Restructuring Versus Automaticity. Alternative Accounts of Skill Acquisition

Abstract

Disagrees with the view of W. Schneider and R. M. Shiffrin (see PA, Vol 58:305; see also Shiffrin and Schneider, PA, Vol 57:4785) that improvement in performance is due to the development of automatic (capacity-free) processes. The present author notes that Schneider and Shiffrin's evidence was confounded by a category difference between memory-ensemble and distractor-set characters and argues that attempts by Shiffrin and Schneider to resolve the confounding failed. An alternative explanation of their results is presented in terms of the restructuring of task components. Other evidence regarding capacity-free processing is reviewed, and the implications of the critique for limited-capacity models of attention are discussed. (34 ref) (PsycINFO Database Record (c) 2012 APA, all rights reserved)

Full text

Psychological
Review
1985.
Vol.
92,
No.
3.
414-423
Copyright
1985
by the
Am
n
Psychological
Association,
Int.
0033-295X/85IS00.75
Restructuring
Versus
Automaticity: Alternative
Accounts
of
Skill Acquisition
Patricia
W.
Cheng
Carnegie-Mellon
University
Limitations
in
human
performance have frequently
been
explained
by the
concept
of
limited
capacity.
Consonant
with this view
is the
assumption
that
improvement
in
performance
is due to the
development
of
automatic
(i.e., capacity-free)
processes.
An
important,
putative
source
of
evidence
for
automaticity
has
been
Schneider
and
Shiffrin's
(1977;
Shifrrin
&
Schneider,
1977)
findings. The
present
article
notes
that
Schneider
and
Shiffrin's
evidence
was
confounded
by a
category
difference
between
memory-ensemble
and
distractor-set
characters,
and
argues
that
attempts
by
Shiffrin
and
Schneider
to
resolve
the
confounding
failed.
The
article
presents
an
alternative
explanation
of
their
results
in
terms
of the
restructuring
of
task
components,
reviews
other
evidence
regarding
capacity-free
processing,
and
discusses
the
implications
of the
critique
for
limited-capacity
models
of
attention.
The
concept
of
limited capacity
has
fre-
quently
been invoked
to
explain limitations
in
human performance.
Shiffrin
and
Schnei-
der
(1977,
p.
186),
for
example, concluded
that divided-attention
deficits
arise
from
lim-
itations
on
controlled processing,
a
mode
of
processing that
is
capacity demanding.
La-
Berge
and
Samuels
(1974)
attributed
the
dif-
ficulty
of
a
complex task such
as
reading
to
the
need
to
distribute limited-processing
ca-
pacity across many task components
at the
same time. Within this theoretical
framework
of
limited capacity,
how
would improvement
in
performance
be
explained? Schneider
and
Shiffrin
(1977;
Shiffrin
&
Schneider, 1977)
proposed
a
dichotomy
of
processes: automatic
and
controlled. Whereas controlled processing
is
capacity demanding
(i.e.,
load dependent),
automatic processing
is not
(i.e.,
an
automatic
task
can be
done without degrading perfor-
mance
on
concurrent, physically compatible
tasks).
The
development
of a
capacity-free
process provides
an
attractive explanation
for
improvement
in
performance: When
a
corn-
Preparation
of
this
article
was
supported
by
National
Science
Foundation
Grant
BNS-7904730.
Keith
Holyoak,
John
Jonides,
John
Palmer,
Keith
Smith,
and
several
reviewers
provided
helpful
comments
on
an
earlier
draft.
Requests
for
reprints
should
be
addressed
to
Patricia
Cheng,
Computer Science Department,
Carnegie-Mellon
University,
Pittsburgh,
Pennsylvania
15213.
ponent
of a
task becomes capacity
free,
one's
attention
can be
devoted
to
other components
of
the
task.
As
enough components become
capacity
free,
a
previously
difficult
or
impos-
sible
task becomes easy
or
possible.
The
dichotomy
of
capacity-demanding
and
capac-
ity-free
processing—simple,
intuitively
ap-
pealing,
and
potentially neatly
definable
with
the aid of
converging
attributes—lies
at the
heart
of
Schneider
and
Shiffrin's
frequently
cited automaticity-control distinction.
But
improvement
in
performance
can be
explained
in
other
ways.
In
particular,
it can
be
due to a
restructuring
of the
task
compo-
nents
so
that they
are
coordinated, integrated,
or
reorganized into
new
perceptual, cognitive,
or
motor
units,
thereby allowing
the
proce-
dure involving
the old
components
to be
replaced
by a
more
efficient
procedure
in-
volving
the new
components.
For
example,
consider
two
alternative procedures
for
solving
the
following
arithmetic
problem:
Find
the
sum of ten 2s. A
problem solver
who
knows
addition
but not
multiplication would have
to
perform
nine addition operations
to
arrive
at
the
answer.
Now
suppose that
the
problem
solver learns
the
multiplication table.
She
could then solve
the
same problem
by
looking
up the
entry
2 X 10 in her
memory.
A
single
multiplication
operation would thus
be
equivalent
to
nine addition operations. This
gain
in
efficiency,
however,
is
obviously
not
achieved
by
performing
the
nine addition
414
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

RESTRUCTURING VERSUS
AUTOMATICITY
415
operations
in a
capacity-free manner. Rather,
those operations
are not
performed
at
all.
Neither
is the
gain
in
efficiency
achieved
by
performing
an
"automatic"
multiplication
operation. More generally, limitations
in
per-
formance
are not
necessarily overcome
by
shifting
from
a
capacity-demanding
into
a
capacity-free
mode
of
processing,
but may
also
be
overcome
by
restructuring
the
task
procedure.
The
present article questions
the
soundness
of
Schneider
and
Shiffrin's
evidence
for
load
independence, presents
an
alternative expla-
nation
in
terms
of
restructuring, reviews other
evidence
regarding load independence,
and
discusses
the
implications
of the
critique
for
limited-capacity
models
of
attention. (The
issue
of
load independence
was not
addressed
by
Ryan's, 1983, critique
of
Schneider
and
Shiffrin.)
Because
I aim to
examine auto-
maticity
in the
context
of
capacity limitations,
I
do not
address other aspects
of
attention,
such
as the
effect
of
expectation
(or the
lack
of
it) on
processing time (e.g.,
LaBerge,
1973;
Posner
&
Snyder,
1975),
voluntary control
(e.g.,
Shiffrin
&
Schneider, 1977, Experiment
4),
or
interference
as
denned
by the
statistical
interaction between
the
load parameters
of
concurrent
tasks (e.g., Logan,
1978).
Although
automaticity
and
restructuring
are not mu-
tually
exclusive
(processes
can
conceivably
be
both restructured
and
automatized),
I
contend
that beyond
the
early processing stage
of
physical-feature
detection there
is as yet
no
empirical evidence
for
capacity-free pro-
cessing.
Critique
of
Schneider
and
Shiffrin's
Initial
Studies
Schneider
and
Shiffrin
(1977,
Experiments
1
and 2)
reported that performance showed
little
effect
of
load (i.e.,
no
memory-
or
display-set size
effects)
in a
consistent mapping
(CM)
condition
in
which targets
and
distrac-
tors
in
search tasks remained
constant
across
trials,
whereas performance showed enormous
effects
of
load
in a
varied mapping (VM)
condition
in
which targets
and
distractors
were
varied
from
trial
to
trial. They concluded
that
CM
produced automatic detection,
which
is
capacity
free,
whereas
VM
produced con-
trolled
search,
which
is
capacity demanding.
However,
the
implications
of
their
clear-
cut
results
are
unfortunately
ambiguous,
for
as
they themselves
pointed
out,
consistency
of
mapping
was
confounded
with
whether
or
not the
memory ensemble
and
distractor
set
were
categorically distinct. Under
CM
con-
ditions memory ensemble items were digits,
whereas
distractors
were
letters
(or
vice versa).
This distinction allows
a
possible alternative
explanation
of the flat
set-size
functions
ob-
tained
under
CM
conditions: Subjects could
restructure
the
task
by
using
a
category
strat-
egy,
whereby
they
make direct
use of
cate-
gorical
information, bypass item-by-item
processing,
and
accordingly produce
flat
set-
size
functions.
In
other words, although pro-
cessing
of
display
and
memory items
may in
fact
require capacity, subjects
may not
need
to
fully
process display items
or
compare
them
to any
item
in the
memory set.
As
I
argue below,
it is
possible
to
distinguish
variants
of a
category strategy.
In the
case
of
digit-versus-letter
discriminations,
the
strategy
could
be
based
on
perceptual
features
(the
categorical-feature
strategy). Considerable
evidence
indicates that subjects
are
indeed
capable
of
classifying
digits
and
letters,
or
searching
for
digits among letters
(or
vice
versa),
by
using categorically distinguishing
physical
information
directly, without
first
having
to
identify
the
individual items (e.g.,
Brand,
1971;
Gleitman
&
Jonides, 1976;
Jonides
&
Gleitman, 1976; Posner, 1970;
Taylor,
1978). Posner
(1970)
found
that
the
task
of
matching
two
letters
on the
basis
of
their
name
was
slowed considerably
by
visual
and
acoustic similarity among
the
letters,
but
that
the
task
of
matching alphanumeric char-
acters
on the
basis
of
their category
was not
influenced
by
this variable. Posner reasoned
that
if
category information were derived
from
the
name
of a
character, then
any
variable
that
affected
the
time
to
retrieve
the
name should also
be
reflected
in the
time
to
respond
on the
basis
of the
category. Because
his
results
did not
support this prediction,
he
concluded that
identification
and
categoriza-
tion
are
simultaneous processes, each pro-
ceeding
from
the
physical
features
of the
character.
Additional
evidence
is
provided
by
many
studies showing that targets
in a
visual search
task
are
detected
faster
and
more accurately
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

416
PATRICIA
W.
CHENG
in
a
between-category
condition
in
which
the
target
is
among distractors
of a
different
category (digits versus
letters)
than
in a
within-
category
condition
in
which
all
array items
are
from
the
same category (e.g., Brand,
1971;
Ingling,
1972).
In
fact,
between-category
search
sometimes yields
reaction
time (RT)
functions
that
are
independent
of the
number
of
display
items
(e.g.,
Egeth,
Jonides,
&
Wall,
1972).
Gleitman
and
Jonides
(1976;
Jonides
&
Gleitman, 1976)
found
that subjects
in a
between-category
condition were able
to
locate
the
target
solely
on the
basis
of
category,
and
were
generally unaware
of the
identities
of
the
characters
in the
display.
They suggested
that subjects based their responses
on
physical
features
that
differentiate
between categories,
and
that these features required less time
to
extract
than
features
that
identify
a
character.
Arriving
at a
similar conclusion, Duncan
(1983)
presented
and
summarized evidence
showing
that
the
slopes
of
both between-
category
and
within-category
visual searches
are
heavily
influenced
by the
physical resem-
blance between targets
and
nontargets.
In
fact,
controlling
for
this
variable causes
the
between-category
advantage
to
vanish alto-
gether,
and
manipulating this variable
can
even
cause
the
effect
to
reverse.
His
results
suggest
that
the
lack
of
effect
of
display-set
size
sometimes obtained
for
between-category
search
of
digits versus letters
was
probably
due
to
subjects' ability
to
base their responses
on
physical
features
that
differentiate
between
the
categories.
Related results were obtained
by
Rabbitt
(1967).
Rabbitt required subjects
to find a
target
letter
from
a
prescribed
set
among
a
number
of
letter
distractors.
Early
on in
practice
RT
increased with
the
number
of
distractors.
As
practice continued
with
the
same
set of
letters,
the
effect
of
number
of
distractors
decreased;
they
no
longer interfered
as
much.
But
when
the set of
distractors
was
changed, interference re-emerged, unless
the
new
distractors
were
visually similar
to the
old
ones. Subjects
in
these experiments were
clearly
learning
to
differentiate
between
the
classes
of
targets
and
distractors
by
certain
key
visual
features.
If
subjects
are
able
to
extract
category
information
directly
from
the
visual
features
of
display characters, then
the flat
set-size
functions
for CM
conditions
reported
by
Schneider
and
Shiffrin
(1977)
need
not
imply
capacity-free
processing.
By
using
a
categor-
ical-feature strategy
to
detect
a
digit among
letters, subjects need
not
identify
the
display
items
or
compare
any
display item
to any
memory-set item.
As a
result, display-
and
memory-set
size would
not
affect
perfor-
mance.
Such
a
strategy would
redefine
load—
individual
display
and
memory items would
no
longer
be
valid psychological units
of the
process.
Given
the
plausibility
of a
categori-
cal-feature strategy,
the flat
set-size functions
reported
by
Schneider
and
Shiffrin
(1977)
do
not
constitute evidence
for
load-independent
processing.
Subsequent
experiments
(Shiffrin
&
Schnei-
der,
1977, Experiments
2 and 3)
attempted
to
resolve
the
confounding between category
and
automaticity. Experiment
2
tested
the
effects
of
"reversals"
in the
mapping
of
targets
and
distractors
in CM and VM
conditions
with
well-known
and
well-practiced
catego-
ries.
In the CM
condition subjects
who had
been trained
to
search
for
digits among con-
sonant
distractors
now
searched
for
conso-
nants
among digit distractors
(or
vice
versa).
In
the VM
condition these same subjects,
who
had
been searching
for
consonants among
consonant distractors,
now
searched
for
digits
among
digit
distractors.
Results showed
that
the
reversal
in
mapping
had no
effect
on VM
performance,
but
dramatically
lowered detec-
tion
accuracy
in the CM
condition.
Shiffrin
and
Schneider noted that reversal markedly
lowered
accuracy
in the CM
condition despite
the
preservation
of a
well-learned
categorical
difference
between memory ensemble
and
distractor set,
and
proceeded
to
argue
that,
therefore,
the
categorical distinction between
memory ensemble
and
distractor
set
could
not be the
cause
of
phenomena associated
with
automaticity.
It
should
first be
noted
that
the
experiment
completely omitted
any
manipulation
of
load:
Memory-
and
display-set sizes were both kept
constant
at 2. As a
result,
no
inferences
can
be
drawn
from
this experiment regarding
whether
a
category strategy
can
account
for
set-size
functions.
Because load independence
was
not
tested,
the
effect
of
reversal
on CM
performance
warrants only
the
conclusion
that
a
category
strategy alone cannot account
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

RESTRUCTURING VERSUS AUTOMAT1CITY
417
for
levels
of
accuracy—a
conclusion irrelevant
to the
issue
of
load independence,
as
well
as
to the
general issue
of
whether there
are two
qualitatively
different
modes
of
processing.
Relative
accuracy
in
itself
does
not
seem
to
qualify
as a
denning
attribute
for
qualitatively
different
modes
of
processing because
the
criterion
on the
continuum
of
accuracy
was
left
unspecified
and
seems necessarily arbi-
trary.
The
markedly
different
effects
of
"reversal"
on CM and VM
performance might none-
theless lead
one to
infer
that there
are two
dichotomous
modes. Such
an
inference
would
be
specious, however, because
"reversal"
in-
volved
quite
different
operations
for CM and
VM
conditions.
In the CM
condition memory
ensemble
and
distractor
set
were
reversed,
whereas
in the VM
condition,
in
which mem-
ory
ensemble
and
distractor
set
were chosen
from
the
same category
of
characters,
the
two
sets were both
shifted
to
another category
of
characters. Because
the
"reversal"
opera-
tions were
different
for the two
mapping
conditions, there
is no
reason
to
expect
the
operations
to
have
the
same
effect
even
on
identical
modes
of
processing.
One
therefore
cannot
infer
from
the
different
effects
of the
operations that
different
modes
of
processing
exist.
In
sum, Experiment
2 did not
rule
out
the
categorical-feature strategy
as an
alter-
native
explanation
for
flat
set-size
functions
under
CM
conditions.
It did not
provide
any
evidence
for
load independence
nor for a
dichotomy
of
processes.
Experiment
3
(Shiffrin
&
Schneider, 1977)
was
another attempt
to
unconfound category
strategy
and
automaticity.
It was
shown that,
after
extensive training
on
arbitrary collections
of
characters that
were
consistently grouped
in
two
separate categories (the categorical
condition), memory-set size came
to
have
no
effect
on
performance despite
the
characters
being
variably mapped onto
the
responses.
(Display-set size
was not
manipulated.) This
result
is
clearly consistent
with
the
hypothesis
that
a
category strategy could lead
to
flat
set-size
functions.
Curiously,
Shiffrin
and
Schneider
interpreted
the flat
set-size
function
to be
indicative
of a
controlled search.
To
explain
the
absence
of a
positive slope,
they
hypothesized
that
the
search, although con-
trolled,
was for a
category rather than
for
individual
items. Their argument
was
based
on
(a) the
increased accuracy
of
detection
observed
when
subjects were
afterward
trained
under
a CM
condition
and (b) the
patterns
of
performance
in
conditions
in
which
two
targets
were
to be
detected.1
The
latter were
almost
identical
in the
categorical
VM
and CM
conditions
and
could
not
contri-
bute
differentiating
evidence (see
Shiffrin
&
Schneider,
1977, Figures
6 and 7),
while
relative
accuracy,
as
mentioned earlier, does
not
seem
to
qualify
as a
denning
attribute
for
different
modes
of
processing.
How
cate-
gorical
controlled search
is to be
distinguished
from
automatic detection, other than
by
rel-
ative
accuracy,
was
left
unclear.
In
fact,
given
that
a
categorical controlled
search
is
conceptually allowable
in
Schneider
and
Shiffrin's
model, Experiment
3
could
not
possibly resolve
the
confounding
between
the
category
strategy
and
load-independent
pro-
cessing.
The
experiment tested performance
under
three conditions: categorical
VM,
non-
categorical
VM, and
categorical
CM.
Predic-
tions regarding set-size dependence based
on
Shiffrin
and
Schneider's model
and
those
based
on the
category hypothesis coincide
in
all
three conditions: Both models predict that
only
the
noncategorical
VM
condition would
yield
set-size dependent performance.
The
experiment
omitted
the
critical condition
in
1
One
type
of
evidence
that
Schneider
and
Shiffrin
brought
to
bear
on
their
two-process
model
was the
pattern
of
performance when
detection
of two
targets
was
required. Performance
was
measured
as a
function
of the
temporal spacing between
the
targets
and of the
similarity
of the
targets. Such
patterns
seemed
by no
means
differentiating,
inasmuch
as
they were notably
unstable
across
experiments.
For
instance,
CM
sometimes
produced
a
nonidentical-targets
advantage (Schneider
&
Shiffrin,
1977,
Experiment
3), and at
other
times
an
identical-targets advantage
(Shiffrin
&
Schneider, 1977,
Experiment
3).
Similarly,
VM
sometimes
produced
the
greatest
decrement
when
the two
targets
were
one
frame
apart
(Schneider
&
Shiffrin,
1977,
Experiment
3), and at
other
times
when
the two
targets were simultaneous
(Shiffrin
&
Schneider, 1977, Experiment
3). In
fact,
VM
and
CM
sometimes
produced
almost
identical
patterns
of
performance
(ShiflHn
&
Schneider, 1977, Experiment
3).
Furthermore,
the
pattern
could
be
substantially
changed merely
by
increasing
the
frame-exposure
time
(see
Shiffrin
&
Schneider,
1977,
Figure
4). An
increase
in
frame
time
from
60 to
120
ms
caused
changes
in
both
the
relative
spacing
decrements
and the
target-similarity
effect.
The
explanations
offered
for
such
capricious
pat-
terns
were necessarily
post
hoc and
speculative.
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

418
PATRICIA
W.
CHENG
which
mapping
is
consistent
so
that auto-
maticity
should develop,
but in
which
the
memory
ensemble
is not a
category
so
that
a
category strategy
is not
possible.
It
should
be
mentioned
that
the
categories
in
the
above experiment were chosen
to be
visually
confusable.
It was
therefore
unlikely
that subjects could extract category-distin-
guishing
visual
features,
as
they probably
could
for
digits
and
letters. However, another
variant
of the
category strategy, which
can
be
termed
the
category-lag
strategy, could
still
yield
a
flat
set-size
function.
Each mem-
ory-ensemble item
may be
tagged
in
memory
with
a
category label,
so
that
to
decide whether
an
item belongs
to the
memory ensemble,
a
person
has
only
to
identify
the
item
and
check
whether
it has the
category label. This
strategy, like
the
feature
strategy, precludes
the
need
to
compare
a
display item
to
each
memory-set
item.
It
follows
that performance
should
be
independent
of
memory-set size.
Note that
the
category-tag strategy
does
not
predict
that
RT
will
remain
constant
as a
function
of
display-set size. Because
Shiffrin
and
Schneider
did not
manipulate display-set
size
in
Experiment
3,
their results
are
consis-
tent with
the
category-tag hypothesis.
In
summary,
the first two
experiments
in
Schneider
and
Shiffrin
(1977),
those that
showed
a
correlation between
the
consistency
of
mapping
and
load independence,
were
confounded
by a
category distinction between
memory-ensemble
and
distractor-set items,
and
none
of the
subsequent experiments
served
to
resolve
the
confounding.
Whenever
a
flat
set-size
function
was
obtained, memory
ensemble
and
distractor
set
always belonged
to
separate categories, thus allowing
for the
possibility
of a
category strategy.
Other
Evidence Regarding
Capacity-Free Processing
Experiments
on
Category-Based
Search
Schneider,
Shiffrin,
and
their colleagues
(Dumais
&
Shiffrin,
1978;
Fisk
&
Schneider,
1983;
Schneider
&
Fisk, 1984) have sub-
sequently
reported
numerous experiments
on
category-based search. However, unlike
Schneider
and
Shiffrin's
1977
papers,
none
of
the
later reports evidence
any
realization
that
the
existence
of a
category-based strategy
may
obviate
the
need
for the
postulation
of
a
dichotomy
of
processes
differing
on
load
independence.
The
implication
of a
category
strategy
was not
discussed,
and
none
of the
experiments
reported were capable
of
differ-
entiating
between
the
automaticity
and the
restructuring
hypotheses.
Fisk
and
Schneider
(1983)
required subjects
to
detect
whether displayed words were mem-
bers
of
four
taxonomic
categories.
After
more
than
ten
45-min sessions
of CM
training
on
these
categories,
positive trials showed
a
slope
of
2
ms
per
category
and
negative trials,
a
slope
of 10 ms.
Because subjects received
such
prolonged practice
on
only
four
cate-
gories,
it
seems plausible
that
they might
be
able
to
form
an
arbitrary superordinate cat-
egory
encompassing
the
four
target categories,
just
as
subjects
in
Shiffrin
and
Schneider
(1977)
were
able
to
form
categories
consisting
of
arbitrary collections
of
letters.
It
is
interesting that
in
Fisk
and
Schneider's
(1983)
word-search condition,
in
which
the
target
set
consisted
of
eight consistently
mapped words,
the
slope remained
at 20 ms
per
word despite extensive practice. (Note
that each word
in the
word-search condition
received
substantially more practice than
the
same
words
in the
category-search condition.)
In
fact,
practice
had no
effect
at all on the
slope
of
positive trials. Fisk
and
Schneider
offered
no
explanation
for the
word-search
results.
In
contrast, this result
is
readily
ex-
plained
by the
category-tag strategy:
It is
harder
to
form
a
superordinate category
of
eight
than
of
four
items, especially when
the
distractor words were
from
the
same taxo-
nomic categories
as the
target words,
as was
the
case
in the
word-search condition.
Schneider
and
Fisk
(1984)
reported
a
series
of
studies
on
category search
and its
transfer.
The
number
of
to-be-tested
exemplars
in a
taxonomic category
was
found
to
have
no
effect
on the
time required
to
determine
the
location
of the
displayed word that
was a
member
of the
category.
The
mapping
be-
tween
category
and
response
was
consistent.
This result
follows
naturally
from
the
category
hypothesis: Subjects have only
to
look
up the
category
tag of the
displayed words
and
note
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

RESTRUCTURING
VERSUS
AUTOMATICITY
419
the
position
of the
target
to
determine
the
response.
Because there
is no
need
to
compare
the
target word with
the
list
of
exemplars,
the
number
of
exemplars should clearly
be
irrelevant.
The
category-tag hypothesis also
predicts that
the
number
of
exemplars
in a
taxonomic
category would have
no
effect
in
a VM
category
search,
a
condition
that
Schneider
and
Fisk omitted. This condition
would
reveal whether
the
null
effect
of the
number
of
exemplars
is
attributable
to the
type
of
mapping
or to
categorization.
Schneider
and
Fisk
(1984)
also reported
that
CM
training
on a
subset
of
words
refer-
ring
to
exemplars
of a
taxonomic category
transferred
substantially
(up to
92%)
to un-
trained members
of the
category. This result
again
follows
naturally
from
a
category-tag
hypothesis.
To the
extent that untrained
ex-
emplars
are
well
learned, their category
membership
can be
readily
identified.
On
identification,
the
task converges with
the
task
of
detecting trained exemplars.
The
cat-
egory
hypothesis thus predicts transfer
to new
exemplars.
In
fact,
it
predicts
transfer
regard-
less
of the
type
of
mapping.
The
automaticity
hypothesis,
in
contrast, predicts
transfer
only
when
mapping
is
consistent. Transfer
to new
exemplars
under
VM was not
tested
by
Schneider
and
Fisk. Other transfer results
suggesting
a
capacity-free process
(Dumais
&
Shiffrin,
1978)
are
similarly
interpretable
in
terms
of the
category-tag hypothesis.
Besides
the
potentially
differentiating
tests
mentioned
above, varying
display-set
size
in
a CM
taxonomic category search might also
differentiate
between
the
automaticity
and
the
restructuring hypotheses.
As
noted above,
although
the
category hypothesis
can
readily
account
for the
lack
of
effect
of
memory-set
size,
it
could
not
possibly account
for a
lack
of
effect
of
display-set size unless there exist
categorically distinguishing
visual
features,
which
is
unlikely
for
words
referring
to ex-
emplars
of
taxonomic
categories
(e.g., weap-
ons,
fruits,
colors). Schneider,
Shiffrin,
and
their colleagues (Fisk
&
Schneider, 1983;
Schneider
&
Fisk,
1984;
Shiffrin
&
Schneider,
1977,
Experiment
3)
always kept display-set
size
constant whenever detection
was
unlikely
to be
based
on
categorically distinguishing
visual
features,
and
hence
did not
obtain
the
relevant data.
A
series
of
experiments
by
Karlin
and
Bower
(1976), however,
did
vary
display-set
size
in a
task
in
which subjects
had to
detect
whether
a
target word appeared among
a
varying
number
of
distractor
words.
The
targets
were
from
either
the
same
or a
different
taxonomic category
as the
distractors.
The
words
were
consistently
mapped.
Reaction
time
was
found
to be
always
an
increasing
function
of
display-set size. When searching
a
visual
display
for a
single target word,
subjects
searched
at the
same rate whether
the
target
and
distractors
were
in the
same
or in a
different
semantic category, indicating
that
the
visual search could
not
utilize
cate-
gory
information. However, when
the
size
of
the
memory
set was
increased,
the
slopes
of
the RT
functions
of the
same-
and
different-
category conditions increasingly diverged,
in-
dicating
that subjects increasingly used
cate-
gory
information
to
speed
the
memory-search
rate
when
targets
and
distractors belonged
to
different
categories.
The
above pattern
of
results
is
consistent
with
the
category hypothesis. Because words
in
the two
taxonomic categories (either colors
or
trees) used
by
Karlin
and
Bower
(1976)
did
not
have categorically distinguishing visual
features,
only
the
category-tag strategy could
be
used,
but not the
categorical-feature
strat-
egy.
Accordingly, although
the
memory-search
component
of the
task could benefit
from
categorical information,
the
visual-search
component could not. This pattern
of
results
is not
explicable
by
consistency
of
mapping.
In
sum,
the
results
on
category search
reported
by
Schneider,
Shiffrin,
and
their
colleagues
(Dumais
&
Shiffrin,
1978; Fisk
&
Schneider,
1983; Schneider
&
Fisk, 1984;
Shiffrin
&
Schneider, 1977)
as
evidence
for a
capacity-free
process
follow
naturally
from
the
category hypothesis. Experimental con-
ditions that could potentially
differentiate
between
the
category hypothesis
and the au-
tomaticity hypothesis were omitted
in
their
studies,
and
available
differentiating
evidence
favors
the
category hypothesis.
Evidence
From
Dual-Task
Experiments
A
series
of
experiments
reported
by
Schneider
and
Fisk (Fisk
&
Schneider, 1983,
Experiment
3;
Schneider
&
Fisk, 1982a,
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

420
PATRICIA
W.
CHENG
1982b;
Schneider
&
Fisk,
1984,
Experiment
2a)
used
a
dual-task paradigm
to
argue
for
capacity-free
processing.
The
rationale
was
to
show that
a CM
task
can be
performed
concurrently
with another task without
de-
grading performance. Schneider
and
Fisk
(1984)
reported
a
decrement
of 11%
below
the
single-task performance level
in a
primary
task
performed concurrently with
a CM
task,
even
after
prolonged practice, indicating
that
the CM
task
was
capacity demanding
to a
small
extent.
Fisk
and
Schneider
(1983)
re-
ported
a
dual-task decrement
of
only
2%, but
acknowledged
that because performance
in
CM
conditions
was
near ceiling,
the
results
provided only
weak
evidence
for the
hypoth-
esis that
practice
can
make
a CM
task insen-
sitive
to
load.
Schneider
and
Fisk (1982a)
reported
a
dual-task decrement
of at
least
15%
for a
task that
was
always consistently
mapped.
Schneider
and
Fisk
(1982b)
performed
a
further
series
of
experiments that attempted
to
obtain evidence
of
capacity-free processing.
In
all but one of
their dual-task experiments
involving
a CM
task,
a
clear
dual-task dec-
rement
was
observed even
after
extended
practice.
In the
single
remaining experiment
(Experiment
3),
percentage
of
correct detec-
tion
for the CM
task also decreased
in the
dual-task condition. Nonetheless, Schneider
and
Fisk concluded that
CM
search requires
no
capacity, based
on a
lack
of
decrement
in
the
A'
measure
of
detection sensitivity. Despite
their
claim that
the
results were
not
contam-
inated
by
ceiling
or floor
effects,
an
inspection
of
their data reveals that most
of the
data
on
which
the
A'
measures were based consisted
of
very
high
hit
rates
and
very
low
false-
alarm rates
for the CM
task.
If
cases
in
which
the hit
rate
was 94% or
above
or the
false-
alarm
rate
was
7%
or
below
for the
single
CM
condition were discarded,
only
one
pair
of
single-
and
dual-task data would
be
left.
Interpretation
of
A'
seems suspect under such
conditions.
The d'
measures
for the CM
task
reported
by
Schneider
and
Fisk were mean-
ingless
because nearly
all of
their subjects
showed
false-alarm
rates
of 0 for the CM
task
when
it was
paired with
a
concurrent
task.
The
reported
d'
measures were calculated
by
substituting
an
arbitrarily chosen
false-
alarm rate
for 0 in
such
cases.
Had a
slightly
higher
or a
slightly lower false-alarm
rate
been chosen,
the
apparent
pattern
of
results
would
have been completely
different.
Hoffman,
Nelson,
and
Houck (1983)
in-
vestigated whether
an
"automatic"
detection
task requires
the use of
limited-capacity
re-
sources
in a
series
of
experiments that
did
not
suffer
from
ceiling
and floor
problems.
They
required subjects
to
detect
a
consistently
mapped target digit
in a
display
of
letters,
and
paired
the
task with
a
variety
of
concur-
rent
visual
discrimination tasks. Both
the
digit-detection
task
and the
visual-discrimi-
nation task showed large dual-task
interfer-
ence,
in
reaction
time
as
well
as in
d'.
Hoffman
et
al.
concluded that detecting
or
locating
consistently
mapped targets,
a
task thought
to be
automatic, makes extensive
use of
lim-
ited-capacity
processes.
It
should
be
noted that
in all the
dual-task
experiments
reported
by
Schneider
and
Fisk,
combining
two
single
highly
practised
tasks
invariably
incurred
a
substantial initial dec-
rement,
and
several hours
of
dual-task practice
were
required before performance improved
to the final
levels
reported.
Schneider
and
Fisk
(1982a)
conceded that
the
large
effect
of
dual-task practice
was not
attributable
to
further
improvement
in the
highly practised
single
tasks,
and
concluded
(p.
165),
"We
believe
that subjects were learning
to
integrate
the CM and VM
searches" (the
two
tasks
in
their experiment). Because consistency
of
stimulus-to-response mapping
is not
relevant
to it,
this integration
is
outside
Shiffrin
and
Schneider's
framework
of
automaticity.
It
seems that such integration might well involve
a
restructuring
of the
process.
In
sum, evidence
from
dual-task studies
suggests that
highly
practised
CM
tasks
in
fact
require capacity.
Restructuring
Versus
Automaticity:
An
Analogy
Interpretation
of the
above
studies
of
cat-
egory
search
and
dual-task performance
may
be
clarified
by an
analogy
with
the
previously
mentioned arithmetic task
of finding the sum
of
ten 2s.
Assuming that
addition
and
mul-
tiplication
are
done
at
typical
speeds,
a
mul-
tiplication operation would
produce
the an-
swer
much
faster
than would nine
serial
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

RESTRUCTURING VERSUS
AUTOMATICITY
421
addition
operations.
Also, assuming that
the
greater
the
number
of
operations
in a
task
(other things being comparable)
the
higher
the
probability
of
making
an
error,
the
serial
addition
procedure would have
a
greater
probability
of
error. Moreover, whereas
the
number
of
.2s
to be
summed should
not
greatly
affect
the
calculation time
for
multi-
plication (looking
up 2 X 5 in
one's
memory
would
take
about
the
same amount
of
time
as
looking
up 2 X
10),
it
would have
a
pronounced
effect
on the
calculation time
for
serial addition. Furthermore, because
the ad-
dition
procedure involves many more opera-
tions,
interference
on and
from
a
concurrent
task
would
be
much greater
for the
addition
than
for the
multiplication
procedure.
Despite such
differences
in
speed, accuracy,
dual-task interference,
and
load
dependence
between
the
addition
and the
multiplication
procedures
for finding the
sum, probably
no
one
would
care
to
argue
on
that
basis
that
multiplication
was
capacity
free,
whereas
ad-
dition
was
capacity demanding. Yet, argu-
ments
for a
capacity dichotomy based
on
CM
and VM
performance rest
on no firmer
a
foundation. Schneider
and
Shiffrin
(1977;
Shiflrin
&
Schneider, 1977)
do not
distinguish
between
shifts
in
fundamental modes
of
pro-
cessing
and
restructuring
of
task
procedures.
In
fact,
the two
alternatives
are
left
so
vague
in
their theory that
one
might have interpreted
it
to
imply
restructuring
as a
mechanism
for
achieving capacity-free processing.
In
terms
of
the
arithmetic task, such
an
interpretation
would
imply
that multiplication
is a
mecha-
nism
for
performing parallel additions
in a
capacity-free
manner.
Further
Evidence Against
Capacity-Free
Processing
The
above analysis suggests that perfor-
mance
patterns
that
have
often
been taken
as
evidence
for
automaticity
may in
fact
be due
to
restructuring.
Other
evidence
also
argues
against
the
existence
of
capacity-free process-
ing
in
visual
and
memory scanning. Duncan
(1980)
reported
that
merely noting
the
pres-
ence
of two
simultaneous targets
for
later
response impairs performance
on the
detec-
tion
of
either target.
His
subjects were well
practised
on
constant
sets
of
targets
and
distractors,
and
consequently should have
acquired "automaticity."
Yet
such
"auto-
matic"
detection
responses
interfered with
each
other, despite their being internal.
Similar
results were
in
fact
reported
by
Schneider
and
Shiffrin
(1977,
Experiment
3) and
Shiffrin
and
Schneider
(1977,
Experiment
2).
The
only
robust evidence
for
capacity-free
processing concerns
the
ability
of
subjects
to
reject simultaneous
distractors
(e.g.,
Egeth
et
al.,
1972; Schneider
&
Shiffrin,
1977). This
ability,
however, seems
to be
based
on the
detection
of
visual features that distinguish
the set of
targets
from
the set of
distractors.
As
discussed earlier, whether
or not
distractors
can
be
rejected simultaneously
depends
crit-
ically
on the
existence
of
distinguishing visual
features
(Duncan, 1983).
Beyond
the
early processing stage
of
visual-
feature
detection, there
does
not
seem
to be
any
robust evidence
for any
process
that
is
capacity
free.
Letter encoding
has
often
been
hypothesized
to be a
capacity-free activity
(e.g.,
Keele,
1973;
LaBerge
&
Samuels, 1974;
Posner
&
Rogers,
1978),
as it
takes place
at
a
relatively early stage
of
processing
and is
(at
least
for
college students)
a
highly
practised
task
with
consistent
mapping between stimuli
and
responses.
Nonetheless,
it has
been shown
to be
capacity demanding. Paap
and
Ogden
(1981)
demonstrated that Posner
and
Boies's
(1971)
evidence
for
capacity-free encoding
was
based
on an
inappropriate baseline,
and
that when
an
appropriate baseline
was
used
for
defining
interference,
letter
encoding
did
interfere
with
a
secondary task.
Implications
for
Limited-Capacity Models
Automaticity
is
often
seen
as a
mechanism
whereby
capacity limitations
are
overcome
in
the
acquisition
of a
complex skill.
But it
seems that beyond
the
stage
of
physical-
feature
processing, there
is
little
or no
evi-
dence
for a
process that
does
not
interfere
with
other
processes.
The
lack
of
evidence
for
automaticity, however,
is not in
itself
inconsistent with limited-capacity models.
Automaticity
is
only
a
limiting
case
in the
theoretical
continuum
of
capacity demand;
without automaticity,
a
capacity model would
be
less
tidy,
but
nonetheless viable. However,
apparent variation
in
capacity demand caused
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

422
PATRICIA
W.
CHENG
by
restructuring does pose
a
problem even
for
the
general limited-capacity model.
In
order
for a
capacity model
to be
predictive,
either
the
capacity demanded
by a
task must
remain
invariant across task combinations
or
the
model must
specify
rules governing
the
restructuring
of the
strategies people adopt.
No
such rules have
yet
been
proposed.
As
the
above analysis
of
Shim-in
and
Schneider's results reveals, what appears
to
be
an
-automatized
process
may in
fact
be
due
to
restructuring. This
flexibility
does
not
only
pertain
to
simple tasks such
as
memory
scanning,
but
also
to
more complex
skills.
Several
studies
(Allport,
Antonis,
&
Reynolds,
1972;
Hirst, Spelke, Reaves,
Caharack,
&
Neisser, 1980;
Shaffer,
1975) have shown
that
given
prolonged practice, complex tasks such
as
reading
for
comprehension, piano playing,
or
typing
can be
combined with little
or no
interference
with
a
secondary task such
as
writing
sentences
at
dictation
with
compre-
hension
or
repeating heard speech. Hirst
et
al.
(1980)
argued that improvement
in
their
subjects' performance could
not be due to
the
exertion
of
extra
effort,
the
alternation
of
attention between
the two
tasks,
or the au-
tomatization
of any of the
tasks. They sug-
gested that improvement
was due to the
learning
of new
stimulus
patterns
and new
patterns
of
action—in
other words,
due to
restructuring. Whenever
a
combination
of
components
in a
task causes
a
restructuring
of
the
process,
so
that
two
components
de-
fined
by
the
experimenter
to be the
same
no
longer
remain psychologically
the
same
to the
subject,
the
capacity
demanded
by
those
two
components
cannot
be
expected
to
remain
constant.
Unlike
a
centrally limited capacity model,
which
assumes
a
single homogeneous pool
of
resources,
a
multiple
limited-capacity model
assumes
multiple types
of
resources (e.g.,
Navon
&
Gopher, 1979).
The
latter
type
of
model, however, must also address
the
issue
of
variations
in
capacity demand
due to
restructuring. Multiple-capacities models
can
account
for
performance with greater theo-
retical
flexibility. For
example, unlike
the
central-capacity model,
it can
account
for the
following
pattern
of
performance. Suppose
A,
B, and C are
three tasks that
do not
interfere
with
each other peripherally. Task
A
does
not
interfere
with Task
B, but
Tasks
A
and B
each interferes with Task
C,
despite
C
being
no
more
difficult
than
A or B. A
multiple-capacities
model could explain such
a
pattern
by
assuming that Tasks
A and B
do
not tap the
same types
of
capacities; that
is, the
interference caused
by C on A, and by
C on B, are due to
different
types
of
capacity
limits.
Despite
its
greater theoretical
flexibility,
the
multiple-capacities model
can
explain
such
a
pattern
only
if
Tasks
A and B can
reasonably
be
assumed
to be
drawing
on
different
types
of
resources. Consider
the
coordination
of the two
hands
in
piano play-
ing.
Combining
4
even
notes
to a
measure
with
12
even
notes
to a
measure
is
easy,
as
is
combining
3
even notes
to a
measure with
6 or 12
even notes, suggesting that none
of
the
tasks require much capacity. Because
all
of
these tasks
are
apparently
of the
same
basic nature, they presumably cannot
be
drawing
resources
from
different
types
of
resources.
Yet as
piano
players
know,
com-
bining
three even notes against
four
even
notes
(e.g.,
as in the first of
Chopin's
"Trois
Nouvelles
Etudes")
is
extremely
difficult.
This
difficulty
poses
a
problem
for any
capacity
model.
It is
interesting that
the
combination
is
learnable,
indicating that
the
difficulty
does
not
stem
from
any
physiological
incompati-
bility,
but
rather
from
lack
of a
suitably
structured skill.
A
distinction
has
been drawn between
data
and
resource
limitations
(Norman
&
Bobrow,
1975).
Once
data-limited
conditions
are
ruled
out, limitations
in
performance have been
with
few
exceptions (e.g., Hirst
et
al.,
1980)
assumed
to be due to
resource limitations.
Consonant
with
this line
of
reasoning
is the
view
that
improvement
in
performance
is
due to the
development
of
automatic pro-
cesses.
In
terms
of the
pervasive computer
metaphor,
resource
limitations
are
analogous
to
hardware limitations, whereas restructuring
is
analogous
to
program modification.
Auto-
maticity
has
typically been assumed
to be the
mechanism
for
overcoming hardware limita-
tions. However, improvement
in
performance
can
theoretically stem
from
changes
in
either
the
system's hardware
or its
software.
The
suspect existence
of
automaticity,
along with
the
evident
flexibility of
human
software,
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.

RESTRUCTURING VERSUS
AUTOMATICITY
423
should
force
us to
study
alternative mecha-
nisms
of
skill
acquisition.
References
Allport,
D.
A.,
Antonis,
B., &
Reynolds,
P.
(1972).
On
the
division
of
attention:
A
disproof
of the
single
channel
hypothesis.
Quarterly
Journal
of
Experimental
Psychology,
24,
225-235.
Brand,
J.
(1971).
Classification without identification
in
visual
search.
Quarterly
Journal
of
Experimental
Psy-
chology,
23,
178-186.
Dumais,
S.
T,
&
Shifrrin,
R. M.
(1978, November).
Automatic
detection:
target
accentuation
or
distractor
inhibition?
Paper
presented
at the
meeting
of the
Psychonomic
Society,
San
Antonio.
Duncan,
J.
(1980).
The
locus
of
interference
in the
perception
of
simultaneous stimuli.
Psychological
Re-
view,
87,
272-300.
Duncan,
J.
(1983).
Category
effects
in
visual
search:
A
failure
to
replicate
the
"oh-zero"
phenomenon. Per-
ception
&
Psychophysics,
34,
221-232.
Egeth,
H.,
Jonides,
J.,
&
Wall,
S.
(1972).
Parallel
processing
of
multi-element
displays. Cognitive
Psychology,
3,
674-698.
Fisk,
A.
D.,
&
Schneider,
W.
(1983).
Category
and
word
search:
Generalizing
search
principles
to
complex
pro-
cessing.
Journal
of
Experimental
Psychology:
Learning,
Memory.
&
Cognition,
9,
177-195.
Gleitman,
H.,
&
Jonides,
J.
(1976).
The
cost
of
catego-
rization
in
visual search: Incomplete processing
of
targets
and field
items.
Perception
&
Psychophysics,
20,
281-288.
Hirst,
W.,
Spelke,
E.
S.,
Reaves,
C.
C,
Caharack,
G., &
Neisser,
U.
(1980). Dividing attention without alter-
nation
or
automaticity.
Journal
of
Experimental Psy-
chology:
General.
109,
98-117.
Hoffman,
J.
E.,
Nelson,
B.,
&
Houck,
M. R.
(1983).
The
role
of
attentional
resources
in
automatic
detection.
Cognitive
Psychology,
15,
379-410.
Ingling,
N.
(1972).
Categorization:
A
mechanism
for
rapid
information processing. Journal
of
Experimental
Psychology,
94,
239-243.
Jonides,
J.,
&
Gleitman,
H.
(1976).
The
benefit
of
categorization
in
visual search: Target location without
identification.
Perception
&
Psychophysics,
20,
289-
298.
Karlin,
M.
B.,
&
Bower,
G. H.
(1976).
Semantic category
effects
in
visual word search.
Perception
&
Psycho-
physics,
19,
417-424.
Keele,
S. W.
(1973).
Attention
and
human
performance.
Pacific
Palisades,
CA:
Goodyear.
LaBerge,
D.
(1973). Attention
and the
measurement
of
perceptual learning.
Memory
&
Cognition,
1,
268-276.
LaBerge,
D.,
&
Samuels,
S. J.
(1974).
Toward
a
theory
of
automatic information processing
in
reading. Cog-
nitive
Psychology,
3,
293-323.
Logan,
G. D.
(1978). Attention
in
character-classification
tasks:
Evidence
for the
automaticity
of
component
stages.
Journal
of
Experimental
Psychology:
General,
107,
32-63.
Navon,
D.,
&
Gopher,
D.
(1979).
On the
economy
of the
human
processing
system.
Psychological
Review,
86,
214-255.
Norman,
D.
A.,
&
Bobrow,
D. G.
(1975).
On the
analysis
of
performance operating
characteristics.
Psychological
Review,
83,
508-519.
Paap,
K.
R.,
&
Ogden,
W. C.
(1981).
Letter
encoding
is
an
obligatory
but
capacity-demanding
operation.
Jour-
nal of
Experimental
Psychology:
Human
Perception
and
Performance,
7,
518-527.
Posner,
M. I.
(1970).
On the
relationship between letter
names
and
superordinate
categories.
Quarterly
Journal
of
Experimental
Psychology,
23,
279-287.
Posner,
M.
1.,
&
Boies,
S. J.
(1971).
Components
of
attention.
Psychological
Review,
78,
391-408.
Posner,
M.
I.,
&
Rogers,
M. G. K.
(1978).
Chronometric
analysis
of
abstraction
and
recognition.
In W. K.
Estes
(Ed.),
Handbook
of
learning
and
cognitive processes,
Vol.
5:
Human
information
processing (pp.
143-188).
Hillsdale,
NJ:
Erlbaum.
Posner,
M.
I.,
&
Snyder,
C. R. R.
(1975). Facilitation
and
inhibition
in the
processing
of
signals.
In P. M.
A.
Rabbitt
& S.
Domic
(Eds.),
Attention
and
perfor-
mance
V
(pp.
669-682).
New
York:
Academic
Press.
Rabbitt,
P. M. A.
(1967). Learning
to
ignore irrelevant
information.
American Journal
of
Psychology,
80, 1-
13.
Ryan,
C.
(1983).
Reassessing
the
automaticity-control
distinction:
Item recognition
as a
paradigm
case.
Psy-
cho/ogical
Review,
90,
171-178.
Schneider,
W.,
&
Fisk,
A. D.
(1982a). Degree
of
consistent
training
and the
development
of
automatic
processing.
Perception
&
Psychophysics,
31,
160-168.
Schneider,
W.,
&
Fisk,
A. D.
(1982b).
Concurrent
auto-
matic
and
controlled visual
search:
Can
processing
occur without resource cost? Journal
of
Experimental
Psychology:
Learning,
Memory,
&
Cognition,
8,
261-
278.
Schneider,
W.,
&
Fisk,
A. D.
(1984). Automatic
category
search
and its
transfer. Journal
of
Experimental
Psy-
chology:
Learning,
Memory,
<fe
Cognition,
10,
1-15.
Schneider,
W.,
&
Shiffrin,
R. M.
(1977). Controlled
and
automatic human information processing:
I.
Detection,
search
and
attention.
Psychological
Review,
84,
1-66.
Shaffer,
L. H.
(1975).
Multiple attention
in
continuous
verbal tasks.
In P. M. A.
Rabbitt
& S.
Domic
(Eds.),
Attention
and
performance
V
(pp.
157-167).
New
York:
Academic
Press.
Shiffrin,
R.
M.,
&
Schneider,
W.
(1977).
Controlled
and
automatic human information
processing:
II.
Perceptual
learning,
automatic
attending,
and a
general
theory.
Psychological
Review.
84,
127-190.
Taylor,
D. A.
(1978).
Identification
and
categorization
of
letters
and
digits.
Journal
of
Experimental
Psychology:
Human
Perception
and
Performance,
4,
423-439.
Received
April
16,
1984
Revision
received October
15,
1984
•
This document is copyrighted by the American Psychological Association or one of its allied publishers.
This article is intended solely for the personal use of the individual user and is not to be disseminated broadly.