Automatic Semantic Processing in a Picture-Word Interference Task Author(s): Richard R. Rosinski, Roberta Michnick Golinkoff, Karen S.

Kukish Reviewed work(s): Source: Child Development, Vol. 46, No. 1 (Mar., 1975), pp. 247-253 Published by: Blackwell Publishing on behalf of the Society for Research in Child Development Stable URL: http://www.jstor.org/stable/1128859 . Accessed: 18/01/2012 05:09
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at . http://www.jstor.org/page/info/about/policies/terms.jsp JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Blackwell Publishing and Society for Research in Child Development are collaborating with JSTOR to digitize, preserve and extend access to Child Development.

http://www.jstor.org

AutomaticSemanticProcessingin a

Picture-Word Interference Task

Richard R. Rosinski, Roberta Michnick Golinkoff, and Karen S. Kukish

University of Pittsburgh

RICHARD R.; GOLINKOFF, ROBERTA MICHNICK; and KUKISH, KAREN S. Automatic Semantic Processing in a Picture-WordInterference Task. CHILD DEVELOPMENT, 1975, 46, 247253. While semantic development has been alleged to proceed slowly, reading instruction begins early in the child's school career. Yet to date, little research has been addressed to understandinghow beginning readers extract meaning from the printed word. This paper reports 2 experiments which measured latencies in a picture-word interference task to assess semantic processing. Results suggest that picture-word interference is partly semantically based and that children and adults experience an equivalent amount of semantic interference. The results are interpreted as indicating that even by the second grade, children are sensitive to the meaning of printed words.
ROSINSKI,

Until recently, psychologists often assumed that learning to read was a simple matter of decoding graphic symbols into auditory (or articulatory) representations. These representations were then interpreted by processes involved in aural language comprehension. Thus, reading for meaning was assumed to be mediated by auditory linguistic processes (Gough 1972), and improvements in reading ability were partly attributed to learning spelling-to-sound correspondences (Gibson, Osser, & Pick 1963). It is now known, however, that although auditory mediation may play a role in initial learning, it is neither necessary nor sufficient for reading comprehension to occur. Such mediation cannot be sufficient, since children and adults can decode accurately and still not understand what they are reading (Cromer 1970). Nor is auditory mediation necessary, since deaf children learn to read with little opportunity to experience spelling-to-sound correspondence (Gibson, Shurcliff, & Yonas 1970). Although newer theories of reading for meaning have repudiated the decoding hypothesis (Gibson & Levin 1974; Neisser 1969), little is known about the extraction of meaning from the printed word. One means by which an individual can

read for meaning has been suggested by Kolers (1970). He has argued that the skilled reader does not operate in terms of words but, rather, operates on semantic relations even to the point of disregarding, in a certain sense, the actual printed text. In two studies (Kolers 1966a, 1966b) using bilingual adults, it was shown that both isolated words and connected discourse printed in two languages are perceived and remembered preferentially in terms of their meaning rather than in terms of their appearance or sound. Perfetti and Garson (1973) also found that semantic relations contained in printed text were remembered far more than either lexical or superficial grammatical information. One implication of Kolers's hypothesis that printed words directly access their concepts in semantic memory is that it should be difficult to ignore the meanings of words. Willows (1974) found evidence which supports this implication. Using Neisser's (1969) selective reading task in which subjects are instructed to ignore printed material appearing between the lines of relevant text, Willows found that skilled readers committed more semantic intrusion errors than did unskilled readers. Willows concluded that poor readers focus more on the

The authorswish to thank Dr. Charles A. Perfetti for his comments on the manuscriptand Mr. Tom Ritchey and the staff and pupils of Skyline School for their cooperation.First author's address: Department of Psychology, University of Pittsburgh, Pittsburgh, Pennsylvania 15260. Second author now at the University of Delaware.
[Child Development, 1975, 46, 247-253. ? 1975 by the Societyfor Research Child Development, in Inc. All rightsreserved.]

248

Child Development

visual aspects of the display, whereas good readers access word meanings more directly. Research using the Stroop color-word interference test may also support the direct semantic access notion. With both children and adults, Schiller (1966) found that it took longer to label colors when conflicting color words were to be ignored than in the absence of such conflict. Although this interference effect may have several other bases, the inability to ignore the meaning of the printed words may play a large role. Although skilled readers may directly access the meaning of a word, such accessibility may develop over the course of learning to read. Children may not find the meanings of words as compelling as do adults. Gibson, Barron, and Garber (1972) suggested that for children the meaning of pictures is more readily accessible than the meaning of words. The purpose of the present experiment was to chart the course of children's automatic pickup of meaning from printed words and pictures.

In the three experimentalconditions, the 100%, in which each word matched its drawing; 50%, in which only half the words matched their drawings (fig. 1 depicts the 50% stimulus sheet); and 0%, in which no 0% conditions, incongruentwords and drawings were randomlypaired, with the constraint that semantic categories were not crossed. Procedure.-Each subject was tested individually. Half of the subjects in each grade were assigned either to read the words or to label the drawings and at the beginning of testing were given the correspondingwarm-up task. Then subjectswere instructedto continue reading the words (or labeling pictures) and to ignore the pictures (or the words). All subjects received the three experimentalconditions in counterbalancedorder and were instructed to read the words or label the picturesas fast as they could. Latencies were measured with a stopwatch (to the nearest0.1 second) and were recorded,as were errors. Results The latency data for each condition as a function of grade and labeling task are prewords and pictures matched. In the 50% and words were superimposed on the pictures in each cell under three degrees of congruence:

Experiment I To assess the extent to which semantic content wordsandpictures automatically of was a interference was task registered, picture-word used. Subjectswere requiredeither to label picturesor to read wordswhen picturesand wordswere superimposed. the meaningof If the distractor items was automatically picked the up, time to perform taskshouldhave been longerwhen the meaningof the picturesand wordsdid not agree.
Method 24 sixth graders (mean age, 11-4), and 24 adult college volunteers. An equal number of males and females served at each combination of grade and condition.All childrenwere classified by the school as average or above in reading ability. Subjects.-The subjects in the first experiment were 24 second graders (mean age, 7-7),

sented in figure 2. A preliminary analysis revealed no effect of sex, so the data were pooled. A 3 (grade level) X 2 (label task) X 3 (experimental condition) analysis of variance with repeated measures on the last factor was performed on these latencies. The effect of age was significant, F(2,66) = 141.13, p < .01, and accounted for a major portion (51%) of the between-subject variance. This effect may merely show that younger subjects take longer to respond than older ones. The effect of labeling was significant, F(1,66) = 183.8, p < .01, increased as a function of decreasing congruence between words and pictures, F (2,132) = 102.36, p < .01. A significant grade x labeling task interaction, F(2,66) = 12.05, p < .01, revealed an age-related effect of picture labeling versus word reading, with picture-labeling latencies higher at the lower grade levels. Thus, word interference in the picture-labeling task was

with picture labeling taking longer than word readingat all grade levels. In addition,latencies

Materials.-Prior to the experimentalconditions, subjects received practice on an appropriate warm-up sheet. Stimulus sheets were X 11-inch paper divided into 20 made of 8 cells of equal size. On the picture warm-up, each cell contained a line drawing of either a

one of the animals or objects.

common animal or an object. On the word warm-up, each cell contained a word naming

greater than picture interferencein the wordreading task, and this differencein interference varied with age. Analysis of the simple main effects, however, indicated that significant

Rosinski, Golinkoff, and Kukish 249

FIG. 1.-Sample stimulus sheet from Experiment I; 50% congruence between pictures and words grade and task effects were obtained at all levels. Although the magnitude of the grade effect was larger in the picture-labeling than in the word-reading task, grade effects were substantial in both tasks, p's < .01. Similarly, although the task effect decreased over increasing grade level, it was significant at all grade levels, p's < .01. A significant labeling task X condition interaction, F(2,132) = 49.83, p < .01, indicated that latencies increased more in the picture-labeling than in the word-reading task as the percentage of congruence decreased. There were no labeling task differences in the 100% condition, F(1,33) = 6.38, .01 < p < .05, but significant labeling task effects were observed in the 50% condition, F(1,33) = 99.84, p < .01, and the 0% condition, F (1,33) = 134.95, p < .01. Congruence condition effects were not found for the word-reading task, F(1,66) = 4.70, p > .01, but were found when the task = required the picture to be labeled, F(1,66) 147.49, p < .01. The presence of conflicting words substantially increased picture-labeling latency. In addition, a significant grade x congruence condition interaction, F (4,132) = 20.98, p < .01, was observed. Although simple effects tests showed a significant grade effect in all conditions, p's < .01, the grade effect increased as the lack of congruence between picture and word increased. An effect of congruence condition was observed at second grade, F (1,66) = 119.76, p < .01, and at sixth grade, F(1,66)

250

Child Development 46 GRADE " 2 6 ...GRADE

42,
z
C C-

.-ICTURE

34m
26

"*"26 0%22SW

mmmm PICTURE ........ ORD

2
I I

10%

50% CONDITION

0%

FIG. 2.-Mean latencies for three grade levels of picture-labeling and word-reading tasks as a function of stimulus congruence. = 19.57, p < .01, but not at the adult level, F(1,66) = 4.98, .01 < p < .05. This pattern of results indicated that the lack of pictureword congruence disrupts the performance of younger subjects more than that of older subjects. The significant main effect of grade, as well as part of the grade X condition interaction, may merely result from the fact that children have longer response times in most tasks than adults. To explore the effect of grade level, a further analysis of interference was carried out using the difference between each subject's performance in the 100% and 0% condition over labeling tasks. The 100% congruence condition can be affected by age differences in response execution, label availability, scan rate, reading speed, etc. The 0% congruence condition can be affected by all these plus interference between picture and word processing. Consequently, the difference between the 100% and 0% conditions provides a measure of processing interference. A one-way analysis of variance of these differences over grade confirmed that younger subjects showed more processing interference than older subjects, F(2,48) = 6.47, p < .01. This indicates that the grade effects were not merely due to differences in response times. Newman-Keuls comparisons showed that processing interference at second grade was significantly greater than at either the sixth grade or adult level (p's < .01), which did not differ between themselves (p > .01). The three-way interaction (grade X task X congruence condition) (see fig. 2), F (4, 132) = 6.59, p < .01, indicates that the effect of condition is less for word reading than picture labeling and that it increases more for younger subjects than for older. Few errors were made at any grade level. Mean number of errors ranged from 0 to 1.1 across all conditions and grades. All were errors of intrusion in which the word in a cell was reported rather than the picture that was to be labeled. In the word-reading task, no errors were made. Discussion The presence of noncongruent words and pictures substantially increased the amount of time taken to perform the task. These results suggest that some processing of the distractor items occurred at all age levels. Furthermore, the amount of interference for the two tasks was not equivalent; picture labeling in the absence of word-picture congruence was more difficult for all ages than was word reading.

Rosinski, Golinkoff,and Kukish 251

This conclusion is supported by the error data, which showed that words intruded more than pictures. It has been suggested (Gibson et al. 1972) that picture processing is more highly developed in young children than word processing and that several years of experience with words might be necessary to reverse this difference. Our results do not support this expectation, since even for the youngest subjects words interfered more than pictures. This is further supported by error data: words intruded a number of times, but pictures did not intrude at all. Clearly, words enjoy either an attentional or a processing advantage over pictures, even after a single year of learning to read. The difference score analysis revealed that the effect of interfering stimuli was greater for younger than for older subjects. If the connection between the printed word and its semantic referent develops over the course of learning to read, one would expect that interference would increase over age. One explanation for decreasing interference over age may be that interference, in the present experiment, is not the result of a single process but of a combination of factors. For example, it is well known that young children's ability to filter out extraneous material is relatively poor (Gibson 1969). Given this difference in attentional ability, ignoring the presence of the superimposed words on pictures may have been more difficult for younger subjects than for older. Younger subjects, since they were less able to ignore the distractor items, may have adopted a strategy of proceeding slowly so as not to make errors. Our results may also have been influenced by developmental differences in response competition. Perceptual processing of both items in a cell may have resulted in the initiation of motor responses to both. In the 100% congruent condition, either of these responses would be correct and appropriate to the task. In the noncongruent conditions, however, only one of the responses would be correct and the subject would have to decide which to make. Alternatively, the locus of the interference may involve cognitive processing mechanisms rather than input or response processes. It could be argued, for example, that as the picture was being labeled, a connection was made with its conceptual category in semantic memory. The meaning of the distractor word was also picked up, and in the noncongruent conditions the discrepancy between these two conceptual categories resulted in interference.

Experiment II
Experiment II was designed to determine whether picture-word interference was semantically based by manipulating the semantic content of the distractors. Since our primary concern was with the automatic processing of the words and the pickup of word meaning, only the picture-labeling task was used. The 100% and 0% congruent conditions of the previous experiment were used. In addition, a third condition superimposing CVC nonsense words on the pictures was also constructed. Three competing expectations can be offered based on the hypotheses discussed above. If interference is the result of input competition (attention), the two conditions would not differ, since the trigrams would also increase the processing load. If response competition occurred, the trigram condition would take longer, since their unfamiliarity would make them harder to read. If, however, the interference was semantically based, the trigram condition latencies should be significantly lower than the 0% condition latencies, since trigrams are less meaningful than the words.

Method
second graders (mean Subjects.-Twelve age, 7-5), 12 sixth graders (mean age, 11-7) (from the same classroom as those used in Experiment I), and 12 adult college student volunteers, half male and half female, served as subjects. Materials.-From Experiment I, the stimulus sheets from the 100% and 0% conditions were used. In addition, a third test sheet was constructed in which CVC pronounceable nonsense trigrams of association values ranging from 35% to 65% (Archer 1960) were paired with the pictures. The trigrams were selected so that they rhymed with the real words (with two exceptions). The warm-up sheets in this experiment, as in the preceding one, consisted of 20 cells, each containing one of the words, trigrams, or pictures. Procedure.-Each subject was tested individually. The word and trigram warm-up conditions were given to all subjects in counterbalanced order. Children were tested 1 day after the administration of these warm-up conditions, adults 1 hour after. In testing, subjects were first given a picture warm-up sheet and

252

Child Development
though an effect of grade was observed at both the 0% condition, F(2,33) = 52.75, p < .01, and the trigram condition, F(2,33) = 62.47, p < .01, there was no differential effect in the 100% condition, F (2,33) = 4.47, p > .01. The fact that latencies over grade did not differ when there was congruence between pictures and words (100% condition) but did differ over age when the pictures and words were noncongruent (0% and trigram conditions) suggests that the magnitude of overall interference is a function of age, as was found in Experiment I. To evaluate the magnitude of the semantic interference effect, analyses of variance were performed on difference scores used as measures of semantic interference. Both the 0% condition and the trigram condition have noncongruent picture-word combinations. Differences between these two conditions, then, must be due to the differential semantic loadings of words and CVC trigrams. Although there was a significant semantic effect of the distractor words, the magnitude of the difference between the 0% and trigram conditions did not change over grade, F < 1, p > .01.

then presented with the three experimental conditions in counterbalanced order. Subjects were instructed to label the pictures as fast as they could and to ignore the words. Latencies (to the nearest 0.1 second) and errors were recorded. Results The latency data as a function of grade are presented in figure 3. A 3 (grade level) X 3 (condition) analysis of variance with repeated measures on the conditions factor was performed. There was an overall effect of age, F (2,33) = 61.83, p < .01, with older subjects faster at the task than younger ones. The overall effect of condition was also significant, F (2,66) = 141.99, p < .01. A Newman-Keuls test showed that all conditions were significantly different from each other. Thus, as predicted by the semantic interference hypothesis, the latencies for the trigram condition were significantly less than those for the 0% condition. In addition, the interaction between grade and condition was significant, F (4,66) = 14.54, p < .01. Analysis of simple effects indicated a significant effect of condition at all three grade levels (p's < .01). However, al-

4 26 2 2

=18 LU

0%

10

2 GRADE

6 GRADE

ADULT

FIc. 3.-Mean picture-labeling latencies in Experiment II as a function of stimulus condition and grade

Rosinski, Golinkoff, and Kukish 253

As in the preceding experiments, few errors were made by the subjects. Discussion As predicted, the semantic content of the words results in greater interference compared with nonsense syllables. The results obtained support the hypothesis that the interference effect in Experiments I and II had a significant semantic component. Consequently, the meaning of the common words used in this experiment must be automatically processed even at the youngest age level used. The finding from Experiment I that overall interference decreased with increasing age was replicated. Our results indicate that this decrease in interference may be ascribed to developmental changes in attentional or response processes. In both studies, subjects were instructed to ignore the distractor items and not choose between items. Yet words provided more interference than pictures, meaningful words more than nonsense syllables. In addition, although overall interference decreased with age, that portion of the effect due to differing semantic content did not. Thus, these results indicate that the development of picture-word interference reflects the action of two processes: an apparently stable semantic access and an increasingly efficient stimulus and/or response filter. Our results show that, within the limitations of this paradigm, the magnitude of the semantic interference effect is constant over age. Although semantic knowledge must involve continual refinement and differentiation as children learn about the domains to which words apply (Clark 1973; McNeill 1970), such refinement was not assessed in our task. Our interference task does indicate that the meaning of printed words is extracted by children after only 1 year of reading experience. The mapping of printed word to meaning must take place in the early stages of learning to read. This is true, at least for average and aboveaverage readers. In conclusion, these experiments suggest that the extraction of meaning from unrelated printed words may be an automatic process even in young readers. ment and the acquisition of language. New York: Academic Press, 1973. Cromer, W. The difference model: a new explanation for some reading difficulties. Journal of Educational Psychology, 1970, 61, 471-483. Gibson, E. J. Principles of perceptual learning and development. New York: Appleton-CenturyCrofts, 1969. Gibson, E. J.; Barron,R. W.; & Garber,E. E. The developmental convergence of meaning for words and pictures. In The relationship between perceptual development and the acquisition of reading skill, appendix to final report. Project No. 90046, Grant No. OEG2-9-42-446-1071 (010) between Cornell University and the U.S. Office of Education, 1972. Gibson, E. J., & Levin, H. The psychology of reading. Cambridge, Mass.: M.I.T. Press, 1974. Gibson, E. J.; Osser, H.; & Pick, A. A study in the development of grapheme-phoneme correspondence. Journal of Verbal Learning and Verbal Behavior, 1963, 2, 142-146. Gibson, E. J.; Shurcliff, E.; & Yonas, A. Utilization of spelling patterns in deaf and hearing subjects. In H. Levin & J. P. Williams (Eds.), Basic studies in reading. New York: Basic, 1970. Gough, P. B. One second of reading. In J. F. Kavanagh& I. G. Mattingly (Eds.), Language by ear and by eye. Cambridge, Mass.: M.I.T. Press, 1972. Kolers, P. A. Interlingual facilitation of short-term memory. Journal of Verbal Learning and Verbal Behavior, 1966, 5, 314-319. (a) Kolers, P. A. Reading and talking bilingually. American Journal of Psychology, 1966, 79, 357-376. (b) Kolers, P. A. Three stages of-reading. In H. Levin & J. P. Williams (Eds.), Basic studies in reading. New York: Basic, 1970. McNeill, D. The acquisition of language. New York: Harper, 1970. Neisser, U. Selective reading: a method for the study of visual attention. Presented at the Nineteenth International Congress of Psychology, London, 1969. Perfetti, C. A., & Garson, B. The forgetting of linguistic information after reading. Journal of Educational Psychology, 1973, 65, 135139. Schiller, P. H. Developmental study of colorword interference. Journal of Experimental Psychology, 1966, 72(1), 105-108. Willows, D. M. Reading between the lines: selective attention in good and poor readers. Child Development, 1974, 45, 408-415.

References
Archer, E. J. Re-evaluation of the meaningfulness of all possible CVC trigrams. Psychological Monographs, 1960, 74, No. 497. Clark, E. V. What's in a word? On the child's acquisition of semantics in his first language. In T. E. Moore (Ed.), Cognitive develop-