reading rotated words - university of...

Journal of Experimental Psychology:Human Perception and Performance1985, Vol. 11, No- 4, 490-508

Copyright 1985 by the American Psychological Association, Inc.0096-l523/85/$00.75

Reading Rotated Words

Asher Koriat and Joel NormanUniversity of Haifa, Haifa, Israel

Hebrew-speaking students performed lexical decisions on Hebrew letter stringsthat appeared at different orientations. Response times evidenced a stronginteraction between string length and orientation. At angular deviations of lessthan 60° from the upright, neither orientation nor string length had any effect,suggesting that words were directly, and probably holistically, recognized. Theresults for the 60° deviation, while also exhibiting no effects of word length,yielded slower response times, suggesting a holistic rectification process. Fordeviations between 60° and 120°, the effects of disorientation increased sharplywith increasing string length, suggesting piecemeal processing that may be due tothe utilization of reading units smaller than the whole word or to piecemealrectification. In this region, stimulus disorientation appears to impair wordrecognition by disrupting transgraphemic information rather than by interferingwith letter identification. Extreme disorientations, 120° or more, exhibited nofurther impairment with increased disorientation, and all evidenced strong andsimilar length effects, suggesting letter-by-letter reading. The implications of theresults for the reading of normal and transformed text are discussed.

This study combined a mental rotationtask with a reading task in an attempt toaddress two separate but related issues. First,is the rectification of disoriented stimuli car-ried out holistically or in a piecemeal fashion?Second, are words read letter by letter or aswhole units? In a previous study (Koriat &Norman, 1984) we asked subjects to performa lexical decision task on five-letter rotatedstrings. Response times increased systemati-cally with angular deviation from the upright,suggesting that words might be mentally ro-tated in a manner similar to that of otherstimuli. In the present study we varied thenumber of letters in a string and examinedthe combined effects of degree of rotationand string length on performance.

The issue of holism in mental rotation hasrecently received a great deal of attention (seeRobertson & Palmer, 1983; Shepard &Cooper, 1982). The question is whether amisoriented figure is mentally rotated as an

We gratefully acknowledge the assistance of Edna Ben-Zvi, Ofra Ziv, Abraham Papo, and Zvia Kaplan. We alsothank William Epstein, Raymond Klein, David Navon,and two anonymous reviewers for their helpful commentson an earlier draft of this article.

Requests for reprints should be sent to Asher Koriat,Department of Psychology, University of Haifa, Haifa31999, Israel.

integral whole with all its parts rotating si-multaneously around the same point or isrotated segment by segment. If mental rota-tion is piecemeal and serial, the effect ofdisorientation might be expected to increasewith the complexity of the figure or thenumber of "parts" it contains. Studies of thisissue have yielded somewhat equivocal results,the majority leaning toward the nonholisticposition (Cooper, 1975; Cooper & Podgorny,1976; Hochberg & Gellman, 1977; Pylyshyn,1979; Yuile & Steiger, 1982). More recently,however, Robertson and Palmer (1983) ob-tained data consistent with the hypothesis ofholistic mental rotation. They used large(global) letters constructed from spatial ar-rangements of small (local) letters. Rate ofmental rotation was not affected by the num-ber of levels that were rotated, suggesting aholistic rotation process in which rotation ofglobal and local levels occurs together.

In their recent reassessment of the holistic-nonholistic issue in mental rotation, Shepardand Cooper (1982) offered a view that appearsto accommodate the divergent results. Mentalrotation is assumed to be holistic when thetask involves only one stimulus that is to becompared to a memory representation andwhen this stimulus constitutes a well-learned,integrated figure. On the other hand, mentalrotation might be piecemeal when the task

490

READING ROTATED WORDS 491

calls for the matching of two stimuli, and/orwhen the stimulus lacks mental integration.

When the recognition of misoriented wordsis examined within this framework, the fol-lowing might be proposed. If familiar wordsconstitute well-integrated wholes, they shouldbe mentally transformed as integral units.Thus, the magnitude of the rotation effectshould be the same for words of differentlength. On the other hand, if they are repre-sented as composites of distinct units, theymight exhibit piecemeal rotation, with "rateof mental rotation" decreasing with numberof letters in the string.

This leads directly to the second issue, thatof holism in word recognition (see e.g., Hen-derson, 1980). This issue has a far longerhistory than the issue of holism in mentalrotation (e.g., Cattell, 1886; Huey, 1908) andhas attracted a vast amount of research effort(see e.g., Henderson, 1982, for a review). Thequestion is whether word recognition is basedon the visual configuration of the word as awhole or is mediated by processes of letteridentification. One experimental approach tothis question has involved the use of nonstan-dard typographic variations, particularly thoseassumed to disrupt the familiar visual config-uration of the word without destroying theidentity or the arrangement of its constituentletters (e.g., Adams, 1979). Such orthographicvariations have been found to impair perfor-mance (e.g., Coltheart & Freeman, 1974),suggesting that word recognition normallycapitalizes on supraletter features (but thismay also be due to the impairment of letteridentification, see e.g., Paap, Newsome, &Noel, 1984). On the other hand, these varia-tions did not reduce the size of the wordsuperiority effect (McClelland, 1976), consis-tent with letter-mediated models of reading(e.g., McClelland & Rumelhart, 1981).

The rotation of word stimuli utilized inthe present study is another form of visualdistortion that has the advantage of allowingexamination of continuously graded degreesof transformation. In the previous study wefound degree of disorientation to exert strongand systematic effects on word recognition(Koriat & Norman, 1984). This effect is noteasy to interpret, particularly because disori-entation does not seem to affect the latencyof identifying single letters (e.g., Corballis &Nagourney, 1978). One possibility is that

word recognition is based on whole-wordunits and that disorientation disrupts thetotal configuration of the word. A secondpossibility is that word recognition does reston letter identities and that when letters areembedded in a letter string, disorientationdoes impair their identities or the perceptionof their arrangement. The study of the com-bined effects of word rotation and stringlength on word recognition might shed somelight on this issue. We know that for uprightwords lexical decision latency is generallyindifferent to word length (Frederiksen &Kroll, 1976), suggesting the possibility ofwhole-word reading. If word recognition isfound to be indifferent to string lengththroughout the entire range of orientations,this would further substantiate the idea thatwords are generally processed as integral units.

There is some indication, however, thatvisual transformations that apparently do notdisrupt letter identities do yield increasedlength effects relative to standard format(Bruder, 1978: Navon, 1978). We may, there-fore, suspect that the effect of string lengthshould increase with increased disorientation,suggesting that some piecemeal sequentialprocessing occurs. If such is the case, thedetailed examination of the manner in whichlength effects emerge with increased disori-entation might provide some insight into thequestion of holism in word recognition. Thus,in our previous study small deviations fromthe upright appeared to have little effects onword recognition. Perhaps, word length effectscan be found only for those disorientationslarge enough to disrupt direct activation ofwhole-word codes.

The following four experiments all involvedlexical decisions for letter strings presentedat different angular disorientations. In thefirst experiment the length of the letter stringswas varied systematically, with the aim ofdetermining how the effects of disorientationmight interact with those of string length inaffecting performance.

Experiment 1

Method

Subjects. Twenty-four volunteers participated in thestudy. They were all students at the University of Haifawhose primary language was Hebrew.

Stimulus materials. Ninety-six Hebrew words werecompiled, 24 of each of the 2-, 3-, 4-, and 5-letter lengths.

492 ASHER KORIAT AND JOEL NORMAN

They contained only consonantal letters and had onlyone pronunciation when unpointed (see Koriat, 1984).Twelve words of each length were transformed intononwords by replacing one letter with another consonantalletter. The order of the letter strings was the same for allsubjects. This order was random except for the constraintthat each block of eight successive strings included oneword and one nonword of each length.

Apparatus and procedure. The experiment was con-trolled by a PDF 11/34 minicomputer, and the stimuliwere presented on a VT-11 CRT Graphic Display Unit.The subjects sat at a viewing distance of 80 cm, withtheir heads resting on a chin-and-head rest, preventinghead rotations. They were asked to classify letter stringsas words or nonwords as quickly as they could withoutmaking errors. They responded by pressing one keylabeled word with their right index finger or a second keylabeled nonword with their left index linger. They wereinformed that the strings would appear in one of sixorientations. The session began with 32 practice trials,followed by two repetitions of the 96 experimental stringsin the same order. There was a 2-min break between thetwo presentations.

All strings appeared unpointed (see Koriat, 1984) atthe center of the screen and were rotated about theircenter. The letter strings subtended 0.8 cm vertically(when upright) and ranged between 2 cm {two-letterstrings) and 5 cm (five-letter strings) horizontally. Each

string remained in view until the subjects responded, andit was replaced by the subsequent stimulus after 500 ms.String orientations were preprogrammed so that for eachsubject and repetition two words and two nonwords ofeach length appeared at each orientation, and for eachrepetition each string appeared equally often at eachorientation across all subjects. The orientation of eachletter string varied between repetitions.

Results

Response times in excess of 5,000 ms orless than 250 ms (0.8%) were eliminated fromthe analyses. The two repetitions yielded verysimilar results and were therefore combined.Mean latencies for correct responses areshown in Figure 1, where the effects of angulardisorientation are seen to vary systematicallywith string length. These results were analyzedin terms of the four angular deviations fromupright, 0°, 60°, 120°, and 180°, (i.e., col-lapsing over equivalent deviations, 60° +300°, 120° + 240°). A three-way analysis ofvariance (ANOVA), Angular Deviation X String

WORDS NONWORDS/

. 2 LETTERS

- 3 LETTERS

. 4 LETTERS

- 5 LETTERS

«? n o 5 n i f ~ 2 4 t f » t f tf tf * 7 i a f f iiff Mtf io^

ORIENTATION

Figure 1. Response time (in milliseconds) for words and nonwords as a function of orientation with stringlength as a parameter (Experiment 1).


2000

leoo

WORDS

- ISO"

NONWORDS

STRING LENGTHFigure 2. Response time (in milliseconds) for words and nonwords as a function of string length fordifferent angular deviations from the upright (Experiment 1).

Length X Lexicality (words vs. nonwords) onresponse times yielded significant main effectsfor angular deviation, F(3, 69) = 86.05, p <.0001, and for string length, F(3, 69) = 86.43,p < .0001. Response times were longer fornonwords than for words, F(l, 23) = 75.43,p < .0001, and nonwords exhibited strongereffects of deviation, F(3, 69) = 3.69, p < .02,as well as stronger effects of length, F(3, 69) =3.81, p < .02, compared with words.

The interaction between deviation andstring length was highly significant, F(9,207) = 28.06, p < .0001. The nature of thisinteraction may be best seen in Figure 2. Forboth words and nonwords the effect of stringlength increases substantially with increaseddeviation from the upright. Normally orientedwords (0°) show no effect of word length(F < 1), consistent with previous findings,whereas for the 120° and 180° deviationsresponse times increased by about 700 msfrom two-letter to five-letter words, F(3, 69) =63.45, p < .0001.

Interestingly, the results for the 60° devia-tion for words exhibit the same independenceof length as found for the 0° orientation, butat the same time yield longer response timesby about 85 ms. A two-way ANOVA, 0° versus60° X String Length yielded a significant ef-fect for the 0° versus 60° comparison, F(l,23) = I1.16,p< .0005, but neither the effectsof length nor the interaction was significant.This appears not to be the case for nonwords,where the 60° deviation yielded a signifi-cant effect of string length, F(3, 69) = 6.56;p < .001.

A three-way ANOVA on the error ratesyielded significant main effects for angulardeviation, F(3, 69) = 3.24, p < .05, and forstring length, F(3, 69) = 12.42, p < .0001,but no effects for lexicality. Although theeffects of deviation mimic those of the re-sponse times, those for string length do not.Mean percent errors for string lengths 2, 3,4, and 5, were 15.1%, 11.4%, 8.4%, and 9.6%,respectively. The higher error rates for the


shorter strings point to the possibility of aspeed-accuracy trade-off for the length effects.

Discussion

In Experiment 1 we examined the questionwhether in reading disoriented words thetransformation process is holistic, operatingon the entire word pattern as an integralunit, or is conducted piecewise, with segmentsof the word transformed in turn. On the faceof it, the results seem to suggest the operationof both types of transformation processes.First, response latency increases sharply withstring length. The extent of orientation effectsfrom 0° to 180°, calculated across words andnonwords, was 339 ms, 508 ms, 909 ms, and1,100 ms for two- to five-letter strings, re-spectively. This systematic increase suggeststhat whatever transformation is applied todisoriented strings, it operates in a piecemeal,apparently letter-by-letter, fashion. Second,however, there is a range of orientations overwhich words seem to be processed as integralwholes, and when transformation is required,it operates on the word as a single unit. Thisis suggested by the results for word stimuliat deviations 0° and 60°. Although the 60°disorientation required significantly longerresponse times than did the upright (0°)orientation, this effect was independent ofword length. Apparently for deviations of upto 40° or 60° from the upright, holistictransformation processes take place.

The increase in response time from 0° to60° deviations is relatively slight. A similartype of relative indifference to small disori-entations from the upright has been discussedby Hock and Tromley (1978) in terms of thenotion of a perceptual upright. They proposedthat each stimulus has a characteristic rangeof orientations over which it is perceived asupright and that mental rotation need proceedonly until the stimulus attains its perceptualuprightness. It is tempting to speculate thatwords are perceptually upright in a range of±60° about their normal orientation, andwithin this range they are processed holisti-cally. But, on the other hand, it remainsunclear why the 60° deviations yield signifi-cantly longer response times than do the 0°deviations.

Cooper and Shepard (1973), using reflectiondecisions on single characters, found a non-linear increase in response time with degreeof disorientation, with smaller disorientationsyielding relatively small effects. They sug-gested that the memory representation of thenormal, upright character may be broadlytuned so that a response can be made fortilts of 60° or more without any need formental rotation. This concept of broad tuningis similar to the idea of a perceptual uprightbut appears better fitted to accommodatefindings of a small increase in response timewith increasing disorientation like that foundin the present study.

As a final comment, it should be notedthat the results appear dichotomous, with thethree smaller disorientations (0°, 60°, and300°) yielding relatively fast responses andthe three larger disorientations (120°, 180°,and 240°) yielding considerably slower re-sponses that are quite similar in their mag-nitude. We shall return to this point inconnection with Experiment 3.

Experiment 2

The results of Experiment 1 can be inter-preted as indicating that letter strings atdisorientations greater than 60° tend to beprocessed nonholistically. This interpretationrests on the assumption that the critical factoraffecting speed of processing is the numberof letters in the string.

There are, however, two additional inter-pretations. The first is that longer strings arevisually more complex, and complex figuresmight require slower rotation rates than dosimple figures. This account is hard to refute,though it does not seem likely in view of thefact that word length did not have any effectsat the 0° and 60° deviations. If complexitywere the critical factor, it would have beenexpected to affect performance at all orien-tations. The second account involves anotherfactor that is correlated with string length,namely, the visual extent of the stimulus. Itmight be argued that longer strings are rotatedmore slowly than shorter strings simply be-cause they are larger. Indeed, Shwartz (1979;see Kosslyn, 1980, pp. 290-292) found thesize of a visual image to affect rotation rate.


Subjects were presented with an Attneave-type polygon and were asked to rotate itsimage until it reached a specified orientationand to press a button when the desired ori-entation was attained. Larger polygons werefound to yield slower rotation rates thansmaller polygons.

In Experiment 2 we varied number ofletters in a string while holding its visualextent constant. The stimuli were three- andfour-letter strings that subtended the samevisual angle. If the two string lengths exhibitsimilar effects of rotation, the results obtainedin Experiment 1 might be interpreted interms of the size effect of Shwartz (1979),namely that the slower "rate of rotation"obtained for longer strings stems from aslower but holistic rotation process, ratherthan from a nonholistic, piecemeal process.

Method

Subjects. Twelve Univeisity of Haifa students partic-ipated in the study for course credit. None had participatedin Experiment 1.

Stimulus materials. Seventy-two three-letter and 72four-letter Hebrew words were compiled, containing onlyconsonantal letters and possessing only one pronunciation.Half of the words of each length were transformed into

nonwords by replacing one letter by another consonantalletter. Strings of both lengths subtended 4.3 cm horizon-

tally (when upright) and 1.4 cm vertically. In the three-letter strings letters subtended 1.3 cm maximally with

about 0.2 cm between letters. The respective values forthe four-letter strings were 1.0 and O.I cm. Hebrew

readers judged the two types of strings to be equallysimilar to normal printed Hebrew.

Apparatus and procedure. The apparatus and generalprocedure were the same as in Experiment 1. Eachsubject was presented with four blocks, each block ho-mogeneous with respect to string length. Half the subjects

received the four blocks in the order three-, four-, four-,and three-letter strings, and the other half in the order

four-, three-, three-, and four-letter strings. Each blockincluded 72 experimental trials preceded by 12 practicetrials. The third and fourth blocks were replications ofthe second and first blocks, but the string orders weredifferent, randomly determined for each subject and foreach block. In each block six words and six nonwordsappeared in each of the six orientations—0°, 60°, 120°,180°, 240°, and, 300°—and each string appeared equallyoften in each orientation across all subjects.

Results and Discussion

Responses in excess of 5,000 ms or lessthan 250 ms were discarded (0.3%). Figure 3presents mean response times for correct

responses for words and nonwords as a func-tion of orientation and string length. Theeffect of disorientation on response time isclearly stronger for four-letter strings than forthree-letter strings. A three-way ANOVA yieldedsignificant main effects for string length, F(\,11) = 99.87, p < .0001; for orientation, F(5,55) = 31.77, p < .0001; and for-lexicality,F(\, 11) = 53.23, p < .0001. The orientationeffects were larger for the four-letter stringsthan for the three-letter strings, F(5, 55) =12.27, p < .0001, and larger for nonwordsthan words, F(5, 55) = 4.22, p < .002. Theother interactions were not significant.

When the results were analyzed in termsof angular deviation from the upright, meanresponse times for 0°, 60°, 120°, and 180°deviations were 690 ms, 735 ms, 1,125 ms,and 1,125 ms for three-letter words and 725ms, 789 ms, 1,378 ms, and 1,426 ms forfour-letter words, respectively. Percent errorsfor these four deviations were 1.4%, 2.8%,6.6%, and 7.6% for the three-letter words,and 0.7%, 0.3%, 10.1%, and 6.9% for thefour-letter words.

It is interesting to note that, as in Experi-ment 1, although words at 60° deviationsyielded significantly longer response timesthan at the 0° deviation, F(l, 11) = 25.63,p < .0005, the interaction between this com-parison and word length was not significant(F < 1). The same was true for nonwords:The 0° versus 60° comparison yielded F(l,11) = 7.08, p < .05, but the interaction wasnot significant, F(l, 11) = 1.65.

All in all, the results of Experiment 2indicated significantly larger orientation effectsfor the four-letter strings than for the three-letter strings even when they subtended thesame visual angle.

Experiment 3

Taken together, the results of Experiments1 and 2 indicate stronger orientation effectsfor longer letter strings, and the critical factoris the number of the letters in the stringrather than its visual extent. These resultssuggest nonholistic processing beyond 60°angular deviation from the upright.

Experiment 3 was designed to obtain amore fine-grained picture of the combined


effects of angular disorientation and stringlength by sampling a larger number of ori-entations. Specifically, the results of Experi-

ment 1 suggest the following trends thatdeserve detailed examination. First, there ap-pears to be a range of orientations over which

2000

1800

1600

LU

5

1400

1200

1000

600

600

O 3 LETTERS

• 4 LETTERS

NON WORDS

240" 300"

ORIENTATION

Figure 3. Response time (in milliseconds) as a function of orientation for three- and four-letter words andnonwords (Experiment 2).


words are processed and transformed holis-tically, as suggested by the length-independentincrease in response time from 0° to 60°deviations. In Experiment 3 we hope to de-limit the boundaries of this range of orien-tations. Second, the results for words (Figure2) suggest a sharp dichotomy between the 0°and 60° deviations, which evidence no lengtheffects at all, and the 120° and 180° devia-tions, which evidence strong and nearly iden-tical length effects. The inclusion of inter-mediate deviations might help determinewhether the transition between these tworegions is sharp or gradual. Third, the com-parison between the 180° and the 120° de-viations suggests that beyond a certain degreeof disorientation there is no further decrementin performance and no further increase inthe size of the length effect with increasingangular deviation. In order to examine thesetrends, 18 different orientations were em-ployed in Experiment 3, 0° to 340° in 20°steps, yielding a total of 10 different deviationsfrom the upright.

Subjects. Eighteen students at the Univer-sity of Haifa, whose primary language wasHebrew, took part in the experiment forcourse credit. None had participated in Ex-periments 1 or 2.

Stimulus materials. Two hundred andeighty-eight 2- to 5-letter Hebrew strings wereused. These represented an equal number ofwords and nonwords of each of the fourlengths. The selection of words and the con-struction of nonwords were according to thesame criteria as used in Experiment 1.

Apparatus and procedure. The apparatuswas the same as in Experiment 1. The pro-cedure was also the same except for thefollowing. The strings appeared in 18 differentorientations, 0° to 340° in 20° steps. Thesession began with 36 practice trials, followedby two repetitions of the 288 experimentalstrings. There was a short break after eachblock of 144 trials and a 2-min break betweenthe two repetitions. Each block began withfour filler items to allow warm-up. Stringorientations were preprogrammed so that foreach subject and repetition two words andtwo nonwords of each length appeared ateach orientation, and for each repetition eachstring appeared equally often at each orien-tation across all subjects. The orientation of

each letter string varied between repetitions.The order of the stimuli was randomly deter-mined for each subject and for each repetition.

Results

Response latencies in excess of 5,000 msand less than 250 ms were discarded (0.3%).The two repetitions yielded similar resultsand were therefore combined.

First, we shall examine in some detail themain effects of rotation, after which we shallturn to the combined effects of orientationand string length. The curves for strings ofdifferent lengths were all quite similar inshape, and Figure 4, which presents responsetimes and percent errors averaged across stringlengths, seems to best convey the major trends.

One noteworthy feature of the curves pre-sented in Figure 4 is their noticeable departurefrom linearity. Although response time gen-erally increases with angular deviation fromthe upright, the shapes of the functions aredifferent from those obtained for mental ro-tation of block figures (Shepard & Metzler,1971) or single letters (Cooper & Shepard,1973) and are not consistent with the notionof a constant rate of mental rotation. Interms of angular deviation from the upright(regardless of direction), the curves appear tobe S-shaped, with the range between the 60°and 120° accounting for the bulk of therotation effect. Outside this range, that is, fordisorientations of less than 60° or more than120°, response time is relatively indifferentto the extent of angular deviation.

In the lower range, 0°, 20°, and 40° devia-tions, rotation appears to have little effect,with rotation effects emerging at around 60°.Comparing only the five orientations between±40°, a one-way ANOVA yielded F < 1 forwords and F(4, 68) = 1.70, ns, for nonwords.These results are consistent with the notionof a perceptual upright and suggest that wordsthat are disoriented up to about ±40" neednot be mentally rotated or rectified beforerecognition.

The upper range of disorientations, 120°-240°, represents perhaps the largest departurefrom what is customarily found for single-letter mental rotation tasks. Response timesfor the 180° orientation are far lower thanwould have been expected. In fact, further


increases in angular deviation beyond 120°do not result in any additional impairmentin performance.

Three additional features deserve mention:the word-nonword differences, the asymmetryeffects, and the upside-down effect, all previ-ously noted by Koriat and Norman (1984).First, nonwords exhibited stronger rotationeffects than did words: The increase in re-

sponse time from 0° to 180° was 550 ms forwords and 800 ms for nonwords, t(\7) =4.63, p < .001. This is consistent with theresults of Experiment I (see Figure 1) andmight suggest that the transformation andrecognition processes partly interact.

Second, the curves for both words andnonwords exhibit a certain degree of asym-metry that is most evident in the upper range

2000

1800

LU

1400

UJCO

OQ.COUJ 1200

1000

12C? 180° 240°

ORIENTATION

300°

Figure 4. Response time (in milliseconds) and percent errors as a function of orientation for words andnonwords (Experiment 3).


of disorientations. Performance on the 120°orientation is considerably poorer than thaton the 240° orientation, both being of equalangular disparity from the upright, t(\l) =4.54, p < .001. One explanation is that, be-cause Hebrew is printed from right to left, atthe 120° orientation strings are read in anupward direction whereas at the 240° orien-tation they are read in a downward direction.Reading in a downward direction might beeasier than reading in an upward direction.A second explanation is that mental rotationin a clockwise direction might be easier thanrotation in a counter-clockwise direction.Robertson and Palmer (1983) reported a sim-ilar asymmetry for compound Latin letters,a finding that is inconsistent with the firstexplanation. However, because there are majordifferences between their study and the present

one, it is not clear that the same processesare involved.

Third, note the relative ease of processingupside-down words in comparison to all otherorientations in the 120° to 240° range. The140-ms drop from 120° to 180° was signifi-cant, t(\l) = 3.36, p < .005. This effect wasfound to hold for each of the four wordlengths. The respective drops (from 120° to180°) for words of lengths 2 to 5, were 136ms, 189 ms, 113 ms, and 229 ms, respectively.A similar trend appears in the error data(Figure 4). This upside-down effect, previouslyreported by Koriat and Norman (1984), seemsconsistent with other findings on the readingof misoriented text.

Turning to the effects of string length,Figure 5 presents evidence for very stronginteractions between string length and angular

E

in

(fl 1400

O

'

20°-

40°-

6tf-

80

-180°

-160°

-140°

ISO'

-100°

WORDS

NONWORDS

STRING LENGTHFigure 5. Response time (in milliseconds) for words and nonwords as a function of string length fordifferent angular deviations from the upright (Experiment 3).


disorientation, for words, F(27, 459) = 11.68,p < .0001, and for nonwords, F(27, 459) =15.51, p < .0001. The results presented inthis figure are very similar to those of Figure2 with respect to the four values in common,0°, 60°, 120°, and 180°. The smallest angulardisorientations, 20° and 40°, evidence littleincrease in response time with word length,the curves virtually overlapping that of theupright position (0°). Thus, word recognitionappears to be indifferent to either word lengthor angular disorientation for deviations of40° or less from the upright position.

As for the error data, a two-way ANOVAyielded significant effects of word length, withpercent errors for two- to five-letter wordsbeing 9.6%, 4.0%, 6.0%, and 5.7%, respec-tively. The somewhat high error rate for two-letter words may be due to the inclusion oftwo very rare two-letter words. This may alsoexplain the slight increase in response timesfor two-letter words (see Figure 5), as well asthe small but significant interaction betweenlength and angular deviation found for percenterrors. Neither of these effects was significantfor nonword errors: The percent errors fortwo- to five-letter strings were 9.5%, 5.6%,5.2%, and 7.0%, respectively. Thus, the sus-picion of a speed-accuracy trade-off for lengtheffects seen in Experiment 1 is not corrobo-rated here.

In comparison to 0°-40° deviations, the60° deviation, although evidencing a slightincrease in response time overall—the differ-ence between 60° and the average of thethree smallest deviations yielded F(\, 17) =15.86, p < .001 for words and f\l, 17) =22.36, p < .0002 for nonwords—shows littleincrease in the effects of string length (theinteraction between the previous comparisonand length yielded F < 1 for both words andnonwords). This is similar to what was foundfor words in Experiments 1 and 2. If mentalrotation occurs in this region, it would appearto be holistic in nature (or piecemeal butparallel, see Funt, 1983).

The results for the extreme disorientationsyield very similar effects of string length forthe four deviations between 120° and 180°.For these deviations, the overall increase inresponse time for each extra letter is about220 ms for words and about 315 ms fornonwords.

The two intermediate deviations, 80° and100°, evidence intermediate effects of stringlength. For words, the slopes of response timeon string length are 55 ms per letter and 145ms per letter for 80° and 100°, respectively.The respective values for nonwords are 109ms per letter and 236 ms per letter.

Discussion

The results of Experiment 3 corroborateand extend the findings of Experiments 1 and2. When the effects of both rotation andstring length are examined conjointly, theangular orientations included in Experiment3 appear to be roughly divided into threeregions. These regions are clearly evident inFigures 4 and 5. First, there is a region inwhich neither angular deviation nor stringlength affects word recognition latency. Itincludes the upright (0°) orientation andextends to disorientations of up to 40° or60°. For the second region, from about 60°to 120° deviations, marked orientation effectsoccur that are strongly dependent on stringlength. To illustrate, orientation effects in the60°-120° region (for words and nonwordscombined) increase by about 220 ms for eachadditional letter. The third region includesdeviations in excess of 120°. These yield nofurther increases in response time nor anOrientation X String Length interaction.

In addition to the three regions mentionedabove, the results for word stimuli at 60°deviations represent an interesting case intheir own right. In Experiment 3, as inExperiments 1 and 2, they yielded significantlyslower response times while at the same timeindicating no effects of word length.

In sum, the results are inconsistent withthe idea that words are processed and trans-formed as integral units over the entire rangeof disorientations. If we assume that therecognition of rotated words indeed involvesa mental rotation process like that discussedby Shepard and Cooper (1982), the resultssuggest that, except for a narrow zone (around60° deviation) in which words are rotated asintegral wholes, in all other orientations inwhich mental rotation occurs, it is by andlarge piecemeal. In fact, for the range ofangular deviations between 60° and 120° thesystematic decrease in "rate of mental rota-


tion" with increasing string length agreesrather well with a model in which singleletters are mentally rotated in turn at aconstant rate.

These results contrast those of Robertsonand Palmer (1983), who, using hierarchicallystructured patterns (large letters made up ofmany small letters), found rotation rate to beunaffected by the number of levels that hadto be rotated. In their task, however, decisionsat the local level could be carried out on thebasis of a single element, which, of course, isnot possible with the lexical decision task.Our results are also not consistent with Shep-ard and Cooper's (1982) assumption thatmental rotation is holistic in tasks involvingthe matching of a single, well-integrated fa-miliar figure to an internal representation.Alternatively, we may conclude that in factwords do not constitute "well-integrated fa-miliar figures" in the sense used by Shepardand Cooper.

It should be stressed, however, that therotation function obtained (Figure 4) is no-tably different from that normally observedin mental rotation studies. This raises thequestion whether the recognition of rotatedwords involves the same sort of mental ro-tation process observed with other stimuli.At the least, it would seem that additionalprocesses must be postulated to account forthe results obtained over the entire range oforientations, and these processes might bespecific to word recognition and reading.

From the point of view of reading pro-cesses, the region of intermediate orientationsposes a difficult problem when contrastedwith the other two regions. In the first regionof relatively small deviations, words appearto be recognized directly, that is, with noneed for additional processes over and abovethose required by upright words. We mayassume that word recognition in this regionis holistic or else that it rests on the processingof all its elements in parallel. Following Hockand Tromley (1978), we shall label this regionthe region of the perceptual upright in wordrecognition.

At the other extreme, beyond 120° devia-tions, there is no increased impairment withincreased angular deviations but a strongeffect of word length. This pattern suggests aprocess in which words are read letter by

letter. If letter identity is not disrupted bydisorientation (e.g., CorbaUis & Nagourney,1978; however, see Jolicoeur & Landau, 1984),such a process should indeed result in ori-entation-independent length effects.

What, then, happens in the middle region?Word recognition in this region apparentlyinvolves neither whole-word reading nor letter-by-letter reading. Experiment 4 was designedto examine what type of visual informationis lost or disrupted in this region. If visualtransformation does occur in this region, wemay assume that it is intended to recoversome of this information.

Experiment 4

If we assume that disorientations beyondthe upright impair letter identification, theresults for the intermediate region of angulardeviations can be interpreted in terms of aletter rotation model: Words in this regionmust be rotated letter by letter in order to berecognized. In this model, mental rotation isassumed to recover letter orientation. Thissimple and appealing model encounters thedifficulty that at least as far as single-letteridentification is concerned, it is not clear thatletters must be rotated to be recognized (seeCorballis & Nagourney, 1978; Jolicoeur &Landau, 1984). A further difficulty is thatthis model fails to distinguish between thesecond and third regions, both of whichapparently involve some piecemeal processing.

An alternative model stresses the role oftransgraphemic features in word recognition.It has been proposed that familiar visualpatterns tend to be processed as a unit, withglobal word shape features dominating wordprocessing (Haber & Schindler, 1981; Lawry,1980). Accordingly, we may assume thatwithin the perceptual upright, word recogni-tion capitalizes on configurational, transgra-phemic features (Garner, 1981; Monk &Hulme, 1983). These features are disruptedwhen the word is tilted beyond the perceptualupright. If a transformation process doesoccur in the intermediate region, then itsfunction might be that of recovering some ofthe supraletter orthographic features.

Experiment 4 evaluated the two models bycontrasting two types of stimulus transfor-mations. In the global transformation, letter


strings were rotated in their entirety abouttheir center (as in the previous experiments),whereas in the letter-upright transformation,only the string-array direction was rotated,but each letter remained upright. If the letterrotation model is correct, the global transfor-mation should prove the easier within theperceptual upright, whereas the letter-uprighttransformation should prove the easier beyondthe perceptual upright. This prediction as-sumes that within the perceptual upright itis the orientation of the entire string thatcounts, whereas beyond the perceptual uprightit is the orientation of the letters that matters.The second model stressing transgraphemicfeatures, on the other hand, predicts an ad-vantage for the global transformation evenbeyond the perceptual upright.

Method

Subjects. Eighteen University of Haifa students par-ticipated in the study for course credit. Hebrew was their

primary language.

Stimuli. Two hundred and eighty eight words wereselected. Half of the words were transformed into non-

words. An additional practice list consisted of 18 wordsand 18 nonwords. They appeared at one of five orienta-tions—0°, 30°, 70°, 290°, and 330° (labeled in a clock-wise direction as in Cooper & Shepard, 1973). When the

strings were misoriented, they appeared in either one oftwo formats—global, with the entire string rotated as in

the previous experiments, or letter-upright, with eachletter occupying the same location as in the global formatbut presented upright.

Apparatus and procedure. The apparatus was thesame as that used in the previous experiments. Theprocedure was also the same except that the subjects

were instructed about the stimuli and were shown ex-amples of the different formats and orientations. Thesession began with 36 practice trials, followed by tworepetitions of the 288 experimental strings in differentrandom orders. There was a short break after each blockof 144 trials and a 2-min break between the two repeti-tions. Each block began with four filler items to allowwarm-up. String orientations were preprogrammed sothat for each subject and repetition 16 words and 16nonwords appeared at the upright orientation and ineach of the eight Disorientation X Format combinations.Each string appeared equally often in each of these ninearrangements across all subjects. The arrangement ofeach letter string varied between repetitions. The orderof the stimuli was randomly determined for each subjectand for each repetition. Each letter was about 1.0 X 1.0cm, with a 0.1 cm gap between letters (in the globalformat).

Results

Response latencies in excess of 5,000 msand less than 250 ms were discarded (0.4%).

The two repetitions yielded similar resultsand were therefore combined.

Figure 6 presents mean response times forwords and nonwords as a function of orien-tation and format. It may be readily seenthat the results do not conform to the letterrotation hypothesis. For both the 30° and70° angular deviations the letter-upright for-mat yields longer response times than doesthe global format, and in fact this differenceis more pronounced for the 70° than for the30° deviations. Considering only the 30° and70° angular deviations and combining wordsand nonwords, a three-way ANOVA on re-sponse times, Deviation X Direction of Ro-tation (clockwise, 30° and 70° vs. counter-clockwise, 290° and 330°) X Format, yieldedsignificant effects for format, F(\, 17) =105.96; p < .0001; for deviation, F(l, 17) =104.52; p < .0001; and for the Format X De-viation interaction, F(l, 17) = 57.56), p<.0001. The results also indicated a significanteffect of direction, F(l, 17) = 77.83; p<.0001, which interacted with the effects ofdeviation, F(l, 17) = 40.33; p < .0001, andwith the effects of format, F(l, 17) = 29.31,p < .0001. The effects of direction are prob-ably due to the same type of asymmetryfound in Experiment 3, and this asymmetryis stronger for the letter-upright format thanfor the global format, and for the 70° devia-tions than for the 30° deviations. It shouldbe recalled that since Hebrew is read fromright to left, the direction of reading is down-ward in the 290° orientation and upward inthe 70° orientation. The triple interactionwas also significant, F(l, 17) = 23.31; p <.0001.

The error data indicated a similar patternto that found for the response time data, buta similar ANOVA yielded no significant effectsapart from that due to angular deviation.

Discussion

The results of Experiment 4 indicate thatstimulus transformations that disrupt wholeword features while leaving letters uprightimpair word recognition to a greater extentthan do global transformations in which someof the transgraphemic features are spared.This appears to be true within the perceptualupright and even more so beyond it.

These findings seem to argue against the


idea that in the intermediate angular devia-tions, words are read letter by letter, witheach letter first rotated to an upright orien-tation. If this were the case, the letter-uprightformat should have yielded better perfor-mance (in the 70° deviation) because in thisformat letters need not be rotated. Rather,the results are consistent with the view thateven in the intermediate angular deviations,word recognition tends to capitalize, to someextent, on supraletter features.

This conclusion can lead to two ratherdifferent interpretations of the results for theintermediate region. The first assumes that

beyond the perceptual upright some transfor-mational process is applied to letter stringsin an attempt to recover some of the supra-letter features disrupted by string disorien-tation. The nature of this process is not clear,except that it appears to involve some piece-meal processing and that its duration dependson the extent of angular deviation from theupright.

A second account does not postulate anyvisual transformation process. It assumes thatas a word departs from perceptual uprightnessmore of its transgraphemic features are lost,rendering it more difficult to recognize. Thus,

1200

"Si

uj

1 1000

NONWORDS

WORDS

Global

Letter Upright

OT 30" 70° 290° 330* 0"

ORIENTATION

Figure 6. Response time (in milliseconds) for words and nonwords for letter-upright and global transformationsas a function of disorientation (Experiment 4).

504 ASHER K.ORIAT AND JOEL NORMAN

orientation effects in the intermediate regionsimply reflect the extent to which word rec-ognition takes advantage of whole-word fea-tures. This account must, of course, makeadditional assumptions to explain why ori-entation effects are length dependent.

The observation that beyond the 120° de-viation there is no increased impairment ofperformance with increased disorientationmay be accommodated by both these ac-counts. Apparently, transgraphemic featuresare lost at extreme disorientations, and wordrecognition must rely solely on letter-by-letterreading.

General Discussion

The research reported in this article com-bined a reading task with a mental rotationtask with the aim of obtaining informationrelevant to two different theoretical issues—first, whether disoriented visual stimuli are"mentally rotated" as integral wholes orpiecemeal and, second, whether words areread holistically or in a letter-by-letter fashion.It was proposed that if words are processedand transformed as integral wholes, the effectsof rotation should be the same for words ofdifferent length, but if they are transformedpiecemeal, the effect should increase withincreasing word length.

The results of Experiments 1 and 3 indi-cated that the effects of angular disorientationvary strongly with word length, and those ofExperiment 2 suggested that these effects arenot due to the physical extent of the longerstrings but to the greater number of lettersthey contain. On the basis of these results wemay safely reject the hypothesis that wordsare processed and transformed as integralunits over the entire range of orientations.

Apart from this general conclusion, how-ever, it is difficult to offer a complete accountof the complex pattern of results obtained.Most important, it is not clear whether thereading of rotated words involves the samesort of mental rotation process envisioned byShepard and his colleagues or any otherrectification process, for that matter. Therotation functions obtained depart consider-ably from linearity. They are also quite dif-ferent from those found for single-letter re-flection decisions, especially in the region ofextreme deviations. The results, on the whole,

present a rather nonuniform pattern withregard to the effects of increased disorienta-tion both on mean response time (Figures 1,3, and 4) and on the extent of string lengtheffects (Figures 2 and 5). Consequently it isdifficult to propose an interpretation of theresults in terms of one unitary principle.

In what follows we shall examine possibleaccounts for the results obtained relying onthe subdivision into three regions discussedabove. Because in the first and third regionsthere was no effect of rotation, we shall firstexamine the results for these two regions,relying solely on what is known about mech-anisms of reading and word recognition.

The results for the first region of relativelysmall tilts about the upright are the mostpertinent to normal reading. It is of consid-erable theoretical interest that word recogni-tion is largely indifferent to disorientations ofup to 40° or 60° from the upright and thatwithin this range it is also indifferent to wordlength. The indifference of lexical decisionlatencies to word length has been reportedfor upright words (e.g., Frederiksen & Kroll,1976; Koriat, 1984), and the present studyextends this finding to words tilted up to 60°in either direction. The relative indifferenceto small disorientations has been reported byCooper and Shepard (1973) for reflectiondecisions on single letters. Furthermore, Ko-riat and Norman (in press) found that this istrue only for normal letters, whereas reflectedletters evidence a remarkably linear increasein response time across the entire range oforientations. This was seen to suggest thatthe memory representations of familiar visualpatterns (e.g., normal letters) are characterizedby broad orientation tuning, allowing directrecognition despite small tilts. The indiffer-ence of word recognition to both word lengthand word orientation in the present studymay similarly suggest, first, that word recog-nition in this region is based on the directactivation of whole-word visual patterns (seeJohnson, 1975, 1981), and, second, that theinternal visual representation of a familiarword is broadly tuned and can be efficientlyactivated even when the visual stimulus de-parts rather noticeably from its upright ori-entation. Following Hock and Tromley (1978),we label this range of orientation indifferencethe region of perceptual uprightness in wordrecognition. Thus, as long as a word is per-


ceptually upright, it appears to be recognizedholistically.

It should be noted, though, that the indif-ference of word recognition to word length isalso consistent with the view that word rec-ognition involves parallel processing of singleletters (e.g., McClelland & Rumelhart, 1981).It is not clear, however, why such parallelprocessing should hold only within the con-fines of the perceptual upright and not be-yond.

The finding that words at the 60° deviationevidenced length-independent orientation ef-fects in all of the first three experimentsdeserves particular attention. It suggests theexistence of a transitional zone on the fringeof the perceptual upright where words appearto be transformed holistically. This phenom-enon lends further support to the idea thatwithin the perceptual upright, word recogni-tion is based on whole-word patterns.

Turning next to the results for the thirdregion (beyond 120° deviations), these seemto suggest a rather different reading strategyin which words are processed letter by letterwithout mental rotation or transformation.Single letters are identified serially and thestrings unitized before (or partly in unisonwith) lexical access. If the speed of identifyingsingle letters is indifferent to their orientation,a process of this sort should yield no effectof word orientation but strong effects of wordlength as was found to be the case. It wouldseem that these nearly horizontal invertedletter arrays induce serial scanning and resultin the type of processes investigated by Kolersand Perkins (1975). This serial processing isreminiscent of the reading disorder referredto as word-form dyslexia (Warrington & Shal-lice, 1980) or letter-by-letter reading (Patterson& Kay, 1982). Indeed, the most distinctivecharacteristic of this syndrome is that readingtime increases monotonically with wordlength. This syndrome has been assumed toreflect the inaccessibility of visual word formsthat normally allow whole-word recognition.The patient is forced to reconstruct the wordfrom the names of the letters (Warrington &Shallice, 1980). It is not clear, however,whether the letter-by-letter reading of thepresent study also relies on phonemic codes(see Shallice, 1981).

Summing up at this point, when the firstand third regions are compared, they appear

to involve two contrasting reading strategies,whole word and letter by letter. Because inboth response time does not increase withincreased disorientation, this may suggest di-rect activation of the respective codes, whole-word codes or letter codes.

Before examining the results for the inter-mediate region, let us note that the contrastbetween the two reading strategies presentedabove may help resolve the paradox notedby Corballis, Zbrodoff, Shetzer, and Butler(1978): Although stimulus disorientation hasa negligible effect on the identification ofsingle letters (e.g., Corballis & Nagourney,1978), it has considerable disruptive effectson the reading of longer strings (Kolers &Perkins, 1975). Why would an upside-downletter be identified as quickly as an uprightletter, whereas an upside-down word takeslonger to recognize than an upright word? Interms of the propositions offered above, thisoccurs because reading an upright word in-volves the activation of a single, whole-wordcode, whereas reading an upside-down wordinvolves the processing of several units insequence. If this interpretation is correct, therecognition of single-letter words should beindifferent to orientation. The results of Ex-periments 1 and 3 suggest that this mightindeed be the case: If we were to extrapolatethe curves depicted in Figures 2 and 5, theywould intersect at a value nearly equivalentto a one-letter word. The same is true fornonwords as well. No single-letter words existin Hebrew, but it would be instructive to seewhether lexical decision times for single-letterEnglish words are indeed indifferent to dis-orientation (but see Samuel, van Santen, &Johnston, 1982). The interesting question,however, still remains: Why are words rec-ognized directly only within the perceptualupright, whereas single letters (and probablysingle-letter words) are recognized directlyover the entire range of disorientations?

We shall turn now to the intermediateregion involving length-dependent orientationeffects. The interpretation of these effectsmust take into account the results of Exper-iment 4, where the global transformation waseasier than the letter-upright transformationat 70° angular deviations. This suggests thatthe reading of rotated words in the interme-diate region can take advantage of informationpertaining to units that are larger than a


letter. Thus, although reading in this regionappears to involve piecemeal processing, itdiffers from the letter-by-letter reading char-acteristic of the third region.

The major difficulty in explaining the re-sults for the intermediate region is that it isnot clear whether the piecemeal processingthat apparently takes place in this regionderives from mechanisms of visual transfor-mation, from mechanisms of word recogni-tion, or from both. One type of interpretationrelies solely on principles of word recognition.In terms of the contrast between the tworeading strategies that seem to characterizethe first and third regions, perhaps the sim-plest account of the results for the interme-diate region is that they reflect either a prob-abilistic mixture of the strategies of whole-word and letter-by-letter reading, or better, areliance on units of intermediate size. Con-sistent with the idea of multiple units in wordrecognition (cf. Santa, Santa, & Smith, 1977),the increase in length effects with increasingangular deviations may be assumed to stemfrom the size of the unit involved in wordrecognition: For small disorientations, theunit is the whole word, for extreme disori-entations it is the single letter, and for inter-mediate disorientations it is letter clusters,the size of which decreases with increasingdisorientation. In this interpretation, orien-tation and length effects are seen to reflectthe same process: Increased disorientationreduces the size of the perceptual unit, andthus the number of effective units becomes ajoint function of disorientation and wordlength. If this interpretation is correct, it mayhelp explain the difference between wordsand nonwords. Nonwords exhibited strongerorientation effects in Experiments 1 ,2 , and3, and stronger length effects in Experiments1 and 3. These differences may be accom-modated by assuming that at any given ori-entation, words require fewer units than non-words of the same length.

A second type of interpretation attributesthe length-dependent orientation effects torectification processes. The best candidate forsuch a process is some sort of piecemealmental rotation in which letters are rotatedin turn at a constant rate. This model fits thedata because it is conjointly sensitive to bothnumber of letters and degree of disorienta-

tion. The results of Experiment 4 suggestthat the aim of letter rotation is not simplyto aid in letter identification but to assist inrecovering transgraphemic features, thus al-lowing word recognition to capitalize on largerunits, even whole words. The weaker effectsof orientation and length found for wordsthan for nonwords may result from the inter-play between activations at the letter andword levels (see McClelland & Rumelhart,1981; Rudnicky & Kolers, 1984). We shouldpoint out that the emphasis on transgra-phemic features is open to several objections(see Henderson, 1982), among them its in-consistency with the finding that word supe-riority is rather immune to the disruption ofsupraletter features (e.g., Adams, 1979;McClelland, 1976).

An alternative account emphasizes the im-portance of letter-order permutations. Severalletter-based theories of word recognition (e.g.,McClelland & Rumelhart, 1981; Paap, New-some, McDonald, & Schvaneveldt, 1982) as-sume position-specific parallel activation ofthe appropriate codes for all the letters in aword. Perhaps disorientation beyond the per-ceptual upright disrupts the letter-positionmapping, and it is this mapping that is re-stored by the rectification process.

In conclusion, this study of the reading ofrotated words has yielded a rather complexpattern of results pertaining to the issue ofholism in mental rotation and word recog-nition. As far as mental rotation is concerned,the results do not generally conform to thehypothesis that word stimuli are mentallyrotated as integral units. This is inconsistentwith Shepard and Cooper's assumption con-cerning the rotation of well-integrated, famil-iar stimuli. Thus, we must either reject thehypothesis of holistic mental rotation alto-gether, admit that words do not constitutewell-integrated wholes, or conclude that ro-tated letter strings induce different types ofprocesses than those normally observed forother disoriented visual stimuli. The resultsfor words at 60° deviations do hint, however,at the possibility of holistic mental rotation,and this deserves further research scrutiny.

As for the issue of holism in word recog-nition, the results appear to indicate thatwithin the perceptual upright (±60° aboutthe upright), word recognition relies on whole-


word units, whereas at extreme disorientations(120°-240°) it appears to be mediated bysequential letter identification. The interpre-tation of what occurs in the intermediateregion of deviations (60°-120°) is more dif-ficult, though it appears that word recognitionin this region may rely on units larger thansingle letters. Further research is needed todetermine whether the length-dependent ori-entation effects in this region are best inter-preted in terms of mechanisms of visualrectification, mechanisms of word recognition,or both.

References

Adams, M. J. (1979). Models of word recognition. Cog-nitive Psychology, 11, 133-176.

Bruder, G. A. (1978). Role of visual familiarity in theword-superiority effects obtained with the srmultaneous-matching task. Journal of Experimental Psychology:Human Perception and Performance, 4, 88-100.

Cattell, J. M. (1886). The time taken up by cerebraloperations. MM, 11, 220-242.

Coltheart, M., & Freeman, R. (1974). Case alternationimpairs word identification. Bulletin of the Psychonomic

Society, 3, 102-104.

Cooper, L. A. (1975). Mental transformations of randomtwo-dimensional shapes. Cognitive Psychology, 7, 20-43.

Cooper, L. A., & Podgorny, P. (1976). Mental transfor-mations and visual comparison processes: Effects of

complexity and similarity. Journal of ExperimentalPsychology: Human Perception and Performance, 2,503-514.

Cooper, L. A., & Shepard, R. N. (1973). Chronometricstudies of the rotation of mental images. In W. G.Chase (Ed.), Visual information processing (pp. 75-176). New York: Academic Press.

Corballis, M. C, & Nagourney, B. A. (1978). Latency tocategorize disoriented alphanumeric characters as lettersor digits. Canadian Journal of Psychology, 32, 186-188.

Corballis, M. C., Zbrodoff, N. J., Shetzer, L. I., & Butler,P. B. (1978). Decisions about identity and orientationof rotated letters and digits. Memory & Cognition, 6,98-107.

Frederiksen, J. R., & Kroll, J. F. (1976). Spelling andsound: Approaches to the internal lexicon. Journal ofExperimental Psychology: Human Perception and Per-formance, 2, 361-379.

Funt, B. V. (1983). A parallel-process model of mentalrotation. Cognitive Science, 7, 67-93.

Garner, W. R. (1981). The role of configuration in theidentification of visually degraded words. Memory &Cognition, 9, 445-452.

Haber, R. N., & Schindler, R. M. (1981). Error inproofreading: Evidence of syntactic control of letterprocessing? Journal of Experimental Psychology: Hu-man Perception and Performance, 7, 573-579.

Henderson, L. (1980). Wholistic models of feature analysis

in word recognition: A critical examination. In P. A.

Kolers, M. E. Wrolstad, & H. Bouma (Eds.), Processingof visible language 2 (pp. 207-218). New York: Plenum.

Henderson, L. (1982). Orthography and word recognitionin reading. London: Academic Press.

Hochberg, J., & Gellman, L. (1977). The effect of land-mark features on "mental rotation" times. Memory &

Cognition, 5, 23-26.Hock, H. S., & Tromley, C. L. (1978). Mental rotation

and perceptual uprightness. Perception & Psychophysics,24, 529-533.

Huey, E. B. (1908). The psychology and pedagogy ofreading (Reprinted 1968). Cambridge, MA: MIT Press.

Johnson, N. F. (1975). On the function of letters in wordidentifiction: Some data and a preliminary model.

Journal of Verbal Learning and Verbal Behavior, 14,

17-29.Johnson, N. F. (1981). Integration processes in word

recognition. In O. J. L. Tzeng &. H. Singer (Eds.),

Perception of print: Reading research in experimentalpsychology (pp. 29-63). Hillsdale, NJ: Erlbaum.

Jolicoeur, P., & Landau, M. J. (1984). Effects of orientationon the identification of simple visual patterns. Canadian

Journal of Psychology, 38, 80-93.Kolers, P. A., & Perkins, D. N. (1975). Spatial and

ordinal components of form perception and literacy.Cognitive Psychology, 7, 228-267.

Koriat, A. (1984). Reading without vowels: Lexical accessin Hebrew. In H. Bouma & D. G. Bouwhuis (Eds.),Attention and performance X: Control of language

processes (pp. 227-241). Hillsdale, NJ: Erlbaum.Koriat, A., & Norman, J. (1984). What is rotated in

mental rotation? Journal of Experimental Psychology:

Learning, Memory, and Cognition, 10, 421-434.Koriat, A., & Norman, J. (in press). Mental rotation and

visual familiarity. Perception & Psychophysics.Kosslyn, S. M. (1980). Image and mind. Cambridge,

MA: Harvard University Press.

Lawry, J. A. (1980). The interfering effect of wordperception on letter identification. Perception & Psy-chophysics, 28, 577-588.

McClelland, J. L. (1976). Preliminary letter identification

in the perception of words and nonwords. Journal ofExperimental Psychology: Human Perception and Per-

formance, 2, 80-91.McClelland, J. L., & Rumelhart, D. E. (1981). An

interactive activation model of context effects in letterperception: Part 1. An account of basic findings.Psychological Review, 88, 375-407.

Monk, A. F., & Hulme, C. (1983). Errors in proofreading:Evidence for the use of word shape in word recognition.Memory & Cognition, 11, 16-23.

Navon, D. (1978). Perception of misoriented words andletter strings. Canadian Journal of Psychology, 32,129-140.

Paap, K. R., Newsome, S. L., McDonald, J. E., &Schvaneveldt, R. W. (1982). An activation-verificationmodel for letter and word recognition: The word-superiority effect. Psychological Review, 89, 573-594.

Paap, K. R., Newsome, S. L., & Noel, R. W. (1984).Word shape's in poor shape for the race to the lexicon.Journal of Experimental Psychology: Human Perceptionand Performance, 10, 413-428.

Patterson, K., & Kay, J. (1982). Letter-by-letter reading:


Psychological descriptions of a neurological syndrome.Quarterly Journal of Experimental Psychology. 34A,411-441.

Pylyshyn, Z. (1979). The rate of "mental rotation" ofimages: A test of a holistic analogue hypothesis. Memory& Cognition, 7, 19-28.

Robertson, L. C, & Palmer, S. E. (1983). Holistic processesin the perception and transformation of disorientedfigures. Journal of Experimental Psychology: HumanPerception and Performance, 9, 203-214.

Rudnicky, A. I., & Kolers, P. A. (1984). Size and case oftype as stimuli in reading. Journal of ExperimentalPsychology: Human Perception and Performance, 10,

231-249.Samuel, A. G., van Santen, J. P. H., & Johnston, J. C.

(1982). Length effects in word perception: We is belterthan I but worse than they or them. Journal ofExperimental Psychology: Human Perception anil Per-

formance, 8, 91-105.Santa, T. L., Santa, C. A., & Smith, E. E. (1977). Units

of word recognition: Evidence for use of multipleunits. Perception & Psychophysics, 22, 585-591.

Shallice, T. (1981). Neurological impairment of cognitiveprocesses. British Medical Bulletin, 37, 187-192.

Shepard, R. N., & Cooper, L. A. (1982). Mental images

and their transformations. Cambridge, MA: M. I. T.Press.

Shepard, R. N., & Metzler, J. (1971). Mental rotation ofthree dimensional objects. Science, 171, 701-703.

Shwartz, S. P. (1979). Studies of mental image rotation:Implications for a computer simulation of visual im-

agery. Unpublished doctoral dissertation, Johns HopkinsUniversity, Baltimore.

Warrington, E. K., & Shallice, T. (1980). Word-formdyslexia. Brain, 103, 99-112.

Yuille, J. C., & Steiger, J. H. (1982). Nonholistic processingin mental rotation: Some suggestive evidence. Perception& Psychophysics, 31, 201-209.

Received May 31, 1984Revision received November 26, 1984

Psychological Documents to Be Discontinued

At its February 2-3, 1985, meeting, the Council of Representatives voted to ceasepublication of Psychological Documents (formerly the Journal Supplement AbstractService) as of December 31, 1985, with the publication of the December 1985 issueof the catalog. Continued low submissions, decreasing usage, and rising costs forfulfillment of paper and microfiche copies of documents were reasons given for dis-continuing publication of the alternative format publication, which was begun in 1971as an "experimental" publication.

Authors who wished to submit documents for publication consideration in 1985were required to do so by July 1. Authors revising documents were required to completeall revisions and submit them for final review no later than July 1. Fulfillment oforders for paper and microfiche copies of documents presently in the system and ofthose documents entered during 1985 will continue through December 31, 1986.