Journal of Reading Behavior 1979, Vol. XI, No. 2 WORD RECOGNITION BY SECOND GRADERS: THE UNIT OF PERCEPTION AND INTERRELATIONSHIPS AMONG ACCURACY, LATENCY, AND COMPREHENSION 1 Christine McCormick and S. Jay Samuels University of Minnesota Abstract Two issues were investigated: the first examined the relationships among accuracy and latency of word recognition and comprehension by non-fluent readers, and the second examined whether component letter or holistic processing was used in word recognition by these same readers. Speed and accuracy of word recognition were measured on individual words, Literal comprehension was measured for the same words presented in meaningful context. The unit of perception was measured by the relationship between latency of word recognition and word length. If students were using component processing, latency would increase with word length, but if holistic processing were used, there would be no increase in latency with length. Results of this study indicated that accuracy and latency were each significantly related to comprehension for both first- and second-grade words, with evidence for latency influencing comprehension scores to a greater extent on the first-grade words. With regard to the unit of perception, in general there was evidence of component letter processing for all subjects with the highly accurate readers showing a tendency towards more holistic processing. This research investigates two problems in reading. One problem undertaken in this study pertains to the relationships among word recognition latency, word recognition accuracy and literal comprehension by beginning readers. The second problem pertains to the size of the perceptual unit used by beginning readers in word recognition. This question focuses on the component letter-holistic processing controversy which has been with us for nearly a century (Cattell, 1947). Under component letter processing it is assumed that the perceptual unit may be as small as the letter; under holistic processing it is assumed that the unit is considerably larger, perhaps as large as the word itself. We have chosen to study these two seemingly disparate problemsthe question of the unit of perception for beginning readers and the relationships among word recognition speed, accuracy and literal comprehension in these same readersbecause both may be examined using word recognition latency.
108 Journal of Reading Behavior Accuracy, Latency and Literal Comprehension While the goal of reading is comprehension, one can assume that decoding is a prerequisite (Gibson & Levin, 1975). Decoding includes two componentsaccuracy of word recognition and also speed of word recognition. Although factors such as accuracy have been previously explored (Shankweiler & Liberman, 1971), the role of word recognition latency has not been extensively investigated. The LaBerge and Samuels (1974) theory of automatic information processing suggests that fluent readers are able to decode words without attention to decoding. That ability frees attention for the crucial task of comprehension. The importance of response latency is that it may be an indicator of automatic processing or automaticity. Terry, Samuels, and LaBerge (1976) assume that slowness hi word recognition suggests that the reader has had to use attention in order to decode the words, and that, conversely, rapidity of word recognition suggests that the decoding has been done automatically. There is evidence to support the view that recognition latency is an indicator of automaticity. LaBerge (1972) had subjects press a button if they perceived two highly familiar visual forms to be identical. He assumed that making recognition responses to highly familiar visual forms was automatic. He then asked the subjects to press the button if they perceived a set of unfamiliar forms to be identical. After extended practice on these unfamiliar forms, speed of response became as fast as to the familiar forms. At this point, LaBerge assumed that perceptual learning had reached the automatic stage. Another piece of evidence is provided by Perfetti and Hogaboam (1975) who found that beginning readers with high comprehension scores had shorter vocalization latencies to single printed words than beginning readers with lower comprehension scores. Judd and Glaser (1969) reported that latency of response could be used as an indicator of overlearning, or learning beyond mere accuracy. Perceptual Unit The size of the perceptual unit used by the reader is relevant both to automaticity in decoding and to the serial-parallel processing controversy in word recognition. An assumption is made that as one develops skill in decoding and becomes automatic at the task, the size of the unit of perception increases. Thus, beginning readers who are non-automatic should have smaller units of perception, and their processing should appear to be serial. On the other hand, skilled readers who are automatic at decoding should have larger perceptual units and their processing should appear to be parallel. Recently analyzed data support this assumption (Samuels, LaBerge & Bremer, 1978). Generally, component letter models of word recognition (Gough, 1971) postulates that letters are processed prior to the activation of a word code, while holistic processing models (Cattell, 1947, LaBerge & Samuels, 1974) postulate that letters need not be processed in order for a word code to be activated. In reviewing the literature of the serial-parallel issue, Bradshaw (1975) has found considerable evidence in favor of a model of word perception involving perceptual units larger than the individual letter. The unit used in perceptual processing he concludes, is influenced by both the skill of the reader and the nature of the reading task. Terry's et al. (1976) study demonstrated how the size of perceptual unit is
Word Recognition by Second Graders 109 related to the nature and difficulty of the reading task. In this study, fluent readers had to recognize animal words presented either in regular orthography or in mirrorimage. She found evidence for holistic processing with regular orthography and component processing with mirror-image text, and concluded that the unit of perception for skilled readers might be at the word level when familiar words were presented but drop down to the letter level when the decoding task became difficult. A developmental study of the unit of perceptual processing in word recognition for grades two, four, six and college indicated that as the students increase in reading skill, the unit of perception increases from the letter level to the word level (Samuels, etal., 1978). Subjects METHOD Twenty-six second-grade children from a midwestern public school were used as subjects. Their ages ranged from 6.5 to 8.25 years with an average of 7.6 years; their IQ's ranged from 85 to 134 with an average IQ of 111. The study took place midway through the school year. Design Correlational analyses including partial correlation and multiple-regression were used to explore the relationships among word recognition accuracy, word recognition latency and comprehension. The unit of processing question was examined by dividing the subjects, post-hoc, on the basis of accuracy scores and then running a one-way analysis of variance for each accuracy category separately across increasing word lengths. In this one-way ANOVA, latency of words accurately recognized was the dependent variable. Following the over-all F test, trend analyses and Neuman-Keuls tests were used to examine the latency data in more detail. Materials The stimuli used to collect data on word recognition accuracy and latency were prepared by making slides of each different word in the first-grade and second-grade Gray Oral Reading Tests (1967) as well as each different word in two selections from the "reading for comprehension" section óf the Science Research Associates Achievement Series, grades 1-2, Form D (1963). The slides of words were placed in a carousel tray so that word length was systematically balanced between blocks of words ranging in length from two to nine letters. A total of 69 first-grade and a total of 87 second-grade words were presented to each child; the order of presentation was the same for all children. A P-L Systems Audio-Visual Trainer presented the words to the child being tested. This apparatus contained a carousel projector which projected the word slides individually onto a small screen about eight inches by eight inches. The A-V Trainer recorded the latency (in tenths of seconds) between the projection of the word and the child's naming of that word into the voice activated timer. The A-V Trainer printed out the latency data for each word presented on the screen. Materials used for measuring comprehension consisted of the first- and second-
110 Journal of Reading Behavior grade stories from the Gray Oral Reading Tests and the Science Research Associates Achievement Series mentioned above. The words which each child read in context for the comprehension measures were the same words which were presented individually for the data on accuracy and latency. Procedure The children were individually brought into a testing room in their school and asked to sit in front of the A-V Trainer. The experimenter then said to the child: I have some words that I would like you to read for me when they are shown on this screen. Please say the word as soon as you know it into this microphone. (The child was given the microphone to hold.) Some of the words are easy and some may be hard. Let's practice a little bit. At this point the practice words [I, a, an) were shown on the screen, one at a time, and the child named them into the microphone. Following this, each different word from the first-grade Gray Oral and the first-grade SRA stories was shown individually on the A-V Trainer. The experimenter sat near the child and recorded the accuracy of word recognition while the A-V Trainer recorded the response latency. A limit of seven seconds was programmed into the machine; if the word was not named before seven seconds elapsed, the word was considered not recognized and the machine advanced. If the child could not read the word, the experimenter told him the word after seven seconds. No child required close to seven seconds to accurately recognize a word. After the first-grade words were presented individually, the experimenter said, "Now 1 have a short story I would like you to read out loud. I'm going to ask you to tell me all about it when you finish." The child was given a sheet of paper with, the first-grade Gray Oral Reading Test typed in primary type. No help of any type was given to the child during this portion of the procedure. The experimenter took the story back when the child finished reading. The child was then asked to tell all about the story. The child's reply was recorded on a cassette tape recorder and later transcribed. After a non-specific prompt ("Can you tell me anything more?"] the child was asked the literal comprehension questions accompanying the Gray Oral. The child was then asked to read the SRA story to himself and was reminded that he would be asked to tell the experimenter all about it. He gave this first-grade SRA story back to the experimenter as soon as he finished reading it. Again the child was asked to tell everything he could remember into the tape recorder. Following this, he was asked literal comprehension questions covering the story. The free recall of the stories was scored by giving one point for each phrase remembered, using a procedure similar to the scoring used in the Durrell Analysis of Reading Difficulty (1955). The scoring procedure gave an estimate of the amount of literal information comprehended by the child for the oral and silent passages. First-grade words from the Gray Oral and the SRA Achievement Series were used in the initial testing session. If a child could not answer at least two of the questions from this Gray Oral story he was not given the more difficult second-grade words and stories, N = 24 for the second-grade words. Second-grade words from the Gray oral and the SRA Achievement Series were used in the second testing session.
Word Recognition by Second Graders 111 The same procedure was used for the second testing session on the following day but without the warm-up trials. Words in the second-grade stories which had already been presented in the first-grade list were not repeated, but the accuracy and latency scores of these 13 words were included in the second-grade accuracy and latency data. To summarize the procedure, the accuracy score was the percentage of all words the child named correctly when these were presented on the A-V Trainer. The latency score was handled by the A-V Trainer which measured voice onset. Only latency scores of words accurately recognized were used in all the data analyses. Comprehension was figured from free recall and responses to literal comprehension questions. Each child responded to both free recall and questions on both oral and _silent stories. The comprehension score was a total score, achieved by summing the points on both modes of recall on both oral and silent stories. Da fa Analysis Data from the first-grade and second-grade words were analysed independently. The relationships among speed and accuracy of word recognition and comprehension were explored using correlation, partial correlation, and regression. Correlations among accuracy, latency and comprehension were calculated. Partial correlations were also calculated with latency and with accuracy as the controlled variable. The percentage of words accurately recognized and the latency scores for these words entered the regression singly and then together. ' The question concerning the unit used in processing words was examined by dividing the group into "greater than or equal to 95 7o total accuracy" and "less than 957o total accuracy" on first- and second-grade words. A one-way analysis of variance was run on average latencies of words accurately recognized across increasing word lengths for each accuracy category. Neuman-Keuls tests were used to discover which specific word lengths significantly differed in response latency. Trend analysis was carried out to determine which function (linear or quadratic] best described the latency data. The rationale for dividing the group on the basis of accuracy scores for the unit of perception analyses was based on suggestive data from Terry et al. (1976), our hypothesis being that good readers would process words more rapidly using units larger than the letter, whereas less able readers encountering difficulty in decoding would process words less rapidly using units as small as the letter. RESULTS Relationships Among Speed and Accuracy of Word Recognition and Comprehension Table 1 shows the Pearson correlation coefficients among accuracy, latency of words accurately recognized, and comprehension. The correlations indicate that latency and accuracy were significantly correlated with each other on both firstgrade and second-grade words. The absolute value of the correlation was.88 for both grade levels, significant beyond the.001 level. Accuracy and latency were each significantly correlated with the total comprehension score on both first- and second-
112 Journal of Reading Behavior TABLE 1 Correlation Coefficients First Grade Words Second Grade Words Variables N Pearson r Sig. Level N Pearson r Sig. Level Latency with Accuracy Latency with Total Comprehension Accuracy with Total Comprehension Silent Comprehension with Oral Comprehension 26 26 26 26 -.88 -.54.49.52.001.002.006.003 24 24 24 24 -.88 -.56.70.64.001.001.001.001 Latency: average latency between onset of word and correct recognition Accuracy: Proportion of word list accurately recognized Total Comprehension: total oral and silent comprehension scores on Gray Oral and SRA stories grade words. The absolute values of the correlations ranged from.49 to.70 and all were significant beyond the.006 level. In summary, high accuracy and rapid word recognition were associated with high comprehension. Partial correlation. The partial rbetween accuracy and comprehension, controlling on latency, was.11, p =.306, on first-grade words and.53, p =.003, on secondgrade words. The partial r between latency and comprehension, controlling on accuracy, was.28, p =.084, on first-grade words and.17, p =.203, on second-grade words. When latency was controlled the association between latency and comprehension was not significant for first-grade words but was significant for secondgrade words. When accuracy was controlled the association between latency and comprehension approached significance more closely on first-grade words. Even though latency and accuracy were highly correlated, the influence of latency independent of accuracy was greater on first-grade words. The partial r can also be interpreted by noting that the smaller the partial r, the more important the controlled variable, (Ferguson, 1967). When latency was controlled the impact on the association between accuracy and comprehension was much greater for the first-grade words than for the second-grade words, indicating that latency was more influential on the data from first-grade words. Regression. Regression analyses show that on first-grade words the R 1 for accuracy alone was.24, p =.Oil, and the R 1 for latency alone was.30, p =.016. Entering both accuracy and latency into the regression, the R 2 =.30, p =.016. Using second-grade words the R 1 =.49, p <.001, when accuracy alone was entered, and the R 1 =.32, p =.003, when latency alone was entered. Entering both accuracy and latency, the R 2 =.50, p <.001. Since the correlation between accuracy and latency
Word Recognition by Second Graders 113 was high (r = -.88 with either first- or second-grade words), the combination of accuracy and latency did not add much to the R 1 for comprehension. Unit of Processing Figure 1 provides graphic illustration of how accuracy and latency variables can be examined in terms of the unit of processing issue. The group was split into "greater than or equal to 95% total accuracy" and "less than 95% total accuracy" categories on both first- and second-grade words. Figure 1 and Table 2 indicate that the less accurate students show more of an increase in latency as words increase in length. Thus, the graph suggests that students who are less accurate tend to process words more serially; more accurate students show less of this tendency. OO 295% 1st grade words N = 11 DD <95% 1st grade words N = 15 «295% 2nd grade words N=5 <95% 2nd grade words N =19 CO 1500- - 1800-1600- 1400-1300- 1200-, 1000H o o CU 900- rr "E 800-700- 600-2 3 4 5 6 Word Length in Letters
114 Journal of Reading Behavior TABLE2 First-grade words a 95% accuracy < 95% accuracy Second-grade words a 95% accuracy < 95 % accuracy Average percent accuracy for each word length 2 letters 100 97.8 3 letters 99.3 97.1 4 letters 97.3 68.9 5 letters 86.9 75.6 6 letters 100 85.3 100 89.5 98.1 95.5 97.8 88.2 Average response latency in seconds for each word length First-grade words S 95% accuracy < 95% accuracy Second-grade words a 95% accuracy < 95% accuracy 2 letters.65.80.67.88 3 letters.64.74.63.85 4 letters.67.91.66 1.10 98.7 81.2 5 letters.76 1.09.80 1.43 95.0 76.3 6 letters.72 1.15.76 1.60 7+ letters 100 65.3 91.0 63.1 7 + letters.84 1.42.75 1.69 Total 97.3 85.0 96.8 82.3 We assumed that if latency increased significantly with increasing word length, smaller units of processing would be indicated, and that conversely, if latency did not increase significantly with increasing word length, larger units of processing would be indicated. Subject word-length analysis of variance was performed separately for each accuracy category to determine if word lengths differed significantly. Three of the ANOVA's were highly significant, ranging from F = 7.27 to 11.90, p <.001. The fourth ANOVA involving the highly accurate students on second-grade words was not significant, F= 1.26. The word-length category of seven-plus contained several words eight or nine letters long, but these words were recognized in the same average time as the sevenletter words. They were included in this category since we wanted accuracy and latency data on all the words read in context. The ANOVA's indicated that the latency of word recognition increased significantly across word lengths for the more accurate students on first-grade words, as well as for the less accurate students on both first and second-grade words. Since we expected the more accurate group to demonstrate a non-significant increase in latency across word lengths on the f irst-grade words, we were surprised to find the ANOVA significant for this group on first- but not second-grade words. However, since the ANOVA for the more accurate students on first-grade words was highly significant, we believe the non-significant ANOVA for the more accurate students on second-grade words may be a result of group size, N = 5, rather than an indication of holistic processing across all word lengths. Neuman-Keuls tests were used to further examine the significant increases in latency across word length. Neuman-Keuls tests Neuman-Keuls tests (Tables 3 and 4) were applied to determine which specific word-length latencies differed significantly. The Neuman-Keuls tests show that the longer words-were taking significantly longer to recognize than the shorter words for all three groups in which the overall ANOVA was significant.
TABLE3 Neuman-Keuls Tests on Latency Differences Between Words of Differing Lengths: First Grade Data (Each cell = latency sum difference) "Greater than or equal to 95% accuracy," N= 11 3 letters = 2 letters = 7.095 7.205 4 letters = 7.337 6 letters = 7.997 5 letters = 8.437 7+ letters = 9.284 Critical Difference.05 3-letter words = 7.095 2-letter words = 7.205 4-letter words = 7.337 6-letter words = 7.997 5-letter words = 8.437 7-letter words = 9.284 "Less than95% accuracy," JV= 15 3 letters = 11.175 3 letter words = 11.175 2-letter words = 12.045 4-letter words = 13.695 5-letter-words = 16.47 6-letter words = 17.25 7+ letter words = 21.33.11 2 letters = 12.045.87.242.132 4 letters = 13.695 2.52 1.65.902.792.66 5 letters = 16.47 5.295* 4.425 2.775 1.342* 1.232* 1.1*.44* 6 letters = 17.25 6.075* 5.205 3.555.78 2.189* 2.079* 1.947* 1.287*.847* 7+ letters = 21.33 10.155* 9.285* 7.635* 4.86* 4.08* 1.03.98.92.84.70 Critical Difference.05 5.84 5.59 5.25 4.77 3.97 f ta 8 I g" a 8» go. I 1 S- % h-fc en
TABLE4 Neuman-Keuls Tests on Latency Differences Between Words of Differing Lengths: Second Grade Data (Each cell = latency sum difference) "Greater than or equal to 95 % accuracy, " N = 5 3 letters = 4 letters = 2 letters = 7+letters = 6 letters = 5 letters = Critical 3.131 3.3269 3.34 3.7397 3.81 4.0054 Difference.05 I CO 3 TO CO IC3 3-letter words = 3.131 4-letter words =.1959.209.6287.679.8744 3.3269 2-letter words =.0131.4328.4831.6785 3.34 7-letter words =.4197.470.6654 3.7597.0503.2457 6-letter words = 3.81 _.1954 5-letter words = 4.0054 1.066 1.017.951.861.712 "Less than 95% accuracy," N= 19 3 letters = 2 letters = 4 letters = 5 letters = 6 letters = 7+letters = Critical 16.178 16.656 20.958 27.109 30.327 32.099 Difference.05 CO 3-letter words = 18.178 4.78 2-letter words = 16.658 4-letter words = 20.958 5-letter words = 27.109 8-letter words = 30.327 7+ letter words = 32.099 4.780 4.302 10.931* 10.453* 6.151 14.149* 13.671* 9.388* 3.217 15.920* 15.442* 11.140 4.989 1.771 9.52 9.11 8.56 7.78 6.48
Word Recognition by Second Graders 117 Trend analysis. Trend analysis was applied to the latency data when the ANOVA was significant to discover which function (linear, quadratic, cubic) best described the trend in response latency across words increasing in length. The procedure suggested by Keppel (1973) was followed. The linear function was highly significant, Fs ranging from 23.5 to 47.0, p <.001. The quadratic and cubic trends were not significant. DISCUSSION Two questions were investigated in this study; the first dealt with the relationships among word recognition accuracy, latency and comprehension, and the second graders reading first- and second-grade words. Automaticity theory (LaBerge & Samuels, 1974; Samuels, 1976) emphasizes the importance of rapid word recognition in reading. The theory suggests that both speed and accuracy develop during the early stages of reading. However, after one attains accuracy then the next step would be to develop increasing speed. That is, after accuracy has been achieved, speed of word recognition continues to develop, and rapid word recognition frees the attention for comprehension. In the present study, when second-grade students read first and second-grade words the results indicate that latency of word recognition influenced the association between accuracy and comprehension to a greater extent on first-grade than on second-grade words. We suggest that with the second-grade words the students were primarily at the accuracy stage, whereas with the more familiar first-grade word a sufficient amount of practice had enabled these students to increase their speed of word recognition beyond accuracy. Consequently, the first-grade words were somewhat more sensitive to the effects of latency. An implication suggested by our findings is that teachers should be aware that word recognition skills develop beyond accuracy and that speed of word recognition is related to comprehension. In regard to the unit of processing issue, we had expected the highly accurate students to show a non-significant increase in latency across word lengths on the first-grade words. The Neuman-Keuls test determined that the longer words were taking significantly more time to recognize for this group. However, evidence for the beginnings of more holistic processing can be seen in the non-significant differences in response latencies among two, three and four-letter words and in Figure 1 which indicates that the more accurate readers were recognizing both first- and secondgrade words more quickly and with much smaller differences among word-length latencies than the less accurate readers. Thus, the results provide some evidence for more holistic processing on the shorter words by the more accurate students. The results of the Neuman-Keuls tests for the less accurate students were also somewhat surprising in that the latency differences among the shorter words were not significant. Figure 1 shows that the difference in latencies among the shorter words were much greater for the less accurate than for the more accurate students. Although the differences were not significant in either case, there is stronger evidence for what may be considered component letter processing by the less accurate students. In summary, our results concerning the unit of processing indicate that for these beginning readers the unit of perception was small. However, the findings also sug-
118 Journal of Reading Behavior gest that the more accurate readers were beginning to process units larger than the letter. FOOTNOTES 1 Requests for reprints should be addressed to: S. Jay Samuels, Minnesota Reading Research Project, The University of Minnesota, Minneapolis, Minnesota 35455. REFERENCES BRADSHAW, J. Three interrelated problems in reading: A review. Memory and Cognition, 1975, 3, 123-134. CATTELL, J. On the time required for recognizing and naming letters and words, pictures and colors. In James McKeen CattellMan of Science. Psychological Research, I, Lancaster, PA: Science Press, 1947. DURRELL, D. D. Durrell Analysis of Reading Difficulty. New York: Harcourt, Brace and World, Inc., 1955. FERGUSON, G. A. Statistical Analysis in Psychology and Education. New York: McGraw-Hill Book Company, 1967. GIBSON, E. J. & LEVIN, H. The Psychology of Reading. Cambridge, MA: The MIT Press, 1975. GOUGH, P. B. One second of reading. In J. Kavanaugh and I. Mattingly (Eds.), Language by Ear and by Eye. Cambridge, MA: The MIT Press, 1971. JUDD, W. A. & GLASER, R. Response latency as a function of training, method, information level, acquisition and overlearning. Journal of Educational Psychology, 1969, 60,1-30. KEPPEL, G. Design and Analysis: A Researcher's Handbook. Englewood Cliffs, NJ: Prentice Hall, 1973. * LABERGE, D. Attention and the measurement of perceptual learning. Technical Report 1, Minnesota Reading Research Project, University of Minnesota, December, 1972. LABERGE, D. & SAMUELS, S. J. Toward a theory of automatic information processing in reading. Cognitive Psychology, 1974, 6, 293-323. PERFETTI, C. A. & HOGABOAM, T. Relationships between single word decoding and reading comprehension skill. Journal of Educational Psychology, 1975, 67,461-469. SAMUELS, S. J. Automatic decoding and reading comprehension. Language Arts, 1976, 53, 323-325. SAMUELS. S. J., LABERGE, D. & BREMER, C. Units of word recognition: Evidence for developmental changes. The Journal of Verbal Learning and Verbal Behavior, 1978, 17, 715-720. SHANKWEILER, D. & LIBERMAN, I. Misreading: A search for causes. In J. Kavanaugh and I Mattingly (Eds.), Language by Ear and by Eye. Cambridge, MA: The MIT Press, 1971. TERRY, P., SAMUELS, S. J. & LABERGE, D. The effects of letter degradation and letter spacing on word recognition. Journal of Verbal Learning and Verbal Behavior, 1976, 15, 577-585. TEST REFERENCES Gray Oral Reading Tests. New York: Bobbs-Merrill Co., Inc., 1967. Science Research Associates Achievement Series, 1-2, Form D. Chicago, IL: Science Research Associates, Inc., 1963.