,
Webb,
J.
M.,
&
Fournier,
L. R.
(1990).
How
much
processing
do
nonattended stimuli receive? Apparently very
little,
but...
Perception
and
Psychophysics,
47,
477^88.
Fallshore,
M.,
&
Schooler,
J. W.
(1993).
The
verbal vulnerability
of
perceptual expertise. Manuscript submitted
for
publication.
Finke,
R. A.
(1990).
Creative imagery: Discoveries
and
inventions
in
visualization. Hillsdale,
NJ:
Erlbaum.
Gagne,
R.
H.,
&
Smith,
E. C.
(1962).
A
study
of the
effects
of
verbalization
on
problem solving. Journal
of
Experimental Psy-
chology,
63,
12-18.
Ghiselin,
B.(1952).
The
creative process:
A
symposium.
New
York:
Mentor
Books.
Grosswirth,
M.,
&
Salny,
A.
(1983).
Mensa
genius quiz book
2.
Reading,
MA:
Addison-Wesley.
Hadamard,
J.
(1954).
An
essay
on the
psychology
of
invention
in the
mathematical field.
New
York: Dover.
Hunt,
E.,
&
Agnoli,
F.
(1991).
The
Whorfian
hypothesis:
A
cognitive
psychology perspective. Psychological Review,
98,
377-389.
Jacoby,
L.
L.,
Ste-Marie,
D.,
&
Tom,
J. P. (in
press). Redefining
automaticity:
Unconscious
influences,
awareness
and
control.
In

THOUGHTS BEYOND WORDS
181
A.
D.
Baddeley
& L.
Weiskrantz
(Eds.),
Attention, selection,
awareness
and
control:
A
tribute
to
Donald Broadbent.
James,
W.
(1890).
The
principles
of
psychology.
New
York: Holt.
Kahneman,
D.
(1973).
Attention
and
effort.
Englewood
Cliffs,
NJ:
Prentice-Hall.
Kaplan,
C. A.
(1989).
Hatching
a
theory
of
insight. Does
putting
a
problem aside
really
help?
If
so,
why? Unpublished doctoral dis-
sertation,
Carnegie-Mellon
University,
Department
of
Psychol-
ogy, Pittsburgh,
PA.
Kaplan,
C.
A.,
&
Simon,
H. A.
(1990).
In
search
of
insight. Cog-
nitive Psychology,
22,
374-^19.
Kaplan,
P.
(1964).
More posers:
80
intriguing hurdles
for
reason-
ably
able minds.
New
York: Harper
&
Row.
Kimble,
G.
A.,
&
Perlmuter,
L. C.
(1970).
The
problem
of
volition.
Psychological Review,
77,
361-384.
Koestler,
A.
(1964).
The act
of
creation.
New
York: Dell.
Kohler,
W.
(1921).
Intelligenzprufungen
an
Menschenaffen
[The
mentality
of
apes].
Berlin, Federal Republic
of
Germany:
Springer-Verlag.
Langer,
E. J., &
Imber,
L. G.
(1979). When practice makes
im-
perfect:
Debilitating
effects
of
overlearning.
Journal
of
Person-
ality
and
Social Psychology,
37,
2014-2024.
Langley,
P., &
Jones,
R.
(1988).
A
computational model
of
scientific
insight.
In R. J.
Steinberg
(Ed.),
The
nature
of
creativity: Con-
temporary
psychological perspectives (pp.
177-201).
Cam-
bridge, England: Cambridge University Press.
Luchins,
A. S.
(1942). Mechanization
in
problem solving:
The ef-
fect
of
Einstellung. Psychological Monographs,
54, (6,
Whole
No.
248).
Maier,
N. R. R
(1931). Reasoning
in
humans.
II. The
solution
of a
problem
and its
appearance
in
consciousness. Journal
of
Com-
parative Psychology,
12,
181-194.
Metcalfe,
J.
(1986a).
Feeling
of
knowing
in
memory
and
problem
solving. Journal
of
Experimental Psychology: Learning, Mem-
ory,
and
Cognition,
12,
288-294.
Metcalfe,
J.
(1986b). Premonitions
of
insight predict impending
error. Journal
of
Experimental Psychology: Learning, Memory,
and
Cognition,
12,
623-634.
Metcalfe,
J.,
&
Wiebe,
D.
(1987).
Intuition
in
insight
and
noninsight
problem solving. Memory
&
Cognition,
15,
238-246.
Newell,
A.,
&
Simon,
H. A.
(1972).
Human problem solving.
En-
glewood
Cliffs,
NJ:
Prentice-Hall.
Nisbett,
R.
E.,
&
Wilson,
T. D.
(1977). Telling more than
we can
know:
Verbal reports
on
mental
processes.
Psychological Review,
84,
231-259.
Norman,
D.
A.,
&
Shallice,
T.
(1986). Attention
to
action: Willed
and
automatic
control
of
behavior.
In R. J.
Davidson,
G. E.
Schwartz,
& D.
Shapiro
(Eds.),
Consciousness
and
self-
regulation: Advances
in
research
and
theory (Vol.
4, pp.
1-18).
New
York: Plenum
Press.
Ohlsson,
S.
(1984a).
Restructuring
revisited.
I.
Summary
and
cri-
tique
of the
Gestalt theory
of
problem solving. Scandinavian
Journal
of
Psychology,
24,
65-78.
Ohlsson,
S.
(1984b). Restructuring revisited.
II. An
information
processing theory
of
restructuring
and
insight. Scandinavian
Journal
of
Psychology,
25,
117-129.
Ohlsson,
S.
(1990).
The
mechanism
of
restructuring
in
geometry.
Proceedings
of
the
12th Annual
Conference
of
the
Cognitive Sci-
ence Society (pp.
237-244).
Cambridge,
MA: MIT
Press.
Ohlsson,
S.
(1992).
Information-processing
explanations
of
insight
and
related phenomena.
In M.
Keane
& K.
Gilhooly
(Eds.),
Ad-
vances
in the
psychology
of
thinking (Vol.
1, pp.
1-44).
London,
OK:
Harvester Wheatsheaf.
Olton,
R.
M.,
&
Johnson,
D. M.
(1976). Mechanisms
of
incubation
in
creative problem solving. American Journal
of
Psychology,
89,
617-630.
Orne,
M. T.
(1962).
On the
social psychology
of the
psychological
experiment: With particular reference
to
demand
characteristics
and
their implications. American Psychologist,
17,
776-783.
Perkins,
D. N.
(1981).
The
mind's best work. Cambridge, England:
Cambridge University
Press.
Poincare,
H.
(1952).
Mathematical
discovery. In H.
Poincare
(Ed.),
Science
and
method (pp.
46-63).
New
York: Dover.
Polanyi,
M.
(1967).
The
tacit dimension. Garden City,
NY:
Anchor
Books.
Reber,
A. S.
(1989). Implicit learning
and
tacit knowledge. Journal
of
Experimental Psychology: General, 118,
219-235.
Russo,
J.
E.,
Johnson,
E.
J.,
&
Stephens,
D. L.
(1989).
The
validity
of
verbal protocols. Memory
&
Cognition,
17,
759-769.
Schacter,
D. L.
(1987).
Implicit memory: History
and
current status.
Journal
of
Experimental Psychology: Learning,
Memory,
and
Cognition,
13,
501-518.
Schlipp,
P. A.
(Ed.).
(1949). Albert Einstein: Philosopher-scientist.
Cambridge, England: Cambridge University Press.
Schneider,
W.,
Dumais,
S.
T.,
&
Shiffrin,
R. M.
(1984).
Automatic
and
control processing
and
attention.
In R.
Parasuraman
& D. R.
Davies
(Eds.),
Varieties
of
attention (pp.
1-27).
San
Diego,
CA:
Academic
Press.
Schooler,
J. W.
(1989, November). Verbalization
can
overshadow
the
non-verbal
components)
of
visual memories. Paper presented
at the
Annual
Meeting
of the
Psychonomic
Society, Atlanta,
GA.
Schooler,
J.
W.,
&
Engstler-Schooler,
T. Y.
(1990). Verbal over-
shadowing
of
visual
memories: Some things
are
better
left
unsaid.
Cognitive Psychology,
17,
36-71.
Schooler,
J. W,
Ryan,
R.,
&
Reder,
L. M.
(1991,
July). Better
the
second time around: Re-presentation reverses
verbalization's
im-
pairment
of
face recognition. Paper presented
at the
International
Conference
on
Memory, Lancaster, England.
Simon,
H. A.
(1986).
The
information processing explanation
of
Gestalt phenomena. Computers
in
Human Behavior,
2,
241-255.
Smith,
S.
M.,
&
Vela,
E.
(1991). Incubated reminiscence
effects.
Memory
&
Cognition,
19,
168-176.
Sokolov,
A. N.
(1972).
Inner speech
and
thought.
New
York: Ple-
num
Press.
Tetlock,
P.
E.,
&
Kim,
J. I.
(1987).
Accountability
and
judgment
processes
in a
personality prediction task. Journal
of
Personality
and
Social Psychology,
52,
700-709.
Vygotsky,
L.
(1989).
Thought
and
language.
(A.
Kozulin,
Ed. and
Trans.).
Cambridge,
MA: MIT
Press.
(Original work published
1934)
Wallas,
G.
(1926).
The art
of
thought.
New
York:
Harcourt,
Brace.
Weisberg,
R. W.
(1986).
Creativity, genius
and
other myths.
New
York:
Freeman.
Weisberg,
R.
W.,
&
Alba,
J. W.
(1982).
Problem solving
is not
like
perception: More
on
Gestalt theory. Journal
of
Experimental
Psy-
chology: General, 111,
326-330.
Weisberg,
R.,
&
Suls,
J. M.
(1973).
An
information-processing
model
of
Duncker's
candle problem. Cognitive Psychology,
4,
255-276.
Wertheimer,
M.
(1959). Productive thinking (2nd ed.).
New
York:
Harper
&
Row.
Wilder,
L.,
&
Harvey,
D. J.
(1971). Overt
and
covert verbalization
in
problem solving. Speech Monographs,
38,
171-176.
Wilson,
T.
D.,
Lisle,
D.
J.,
Schooler,
J.
W.,
Hodges,
S.
D.,
Klaaren,
K.
J.,
&
Lafleur,
S. J. (in
press).
Introspecting about reasons
can
reduce post-choice satisfaction. Personality
and
Social Psychol-
ogy
Bulletin.
Wilson,
T.
D.,
&
Schooler,
J. W.
(1991). Thinking
too
much:
In-
trospection
can
reduce
the
quality
of
preferences
and
decisions?

182
J.
SCHOOLER,
S.
OHLSSON,
AND K.
BROOKS
Journal
of
Personality
and
Social Psychology,
60,
181-192.
Woodworth,
R. S.
(1938). Experimental psychology.
New
York:
Holt.
Yaniv,
I.,
&
Meyer,
D. E.
(1987).
Activation
and
metacognition
of
inaccessible
stored information: Potential bases
for
incubation
effects
in
problem solving. Journal
of
Experimental
Psychology:
Learning, Memory,
and
Cognition,
13,
187-205.
Zajonc,
R. V.
(1965).
Social facilitation. Science, 149,
269-274.
Appendix
A
Insight Problems
1.
(Experiments
1-4)
Show
how you can
make
the
triangle below
point downward
by
moving only three
of the
circles,
[see Figure
Al.]
2.
(Experiments
i^)
A
prisoner
was
attempting
to
escape
from
a
tower.
He
found
in his
cell
a
rope
that
was
half long enough
to
permit
him to
reach ground safely.
He
divided
the
rope
in
half, tied
the
two
parts together,
and
escaped.
How
could
he
have done this?
Solution:
He
unraveled
the
rope
and
tied
the two
pieces together.
3.
(Experiments
1-4)
A
dealer
in
antique coins
got an
offer
to buy
a
beautiful bronze coin.
The
coin
had an
emperor's head
on one
side
and
the
date
544
B.C.
stamped
on the
other.
The
dealer examined
the
coin,
but
instead
of
buying
it, he
called
the
police. Why?
o o
o
o o
O
Solution
Solution:
Figure
Al.
Diagram
and
solution
for the
"Triangle"
problem.
Figure
A2.
Diagram
and
solution
for the
"Pigs
in a
pen'
problem.

THOUGHTS BEYOND WORDS
183
Solution:
In 544
B.C.
Christ
had not
been born,
so a
coin from that
time would
not be
marked
"B.C."
(before Christ).
4.
(Experiments
1 and 2)
Nine pigs
are
kept
in a
square pen. Build
two
more square enclosures that would
put
each
pig in a pen by
itself
[See Figure
A2.]
5.
(Experiments
1
and
2)
Describe
how
to
cut
a
hole
in
a
3-
X-5-in.
card that
is big
enough
for you to put
your head through.
Solution: First
cut a
spiral path from
the
outside
of the
card
to the
inside.
Then
cut a
long slit down
the
middle
of the
spiral strip
leaving
the
ends
of the
strip intact.
A
number
of
similar variations
to
this solution were also accepted.
6.
(Experiments
1 and 2) A
giant inverted steel pyramid
is
per-
fectly
balanced
on its
point.
Any
movement
of the
pyramid will
cause
it to
topple over. Underneath
the
pyramid
is a
$100 bill.
How
would
you
remove
the
bill without disturbing
the
pyramid?
Solution: Burn
or
tear
the
dollar bill.
7.
(Experiment
1
only) Water lilies double
in
area
every
24 hr.
At
the
beginning
of the
summer, there
is one
water lily
on the
lake.
It
takes
60
days
for the
lake
to
become completely covered
with
water lilies.
On
which
day is the
lake half-covered?
Solution:
The
lake
is
half-covered
on the
59th day.
Appendix
B
Noninsight
Problems
Used
in
Experiments
3 and 4
1.
(Practice) Mary won't
eat fish or
spinach, Sally won't
eat
fish
or
green beans, Steve won't
eat
shrimp
or
potatoes, Alice won't
eat
beef
or
tomatoes,
and Jim
won't
eat fish or
tomatoes.
If you are
willing
to
give such
a
bunch
of
fussy
eaters
a
dinner party, which
items
from
the
following
list
can you
serve: green beans, creamed
codfish,
roast beef, roast chicken, celery,
and
lettuce.
Solution: roast chicken, celery,
and
lettuce.
2.
Three cards
from
an
ordinary deck
are
lying
on a
table, face
down.
The
following information (for some peculiar reason)
is
known
about those three cards (all
the
information below refers
to
the
same three cards):
• To the
left
of a
queen there
is a
jack
• To the
left
of a
spade there
is a
diamond
• To the right of a
heart there
is a
king
• To the right of a
king there
is a
spade
Can you
assign
the
proper
suit
to
each picture card?
Solution: jack
of
hearts, king
of
diamonds, queen
of
spades.
3. The
police
were
convinced
that
either
A, B, C, or D had
com-
mitted
a
crime. Each
of the
suspects,
in
turn, made
a
statement,
but
only
one of the
four
statements
was
true.
• A
said,
"I
didn't
do
it."
• B
said,
"A is
lying."
• C
said,
"B is
lying."
• D
said,
"B did
it."
Who is
telling
the
truth?
and Who
committed
the
crime?
Solution:
B is
telling
the
truth,
and A
committed
the
crime.
4.
There
are
four
coins—two
heavier coins
of
equal weight
and
two
lighter coins
of
equal weight,
all of
which
are
indistinguishable
in
appearance
or by
touch (you cannot tell them apart
by
looking
at
them
or
holding them).
How can you
tell which coins
are the
heavy
ones
and
which coins
are the
light ones
in two
weighings
on
a
balance
scale?
(You
may
only
use the
scale twice.)
Solution: Begin
by
placing
one
coin
on
each side
of the
scale.
If
they
do not
balance, then
you
have already identified
one
heavy
and
one
light coin. Repeating
the
procedure with
the
remaining
two
coins will
identify
the
other light
and
heavy coins.
If the
initial
two
coins balance, simply remove
one of the
coins
and
replace
it
with
one of the
remaining coins. This weighing will provide
the
remain-
ing
information needed
to
determine which coins
are
heavy
and
which
are
light.
Received January
13,
1992
Revision received September
14,
1992
Accepted September
24,
1992