Journal of Experimental Psychology.Human Learning and MemoryVol. 2, No. 5, 554-565

Reading A Year Later

Paul A. KolersUniversity of Toronto

Two sets of measurements evaluated performance on typographically in-verted text that students had learned to read 13 to IS months earlier. In oneset, speed of reading was compared for pages read for the first and secondtimes. Reread pages were read more quickly, thereby revealing an exceptionaldegree of memory at the pattern-analyzing level. In the second set of mea-surements, the readers classified the pages as to occasion of reading. Com-paring the two sets of measurements showed that different aspects of memorywere measured by the different tests, and they were not well correlated. Per-formance is accounted for in terms of encoding operations directed at thelinguistic patterns, in contrast to the more familiar notion of manipulatingsemantic representations.

When college students read a large num-ber of pages that were in an unfamiliar typog-raphy, they quickly mastered its intricacies,and after 160 pages of practice with the un-familiar typography they read it almost asrapidly as they read normally oriented text.Their ability subsequently to recognize whatthey had read fell away with an improvementin their reading skill (Kolers, 1975a). Thedata were accounted for in terms of proced-ures available to readers for encodinggraphemes; recognition was said to proceedby reinstituting the encoding procedures thatacquired the objects initially. The less skilledthe processing, the more operations requiredfor its encoding, the greater was the likeli-hood of subsequent recognition; conversely,the more skilled the processing, the fewerwere the component operations available toaid recognition.

These views were put forward in contrastto the more popular view of sentence memorythat is derived from transformational linguis-

This work was supported by Grant A76SS fromthe National Research Council of Canada. I thankDavid F. Andrews for advice on statistical trans-formations and analysis of variance, BruceSchneider and C. D. Creelman for discussion andadvice on TSD, and Paul W. Smith for collectingthe data and carrying out their analysis. A briefreport of Experiment 1 has been published else-where (Kolers, 1976).

Inquiries concerning this article may be sent toPaul A. Kolers, Department of Psychology, Uni-versity of Toronto, Toronto, Canada MSS 1A1.

tics. In that view the reader separates thelinguistic base from its superficial embodi-ment and stores only the base. Recentelaborations of this two-stage model ofencoding proposed that the storage is an"abstract meaning" (Bransford & Franks,1972), a reduction of the sentence to itslogical propositional form (Kintsch, 1974),or its decomposition into semantic features(Smith, Shoben, & Rips, 1974), to mentionthree. Only meanings are important in theseviews; and they are derived from their em-bodiments by "extraction" and then are sub-jected either to reduction or to combination.

The contrast with the predominantly se-mantic view of memory is pursued in thepresent paper, where two other sets of mea-surements are described on some of the samereaders, who were retested 13 to 15 monthsafter their first reading. Their performancewas measured both in respect to speed ofreading and to accuracy in classifying whatthey had read. For ease of description themeasurements are treated as two separateexperiments.

EXPERIMENT 1

Method

Over a period of 7 days the students each read7 pages of normally oriented text and 98 pages oftext in inverted typography; a sample of invertedtypography appears in Figure 1. On each testday each reader read 1 page of normal text and14 pages of typographically inverted text, 7 oldones that had been read 13 to IS months earlier,

554

READING A YEAR LATER 555

8S6IU3 so BTwbje ' qTxecf' TUffljeqTSpe' j,p6 SKTITtT btoc633 Tiid

fpgf ujnap ps TUAOjAeq TS wop xfae j^ sbbstsup*

501. <:pe opDecp aseiijs jgwrjrgt 33 soon ss ppe sAe suconirpets rf

IV b33Lf l KeCOdUTfTOlI 366U/3 SO TUni/egTSfS P6C3036 MS 3bSljq UlOSf

OE oni: £Tius TIJ j3u/r]-T33: BnttonuqrjJda' MS Motjc TJJ fps asms xoou/a1

366 pps 33U/G beobje' MSJK ppe 3311/6 B-pteeps* IT^s TU ppe sgiue ponsea

£0i. joud betroqs o^ fiuie' j,pe bexcebpns]- rutotujgfTou we teceTAS

sacp oxqTustA qsrA TS ns/T3XTA tebeprprona sruq j.sqnuqguf MS coiwe

po KUOW fP6 fPTu^B MS MTII suconuper. 6A6U pe^ote M6 euconufei.

na ou nu^gwTjTgi; dtorruq guq M6 3ts wncp ajOMSi; fo xesjrse

M6 JOO)< 36A6t3J

FIGURE 1. Example of geometrically inverted text, Experiment 1.(It begins at the upper left with the word Recognition.)

and 7 new pages taken from the same sources asthe old, described previously (Kolers, 1975a). Theorder of pages was scrambled from reader toreader, with the restriction that the 7 old pagesread each day be selected from the full range of theorder they were read in earlier, thus decouplingskill in reading and content of page read. Thestudents read the normally oriented page at thestart of each session. All reading was aloud intoa tape recorder, under instruction to read forspeed and accuracy. On the first day an experi-menter timed the subjects with a stopwatch; onsubsequent days they timed themselves, writingdown their times on a score sheet, the same pro-cedure followed earlier (Kolers, 197Sa). Only sixof the eight students tested previously were avail-able for this further examination. They weretested three times per week, at least 1 full dayseparating test sessions.

Results

Time scores. Figure 2 plots logarithm ofreading time against logarithm of serial posi-tion of pages. The lines are least-squares fitsof the data points; A' and B' for normallyoriented text, A and B for typographicallyinverted text. The shorter lines, B and B',represent the present reading; A and A'represent performance by these same sixreaders on text read earlier. This figureshows that (a) a linear trend in logarithmsdescribes performance in both studies, (b)the rate of decrease in reading time for in-verted text is less in the present study than

initially (the slope of the line), (c) there issignificant savings from the first test to thesecond (the Y-intercept), and (d) thereaders improved even in reading normallyoriented text aloud.

The average curve for the 98 pages of thepresent experiment, line B in Figure 2, iscomposite; it is made up of scores for pagesread for the first time in the present testand pages reread after 13 to 15 months. Thetwo kinds of pages are separated in Figure 3.The least-squares line is based on all thedata; reading times for pages read for thefirst time are shown by strokes, and readingtimes for pages reread after 13 to 15 monthsare shown by filled circles. (Notice that thevertical scale of Figure 3 is expanded three-fold over the scale of Figure 2. For ease ofreference, remember that zero on the scalesdesignates 1, and successive increments of.3 represent successive doublings on an arith-metic scale; 1.80 is the logarithm of 64, forexample. On the vertical scale .1 designates1.26 min. and .3 and .6 designate 2 min and4 min respectively.) The average values ofFigure 3 are further decomposed in Figure 4,which shows the results for the six readersindividually.

All six of the readers speeded up over thecourse of 98 pages, from about 5.2 min perpage by the slowest (HY) and 3.2 min by

556 PAUL A. KOLERS

PAGE NUMBER ( LOGARITHM)

FIGURE 2. Results for the same six readers in two tests more than a year apart. (The inclinedlines A and B are for reading inverted text, the lines nearly parallel to the abscissa, A' and B',are for normally oriented text. Lines A and A' are for pages read earlier, lines B and B' arefor the pages of this experiment.)

the fastest (OL) to a final time near 1.7min for all readers except EL The differencebetween new and old pages can be seen forthe individual readers: More strokes (pagesread once) appear above each least-squares

line than below it, whereas more filled circles(pages reread) appear below the line thanabove. The obtained difference in readingtime is small, about 5% overall, and so wastested for statistically.

060 -

COUJ

0.40 -

LU

O

0.30 -

UJIT

0.10

0.20 -

PAGE NUMBER ( LOGARITHM )

FIGURE 3. Composite results for six readers in Experiment 1. (Strokes are for pagesread once, filled symbols are for pages reread after 13 to 15 months.)

READING A YEAR LATER 557

It/J

I

.fc a

3 O

ll

( S3ini\lltM 901 ) 3IA1I1 9NIQV38

558 PAUL A. KOLERS

TABLE 1SIGNIFICANCE LEVELS FOR TESTS OF READING SPEED

Regression

Reader

El JO OL HY AH BR Pooled

ParabolicLogarithmic

.021.014

.004

.073.092.063

.030

.080.010.008

.115

.175.0001.001

The logarithmic representation of the dataused in Figures 2 to 4 provides a good fit forthe means of reading time as a function ofpractice; close examination revealed that itdid not eliminate the coupling of mean andvariance that often is found with time scores,however. Several other transformations alsofailed to stabilize the variances in a mannersuitable for analysis of variance. Decomposi-tion of speed of reading into page numberplus the square of page number plus thedifference between old and new pages plusresidual error accommodated both linear andcurvilinear aspects of the data, decoupledmean and variance, and made possible a testof the significance of the difference in timeto read old and new pages. Analysis ofvariance was carried out on the transformedscores for each reader individually. The sig-nificance levels associated with the differencebetween old and new pages, the variable ofmain interest, are displayed in Table 1 in therow labelled parabolic regression. The anal-ysis was carried out on the logarithmic trans-formation also, despite the coupling of meanand variance. The table shows that when thedata were evaluated with the appropriateparabolic transformation, the significance ofthe difference in reading speed for once-readand twice-read pages is p < .03 for 4 of thesix readers, p = .092 for the fifth, and^ = .115 for the sixth; the pooled alphalevel is p < .0001. In the weaker test onlogarithms, five of the readers yielded datawhose alpha level was < .08; the pooledvalue is p < .001. The analysis of variancethus confirmed what the graphs imply:Pages reread after an interval of 13 to 15months tended to be read more rapidly thanpages of similar text read for the first time;and although small, the difference is reliable.

This analysis was designed to maximizethe likelihood of finding a small difference.

Separate functions could have been fitted tothe two distributions, of course, but doing sowould have dispersed the small effect acrossmore degrees of freedom, lessening thechance for its detection. The occurrence ofthe effect rather than its magnitude is thechief datum.

Savings. Two aspects of learning andmemory are revealed by Figures 2 to 4: thegeneral improvement in performance fromthe first occasion to the second, shown bythe smaller Y-intercept of the present data;and the specific enhancement associated withreread pages. Two methods of measuring thepercentage savings of visual skill were ex-plored. Figure 5 illustrates one procedure:Line A is the curve for 160 pages previouslyread, line B is the extrapolation of 98 pagesof present reading, and line C is the averagedvalue for normally oriented text. Percentagesavings is expressed as Area 1 / (Area 1 +Area 2). The second method, illustrated inFigure 6, evaluated reading in the presentstudy (line B) as an extension of readingin the earlier study (line A), extrapolatingfrom the Brunswik ratio for the purpose.If savings were perfect, new pages would beread at a speed and with a rate of learningthat continued the old (top panel). If therewere no savings from the first encounter tothe second, the new pages would start at thesame Y-value, and speed of reading wouldincrease at the same rate, as for the pagesread months earlier (middle panel). In prac-tice, savings was measured as the area intowhich obtained line B divided the space be-tween theoretical lines B' and B" (bottompanel). In this calculation, also, percentagesavings is expressed as Area l/(Area 1 +Area 2). Calculated savings ranged from35% to 70%, average of 59%, for the firstmethod; and from 70% to 85%, average of79%, for the second method. The two meth-

READING A YEAR LATER 559

TABLE 2ASPECTS OF PERFORMANCE

Reader

Feature

Slope, 160 pagesY-intercept, 160 pagesSlope, 98 pagesY-intercept, 98 pagesSavings ASavings BCoefficient old-new

El

-.341.05

-.17.67.35.70

-.029

JO

-.391.06

-.15.53.59.76

-.024

OL

-.411.16

-.14.51.69.83

-.024

HY

-.551.50

-.24.70.61.81

-.021

AH

-.391.12

-.17.52.59.85

-.035

BR

-.491.24

-.16.52.70.77

-.015

Note. All calculations except the coefficient are based on logarithms; the coefficient is derived from the analysis of variance ofthe parabolic regression.

ods make different assumptions and implydifferent notions of pattern recognition. Asneither has special theoretical support, bothare included, especially as they are not per-fectly correlated.

Table 2 summarizes several aspects of per-formance: slopes and Y-intercepts for thetwo readings, percentage savings on the twomethods (A in Figure 5 and B in Figure 6),and the coefficient generated by the analysisof variance discriminating reading speed ofold and new pages (reader AH was best,BR was poorest). Spearman's correlationcoefficient was computed between some ofthe more obvious candidates: rate of acquisi-tion of skill (slope) versus percentage sav-ings, rate of acquisition of skill versus dis-crimination between old and new pages (old-new coefficient), rate of acquisition of skillin the two tests, and slope versus intercept onthe two occasions. In all cases but the last,the correlations ranged from p = —.41 top = +.39, none of them significant. In thereading a year earlier the correlation betweenslope and Y-intercept was perfect; in thecurrent test it tended toward significance,

but, attenuated somewhat by the attenuatedrange, it is only p = —.60.

Discussion

Three aspects of acquisition of skill arerevealed by the curves of Figures 2 and 3;improvement in performance across pages,retention of skill across the months separat-ing the two readings, and specific improve-ment distinguishing performance on pagesread once or twice. The obvious main effectis that extended practice with an unfamiliartypography speeded its reading. The findingof a residual memory expressed in the su-perior reading speed of reread pages is themore impressive fact needing an accounting,for, although small, the effect is reliable andits significance lies with its occurrence, notits size.

FIGURE 5. One calculation of savings ofvisual skill, Experiment 1.

PAGES ( Log )

FIGURE 6. A second calculation of savingsof visual skill, Experiment 1.

560 PAUL A. KOLERS

It might be suggested that the superiorreading speed can be explained by thereaders' recognizing the semantic content ofthe reread pages. Two arguments againstthis proposal can be advanced. First, the pro-posal would have to suppose that readingthe first few lines of a page enabled thereader to recover from memory enough ofthe semantic content of the remaining 20-oddlines accurately enough to be able to use thatmemory to facilitate the reading. (Unless thememory were accurate, of course, no facilita-tion could be expected.) Such an assumptionabout literal accuracy of semantic retrievalwould be difficult to justify on the basis ofcurrent knowledge, which assumes that se-mantic information is receded, not storedliterally. Second, in an experiment designedto test just this issue it was shown that thesemantic content was a less useful aid toreading typographically transformed sen-tences than analysis of the words as typo-graphical objects was (Kolers, 1975b).These two reasons induce skepticism thatmemory of the semantic content of the oldpages explains their being read faster thanthe new pages. Whatever memory is presentseems to be for the pattern-analyzing opera-tions that produce words, not for abstractedsemantic contents. The operations, however,are difficult to identify (Table 2).

One possibility is that the various .mea-sures described belong not to a single com-posite pattern-recognizing skill but are dif-fent aspects of performance that are notnecessarily correlated. This possibility isdiscussed below.

EXPERIMENT 2

In this experiment another aspect of liter-acy was assessed, the reader's ability to clas-sify passages according to various criteria.

MethodThe experiment was divided into three occa-

sions. The first occasion was the reading of 160pages of inverted typography 13 to 15 monthsearlier; the second was the reading described inthe first part of this paper, extending over seventest days; the third occasion was an eighth testday.

On the third occasion the readers classifiedseven sets of 24 pages each, according to whenthey had been read. Each set was made up of

7 pages read only on the first occasion, 7 pagesread only on the second occasion, 7 pages read0n both the first and second occasions, and 3 pagesread only on the third occasion, making 49, 49,49 and 21 = 168 pages in total. The convention torefer to time of reading will be new (third occa-sion only), recent (second occasion only), old(first occasion only), and both (first and secondoccasions). The various sets of pages were readin the same order as on the second occasion; thatis, the pages read on the first test day of the secondoccasion were sorted first, pages read on the sec-ond test day were sorted second, and so on. Thetask was to sort each page into the proper one offour piles describing the occasion of reading, new,old, recent, or both. (In practice, descriptivephrases were used for the piles, such as "Readonly a year ago," "Never read before," and thelike.)

Analysis. New pages were included in the sortingto facilitate ^analysis according to the methods ofsignal-detection theory (TSD). The data ob-tained provided three estimates of each value ofd', corresponding to the three criterion valuesassumed to separate the response categories. Fora unidimensional TSD analysis to be appropriatefor these data, the three values should not differsignificantly; that is, the four frequency judgmentsshould be on a single axis. Analysis of variancewas carried out on the three estimates of eachreader's cut-offs for the distance between recentand new, both and new, and old and new. Theresulting F(2, 10) =6.59, p < .05, implied that thethree estimates of distance did not lie on a singleaxis, so that the assumptions required for a uni-dimensional treatment of these data in a TSD modelwere not satisfied. Efforts to use a higher-dimen-sional TSD model were blocked in turn by the un-availability of any decision rule for positioningcut-off points, or for establishing the angular rela-tions of their axes. The analytical power of theTSD model was, therefore, not available for thisstudy, so the data were evaluated in terms ofXs tests.

Results

The readers maintained nearly constant hitrates across the seven sets of pages; the sort-

TABLE 3COMPOSITE RESULTS ON CLASSIFYING PASSAGES

Response

Stimulus

BothRecentOldNew

Sum

Both

.49

.30

.060

.85

Recent

.29

.55

.03

.04

.91

Old

.17

.08

.56

.301.11

New

.05

.07

.35

.661.13

READING A YEAR LATER 561

ings were therefore collapsed across sets. Theproportion of responses to the four kinds ofpassage are shown in Table 3, averagedacross the six readers. About 95% of bothand 93% of recent passages were recognizedas having been read before, and about halfof them were categorized exactly as to occa-sion. A smaller 65% of old passages wererecognized as read before, while about one-third of them were thought to be new; almostexactly the reverse results were found fornew passages recognized as new or thoughtto be old. There was very little confusion ofboth or recent passages with old or new ones.Inspection reveals that the table does notcontain equal cell values; moreover, biasmarks the responses, as shown by theunequal column sums. As the overall tableis significant, it was partitioned into threeother tables, each testing a particular con-trast. The signficance levels are p — .05when x2(l) = 3.84, and p = .01 when X

2(l)= 6.64.

The contrast old versus new answers thequestion whether the readers distinguishedbetween passages read earlier and passagesnever read before (Table 4). The test isbetween the distribution of responses intothe old pile and into the three other piles

TABLE 4x2 TEST FOR OLD-NEW DISCRIMINATION

TABLE 5x2 TEST FOR RECENT-OLD DISCRIMINATION

Reader and stimulus

EIOldNew

JOOldNew

OLOldNew

HYOldNew

AHOldNew

BROldNew

CompositeOldNew

Re;

Old

201

306

318

3312

226

305

16638

aponse

Other

2920

1915

1813

169

2715

1916

12888

X2

7.46

5.04

2.82

.30

1.02

6.80

23.39

Reader and stimulus

EIRecentOld

JORecentOld

OLRecentOld

HYRecentOld

AHRecentOld

BRRecentOld

CompositeRecentOld

Res

Recent

271

190

322

300

266

270

1619

iponae

Other

2248

3049

1747

1949

2343

2249

133285

X2

31.25

21.15

37.88

40.40

16.75

34.56

188.67

when the passage was old, compared to thedistribution of responses into the old pile andinto the three other piles when the passagewas new. Three readers, AH, HY, and OLyielded x2s whose p values are greater than.05; for JO, .01 < p <.05; and for the re-maining two readers, p < .01. Combiningfrequencies into a composite table yieldedX

2 ( l ) = 23.39, p< .01. The ability to dis-tinguish passages read before was absent inthe performance of three readers, but presentto varying degrees in the other three. Notice,however, that EI and AH discriminated inthe wrong direction; EI regarded old pagesas not old with a frequency greater thanchance.

A second test was between passages readonly on the first occasion and passages readonly on the second, old versus recent, Table5. Readers HY and OL discriminated wellthe passages read recently, reader JO re-garded most passages read recently as havingbeen read at some other time, and the threeother readers equivocated as to the time atwhich they thought they had read a passage,when in fact they had read it recently. Allof the readers recognized that passages readonly on the first occasion had not been readrecently.

562 PAUL A. KOLERS

TABLE 6x1 TEST FOR BOTH-RECENT DISCRIMINATION

Response

Reader and stimulus

ElBothRecent

JOBothRecent

OLBothRecent

HYBothRecent

AHBothRecent

BRBothRecent

CompositeBothRecent

Both

188

2525

1210

3615

2115

3315

14588

Other

3141

2424

3739

1334

2834

1634

149206

x'

4.24

.04

.06

16.35

1.10

11.80

22.29

The third evaluation was between pas-sages read recently and passages read bothrecently and on the first occasion (Table 6).Readers AH, JO, and OL experienced dif-ficulty with the discrimination, for differentreasons; readers HY and BR made the dis-tinctions more readily; and reader El recog-nized recent passages but confused passagesin the both category.

The contrasts can be summarized as fol-lows : The first was a broad distinction be-tween old and new, the second was a distinc-tion with respect to recency, and the thirdwas a distinction with respect to frequency.The first and third comparisons, on disjointsets of data, are clearly independent; the sec-ond, although not independent of the othertwo, yields additional information. The indi-vidual results of the x2 tests are shown inTable 7 for these contrasts; discriminations

in the wrong direction are noted. It can beseen that readers made the various judg-ments with differential success. Reader HYdid well on the recency and frequency dis-criminations but judged poorly on whethera passage had been read before; reader AHdid poorly on the familiarity and frequencydistinctions but did well on the recency dis-tinction ; reader JO made the familiarityjudgment reliably but confused the recencyjudgments and could not make the frequencyjudgment. El scored p < .05 on all threetests, but two results were in the wrongdirection. Only one reader, BR, scored p <.01 on all three tests. Comparing performanceon classification with reading speed is useful ;AH, who did not distinguish old passagesfrom new at the semantic level (Table 4),had the highest coefficient discriminatingbetween old and new pages read aloud(Table 2) ; BR, who classified accurately,distinguished least well between old andnew in reading speed. The lack of associationin criteria can be seen in the independent fre-quency and familiarity tests; it is confirmedby the recency test. Overall, the correlationsbetween criteria are low.

GENERAL DISCUSSION

The fact that the typography of a textwas recognized some weeks after reading(Kolers & Ostry, 1974) and that instructionand bias constrain what the reader reports(Craik & Tulving, 1975; Graesser & Hand-ler, 1975) question the claim that only se-mantic content of text is preserved in mem-ory. The present finding of savings in speedof reading after an interval of more than ayear must lay it to rest. A proposal put for-ward previously (Kolers, 1975a), regardingthe encoding of sentences in terms of pat-tern-analyzing operations directed at the

TABLE 7SUMMARY OF x2 TESTS

Reader

Test El JO OL HY AH BR

Familiarity (old-new)Recency (old-recent)Frequency (recent-both)

7.46»31.25

4.24»

5.0321.15"

.04

2.8237.88

.06"

.3040.4016.35

1.02«16.751.10"

6.8034.5611.80

• Discrimination in the wrong direction.

READING A YEAR LATER 563

lexical objects provides an alternative andsimpler account of the facts. It contrasts withthe emphasis on semantic encoding (Ander-son, 1974; Bransford & Johnson, 1973;Kintsch, 1974) ; its implications are exploredbelow after consideration of some alterna-tives.

What Is Remembered I1

Figures 2 to 4 describe a general increasein speed of reading as a function of practice,and an extra facilitation for reread pages.Some of the speeding up might be attributedto variables extrinsic to the main encoding—practice at reading aloud, smoother motoradjustments to the physical organization ofthe task, an easier and more relaxed attitudetoward it, and the like. Speeded reading ofnormally oriented text might be accountedfor in these terms (Figure 2), but improve-ment in reading inverted text was far moreextensive. The extrinsic-factor interpretationis of limited value, therefore. Another viewemphasizes semantic memory as facilitatingreading, especially of the twice-read pages,Pages read for the first time late on occasiontwo were also read more quickly than pagesread earlier, however (Figure 3), but thisimprovement in speed cannot be due to re-trieval of semantic content, for one cannotretrieve what one has not previously encoded.Thus, emphasizing semantic content requiresthat two different principles be invoked toexplain performance—one, a "learning"principle to account for improvement inreading pages not read before; second, a"semantic memory" principle for rereadpages. Performance can be accommodatedwith a single alternative principle, however.The reader is thought to learn specific pat-tern-analyzing operations directed at thegraphemes and the lexical objects they em-body, such as for coping with direction ofscan and orientation of characters, as well asother analyses whose output or product isthe spoken word; recurrence of the stimuluspattern permits a speeded or facilitated oper-ation of these procedures. Skill in processingthe graphemic patterns accounts both for the

speeded performance as a function of prac-tice (Figure 2) and for the special advant-age to reread pages (Figures 3 and 4). Itmay also account for the speeded recognitionthat underlies the word-superiority effect(McClelland, 1976; Reicher, 1969).

Modeling Memory

Theories of memory are often theories ofmatching, in which a store of information inmind is searched and scanned until featuresof the present object are aligned with storedfeatures (templates, lists, or other represen-tations of prior experience). The data inTables 2 and 7 suggest that memory per-formances, like some perceptual perfor-mances, are not monolithic matchings butare compositions of subprocesses or subskillsthat are not all alike. Some of the variabil-ity of performance may be due to differentstrategies that readers followed, or to otheraspects of circumstance and self-instruction;some of it, however, may be attributable todifference in the structure of the encodingoperations underlying a task, for some recog-nitions are achieved with continuous com-pensations, whereas others are encoded dis-cretely (Keele, 1973; Kolers & Perkins,1975). Speed of reading and judgments offamiliarity, frequency, and recency mayactually be based on different kinds ofoperations. If the operations change with achange in skill, moreover, as this and thepreceding research show (Kolers, 1975a),it becomes quite unclear what interpretationto give to terms such as "memory trace" orwhat is measured by efforts to measure it.

A further point along this line is the im-plication contained in the dissociation of rec-^ognition based on skilled pattern analysis,as in the coefficient describing the differencein speed of reading old and new pages(Table 2), and recognition based on consci-ous judgment of familiarity (Table 7). Ap-parently, knowledge can be expressed asskilled performance without a correspondingrecognition in conscious judgment. Such dis-sociations between automatized function andverbal judgments are well known in the

564 PAUL A. KOLERS

clinical literature (Rapaport, 1951) inamnesias, hysterias, and other states. Inthe present case, ordinary life experienceproduced an effect usually noted only aftera traumatic encounter. This dissociation, in-teresting from the clinical point of view, isadditional evidence for the notion that dif-ferent aspects of an encoding may be withoutcontact, making even less intelligible thehypothesis of some common memory traceas representing an event.

Skill '-

In one account of recognition, it wasassumed that the sense organs were in con-tact with reality and provided pictures orsounds or touches that language then en-coded (Neisser, 1967; Dick, 1974). Actually,skill that the reader brings to his encounterwith text constrains any processing of itthat is carried out.

For example, in one study it was shownthat as,,skill in encoding a symbol systemincreased, the statements embodied in itwere recognized less well; that is, increasedskill led to poorer recognition (Kolers,1975a). In another study, it was shown thatlack of skill in encoding the symbol systemalso was associated with poor performance(Kolers, 1974). The ability to recognizewritten sentences, therefore, turns out tobe a U-shaped function of familiarity with—and skill in manipulating—the symbols em-bodying the sentence. The present workshows that in addition to these relations be-tween competence with a symbol system andmemory for the sentences it embodies, recog-nition of sentences is based on a number ofdifferent constituents that are not necessarilycorrelated. In the present view, memoryis not a place in the mind where things arestored, but a way of responding to eventswith skills acquired in previous encounterswith like circumstances; recognition is nota matter of matching but a process of trans-ferring skills across occasions. The modelthat assumes that encoding begins with lightfrom the stimulus reaching the eye is errone-ous; it begins with the skills available forthe process, which are directed at the many

aspects of the stimulus on which they havebeen trained. The more skills available, otherthings remaining the same, the more differentways in which the encoding can proceed.

REFERENCES

Anderson, J. R. Retrieval of prepositional informa-tion from long-term memory. Cognitive Psy-chology, 1974, 6, 451-474.

Bransford, J. D., & Franks, J. J. The abstractionof linguistic ideas. A review. Cognition, 1972,1, 211-249.

Bransford, J. D., & Johnson, M. K. Considera-tion of some problems of comprehension. In W.G, Chase (Ed.), Visual information processing.New York: Academic Press, 1973.

Craik, F. I. M., & Tulving, E. Depth of processingand the retention of words in episodic memory.Journal of Experimental Psychology: General,197S, 104, 268-294.

Dick, A. O. Iconic memory and its relation toperceptual processing and other memory mecha-nisms. Perception & Psychophysics, 1974, 16,57S-S96.

Graesser, A., II, & Handler, G. Recognition mem-ory for the meaning and surface structure ofsentences. Journal of Experimental Psychology:Human Learning and Memory, 1975, 1, 238-248.

Keele, S. W. Attention and human performance.Pacific Palisades, Cal.: Goodyear PublishingCompany, 1973.

Kintsch, W. The representation of meaning inmemory. Hillsdak, NJ.: Erlbaum, 1974.

Kolers, P. A. Remembering trivia. Language andSpeech, 1974, 17, 324-336.

Kolers, P. A. Memorial consequences of autom-atized encoding. Journal of Experimental Psy-chology: Human Learning and Memory, 1975,1, 689-701. (a)

Kolers, P. A. Specificity of operations in sen-tence recognition. Cognitive Psychology, 1975,7, 289-306. (b)

Kolers, P. A. Pattern analyzing memory. Science,1976,191, 1280-1281.

Kolers, P. A., & Ostry, D. J. Time course of lossof information regarding pattern analyzing opera-tions. Journal of Verbal Learning and VerbalBehavior, 1974,13, 599-612.

Kolers, P. A., & Perkins, D. N. Spatial and ordinalcomponents of form perception and literacy.Cognitive Psychology, 1975, 7, 228-267.

McClelland, J. L. Preliminary letter identificationin the perception of words and nonwords. Jour-nal of Experimental Psychology: Human Per-ception and Performance, 1976, 2, 80-91.

READING A YEAR LATER 565

Neisser, U. Cognitive psychology. New York: nal of Experimental Psychology, 1969, 81, 275-Appleton-Century-Crofts, 1967. 280.

Rapaport, D. Organisation and pathology of Smith, E E., Shoben, E. J. & Rips, LJ Struc-ii i * M 17 1 /- i u- TT • •* -n ture and process in semantic memory: A featuralthought. New York: Columbia University Press, ^^ for ^^^ decisions. Psychological Re-19S1- www, 1974, W, 214-241.

Reicher, G. M. Perceptual recognition as a func- (Received December 16, 197S; revisiontion of meaningfulness of verbal material. Jour- received March 30, 1976)