Journal of Experimental Psychology:
Human Learning and Memory
197S, Vol. 1, No. 6, 689-701
Memorial Consequences of Automatized Encoding
Paul A. Kolers
University of Toronto, Canada
The first experiment tracked the acquisition of skilled reading as college
students read as many as 160 pages of geometrically inverted text. The
logarithm of reading time decreased linearly as a function of the logarithm
of amount of practice, and performance on inverted text approached performance on normal text remarkably rapidly. The second experiment assessed the consequence for memory of skill at reading. Students unpracticed
at reading inverted text remembered for lengthy intervals the inverted sentences they read; when students acquired skill with the typography, their
memory for inverted sentences was poorer. The results are interpreted in
terms that emphasize an operational basis to memory—pattern-analyzing
procedures rather than conscious contents. This view is contrasted with
three other accounts of recognition.
For all strange things seme . . . hard of entertainment at their first arrivall, till theie be acquainted:
but after acquaintance theie be verie familiar and
easy to entreat. . . . Familiaritie and acquaintance
will cause facilitie, both in matter and in words.
(Mulcaster, cited in Baugh, 1957, p. 264)
A standard view of perception and memory emphasizes concepts, ideas, thoughts,
and related cognitive contents as the coin of
mind. Semantically based theories of perception and memory, usually proposing hierarchical organizations of information, underlie such work. In this view, the mind is
full of knowledge of objects and things, full
of concepts, ideas, and images; and it works
by sorting, comparing, and coding them. An
alternate view holds that mind is procedure,
operation, and activity; and that what it
knows is what it knows how to do. Knowledge is expressed by the way the mind has
learned to sort, organize, and analyze, rather
than by the results of those processes (Bartlett, 1958; Schneirla, 1948; Werner, 1948).
In three previous papers the latter view was
put forward and illustrated in the context of
experiments on recognition memory (KolThis work was supported by Grant A76SS from
the National Research Council of Canada and by
Grant 382 from the Ontario Mental Health Foundation. I thank Linda Vettor, Stephen Israelstam,
and Paul W. Smith, who collected and analyzed
the data.
Requests for reprints should be sent to Paul A.
Kolers, Department of Psychology, University of
Toronto, Toronto, Canada MSS 1A1.
ers, 1973, 1974; Kolers & Ostry, 1974).
Some other features of this view are discussed in the present paper.
The experiments just cited required college
students to read normally oriented sentences
and sentences in inverted typography, some
only once, others twice. The inverted sentences were recognized on the second reading
more frequently, more accurately, and over a
longer interval than the normally oriented sentences. The difficulty of the inverted sentences and the longer reading time they
required might be thought to account for
the outcome—hence the importance of a
control experiment to evaluate the roles of
time-on-task and task difficulty (Kolers,
1974). The control experiment found that
subjects not only recognized geometrically
transformed sentences more readily than
normal ones but they also distinguished between two kinds of transformation, both of
which were hard and took long to read.
Thus, time and difficulty, intuition to the
contrary notwithstanding, did not successfully account for the results. An explanation in terms of pattern-analyzing operations
directed at the graphemes was put forward
as an alternative account.
This previous work was directed against
the view that recognition is a two-stage process in which pattern analysis of the surface
representation precedes extraction of an embedded semantic core. It argued for the
689
690
PAUL A. KOLERS
i
OS'I
OS'I
05'0
09'0
OE'O
06'0
09'0
OE'O
0
OS'O
OE'O
0
0
OS'I
OS'I
06'0
09'0
OE'O
0
06'C
09'0
OE'O
0
I i
OS'I
i
|f i
OS'I
i
I i
06'0
OS'I
OS'I
0
9oi) awn 9Niav3a
i
I i
09'0
i
I I
OE'O
5
i
CONSEQUENCES OF AUTOMATIZED ENCODING
view that remembering, and especially recognizing, can go forward as the reinstitution
at the later time of the pattern-analyzing
operations used initially to encode the pattern. On this view, the more extended and
elaborate the original pattern analysis, the
better the memory for the encoded material.
This is not a matter of more receding due
to greater study; rather it is a matter of
more encoding made necessary by an absence
of sophisticated pattern-analyzing skills. It
follows from this notion that increasing a
person's skill in analyzing the relevant patterns might lessen his memory for the contents they embody. One purpose of the
present work was to check this conclusion.
A second purpose was to study the acquisition of pattern-analyzing skill. It was
shown previously that the kind of pattern
studied—geometrically inverted text—is
mastered with some rapidity, but the training was not carried very far (Kolers, Boyer,
& Rosenthal, 1965). The -present work required students to read a large number of
pages in a single geometric rotation and
measured the acquisition of skill at the task.
Skill was attained at a surprisingly rapid
rate, which was the main finding of the
first experiment.
EXPERIMENT 1
The main concern was to trace the acquisition function as otherwise literate readers
mastered the skill of reading geometrically
inverted text. This text challenges the
visual pattern-recognizing system. The text
embodies familiar language, thereby avoiding the problem that absence of understanding creates for study of a written language,
but the graphemes are unfamiliar in appearance. Faced with such a challenge, how
does the visual system respond? The question was studied with two sets of measurements.
Method
In the first study, 12 undergraduates, 6 male
and 6 female, read 8 pages of inverted text on each
691
of 3 days, a total of 24 pages each. In the second
study, 8 male undergraduates each read a total
of 160 pages of inverted text over a period of
about 2 mo. These 8 subjects read 5 or 6 pages
per day for the first several days, then gradually
increased their daily load up to a maximum of
10 pages per day, but never reading for more than
1 hr at a time with rests of 1 min between pages.
As with the preceding group, these readers were
tested only 3 times per week, so that at least 1
rest day intervened between 2 test days. The text
was transformed (inverted) by a 180° rotation of
each line around its horizontal axis, that is, each
line turned top to bottom. It is illustrated as
Sample I in Kolers (1973, 1974) and in Kolers and
Perkins (1975).
Before reading the first page of inverted text,
and every 30 pages thereafter, the subjects read
a page of normally oriented text. All reading
was aloud, into a tape recorder; the subjects measured and recorded their own reading time, using
a stopwatch for the measurements. The text came
from popular psychological sources, such as G. A.
Miller's Psychology of Communication, R. Brown's
Words and Things, and the collections from Scientific American called Contemporary Psychology
and Perception, but the greater part came from
Miller (1962). There were about 300 words to
the page. The order of pages was scrambled
from reader to reader. The subjects were right
dominant, as assessed by sighting tests for vision
and questions regarding handedness and footedness, and their reading skills were restricted to
languages using the roman alphabet. Only students
who could read a page of normal English aloud
in less than 2 min were tested. They were paid
for their participation in the experiment.
Results
The results on 24 pages from the 12 subjects in the first group are very similar to
the results on the first 24 pages of the 160,
so. only the longer duration group will be
described here. Figure 1 plots the logarithm
of reading time against the logarithm of
pages of text for each of the 8 readers individually; their average is shown as "mean
curves." For the individual subjects, each
plotted point represents only a single observation, the time to read one page of
inverted text; hence, the points are somewhat "noisy." Plotted points on the mean
curves represent averages of eight observations. (For ease of reference, remember
that zero on the vertical scale designates 1
FIGURE 1. Reading time as a function of practice. (Inclined lines are for inverted text, the
flatter lines for normal text. Data for eight subjects individually [one observation per plotted
point] and their average [mean curves].) Variations in size of the plotted points are distortions
in printing.
PAUL A. KOLERS
692
TABLE 1
LEAST SQUARES FITS FOR READING TIME DATA
.Subject
F-Intercept
HY
SU
BR
HA
OL
31.65
18.32
17.53
14.85
14.51
13.28
11.56
11.24
15.75
AH
B
Group average
Exponent
-.548
-.504
-.486
-.436
-.414
-.393
-.387
-.337
-.438
PSS
.957
.884
.937
.932
.920
.939
.932
.921
.986
Note. PSS is the proportion sum of squares.
min, .3 designates 2 min, .6 designates 4
min, and each additional increment of .3 log
unit represents another doubling of the arithmetic values. The same is true on the horizontal scale, the unit properly changed, so
that log value of 2.1 marks 128 pages of
text.) The trend of the data is clearly linear
on logarithmic coordinates, following the
form Reading Time = kX~a, where X is
page number, —a is the slope of the line,
and k accommodates differences between
subjects or transformations. The inclined
lines are the slopes for reading inverted
text; the lower lines, usually parallel to the
abscissa, are the slopes for normal text.
One subject (AH) improved in reading
normal text aloud from 1.32 to 1.02 min per
page, another (El) grew somewhat worse
with practice, but most were nearly constant.
All subjects improved markedly in reading
inverted text aloud.
At first, the subjects took between 32.36
(HY) and 10.68 (El) min to read a page of
inverted text compared with 1.46 and 1.31
min for normal text. The average results
for the eight subjects show a gain from
15.75 min per page for inverted text on
Page 1 to 4.07 min on Page 24 and 1.74 min
on Page 160, an overall gain in speed of
about 9 times. The time for normal text
was near constant.
Least squares lines were calculated for
each subject separately and for the eight
subjects as a group. The arithmetic value
of the calculated F-intercept (its antilog),
the slope of the line (exponent of the power
function), and the proportion of the total
sum of squares accommodated by the regres-
sion (proportion sum of squares, PSS) are
shown in Table 1. The linear regression of
the logarithms accommodated the data well
for the subjects individually, and is almost
perfect for the group average. The major
source of perturbation was the first four
pages of text (see Figure 1).
A perfect correlation can be seen in Table
1 between F-intercept and slope, readers
who initially took longer improving at the
greatest rate. Despite almost a threefold
difference between fastest and slowest subjects in initial time to read inverted text
(F-intercepts), all subjects read inverted
text in nearly the same amount of time after
160 pages of practice (see Figure 1).
Discussion
At the beginning of the experiment the
subjects struggled to decode inverted text,
the best of them taking more than 8 times as
long and the poorest more than 22 times as
long to read inverted text as normal text.
Literate college students learned the inverted
typography with readiness; after only 160
pages of practice both the initially fastest
and initially slowest subjects read inverted
text in only 1.3 times the time required to
read normal text. A priori, we could expect that all of the students would learn at
the same rate but differ in ability, or we
could expect that they would learn at different rates but differ in initial ability. The
first expectation would predict equal slopes
with different intercepts, that is, parallel
lines. The data obtained were more nearly
consistent with the second expectation, a
number of different processes converging on
a limiting value: The subjects approached
their asymptotic speed after exposure to
about the same number of pages of text, regardless of their initial level of skill. Why
the rate of learning should bear so impressive a relation to initial difficulty is not
immediately clear; nor is it clear what subjects learned or learned to do.
The proposal developed in detail below
emphasizes the acquisition and use of analytical skills directed at the visual patterns,
that is, the learning of pattern analyzing
operations. This postulated gain in pattern
CONSEQUENCES OF AUTOMATIZED ENCODING
analyzing skill has consequences for other
aspects of performance as well, as shown by
the second experiment.
EXPERIMENT 2
In previous experiments that demonstrated better memory for inverted sentences
than for normal ones, subjects' ability to
read the transformed sentences was poorer
than their ability to read normal sentences.
The present experiment assessed the significance to memory of this difference in reading skill. In this test the difference was
eliminated because the subjects were first
practiced extensively at reading the unfamiliar typography, as described in Experiment 1. The underlying theory holds "that
recognizing occurs in terms of the encoding
operations that acquire sentences, not necessarily in terms of the semantic content of
the sentences themselves; in terms of acquisition procedures rather than acquired contents (Kolers, 1975b). On this notion
protracted practice at reading inverted typography should lessen the difference for
recognition memory of sentences in inverted
and normal typography.
Method
The procedures of this experiment were similar to
those reported previously (Kolers, 1973; Kolers &
Ostry, 1974) and used the same materials. (An error
in computation in the 1973 report is corrected in
Kolers, 1975a). Students read two decks of sentences, 60 sentences in a read deck, then 120 sentences in a recognise deck. Thirty of the sentences in the read deck were in normal orientation
(N), the other 30 were inverted (I). These 60
sentences reappeared in- the recognize deck, but
half of them were then in the alternate typography.
Thus the recognize deck was made up of four kinds
of sentences: IS in normal orientation that had
been in normal orientation in the read deck also,
15 in inverted orientation in both decks, 15 in
normal orientation that had been in inverted orientation in the read deck, and 15 in the reverse order.
These. four pairings are designated NN (normal
orientation in both decks), II (inverted in both
decks), NI (normal in the read deck, inverted in
the recognize deck), and IN (inverted in the read
deck, normal in the recognize deck).
In addition to these four kinds of sentences, the
recognize deck contained 60 new sentences, 30 in
normal orientation and 30 inverted. These are
designated WN and WI, where W represents new.
All of the sentences came from Miller (1962) ; the
693
old sentences of the recognize deck were the oddnumbered ones from a passage and the new sentences were the even-numbered ones. Old and new
sentences were therefore alike in context, theme,
style, and related features. None of them had appeared in the passages read for Experiment 1.
The subject's task was to read the sentences
aloud as rapidly and as accurately as he could.
Immediately after reading each sentence of the
recognize deck he placed it into one of three
piles: (a) new sentence; (b) old sentence, same
form; (c) old sentence, different form. These
categories are for the judgments whether the individual sentences appeared for the first time in
the recognize deck (WN, WI), whether their
appearance in the recognize deck was in the same
typography as in the read deck (NN, II), or in
the alternate typography (NI, IN). Only sentences
read correctly on both encounters were included
in the results; speed of reading sentences was timed
by an experimenter with a stopwatch. The study
was conducted twice—on the 12 subjects described
in Experiment 1 who had first read 24 pages of
inverted typography, and on the 8 subjects who
had first read 160 pages of inverted typography.
Analysis. The methods of signal detection theory were used to assess hits and false alarms.
These were tallied separately for each subject and
converted to values of d', using the tables for
this purpose (Swets, 1964). The occasional instances of 0 and 1 in the proportions were treated
as .02 and .98, respectively, to avoid the tails of
the d' distribution. Four levels of description
were evaluated.
The first level was an overall estimate of the
discriminability of old and new sentences. Old
sentences placed in either of the old sentence piles
were taken as hits; new sentences in the old piles
were false alarms. The corresponding value is
called d' (all).
A second level examined recognition of the sentences as semantic objects, but with respect to
their typography. All of the NN sentences put
into either old pile were hits for NN, and the
new sentences in normal orientation (WN) put
there were false alarms. All of the II sentences
in both old piles were hits, and the new sentences
in inverted orientation (WI) put there were false
alarms. Similarly, for IN in both old piles designated as hits, WN sentences were false alarms,
and for NI sentences designated as hits, WI sentences were false alarms. This measure is called
d' (sem); it evaluates memory for the content
of a sentence without regard to its typographic
features, but it measures with respect to them.
The third level requires for a hit the exactly
correct placement of a sentence. Thus, only the
NN and II sentences that were put into the same
form category, and the IN and NI sentences put into
the different form category were hits. The corresponding false alarms were WN and WI sentences in the same form category, and WN and
WI sentences in the different form category. This
measure is called d' (new) : It describes recogni-
694
PAUL A. KOLERS
TABLE 2
PROPORTION OF SENTENCES THAT SUBJECTS IN THREE EXPERIMENTS
ALLOCATED TO EACH RESPONSE CATEGORY
Kind of sentence*
Experiment and response
cateogry
NN
H
NI
IN
WN
WI
24 pages
Same form
Different form
New
.54
.14
.32
.77
.21
.01
.08
.55
.37
.21
.75
.05
02
02
96
.03
.05
.92
.49
.28
.55
.26
.20
.17
.41
.42
.19
.56
.25
.05
.04
.91
.05
.02
.93
.49
.18
.33
.77
.20
.03
.14
.54
.32
.08
.89
.04
.03
.10
.87
.07
.10
.83
160 pages
Same form
Different form
New
Kolers (197Sa)
Same form
Different form
New
.23
a
NN = normal orientation in both decksu II = inverted orientation in both decks; NI = normal in read deck, inverted ir
recognize deck ; IN = inverted in read deck, normal in recognize deck; WN = new in recognize deck and in normal orientation
WI = new in recognize deck and inverted.
tion of the content of a sentence contingent on
recognition of its typography.
The fourth level of analysis examined only
those sentences that the subjects recognized as
old. The NN and II sentences in the same form
pile were hits, and the IN and NI sentences in
those piles were the corresponding false alarms.
Only two values of d' are available for this computation, which is called d' (old) : I. designates
sentences initially in inverted orientation (II and
IN) and N. is for sentences initially in normal
orientation (NN and NI). This measure describes
recognition of sentences primarily on the basis of
their pictorial or typographic features, for it is
conditionalized on correct recognition of the sentences as statements.
The four analyses span the range from a diffuse
discrimination of old from new, d' (all) ; through
discrimination of the sentences as statements seen
before, d' (sem) ; through memory for a sentence
in respect to its typography, d' (new) ; to memory
largely in terms only of typographic features,
d' (old). These last three values of d' were subjected to separate analyses of variance, typically
a Sentence Pairs X Subjects repeated measures
analysis for each d'. For d' (sem) and d' (new),
four pairs of old sentences entered into the analysis; for d' (old), only two pairs entered.
Results
Table 2 displays the averaged results for
three groups of subjects on the sorting
tests: the 12 subjects who were practiced
with 24 pages of text, the 8 subjects practiced with 160 pages, and, for comparison,
the 12 subjects of Experiment 1 from Kolers
(197Sa), who had only a few lines of practice
at reading inverted typography. The table
shows the kinds of sentence sorted and the
categories into which they were sorted.
Two features are especially notable in the
data for the subj ects who were first practiced
with 24 pages of text. One is the proportion of old sentences improperly categorized
as new. About one third of those in normal
orientation in the read deck were so classified when they reappeared in the recognize
deck (.32 and .37 for NN and NI, respectively), but only a small percentage of those
that had appeared inverted in the read deck
were so classified (.01 and .05 for II and
IN, respectively). This difference shows
that sentences initially in inverted orientation in the read deck were more likely than
the others to be correctly recognized as old.
The second notable feature is the difference
between NN and II in the same form category and between NI and IN in the different form category: .54 and .77 in the first
comparison, .55 and .75 in the second. The
differences show that sentences initially in
inverted orientation were the more likely to
be placed in the correct old pile.
Similar results were obtained from the 12
subjects in Kolers (1975a) who classified
sentences after only a few lines of practice
at reading inverted typography. Their val-
CONSEQUENCES OF AUTOMATIZED ENCODING
695
TABLE 3
VALUES OF d' IN Two EXPERIMENTS
Kind of sentence"
Experiment and
measure
NN
II
NI
IN
SEDM'
24 pages
d' (sem)
d' (new)
d' (old)
2.23
1.99
1.81°
3.49
2.76
2.24d
1.91
1.81
3.49
2.67
.16
.24
.22
160 pages
d' (sem)
d'1 (new)
d (old)
2.18
1.64
1.23°
2.S9
1.79
.97d
1.92
1.54
2.38
1.86
.27
.26
.18
a
NN = normal orientation in both decks; II = inverted orientation in both decks; NI ' normal in read deck, inverted in
recognize
deck; IN = inverted in read deck, normal in recognise deck.
b
Standard
error of the difference between means for rows.
0
Values for sentences initially in normal orientation (NN and NI).
d
Values for sentences initially in inverted orientation (II and IN).
ues for same form hits (.49 and .77) and
different form hits (.54 and .89), and the
values for misses (old sentences classified
as new, .33 and .32, .03 and .04) all came
close to those obtained from subjects practiced with 24 pages of text. But among
subjects practiced with 160 pages, the outcome was quite different. Among those
subjects the differences between NN and II
in the same form category (.49 and .55) and
between NI and IN in the different form
category (.41 and .56) were substantially
less than for the other two groups of subjects. In addition, the proportion of sentences initially in inverted orientation (II
and IN) that were improperly classified as
new increased more than fivefold to .20 and
.25 respectively, whereas the proportion of
NN and NI sentences erroneously classified
as new was about the same as in the other
groups, .28 and .42, respectively. This apparent loss of differentiation between normally oriented and inverted sentences was
confirmed by analysis of variance on the d'
values derived from the data that were averaged for Table 2; these are summarized in
Table 3.
The three rows of each section of Table 3
show the three kinds of d' that were evaluated, each with a separate analysis of variance : d' (sem), d' (new), and d' (old). These
are, respectively, measures of a semantic
component independent of typography, a semantic component contingent on recognition
of typography, and a pictorial typographic
component fairly free of the semantic component. The overall discrimination of old
from new sentences by the subjects who
were first practiced with 24 pages of inverted text ranged from .95 hits and .02
false alarms, d' (all) = 3.73, to .67 hits and .07
false alarms, tf (all) = 1.89 (average d' (all)
= 2.60 [not shown]). Analysis of variance of
the various levels displayed in Table 3
yielded outcomes as follows. Analysis of
tf (sem) yielded F(3, 33) = 55.74, MS* =
.148, for the difference between sentence
pairs; analysis of d' (new) yielded F(3, 33)
= 8.06, MSe = .338, for the difference between pairs, and analysis of d' (old) yielded
F(l, 11) =3.76, MSe= .294, for the difference between I. and N.. The F values for
d' (sem) and d' (new) are both significant
below the .01 level, but the F value for
d' (old) exceeds even .05. Thus, subjects
were well able to tell apart sentences they
had read before from those they had not
(old versus new), and they recognized more
sentences that had initially appeared in inverted typography than sentences initially in
normal typography. However, they did not
memorize the typography in any iconic way.
The corresponding analysis for subjects
practiced with 160 pages of text yielded different results. The overall discrimination
of old from new ranged from .86 hits and
.02 false alarms, d' (all) = 3.16, to .77 hits
and .39 false alarms, d' (all) = 1.03 (aver-
696
PAUL A. KOLERS
age d' (all) =2.20 [not shown]). However, analysis of variance of d' (sem)
yielded F(3, 21) =2.16, MSe = .301, for
the difference between sentence pairs at the
semantic level, analysis of d' (new) yielded
F(3, 21) < 1.00 for the difference contingent on recognition of typography, and F(l,
7) = 2.02, MSe = .134 for the pictorial discrimination evaluated as d' (old). In none
of these three tests did the significance level
reach .05. Thus, subjects were still able to
distinguish sentences they had read before
from those they had not (old versus new),
but they recognized fewer old sentences
whose typography was initially inverted.
The increase in reading skill consequent on
extended practice at reading inverted typography sharply reduced the influence of typographic inversion on performance, and, as a
consequence, reduced memory for inverted
sentences.
GENERAL DISCUSSION
Two kinds of data have been obtained.
One set describes the speed with which a
complex pattern-analyzing skill was acquired. The second set describes the effect
of that skill on recognizing sentences. To
review the data, skill at reading a difficult
typographic transformation was acquired at
a regular rate, expressed by the logarithm
of time to read plotted as a function of the
logarithm of the quantity of practice; this is
similar to previous results (Kolers et al.,
1965; Kolers & Perkins, 1975). In addition to altering the speed of reading, practice also altered the accuracy of recognition
of what had been read: Extended practice
at reading geometrically transformed text
reduced its advantage to memory. The
points requiring discussion are the characteristics of the perceptual learning, and the
operations that mediated the loss of memorability. These will be described and put into
a theoretical framework, and this theory will
be contrasted with some others that are relevant to this task.
Pattern Analytic Aspect of Literacy
The marks read were endowed with meaning in normal orientation, but were disguised
by the typographic transformation; the
learning required was to see through or
compensate for the disguise. Particular pattern-analyzing operations were brought to
bear upon the problem; two components involved are mastering rotation, and mastering
a change in letter order (Kolers & Perkins,
1975). Other operations are also present.
Understanding the role of pattern analysis
in the present task can be aided by considering the effect of extended practice in reading
inverted text, and by comparing performance
after 24 pages of practice and after 160
pages. Consider two hypotheses to account
for the differences.
One hypothesis would hold that all of the
cognitive activities present in initial reading
were also present in practiced reading, but
were carried out more quickly and neatly.
On this view the change in reading time was
due primarily to a change in the speed with
which the component encoding operations
were carried out. A second hypothesis would
hold that the practiced reading was different
in its constituents from the unpracticed in
that some components present initially had
fallen away or been discarded, and others
were modified. On this second view, the
change in reading time indexes a change in
the composition of the constituent operations. The first hypothesis is discussed
later; the following section expands and
illustrates this second point of view.
Three Kinds of Performance
When first faced with the difficult task of
reading inverted text, the otherwise literate
reader engages in visual exploration, problem solving, geometric analysis, and other
tactics to penetrate the disguise that typographic inversion imposes on text. The
successful solution of the problem requires
not a single insight regarding the geometric
relation of inverted to normal text but the
exercise of skills that encode the samples.
(If the problem created by transformed text
were of the classic kind, verbal realization
of the nature of the transformation would
be sufficient for successful reading. But
even telling the subject of that relation rarely
aids performance.) With repeated tries at
CONSEQUENCES OF AUTOMATIZED ENCODING
the typography, the reader rapidly masters
its intricacies (Figure 1). The procedures
and routines that exploration and problem
solving first revealed as the way to analyze
the texts become with practice an institutionalized, routinized way of coping with them.
The problem solving gives way gradually to
mastery, when performance is described as
skilled.
Skilled procedures, in contrast to the unskilled, have little problem solving or exploration about them; they usually run off
smoothly, require only a little monitoring or
control, and typically are resistant to accurate, introspective description. The underlying procedures have lost the excesses, redundancies, and irrelevancies that marked
the early practice. The skilled, automatized
procedure is therefore not the unskilled run
off more rapidly; it is constituted differently,
more efficiently and compactly. On this
view, extended practice enables the patternrecognition system to select a few relevant
properties from a complex sample, and with
those analyses it establishes the corresponding perceptual representation.
The difference between exploration or
problem solving and routinized procedures
can be brought out in the explanations comparing performance after 24 pages of practice and after 160 pages. After 24 pages
of practice, reading time was reduced to
about one fourth of its initial duration, but
it was still about three times longer than
reading time for normal text. After this
limited practice, according to the present
theory, the reader's performance was still
largely dominated by the same activities to
read inverted text as initially; the analytic
operations were more efficient, but the
greater number of them were still present as
at the beginning. After 160 pages of practice, however, greater changes had occurred.
Performance was more skilled; it was not
only more rapid, it was also organized differently, for many procedures that were
irrelevant to the main task of encoding had
been dropped. After 24 pages of practice,
on this analysis, the reader read inverted
and normal text in substantially different
ways, but after 160 pages of practice, he
697
read the two samples in ways that were more
similar. The regularity of the curves of
Figure 1 suggests that the acquisition of
skill went forward by a slow exchange of
newer for older routines.
When procedures become highly skilled
and automatized, they can be applied in or
transferred to different contexts; we speak
then of knowledge. Consideration of such
recombinations of skilled performance will
not be undertaken here, however, except to
note the following. The data used in support of some arguments for immediacy and
directness of visual perception (e.g., Gibson,
1966; Michotte, 1963) may be better accounted for as due to well articulated, automatized procedures created by familiarity of
what were once visual puzzles and problems
that required analysis. In light of the speed
with which skill was attained in the present
task—160 pages of practice at inverted text
brought readers close to the speed attained
after thousands of pages of practice with
normal text—one may wonder how little
practice a child needs to acquire considerable
skill at visual recognition.
On this analysis, unskilled reading of inverted text is carried out by means different
in principle from those marking skilled reading. The difference is this: Extensive analysis of a stimulus is required for its recognition at early stages of practice; with repeated encounters the pattern-recognition
system builds up a catalog of procedures
used in the encoding, so that after some sufficient practice, a subset of features of the
sample is enough to actualize a recognitive
experience of the text. Skill, therefore, not
only enhances and organizes the necessary
feature analysis but also liberalizes feature
selection, allowing many different subsets of
features to trigger the recognitive reaction
required for rapid reading. Depth of understanding can, as a consequence, be associated
with less analysis of the stimulus display
rather than more.
In sum, the view here is that as readers
learn to read inverted text, they become
more selective in sampling features of the
target for analysis, that is, they ignore irrelevancies and redundancies in preference
698
PAUL A. KOLERS
to those features that distinguish the target
from other samples, and they improve the
procedures and articulate them more neatly.
Hence, skilled performance differs from unskilled performance in the composition and
organization of components. The transition
from unskilled to skilled performance and
the further transition to knowledge are accompanied by a change in the subjective
representation of the stimulus.
tunity for subsequent recognition; more skill
implies less analysis and a corresponding
diminution in discrimination. In fact, stereotyping is a common consequence and a recurrent danger in perceiving familiar stimuli
just because they are perceived in more categorial terms rather than as individuals. In
Mulcaster's words, "familiaritie and acquaintance will cause facilitie," but analysis
is directed at strangeness and disparity.
Effect on Memory
Contrast with Alternatives
It has been proposed that recognition is
the reinstitution of actions and analyses
undertaken earlier. A person recognizes a
target by recognizing his analysis of it. The
less skilled his processing of a sample, the
less able the reader is to select the contingent from the constitutive and the significant
from the irrelevant. Consequently, the more
analyses he carries out, the more detail he
encodes. The more encoding the target requires for its initial perception, the more
opportunity the target presents for subsequent recognition. Typography usually is irrelevant to the denotative aspect of text, but
if a reader is not skilled at ignoring its irrelevancies to attend primarily to the message it embodies, he will encode many of its
superficial features, thereby enhancing opportunities for subsequent recognition. With
the onset of routinized procedures for encoding typography, the words and sentences
tend to be processed more in terms of their
categorial features and less as individuals.
We can see how this applies in the present
case. After 24 pages of practice the readers
were still selecting, organizing, and articulating the pattern-analyzing operations, and
they encoded more than the semantic features of the sentences. As a consequence,
memory for inverted sentences was superior
to memory for sentences in normal typography. After 160 pages of practice the pattern-analyzing operations had become more
automatized and fewer analyses were present to distinguish inverted texts from normal
texts in memory. Consequently, performance on the two typographies approached
equivalence. Less skill leads, to more encoding of irrelevancies, hence greater oppor-
The explanation offered here is different
from most currently popular accounts of recognition memory. Brief allusion to three of
them will highlight aspects of the present
proposal. One alternative is the serial analysis or stage-wise decomposition sort of theory. In those schemes various features are
extracted from a stimulus through serial
stages of analysis, with the graphemic the
earliest stage and the semantic the last (see
Craik & Lockhart, 1972, for reviews). The
notion in the present work, in contrast, is
that stimuli are not all put through the same
program of analysis. Quite to the contrary,
the more familiar a stimulus, the fewer the
encoding operations directed at it, the fewer
the features selected for analysis (other
things remaining equal), and the more readily it is seen in terms of practiced skills.
There are at least two ways that selection
could be exercised: (a) by contextual or
other background information extrinsic to
the target operating in support of a difficult
analysis (Kolers, 1974) or (b) by the action of well-practiced and automatized feature analyzing procedures "tuned" to particular displays. In the first case, semantically dominated information would be present at the beginning to aid or even to guide
the pattern analysis. This sort of support
characterizes even machine recognition of
characters in which context limits possibilities and samples are tested in a hypothesischecking procedure. (In some sentence,
context may suggest that the next word is
likely to be a personal pronoun in the nominative case. Relatively little feature analysis
is then required of the reader to tell he from
she, for example.) In the second case, the
CONSEQUENCES OF AUTOMATIZED ENCODING
699
visual system would become sensitized or the perceiver is less than wholly skilled with
"tuned" to certain patterns and would per- a particular stimulus, production may be
form its analysis on a fairly large unit, such present in many aspects of perception; it
as a syllable, word, or even phrase, rather may • not be needed for all aspects (Kolers,
than on a letter or syllable as in the early 1966). For some kinds of reading, substages of practice. Frequently occurring vocalized speech might be the production
words such as and and the, to take two neu- mediating the perception, but this silent
tral examples, are unlikely to need much speech often retards rather than enhances
analysis by a skilled reader. Such practiced, reading (Edfeldt, 1960), and little sign of
automatized analyzers would of course go subvocal speech is found m highly skilled
far toward achieving "direct" interpretation reading (Smirnov, 1973). Moreover, it is
of the visual symbol (Kolers, 1970), "whole not at all clear what the production would
word" recognition that is sometimes de- be for the visual pattern-analyzing composcribed as "direct access to memory" nent'of reading, particularly at the advanced
(Meyer, Schvaneveldt, & Ruddy, 1974).
level; A weaker notion of the productive
An additional consideration is the effect aspect of perception can be formulated in
of expertise. A particular brush stroke, terms of motor commands or implicit speech
curvature of a line, or seam of a casting; acts; this view also assumes that recognireflectance, shade, or crackling of a varnish; tion is not automatized but is always created
choice of a word or its syntactic embed- anew. One consequence of skill may be to
ding—let the expert tell one work from an- free perception from continued dependence
other, one maker from another, often at a on production. Perceiving in this view
glance. These are among the most trivial, would still be a generative process, but one
superficial features of a work: the varnish based on only a limited sampling of the
in a Vermeer compared with the painting stimulus array. The notion of skill can play
itself (Coremans, 1949), the typography of a useful role in the enterprise of formulating
a Shakespeare folio compared with the sen- a more adequate generative theory of pertences it embodies (Friedman & Friedman, ception than analysis by synthesis.
1957). However, their perception can be of
The third point concerns automatization,
the utmost significance to interpretation. In- and the discussion of the hypothesis that was
deed, despite their superficial nature they earlier postponed. The present view is in
can even be the message. Thus, these fea- marked contrast with that developed retures can hardly be considered irrelevant, cently by LaBerge and Samuels (1974). In
nor can the investment in their study be said their theory, the major difference between
to be less or less worthwhile than the invest- unskilled and skilled (automatized) procement in the study of the semantic content of dures is the presence of attention. The unthe material. The ability of semantic infor- skilled action is marked by routines executed
mation to affect encoding of the graphemes, undej- conscious control; the automatized act
the aggregation of graphemic elements into is composed of the same routines, but carlarger units, and the fact that different fea- ried Out more rapidly and without attention.
tures of an object may vary in significance On the present view, in contrast, the wellfor a person, according to the task he is per- practiced skill is thought to be composed difforming and his background with respect to ferently, not only by loss of redundant and
it, all seem to be events that are incompati- irrele|vant procedures present in the unble with or missed by a notion of rigidly skilled act, but also by a greater reliance on
programmed or preprogrammed stages of stored information. Skilled processing on
this view is thus not made up of the same
analysis.
They seem to be missed, too, by an in- actions as unskilled processing, executed
sistence that production always accompany more rapidly and less attentively; skilled
perception, as in analysis-by-synthesis (Mac- processing is a more selective and directed
Kay, 1951; Stevens & Halle, 1967). When analysis.
700
PAUL A. KOLERS
Consciousness and Recognition
The analytical operations that yield recognition are not usually aspects of conscious
judgment; rather, they are actions of the
visual system engaged in solving a task,
whose outcome is sometimes a conscious experience. On the present view, to resort to
conscious analysis of features to encode or
retrieve an event indexes a failure of skilled
processing; typically, one resorts to conscious supplementation as an aid to problem
solving only when the well-inculcated, automatized skills fail to operate properly.
Analysis at the conscious level is not the ordinary but the extraordinary procedure.
In the theory proposed here, conscious selection and evaluation are not required for
memory. There is no deposit of information in some memory bank or store that is
matched or referred to by a later encounter.
Rather, the nervous system in its encounters
with stimuli acquires and uses skills in encoding them; it does so by engaging in a
"dialogue" with the stimuli of such a kind
that repeated encounters modify the encoding operations. Memory then is not traces
that are matched to a stimulus (or vice
versa) but procedures, operations, ways of
encoding the stimulus that are used, and
these change as a function of encounters
with the stimulus.
Substance and Procedure
Relating recognition to skilled analysis
rather than to conscious content lets us see
again the profit in ascertaining an influence
of procedure on performances often assigned
to propositions or statements.
Knowing
how to encode a sentence can be sufficient to
accommodate the knowing that the statement or proposition expresses. It might
prove equally useful to analyze other areas
of knowing that are frequently described in
terms of semantic categories, semantic relations, semantic networks, and the like.
Events taken as instances of semantic skills
and attributed to a semantic memory might
yield to an analysis that sought the transfer
across tasks or contexts of encoding operations—procedural knowledge acquired in one
context and applied in another. The com-
ponents of the skills and the limits of their
transfer then become the subject matter of
inquiry into cognitive processes.
In much contemporary theorizing, consciousness is given an exceptional status, and
theories of memory are often only theories
of memory for conscious contents of mind,
or verbal theories of semantic relations.
Whereas consciousness clearly is characterized by unique properties and conscious experiences can be recalled, consciousness itself
is only a part of cognition. The theory
proposed here supposes that what emerges
in consciousness is the result of many other
cognitive operations, and these, rather than
consciousness, are the basis of recognition.
REFERENCES
Bartlett, F. Thinking. London: Allen & Unwin,
1958.
Baugh, A. C. A history of the English language.
(2nd ed.). New York: Appleton-Century-Crofts,
1957.
Coremans, P. B. Van Meegeren's faked Vermeers
and De Hooghs: A scientific examination. London: Cassell, 1949.
Craik, F. I. M., & Lockhart, R. S. Levels of processing: A framework for memory research.
Journal of Verbal Learning and Verbal Behavior, 1972, 11, 671-684.
Edfeldt, A. W. Silent speech and silent reading.
Chicago: University of Chicago Press, 1960.
Friedman, W. F., & Friedman, E. S. The Shakespearean ciphers examined. Cambridge: Cambridge University Press, 1957.
Gibson, J. J. The senses considered as perceptual
systems. Boston: Houghton Mifflin, 1966.
Kolers, P. A. Reading and talking bilingually.
American Journal of Psychology, 1966, 79, 357376.
Kolers, P. A. Three stages of reading. In H.
Levin & J. Williams (Eds.), Basic studies on
reading. New York: Basic Books, 1970.
Kolers, P. A. Remembering operations. Memory &• Cognition, 1973, 1, 347-355.
Kolers, P. A. Two kinds of recognition. Canadian
Journal of Psychology, 1974, 28, 51-61.
Kolers, P. A. Addendum to "Remembering operations" Memory &• Cognition, 1975, 3, 29-30. (a)
Kolers, P. A. Specificity of operations in sentence
recognition. Cognitive Psychology, 1975, 7, 289306. (b)
Kolers, P. A., Boyer, A. C., & Rosenthal, K. F.
Protracted practice on decoding spatially transformed text. (In Quarterly Progress Report No.
77). Cambridge, Mass.: Massachusetts Institute
of Technology, Research Laboratory of Electronics, 1965.
CONSEQUENCES OF AUTOMATIZED ENCODING
Kolers, P. A., & Ostry, D. Time course of loss of
information regarding pattern analyzing operations. Journal of Verbal Learning and Verbal
Behavior, 1974, 13, 599-612.
Kolers, P. A., & Perkins, D. N. Spatial and
ordinal components of form perception and literacy. Cognitive Psychology, 1975, 7, 228-267.
LaBerge, D., & Samuels, S. J. Toward a theory
of automatic information processing in reading.
Cognitive Psychology, 1974, 6, 293-323.
MacKay, D. M. Mindlike behaviour in artifacts.
British Journal for the Philosophy of Science,
1951, 2, 105-121.
Meyer, D. E., Schvaneveldt, R. W., & Ruddy,
M. G. Functions of graphemic and phonemic
codes in visual word-recognition. Memory &
Cognition, 1974, 2, 309-321.
Michotte, A. The perception of causality." London: Methuen, 1963.
701
Millar, G. A. Psychology: The science of mental
life. New York: Harper & Row, 1962.
Schrieirla, T. C. Psychology, comparative. Encyqlopedia Britannica. Chicago: Encyclopedia
BrStannica, 1948.
Smirlnov, A. A. Problems of the psychology of
memory. New York: Plenum, 1973.
Stevens, K. N., & Halle, M. Remarks on analysis
by] synthesis and distinctive features. In W.
Wathen-Dunn (Ed.), Models for the perception
ofispeech and visual form. Cambridge, Mass.:
MJI.T. Press, 1967.
Swe^s, J. A. (Ed.), Signal detection and recognition by human observers. New York: Wiley,
1964.
Werner, H. Comparative psychology of mental
development (Rev. ed.). Chicago: Follett, 1948.
'. (Received November 21, 1974; revision
received February 17, 1975)