HUMAN LEARNING (now APPLIED COGNITIVE PSYCHOLOGY), VOL. 4, 105-125 (1985)
Learning Logographies and Alphabetic Codes
For a reprint, e-mail firstname.lastname@example.org
PAUL MUTER AND ELIZABETH E. JOHNS
Psychology Department, University of Toronto, Toronto, Ontario, M5S 3G3, Canada
(c) Copyright 1985 John Wiley & Sons, Ltd.
Received 1 June 1984, Revised 21 September 1984
In three experiments, literate English-speaking adults learned either to identify or to extract meaning from either logographs (Blissymbols or Chinese characters) or English words written in an unfamiliar alphabetic code. Performance was substantially and reliably better in the logographic conditions than in the alphabetic conditions. Vocabulary sized ranged from 30 to 240 words per condition. In a fourth experiment, learning was slower with inconsistent mapping of graphemes to phonemes (similar to English) than with consistent mapping. These results indicate that, under a reasonably wide range of conditions, logographic writing systems may be easier to learn to read than alphabetic writing systems.
In logographic writing systems, each symbol tends to represent a meaningful unit, typically a word. In alphabetic writing systems, each symbol tends to represent a phoneme. Much space in books and journals has been devoted to discussion of advantages and disadvantages of logographic and alphabetic systems (e.g., Carroll, 1972; Gibson and Levin, 1975; Smith, 1978; Taylor and Taylor, 1983), and to discussion of differences in cognitive processing with different writing systems (e.g., Besner and Coltheart, 1979); Biederman and Tsao, 1979; Brooks, 1977; Kolers, 1969; Taylor and Taylor, 1983), but there is little experimental evidence regarding whether people are better adapted to learning logographies or alphabets, or which is more optimal under particular circumstances.
Cross-cultural research is inconclusive. Makita (1968) had teachers of over 9000 grade school students in Tokyo fill out questionnaires. The percentage of children reported to have some degree of reading difficulty was only 0.98. This rate is spectacularly low compared to the United States, where the incidence of serious reading problems is estimated to be approximately ten to twenty per cent (Gibson and Levin, 1975). Makita attributed this difference to differences in writing systems. But Stevenson, Stigler, Lucker, and Lee (1982) argued that the discrepancy was probably due to a response bias in reporting instances of 'reading disability'; for various reasons, according to Stevenson et al. (1982), there is a stronger tendency to categorize students as 'disabled' in the west than in the east. Stevenson et al. made a concerted attempt to compare reading achievement in the United States, Japan, and Taiwan by constructing reliable and culturally fair reading tests. They found similar rates of reading disability in the three countries. All of this cross-cultural evidence is contaminated by the fact that these cultures differ in many respects besides the nature of their writing systems.
Rozin, Poritsky, and Sotsky (1971) studied eight second-grade children with 'clear reading disability'. Each child successfully learned 30 Chinese characters in a few hours of individual tutoring. Parallel tutoring in normal English reading yielded relatively little progress. The advantage of the Chinese condition may have been attributable to the nature of the writing systems, but Rozin et al. acknowledged that the interpretation was somewhat in doubt because of the lack of rigorous controls. Differences in motivation or novelty may have affected the results. In addition, Rozin et al. used only simple, concrete words, and it is unknown whether the result would generalize to a more comprehensive vocabulary.
In the present experiments, English-speaking university students learned to identify or to extract meaning from various kinds of symbols over several sessions. In the Alphabet condition, subjects learned to read English words written in an unfamiliar alphabet, much as students in English-speaking countries do in grade 1. In some instances the mapping between phonemes and graphemes was inconsistent, as it is in English; in most instances the mapping was consistent. In some instances, the logographs were arbitrary in appearance (Chinese characters), but for the most part, the particular logography focused on in the present experiments was Blissymbolics. Blissymbolics (Bliss, 1965; Hehner, 1980) is an international writing system composed of pictographs, in which symbols represent their referents in a pictorial manner; ideographs, in which symbols suggest the meaning of their referents, though not necessarily in a pictorial manner; and a few arbitrary symbols such as '+' and '?'. Blissymbolics has been cited as an example of a logography which may become more widely used because of developments in computer and communications technology (Thompson, 1979, 1983).
In Experiment 1 a single trial consisted of the following (see Figure 1). A question, e.g. ' WHICH CAN A DOG DO?', appeared on a cathode-ray screen for 2.5 s. Then two symbols from the same symbol system (e.g., two Blissymbols) were displayed side by side (e.g. the symbols for 'smile' and 'grow'). The subject pressed a key to indicate which of the two symbols provided a better answer to the question ('grow' in the example). Feedback was then provided, and all of this information remained on the screen for several seconds. Thus, each trial was a study trial as well as a test trial.
In addition to the two conditions of primary interest, the Bliss condition and the Alphabet condition, Experiment 1 included another logographic condition, Chinese, and another alphabetic condition, Esperanto. Chinese differs from Blissymbolics in that Chinese characters have evolved over the centuries and have lost most of their pictographic and ideographic properties. For literate adults, Esperanto is a potential alphabetic alternative to Blissymbolics as an artificial international language. The Esperanto condition was of tertiary interest only: It uses the Roman alphabet, and therefore tells us nothing about learning an unfamiliar alphabetic code.
There were four female subjects in Experiment 1. In all four experiments, subjects were students at the University of Toronto. They were native speakers of English, and did not know any of the symbol systems used in the experiments. They were paid for their participation. Each subject completed a number of sessions of approximately 50 min each at the rate of approximately five sessions per week. No subject appeared in more than one experiment.
Apparatus and materials
Experiments 1, 3, and 4 were controlled by an Apple microcomputer. Stimuli were presented on a 30.5 cm green video monitor. The 83 per cent of the screen used for the graphics had a resolution of 280 by 160 picture elements. Subjects indicated their responses by pressing one of two keys.
The 120-word vocabulary used in Experiment 1 was selected at random from the 1000 most frequently appearing words in English as determined by Thorndike and Lorge (1944), with the following constraints. Half of the 120 words were nouns and half were verbs. Each word appeared as a standard vocabulary item in Blissymbols for Use (Hehner, 1980), and each word appeared in the Esperanto Dictionary (Wells, 1969).
Because there were four conditions in the experiment, the pool of 120 words was split up randomly into four groups of 15 nouns and 15 verbs, called sets 1 through 4.
For the Alphabet condition, 40 Devanagari (characters used to write Sanskrit and Hindi), one for each phoneme, were used as the unfamiliar alphabet. Of the set of 47 possible Devanagari, the two most complex and five relatively confusable characters were not included in our set of 40. The entire set of 40 Devanagari is shown in Figure 2. The Devanagari-phoneme mappings were determined randomly, but were consistent throughout the experiment.
Within the constraints of the resolution of the screen, Blissymbols were displayed according to the standards described in Blissymbols for Use (Hehner, 1980). Some examples of the Blissymbols are presented in Figure 3.
The Chinese characters were created for delivery in the following way. First, they were drawn on grids by a native Chinese-speaker. Then each word was transformed into an Apple 'High-Resolution' shape. The Chinese condition included some compound characters. Some examples of the Chinese characters are presented in Figure 4.
Esperanto is an artificial alphabetic language based on several European languages. It uses the Roman alphabet.
The 16 'questions', eight for verbs and eight for nouns, used in the experiment are listed in Table 1.
Each subject learned 30 words in each of the four conditions: Bliss, Alphabet, Chinese, and Esperanto. The assignment of condition to set (of 30 words) was different for each subject and was determined by a 4 x 4 latin square. Thus, each of the 120 words occurred in each condition equally often across subjects.
After the first session, which was a training session, and before the two transfer sessions (to be described later), each subject performed in 28 experimental sessions.
Procedure for the training session
A training trial consisted of the following. A symbol appeared on the screen for 1.5 s. During this 1.5 s, the subject attempted to say aloud into a tape recorder the English translation of the symbol. Whether the subject was successful or not, after the 1.5 s had elapsed the English translation appeared on the screen together with an explanation or other learning aid, where such a learning aid was possible. This information remained on the screen for 5 s. Then the screen cleared and the next trial started. Each subject studied all 120 words three times in the training session.
The learning aids, which appeared only in the training session, were of necessity different in the different conditions. In the Bliss condition, the learning aid consisted of an explanation similar to the explanation provided in Blissymbols for Use (Hehner, 1980). For example, the learning aid for the word 'back' in Figure 2 was 'standing person + point to back'. For some Blissymbols, no explanation was required, because the meaning was obvious. For example, the Blissymbol for 'chair' is totally pictographic.
In the Alphabet condition, the learning aids consisted of a phonetic spelling of the word. For example, for the word 'cause', the learning aid was 'k au z'. These spellings were similar to those of the Initial Teaching Alphabet (Downing, 1965, p. 71). Prior training on individual letters was not given for two reasons. First, evidence suggests that such training may be unhelpful (Brooks, 1977; Kolers and Magee, 1978). Second, such training would have introduced a confounding in terms of time on task. In order to minimize such a confounding, perhaps analogous prior training in the other conditions could have been arranged (for example, the traditional method of teaching Blissymbols includes pretraining on elements), but on balance we felt that no prior training at all resulted in the fairest and cleanest tests.
In the Chinese condition, no learning aids were provided. A common method of teaching Chinese is the look-say method (Taylor and Taylor, 1983). In any event, because of the arbitrary appearance of most of the Chinese characters, verbal explanations could have been helpful in only a handful of the 120 cases in the present experiment.
In the Esperanto condition, the learning aid consisted of a familiar word similar to the Esperanto word. For example, 'angle' was the learning aid for the Esperanto word 'angulo', which means 'corner'.
Before the training session, the procedure was explained in detail to the subjects. They were explicitly told that, in the Alphabet condition, a particular unfamiliar character would always represent the same sound.
Procedure for the experimental sessions
A trial for the experimental sessions consisted of the following. A question written in English appeared on the screen for 2.5 s along with two fixation points. Then two symbols from the same system appeared centred where the fixation points had been. The subject pressed a key to indicate the right or the left symbol immediately following the subject's response. English translations for the two symbols, the message 'CORRECT' or 'WRONG', and an asterisk next to the correct answer appeared on the screen, and a beep sounded if the response was correct. All of this information, including the two symbols, then remained on the screen for 6.5 s after the response. Immediately after the 6.5 s had elapsed, the screen cleared and the next trial began.
Experimental trials were organized into blocks of 16 trials, 4 in each condition. The part of speech, noun or verb, was determined randomly; then which of the eight questions for that part of speech was to be used was determined randomly; then an appropriate word for the left probe was selected randomly; then an appropriate different word for the right probe was selected randomly. (Appropriateness information was stored in a table. This table indicated, for each question, which words were suitable for that question, and which pairs of words should not be tested together because there would be no clearly correct answer. Hence, not all words were tested with all questions. On the average, for each question 47 of the 60 words were potential probes.)
There were 12 blocks of 16 trials in the first experimental session; 13 blocks in the second session, 14 in the third, 15 in the fourth, and 16 in sessions 5 to 28. At the end of each block mean reaction time in milliseconds for that block was displayed. When the subject pressed a key to indicate that she was ready, the next block of trials began. At the end of the session, a message appeared indicating mean reaction time and percentage correct for the session. Feedback was never broken down by condition.
For a particular session number, all four subjects received exactly the same sequence of events, except that the symbol systems differed. That is, for a particular session number, all four subjects received exactly the same sequence of questions and test-words; subjects differed only in the assignment of symbol systems determined by the latin square.
Before the first experimental session, the procedure was explained in detail to the subjects, they were asked to respond as quickly and accurately as possible, and they were asked to guess after a few seconds if they did not know the answer.
Procedure for the transfer sessions
After the 28 experimental sessions, each subject performed for two sessions with a new vocabulary for each condition. Words which had been in the Chinese condition were now in the Alphabet condition; words which had been in the Alphabet condition were now in the Esperanto condition, and so on.
Before the first transfer session, training on the new vocabulary was given as in the original training session, except that only one block of training trials was given instead of three. Subjects were fully informed of the relationship between the previous sessions and the transfer session, including the fact that each character in the alphabet condition still represented the same sound. Then the subjects completed two sessions identical to the original two experimental sessions. The only difference was in the mapping of sets to conditions.
Results and discussion
The percentages of correct anticipations for each training block for each condition are provided in Table 2. Scoring was stringent: only exactly correct responses were accepted. Kolers (1969) has argued that it is unrealistic to expect that any comprehensive logography could be directly understandable without some learning. Table 2 is consistent with this argument: subjects could correctly name only 1.7 per cent of Blissymbols upon first viewing them.
The results of primary interest in Experiment I are that performance in the logographic Bliss condition was dramatically better than in the Alphabet condition. Figure 5 and Figure 6 show mean reaction time and mean percentage correct, respectively. (In Figure 6, chance is 50 per cent.) Each point in Figures 5 and 6 is based on at least 864 observations, The data in Figure 5 represent means of all trials including incorrect trials. (In the present paper, reaction times are of interest primarily as an index of difficulty; hence the reaction times of all responses are relevant.)
Figure 5 suggests that both major variables, session and condition, had an effect on reaction time. The analysis of variance (based on the mean for a block of four sessions) supports this conclusion. The effect of condition was reliable: F (3, 81) = 67.7, p < 0.001; and the effect of block (of 4 sessions) was reliable: F (6,81) = 23.2 p < 0.001. Furthermore, the interaction between these two variables was reliable: F(18,81) = Z58, p < 0.01. It is clear from Figure 5 that the effect of condition was greater in the early sessions than in the late ones.
Tukey tests showed that the Bliss and Esperanto conditions were not reliably different, but that all other pairs of conditions were (alpha = 0.01). Thus, even Chinese logographs, which have few or no pictographic or ideographic properties, were easier to learn than an alphabetic code. The difference might have been even larger if learning aids had been included in the Chinese condition. This result is incompatible with the common assumption that Chinese logographs are extremely difficult to learn. The fact that performance in the Esperanto and Bliss conditions was approximately equal suggests that, for adults who are already skilled readers of English, Bliss and Esperanto are of approximately equal difficulty.
Figure 6 reveals that the accuracy data generally reflect the reaction time data. The analysis of variance on accuracy revealed the following effects: a reliable effect of condition, F (3,81) = 32.7, p < 0.001; a reliable effect of block, F (6,81) = 82.0. p < 0.001; and a reliable interaction, F (18, 81) = 9.25, p < 0.001.
The data from Figures 5 and 6 are re-presented in Table 3 and Table 4 broken down by subject instead of session. Each subject displayed essentially the same pattern.
Though study times in Experiment I were identical in the four conditions, total exposure to the symbols, including the reaction times, differed in the different conditions. For example, total exposure to the Alphabet condition was greater than total exposure to the Bliss condition because of the relatively slow reaction times in the Alphabet condition. Thus, any differences in performance among conditions would tend to be reduced, if the extra exposure aided learning. Therefore, the true differences in difficulty among the conditions may be even greater than suggested by the results of Experiment 1.
Table 5 and Table 6 show the data for sessions 29 and 30, in which subjects were tested on a new mapping of sets to conditions. Data from experimental sessions 1 and 2 are also provided to facilitate direct comparison. For Bliss, Chinese, and Esperanto, mean performance was slightly better on the transfer task than on the original for both speed and accuracy.
For the Alphabet condition, the transfer results are difficult to interpret. Accuracy was better on the transfer task than on the original task, but speed was worse. One interpretation of this result is that, because of a speed-accuracy trade-off, it is unknown whether there was positive transfer. A second interpretation is that the increase in reaction time suggests an effort to translate the symbols into sounds, which indicates positive transfer of earlier learning. Clearly, however, on the transfer task performance was still worst in the Alphabet condition, and accuracy in the Alphabet condition was far from 100 per cent, despite the fact that the alphabet and the phoneme-grapheme correspondences were identical to the ones that subjects had experienced for 28 sessions.
The results in Experiment 1 were obtained with 30 words in each condition. It is possible that the investment of learning an alphabetic code does not pay off until vocabulary size reaches a certain criterion, and that the larger the vocabulary size, the greater advantage the Alphabet condition has. In Experiment 2 four new subjects learned a much larger vocabulary: 240 words per condition.
In addition, in Experiment 2 an identification (naming) task was used instead of a meaning extraction task. It is possible that an identification task would produce different results: pronouncing a word without recognizing it would be possible in the Alphabet condition, but not in the Bliss condition. Potter and Faulconer (1975) found that words were named more rapidly than corresponding pictorial representations, though semantic decisions could be made more rapidly for pictorial representation
than for words. Feldman and Turvey (1980) found that Japanese words written in Kana, which is sound-based, could be identified faster by Japanese subjects than Japanese words written in logographic Kanji.
A third difference between Experiments I and 2 was that the Chinese and Esperanto conditions were not included in Experiment 2.
One male and three female students served as subjects.
Apparatus and materials
The 480-word vocabulary included the 120 words from Experiment 1, and was selected using essentially the same method as in Experiment 1. Because there were only two conditions, the pool of 480 words was split up into two groups of 120 nouns and 120 verbs, called sets 1 and 2.
Experiment 2 was the only experiment not run on the microcomputer. Stimuli were presented by an experimenter on index cards. Blissymbol stamps (obtained commercially) were applied to the cards. For the Alphabet condition, words were created on the microcomputer, as in Experiment 1, and then were printed out and taped onto cards. On the back of each card, the symbol was re-presented along with the English translation. For use in the training session, learning aids like those in Experiment I were typed on index cards.
Each subject attempted to learn 240 words in the Bliss condition and 240 words in the Alphabet condition. Two subjects learned set 1 in the Bliss condition and set 2 in the Alphabet condition, and two subjects did the reverse. After the training session, then were nine experimental sessions per subject. In all sessions, subjects went through set 1 nouns, then set 2 nouns, then set 1 verbs, then set 2 verbs.
In the training session, the nature of the Alphabet condition was explained, as in Experiment 1. Then the subject studied each of the 480 words for 6 s. That is, for 6 s, the subject was exposed simultaneously to the symbol, the English translation, and the learning aid. The 6 s intervals were timed by beeps on a cassette tape.
In experimental sessions, trials were timed by beeps on a cassette tape at 3 s intervals. Upon hearing a beep, the subject picked up a card, looked at the symbol, and tried to speak aloud the English translation within 3 s. At the end of 3 s, another beep sounded. At this point, the subject turned the card over and looked at the symbol along with its English translation for 3 s, On the next beep, the subject would proceed to the next card, and so on. Thus, again, each trial was both a test trial and a study trial. The experimenter incremented a counter whenever the subject emitted a correct response. Reaction time was not measured. Within blocks of 120 trials, the experimenter randomized the order by shuffling the cards.
Results and discussion
As in Experiment 1, for all four subjects, performance was substantially better in the Bliss condition than in the Alphabet condition. Means of the accuracy scores are presented in Figure 7 The analysis of variance (based on the mean for each session) showed a reliable effect of condition, F (1, 51) = 137.4, P < 0.001.
It is possible that a still larger vocabulary, for example 3000 words per condition, would produce different results, but Experiment 2 demonstrates that the advantage of the Bliss condition over the Alphabet condition extends at least to a vocabulary size of 240 words per condition. Experiment 2 also demonstrates that the advantage of the Bliss condition extends even to a naming task, despite the fact that in the Alphabet condition there is the potential to pronounce the words without recognizing them and receive credit for a correct response. This is not possible in the Bliss condition.
The results of Experiments 1 and 2 may have been attributable to the nature of the characters used in the Alphabet condition. For example, if the Devanagari were particularly difficult to distinguish, performance would be impaired in the Alphabet condition. In Experiment 3, a different set of characters (see Figure 8) was used.
In addition, in Experiments 3 and 4 the training session was dropped. This meant that there were no learning aids in Experiments 3 and 4. The learning aids in Experiments 1 and 2 were, of necessity, different in the different conditions, and it is possible that this confounding influenced the results.
In most other respects, Experiment 3 was similar to Experiment 1.
Four female students served as subjects.
Apparatus and materials
Forty characters from the G. B. Shaw alphabet (Shaw, 1962) were used in place of the Devanagari. The entire set is shown in Figure 8.
Each subject learned 30 words in each of the two conditions, Bliss and Alphabet. The assignment of conditions to sets (of 30 words) was different for each subject, and was the same as in Experiment 1, except that the Chinese and Esperanto conditions were not included. Thus, as in the other experiments, every word appeared in each condition equally often across subjects. Each subject performed in 12 sessions. There was no training session.
Before the first session, the general nature of each condition was explained to the subjects, including the fact that in one condition, a particular character always represented the same sound.
The procedure of Experiment 3 was identical to the procedure of sessions 1 to 28 of Experiment 1, with the following exceptions. The blocks of 16 trials consisted of a random sequence of 8 Bliss trials and 8 Alphabet trials. In all sessions, there were 16 blocks of 16 trials.
Results and discussion
Though the difference was not as large as in Experiment 1, in Experiment 3 performance in the Bliss condition was again substantially and reliably better than in the Alphabet condition. The reaction time means are presented in Figure 9. The analysis of variance indicated a reliable effect of condition, F (1, 69) = 50.0, p < 001. The accuracy results (Figure 10) are strikingly similar to Experiment 1. Again, there was a reliable effect of condition, F (1, 69) = 45.3, p < 0.001.
Thus the most important results of Experiment 1 held up in Experiment 3 with a different character set in the Alphabet condition, and with no training session and no learning aids. Bliss (1965) repeatedly stated that Blissymbols are not self-explanatory and need explanations. It is possible that explanations are helpful, but the results of Experiment 3 show that even without explanations, the logographic Bliss condition is easier to learn to read than the Alphabet condition.
In the Alphabet condition in Experiments 1, 2, and 3, the phoneme-to-grapheme correspondence was perfect, in marked contrast to English (Venezky, 1970). Perhaps more extreme results would have been obtained had the phoneme-to-grapheme consistency in the Alphabet condition been as low as that of English. Some crosscultural evidence suggests that learning to read consistent alphabets is easier than learning to read inconsistent ones (e.g. Kyostio, 1980).
On the other hand, subjects may be more likely to use a rule-based strategy, as opposed to a whole-word strategy, when grapheme-phoneme mapping is consistent, and this strategy may be maladaptive. Brooks (1977) and Brooks and Miller (1979) found that under some circumstances, learning to recognize words which were random strings of characters was easier than learning to recognize words written in an alphabetic code.
In Experiment 4, there were two conditions, the Consistent condition, which was basically the same as the Alphabet conditions in the earlier experiments, and the Inconsistent condition, in which an attempt was made to approximate the inconsistency of phoneme-to-grapheme mapping that exists in English. Within a condition, any particular word was, of course, always spelled in the same way. There was no logographic condition. An identification task was used, as in Experiment 2.
Six female and two male students served as subjects.
Apparatus and materials
To determine the spellings of the words in the Inconsistent condition, a computer program was written which embodied the information contained in tables produced by Dewey (1970). These tables contain the relative frequencies of all possible spellings for every phoneme in some samples of English. For example, for the 'p' sound, there are two possible spellings,'p' and 'pp', and Dewey provides the relative frequencies of each of these spellings, broken down by position in the syllable. The computer program accepted a key press which denoted a phoneme, accepted a number which indicated position in the syllable, and printed out a spelling according to the probabilities in Dewey's tables. Every phoneme in every word used in Experiment 4 was fed into this program, and for each word a spelling was obtained. For example, for the word 'salt', the spelling 'saltt' was obtained. (For some words, the program happened to produce the correct English spelling.) Finally, 26 unfamiliar characters (Devanagari or Shaw characters) were substituted for the 26 English letters.
Two pools of unfamiliar characters were used, each consisting of half Devanagari and half Shaw characters. Of the eight subjects in Experiment 4, four received pool A in the Inconsistent condition and pool B in the Consistent condition, and four received the reverse.
The design was similar to Experiment 2. Each of the eight subjects attempted to learn 240 words in the Consistent condition and 240 words in the Inconsistent condition. After the training session, there were eight experimental sessions per subject. Subjects received set 1 nouns, then set 1 verbs, then set 2 nouns, then set 2 verbs on even sessions, and the reverse order on odd sessions.
The procedure involved an identification task similar to that of Experiment 2. Before the training session, subjects read instructions which included the following:
"In this experiment we will ask you to try to learn some words written in unfamiliar alphabets.
"In the Consistent condition, a particular unfamiliar character will always represent the same sound. For example, one character will always represent the 'p' sound, another character will always represent the 'ee' sound, etc.
"In the Inconsistent condition, there will be an inconsistent relation between characters and sounds, as there is in English. For example, in English,'cat' and 'kill' start with the same sound but different letters. Also, 'cat' and 'ceiling' start with the same letter but different sounds. These are examples of the kinds of inconsistencies that will exist in the Inconsistent condition. Any particular word, however, will always be spelled in the same way."
In the training session, as in Experiment 2 the subject studied each word along with its English translation for six seconds. Throughout the experiment, the condition on a particular trial was always indicated by the word 'Consistent' or 'Inconsistent', written in the bottom left corner of the screen. In addition, trials came in blocks of 120 for a condition, and the word 'Consistent' or 'lnconsistent' appeared on the screen before the block of 120 trials. Within a block of 120 trials, order was randomized. There were no learning aids.
In the experimental sessions, a word written in the unfamiliar alphabet appeared on the screen for 3 s. The subject attempted to speak aloud the English translation within the 3 s. When the three seconds had expired, the English translation was added to the screen, and all of this information remained on the screen for another 3 s. If the subject had spoken the correct translation, he or she pressed a key to indicate this. If the subject had not spoken the correct translation, he or she did nothing at this point, beyond studying the information. Thus subjects did their own scoring. (At the same time, however, a tape-recorder was running, making it possible to later check the accuracy and honesty of the subjects. Spot checks revealed better than 99 per cent agreement between subjects' self-scoring and experimenter's scoring.)
In Experiment 4, at the end of each block of trials, a message displayed the number of correct responses in the block.
Results and discussion
Performance in the Consistent condition was substantially and reliably superior to performance in the Inconsistent condition (see Figure 11). The analysis of variance revealed a significant effect of condition, F (1, 105) = 189.2, p < 0.001.
Though it is possible that inconsistency is helpful in later stages of learning to read an alphabetic code (Frith and Frith, 1980), results of Experiment 4 suggest that inconsistency is disruptive in the early stages of learning, at least.
Experiment 4 suggests that the results of the first three experiments would have been even more dramatic if the mapping between phonemes and graphemes in the Alphabet condition had been inconsistent, as it is in English.
Of the four experiments, Experiment 4 is the only one in which not all subjects displayed the same general pattern of results. Table 7 shows that subject 4 performed better in the Inconsistent condition than in the Consistent condition. One interpretation of the results of subject 4 is that the regularity in the Consistent condition induced her to use a rule-based approach in this condition only, and that this approach was maladaptive compared to a whole-word approach. According to this interpretation, the other subjects used the same approach (either rule-based or whole-word) in both conditions. Better performance in the Consistent condition can be accommodated by both rule-based and whole-word models (Glushko, 1979).
In the present experiments, learning a logography was dramatically easier than learning an unfamiliar alphabetic code. Of course, no one series of experiments can determine that a logography is always easier to learn than an alphabetic code, but the present experiments sampled a reasonably wide range of conditions: The same pattern was obtained with a meaning extraction paradigm and with an identification paradigm; with Blissymbols and Chinese characters; with two different unfamiliar alphabets; with and without learning aides; and with vocabulary sizes of 30 and 240 words per condition.
These results were obtained in spite of the fact that the subjects were literate adults, and had long since attained 'linguistic awareness' (Liberman, Liberman, Mattingly, and Shankweiler, 1980; Mattingly, 1972): they were intimately familiar with the general way in which alphabets work. Subjects without such awareness (e.g. six-year olds) might yield even cleaner results. Furthermore, learning in the Alphabet condition in the present experiments was easier than learning to read English in several respects. There was only one font and only one case to learn. In English, students are confronted with upper and lower case, different fonts, and printing and writing. (Blissymbols are typically standardized.)
There are important differences between understanding words in isolation and reading continuous text (Kolers, 1969, 1970). The present study looked at the processing of symbols in comparative isolation. Thus, the conclusions are most applicable to tasks such as reading highway signs and labels. It is probable, however, that the results have relevance for the reading of continuous text. Similarly, the present experiments tap only recognition of words, and do not speak to questions regarding the production of the different orthographies. It is possible that the optimal orthography for input is incompatible with the optimal orthography for output (Frith and Frith, 1984; Smith, 1978). Considerations regarding output may evaporate as automatic speech recognition becomes better and cheaper: most output may soon be in the form of speech.
The relative advantages of different writing systems may depend on the language in question. For example, a logography may be even more advantageous in languages such as Chinese which have a large number of homophones. (Logographs disambiguate homophones.) Similarly, the relative utility of an alphabetic system my depend on the number of phonemes in the particular language, which ranges from 12 (Hawaiian) to 70 (Abkhazian). It would be desirable to replicate the present experiments with speakers of different languages.
The sizes of the symbols were not the same in the different conditions of the present experiments. It is impossible to avoid all confoundings in this respect, because if the mean widths of the symbols had been equated, their mean areas would have been unequal and vice versa. Fortunately, there is evidence, reviewed by Tinker (1965, pp. 150-151), suggesting that within reasonable limits, size of type is unimportant in learning a new written language.
Figure 4 suggests that performance may not have reached an asymptote by the end of 28 sessions, particularly in the Alphabet condition. It is possible that asymptotic performance in the Alphabet condition would surpass performance in other conditions, or that performance would be equal in all conditions. Some evidence suggests that asymptotic performance is essentially the same regardless of the writing system (Gibson and Levin, 1975; Gray, 1956). The important criterion my be the ease of learning of a symbol system in the early stages of the acquisition of the skill.
Them were many differences, between the logographic conditions and the alphabetic conditions in the present experiments. These differences reflect intrinsic differences between logographic and alphabetic writing systems in general. The present experiments do not tell us which of these differences account for the results. It is hoped that future research will explicate the reasons that performance is different with different writing systems.
The majority of the world's languages still have no writing system, and new writing systems are being introduced every year (Grimes and Gordon, 1980). The Chinese are considering changing from a logographic system to an alphabetic system. In the west, it has been argued (e.g., Thompson, 1979, 1983) that new technology may result in increased use of logographs. In spite of these considerations, it has not been established what the optimal writing system is for human users, or which system is optimal under given conditions. A key consideration in identifying the optimal writing system for human users is ease of learning. The present results indicate that, under a reasonably wide range of conditions, logographic writing systems may be substantially easier to learn to read than alphabetic writing systems.
This work was supported by Research Grant U0149 from the Natural Sciences and Engineering Research Council of Canada to the first author. Experiment 1 was partially supported by research contract OSU81-0073 with the Department of Communications of Canada. Elizabeth Johns, who is now at Queen's University, Kingston, Ontario, collaborated in Experiment 1. We are grateful for the help of Irene Rukavina and David Schloen, and for the useful comments of James Alexander, Paul Hearty, Paul Kolers, Lochlan Magee, Ben Murdock, Dorothy Phillips, Peter Reich, Insup Taylor, and especially Lori McElroy.
Besner, D. and Coltheart, M. Ideographic and alphabetic processing in skilled reading of English. Neuropsychologia, 1979, 17, 467-472.
Biederman, 1. and Tsao, Y. On processing Chinese ideographs and English words: some implications from Stroop-test results. Cognitive Psychology. 1979, 11. 125-132.
Bliss, C.K. Semantography (2nd edn). Sydney, Australia, Semantography, 1965.
Brooks, L. Visual pattern in fluent word identification. In A.S. Reber and D.L. Scarborough (Eds), Toward a psychology of reading. New York: Erlbaum 1977.
Brooks, L. and Miller, A. A comparison of explicit and implicit knowledge of an alphabet. In P.A. Kolers, M.E. Wrolstad and H. Bouma (Eds), Processing of visible language (Vol. 1). New York: Plenum, 1979.
Carroll, J.B. The case for ideographic writing. In J.K. Kavanagh and I.G. Mattingly (Eds), Language by ear and by eye. Cambridge: MIT, 1972.
Dewey, G. Relative frequency of English spellings. New York: Teachers College Press, 1970.
Downing, J. The initial teaching alphabet: Reading experiment. Chicago: Scott, Foresman, 1965.
Feldman, L.B. and Turvey, M.T. Words written in Kana are named faster than the same words written in Kanji. Language and Speech. 1980, 23, 141-147.
Frith, U. and Frith, C. Relationships between reading and spelling. In J.F. Kavanagh and R.L. Venezky (Eds), Orthography, reading, and dyslexia. Baltimore: University Park Press, 1980.
Gibson, E.J. and Levin, H. The psychology of reading. Cambridge, Mass.: M.I.T. Press, 1975.
Glushko, R.J. The organization and activation of orthographic knowledge in reading aloud. Journal of Experimental Psychology: Human Perception and Performance. 1979, 5, 674-691.
Gray, W.S. The teaching of reading and writing, an international survey. Paris: UNESCO, 1956.
Grimes, J.E. and Gordon, R.G. Design of new orthographies. In J.F. Kavanagh and R.L. Venezky (Eds), Orthography, reading, and dyslexia. Baltimore: University Park Press, 1980.
Hehner, B. (Ed.). Blissymbolics for use. Toronto: Blissymbolics Communication Institute, 1980.
Kolers, P.A. Some formal characteristics of pictograms. American Scientist. 1969, 57, 348-363.
Kolers, P.A. Three stages of reading. In H. Levin and J. P. Williams (Eds), Basic studies in reading. New York: Basic Books, 1970.
Kolers, P.A. and Magee, L.E. Specificity of pattern-analyzing skills in reading. Canadian Journal of Psychology. 1978, 8, 322-328.
Kyostio, 0.K. Is learning to read easy in a language in which the grapheme-phoneme correspondences are regular? In J.F. Kavanagh and R. L. Venezky (Eds), Orthography, reading, and dyslexia. Baltimore: University Park Press, 1980.
Liberman, I., Liberman, A.M., Mattingly, I. and Shankweiler, D. Orthography and the beginning reader. In J.F. Kavanagh and R.L. Venezky (Eds), Orthography, reading, and dyslexia. Baltimore: University Park Press, 1980.
Makita, K. The rarity of reading disability in Japanese children. American Journal of Orthopsychiatry. 1968, 38, 599-614.
Mattingly, I.G. Reading, the linguistic process, and linguistic awareness. In J. F. Kavanagh and I.G. Mattingly (Eds), Language by ear and by eye. Cambridge, Mass.: M.I.T. Press, 1972.
Potter, M.C. and Faulconer, B.A. Time to understand pictures and words. Nature. 1975, 253, 437-438.
Rozin, P, Poritsky, S. and Sotsky R. American children with reading problems- easily learn to read English represented by Chinese characters. Science. 1971, 171,1264-1267.
Shaw, G.B. Androcles and the Lion. Middlesex: Penguin, 1962.
Smith, F. Understanding Reading (2nd edn). New York: Holt-Rinehart and Winston, 1978.
Stevenson, H.W, Stigler, J.W, Lucker, G.W. and Lee, S. Reading disabilities: The case of Chinese, Japanese, and English. Child Development. 1982, 53, 1164-1181.
Taylor, I. and Taylor, M.M. Psychology of reading. New York: Academic Press, 1983.
Thorndike. E.L. and Lorge, I. The teacher's word-book of 30,000 words. New York: Teachers College, Columbia University, 1944.
Thompson, G.B. Memo from Memory: Information technology is different. Montreal: Institute for Research on Public Policy, 1979.
Thompson, G.B. Visual literacy and the time-varying icon. IEEE Journal on Selected Areas in Communications. 1983, SAC-1, 304-305.
Tinker, M.A. Bases for effective reading. Minneapolis: University of Minnesota, 1965.
Venezky, R.L. The structure of English orthography. The Hague: Mouton, 1970.
Wells, J.C. Esperanto Dictionary. London: Hodder and Stoughton, 1969.