Unlike English-speaking countries like Great Britain, the United States, and Australia, colonialism in developing countries such as India has given rise to multilingual or at least bilingual people who, in addition to their native/first language, can converse in English as well. India is a multilingual society that has 14 languages recognized by the constitution, with Hindi as the national language and English as the second language. In metropolitan cities like Delhi, English plays an important role in everyday life, as many educational institutions follow English as a medium of instruction. With bilingualism being such a common phenomenon, it would be interesting to study the reading accuracy in the two languages (English and Hindi), which differ in overall orthographic transparency. The present study endeavors to examine the Hindi and English word-reading accuracy of bilingual normally progressing and dyslexic readers to understand the nature of reading strategies adopted by them as a consequence of different degrees of transparency of English and Hindi orthographies.
Recent research studies comparing reading in different alphabetic systems have pointed out that factors like orthographic transparency and the characteristics of the spoken language may influence the processes of word recognition (Defior, Martos, & Cary, 2002). The idea that reading acquisition may differ according to the nature of the orthography has been referred to as the “orthographic depth hypothesis” (e.g., Lukatela, Carello, Shankweiler, & Liberman, 1995). Alphabetic orthographies differ in orthographic depth. In transparent orthographies (e.g., German, Italian, etc.) the grapheme–phoneme correspondence (GPC) rules are highly consistent, whereas in opaque orthographies (English, French) they are inconsistent and unpredictable. Although in the former, graphemes generally represent only one phoneme, in the latter, individual graphemes represent a number of different phonemes in different words; for example, in English, letter “a” maps onto a different phoneme in the words hand, hall, hate. English features many words whose spelling does not convey their pronunciation clearly, and has numerous exceptions and many irregular words. Hindi, which has syllabic as well as alphabetic properties, is a transparent orthography with consistent GPC rules (Vaid & Gupta, 2002). The orthographic depth hypothesis postulates that readers adapt their processing strategies to the demands of the orthography they are reading (Goswami, Gombert, & De Barrera, 1998).
Dual-route models of reading typically specify that there exist separate lexical and sublexical mechanisms for processing words (Coltheart, 1978, 1980; Ellis, 1984; Morton & Patterson, 1980). The lexical processing route, which operates at the whole-word level, provides a link between orthography and phonology whereas the sublexical route relies, among other procedures, upon a set of sublexical conversion rules (traditionally grapheme–phoneme rules). At the lexical level, the mapping of the orthographic representations of whole-word forms to their phonological counterparts will allow for the pronunciation of any regularly or irregularly spelled word, provided these words have been previously memorized. On the other hand, the sublexical route generates the pronunciation of regular words, nonwords, and novel words (if regular). The pronunciation of irregularly spelled words, however, cannot be derived by sublexical means because irregular words do not obey traditional sublexical orthography–phonology correspondence rules (Joubert & Lecours, 2000). It follows then that readers of transparent orthographies, such as Hindi, are more likely to succeed in reading by means of sublexical reading strategies than readers of opaque orthographies, such as English.
Cross-linguistic studies carried out in orthographies differing in degree of transparency such as Italian, German, Hindi, Greek, Welsh, and English, have reported important differences in children's reading strategies (Cossu, 1999; Frith, Wimmer, & Landerl, 1998; Gupta, 2003; Porpodas, 1991; Spencer & Hanley, 2003; Wimmer & Goswami, 1994; Wimmer & Hummer, 1990). For example, Wimmer and Hummer (1990) found that reading errors made by German beginning readers in word reading were mostly nonwords, which indicates that these readers use the sublexical route and move into reading by assembling pronunciations through the use of GPC rules. In marked contrast, Seymour and Elder (1986) and Stuart and Coltheart (1988) found that reading errors made by English beginning readers involved reading one word for another, which indicates the use of a lexical–logographic reading strategy. A study by Gupta (2003) comparing reading accuracy among Hindi-speaking dyslexic, reading age (RA) and chronological age (CA)-matched children on word and nonword tasks in Hindi reported that all three groups produced a greater percentage of nonword than word errors in the case of both word and nonword reading tasks. Nonword responses may be expected if children attempt to recode the grapheme sequence of a word from left to right, and fail either because they do not have a complete knowledge of GPCs or because they have difficulties with pronunciation assembly (Wimmer, 1996). In yet another study, Wimmer and Goswami (1994), comparing reading acquisition in German and English, showed that although German children use the same procedure (a phonological, sublexical, or indirect one) to read both words and nonwords, English children use a visual, lexical, or direct procedure to read words and a phonological, indirect procedure to read nonwords. Ellis and Hooper (2001) compared beginning reading in orthographically transparent Welsh and orthographically opaque English using reading tests that were equated for frequency of written exposure. Welsh readers produced more nonword errors and were more reliant on an alphabetic decoding strategy, whereas English children made more real word substitutions and null attempts. In another study, Spencer and Hanley (2003) compared the performance of children learning to read Welsh on tests of reading and phoneme detection with the performance of children learning to read English. The findings reported that, in the case of English word reading, a greater percentage of errors was composed of real-word substitutions and refusals whereas, in the case of Welsh word reading, a greater percentage of errors was composed of nonword responses.
The cross-linguistic comparisons presented here suggest that children learning to read orthographically transparent languages are more likely to rely on an alphabetic reading strategy (Ellis & Hooper, 2001), whereas children learning to read orthographically opaque languages may also supplement sublexical recoding strategies with whole-word (lexical) knowledge (Ziegler & Goswami, 2005). These cross-linguistic comparisons, although extensive in nature, overlook the importance of sociocultural differences (Ziegler & Goswami, 2005). They compare children who have learned to read different languages, who have also been taught by different teachers, in different classrooms, in different schools, using potentially different methods of instruction, and in different cultures (Ellis & Hooper, 2001). Such factors are hard to be controlled.
In this regard, India affords a contrast of written languages with the opacity of English alongside the transparent Hindi orthography. The public schools of metropolitan cities in India such as Delhi may allow control of instructional and cultural potential independent variables. In Delhi, the two languages are spoken and read side by side. Parents can choose whether they want their child to attend English medium or Hindi medium schooling. Generally, parents choose the former. The same education authority administers these schools, and they follow similar curricula and teaching approaches. Such a situation allowed for the assessment of effects of script on reading accuracy and the reading strategies adopted by readers who are matched for geographical area and concomitant socioeconomic variation, broad instructional milieu, and educational background, factors that are not as tightly controlled in many of the studies reviewed here. In the present study, tasks in Hindi and English were devised, for which words were matched for the spoken frequency, age of acquisition (AoA), imageability, and word length.
For interpretation of the results, some information on Hindi and English orthography may be useful. Hindi is written in the Devanagari script, a script that originated to write down Sanskrit. It consists of 48 letters and additional diacritical signs (Bright, 1996). The arrangement of the alphabet is strictly phonetic, with letters classified by place of articulation: vowels and diphthongs first, then consonants with an inherent implicit schwa vowel, which does not have an independent graphemic form (Bright, 1996; Vaid & Gupta, 2002). Consonant clusters are written either one above the other, or by having a special sign added to indicate the absence of the schwa. Vowels are graphemically marked for length and appear in full form in word-initial positions or as diacritical signs (matras) in medial or word-final positions. The Hindi script has syllabic as well as alphabetic properties. The fact that phonemes are graphemically marked aligns it with other alphabetic scripts. However, unlike most alphabetic scripts in which consonants typically stand alone as phonemes, consonants in Hindi have an inherent associated vowel, and as such, Hindi resembles a syllabary (Vaid & Gupta, 2002).
In contrast to Hindi, English is an alphabetic language with a deep orthography. English differs from other alphabetic orthographies due to its low grapheme–phoneme consistency (Venezky, 1970), as exemplified by words with similar spellings but different pronunciations such as hear-bear-heard-beard. Such inconsistencies may make the acquisition and the use of regularities between letter patterns and phonological codes particularly difficult (Wimmer, 1996).
In view of the existing evidence, a study comparing reading accuracy in two orthographies differing in orthographic transparency from within the same education system in the same country by the same children would serve as a meaningful extension to the cross-linguistic studies. Hence, the present study set out to investigate: (a) whether the reading accuracy in Hindi and English was different for bilingual normally progressing and dyslexic readers, (b) whether reading accuracy of the groups varied as a function of spoken frequency of words, and (c) whether the two groups of readers made qualitatively different pattern of errors indexing a difference in reading strategy.
METHOD
Schools
An introductory letter presenting the investigation was sent out to several English medium public schools in Delhi. On the basis of interviews conducted with the schoolteachers, the schools were matched on a number of variables such as their classroom sizes and teaching methods. Three schools were selected for the present study, which followed a similar method of reading instruction. During their nursery school, children were taught the letters of the English and Hindi alphabet, and they were introduced to books that largely contained colorful picturesque illustrations with alphabets. Short stories and poems with emphasis on repetition were read to them daily. In the Kindergarten class, the Interactive English Reader (Macmillan India Limited) and a Hindi book called BalBharti (Delhi Book Bureau, New Delhi) were used by all the schools. These books were used to read aloud to a whole class to develop individual and group reading skills in children. Children were encouraged to read from the books at home. In Grade 1, the National Council for Educational Research and Training (New Delhi) approved English and Hindi textbooks were introduced as reading text in each school. In Grade 1, a phonics teaching approach was followed to develop phonological awareness. Poetry was also introduced at an early grade level in all the three schools. Various activities suggested at the end of each book chapter were used to help children develop phonological skills. Grades 2 and 3 saw the application of the material introduced in the previous class as well as the introduction of new words. Approximately 30 to 40 min were spent each day on reading in Hindi and English, with children taking turns to read. Children were also assessed on a one-to-one basis by their teachers at least four times a year. Giving them home assignments encouraged them further to read at home. Grades 4 and 5 saw the final stages of the primary level reading curriculum, with children performing various activities, for example, essay writing, poem recitation, newspaper reading, and staging skits in Hindi as well as in English. Children were also encouraged to do collage work and poster making by searching for information in different types of magazines and newspapers on particular themes of interest.
All the children were bilingual, with their native language being Hindi. All had Hindi-speaking parents and at least one English-speaking parent. According to the parents and teachers, all the children were exposed to Hindi and English simultaneously from birth onwards at home, and had to study the two languages since the beginning of their schooling. English was spoken in class, except during the Hindi sessions, whereas Hindi and English were spoken at home and with peers in the school. The children from all the schools were exposed to similar kinds of language training activities such as storytelling, singing, dramatics, and so on, both in Hindi and in English. There was similar exposure to media in Hindi and English as newspapers, comics, and TV programs (especially children's programs such as cartoons) were available in both the languages. From the grades obtained on oral activities (such as routine conversation skills, recitation, and story telling), reading and written assignments reported in their annual reports, it was found that children were performing at a similar level in both Hindi and English. All the schools were located in central Delhi, and children in this study lived in areas, which, from a socioeconomic point of view, could be considered as equivalent.
Participants
Participants were 60 right-handed children studying in Grade 3 in three English medium public schools in central Delhi. Two groups were included: 30 dyslexic readers (18 boys, 12 girls) with a mean age of 103.07 months (SD = 6.23 months, range = 90–116 months) and 30 normally progressing readers (17 boys, 13 girls) with a mean age of 102.97 months (SD = 3.55 months, range = 90–109 months). Appropriate consent from parents and school authorities to participate was obtained for all the children. Suspected cases of dyslexia were obtained by asking teachers of third-grade classes whether they had any students with unexpected difficulties in reading and/or spelling in Hindi as well as in English. These children were reported to be facing difficulties right from their early stage of reading and writing, and none of them had received any remediation for their reading difficulties. The CA-matched children were judged by their teachers as showing normal reading and writing progress. All the participants were at least average in arithmetic on the basis of their school records. Their nonverbal intelligence was in the normal range as tested with Raven's Colored Progressive Matrices (Raven, Court, & Raven, 1992), with the mean nonverbal IQ being 102 (range = 90–112) for the dyslexic children, and 104 (range = 95–118) for the CA-matched normally progressing readers.
Children were termed as dyslexics in the present study if they had an intellectual level within the normal range but if their reading and spelling scores fell two standard deviations or more below the mean than that of their CA counterparts. Children with neurological impairment, emotional disturbance, mental retardation, or environmental disadvantages as causes of their reading impairment were excluded from the study. In addition, children whose native language was not Hindi were also excluded from the study. Classification of dyslexia was based on several criteria including children's performance on a single word-reading test developed by Gupta (1997), performance on an English word-reading task, performance on spelling tasks in Hindi and English, teachers' reports, and the authors' observations of the children's spelling errors made in English and Hindi in their school notebooks. Also, schoolteachers and parents of the dyslexic readers reported that attempts to provide extra coaching in reading and spelling to the children were not particularly effective.
Participants were first tested on their graphemic knowledge in Hindi and in English. They were shown in random sequence all the letters of the Hindi alphabet, and were asked to name each letter aloud. Then they were shown in random sequence all the letters of the English alphabet in upper and lower case, and were asked to name each letter aloud. All the children except two were found to read all the single letters in Hindi and English correctly. These two children showed confusion between “
” and “
”, “
” and “
” (“j” and “c,” “m” and “g”) in Hindi and “b” and “d” in English. A standardized single word reading test in Hindi was administered to each child individually (Gupta, 1997). The test consists of presentation of a list of 100 vertically arranged words. The words were drawn from a large pool of words from children's books. Test words were arranged in increasing order of difficulty. Each word had to be read aloud. All clear responses were scored as correct or incorrect. The total score on the Hindi word reading is the number of correctly read words out of a maximum score of 100.
Apart from Hindi word reading, children were also tested on English word reading. An English word-reading task was administered to each child individually. The task consists of presentation of a list of 40 vertically arranged words. All the words were drawn from a large pool of words from children's third- and fourth-grade English reading books, which were a part of the curriculum of the participating schools. Words were arranged in increasing order of difficulty. Each word was to be read aloud. All clear responses were scored as correct or incorrect. The total score on the English word reading was the number of correctly read words out of a maximum score of 40. As expected, dyslexic readers were significantly less accurate than normally progressing readers on individually administered Hindi word reading test as well as on English word reading task, t (58) = 38.90, p<.001; t (58) = 37.73, p<.001, respectively.
The reading accuracy scores of both the groups on Hindi word reading test and on English word reading task were correlated with teachers' ratings of the students' reading achievement on a 5-point scale. In the case of dyslexic readers, the correlation between the reading accuracy scores and teachers' ratings were significant beyond the .01 level of confidence in Hindi as well as in English (r = .50, df = 29; r = .62, df = 29, respectively). In the case of normally progressing readers, the correlation between the reading accuracy scores and teachers' ratings were significant beyond the .05 level of confidence in Hindi as well as in English (r = .38, df = 29; r = .36, df = 29, respectively). Further, the dyslexic readers' word reading in Hindi and English was characterized by distortions, substitutions and/or omissions.
In addition, the two groups in the present study were also tested on spelling in Hindi and in English. To determine the spelling ability of the children, 15 words from the Hindi word-reading test and 15 words from the English word-reading task were selected for the spelling tasks. Dyslexic readers were significantly poorer than normally progressing readers in the case of both Hindi and English spelling tasks, t (58) = 38.82, p<.001; t (58) = 34.87, p<.001, respectively. Further, an examination of the spelling errors from their Hindi and English notebooks revealed that the dyslexic readers often committed errors of substitution, deletion, and insertion, with occasional reversal errors. Characteristics of the participants are presented in Table 1.
Materials
We administered a Hindi word-reading task and an English word-reading task to the normally progressing as well as dyslexic readers. In the case of both tasks, only content words were used. Content words include nouns, verbs, adjectives, and adverbs. All the words in Hindi were regular words. The words in the two tasks were matched as much as possible in terms of spoken frequency of usage, AoA, imageability, word length (number of letters), and syllable length (number of syllables). An outline of Hindi phonetics (Kostic, Mitter, & Rastogi, 1975) is given in Appendix A, and words used in Hindi and English word reading task are given in Appendix B. There were no significant differences between Hindi and English words in terms of spoken frequency of usage, AoA, imageability, and word length. The only significant difference between Hindi and English words was that Hindi words had significantly more syllables than the English words, t (58) = 8.25, p<.001. Further, in the present study, the words selected for Hindi were permitted to contain orthographically complex features (diacritics and multiletter graphemes). In a recent study on literacy acquisition in European Community countries, in the case of deeper orthographies, Seymour, Aro, and Erskine (2003) permitted the inclusion of items that contained orthographically complex features such as, diacritics, multiletter graphemes, and irregularities.
Hindi word-reading task
For this task, 50 words were taken from the Hindi books of the participating children and were rated in terms of spoken frequency of usage, imageability, AoA, word length, and syllable length. The ratings for spoken frequency of usage, imageability, and AoA were obtained from 30 Hindi language teachers teaching Grades 1–5 in English medium schools. They were each given a booklet containing 50 words presented in random order. For rating the words according to spoken frequency of usage, participants were asked to rate the words using a 5-point scale, ranging from 1 to 5 (1–2 = low, 3 = medium, and 4–5 = high spoken frequency of usage). For rating the words for imageability, the instructions given to teachers followed those of Gilhooly and Logie (1980) and Monaghan and Ellis (2002), who asked participants to rate the imageability of a word using a 7-point scale, ranging from 1 (highly imageable) to 7 (poorly imageable), depending on the ease with which the word aroused a mental image-sensory experience. For obtaining AoA ratings, participants were given instructions adapted from Carroll and White (1973), Gilhooly and Logie (1980), and Monaghan and Ellis (2002), which asked them to estimate the age at which they believed that they and others first learned each word and its meaning, in either spoken or written form. Such ratings have been shown to correlate highly with objective measures of word-learning age (Morrison, Chappell, & Ellis, 1997). Ratings were made on a 9-point scale, where the middle 7 points ranged from 2 (1–2 years of age) to 8 (10– 12 years of age), with additional points being added at each end of the scale, 1 (0–1 years of age) and 9 (12+ years of age), to encourage use of the full range. These words were also analyzed in terms of word length and syllable length by the second author. For the present task, an equal number (10 each) of words of high, medium, and low spoken frequency of usage were taken. The words were printed in a mixed order with 10 words appearing on the left-hand side of each A4-sized paper in bold 20-point Devanagari font (this font was similar to the font used in the children's reading books). Reading was evaluated in terms of number of words read correctly.
English word-reading task
For this task, 50 words were taken from the English books of the participating children and were rated in terms of spoken frequency of usage, imageability, AoA, regularity, word length, and syllable length. The ratings for spoken frequency of usage, imageability, and AoA were obtained from 30 English language teachers teaching Grades 1–5 in English medium schools by following the same procedure that was used to rate words in Hindi word-reading task. These words were also analyzed in terms of regularity, word length, and syllable length by the second author. For the present task, an equal number (10 each) of words of high, medium, and low spoken frequency of usage were taken. The words were printed in a mixed order, with 10 words appearing on the left-hand side of each A4-sized paper in bold 20-point Times New Roman font. Reading was evaluated in terms of number of words read correctly. Description of word reading tasks is given in Table 2.
Procedure
Children were tested individually in a quiet room in their school by the second author, a fluent Hindi and English speaker. For both Hindi and English word reading, children had to read the list of items aloud. First, children were assessed on Hindi word reading, for which they were asked to try to read the words aloud so that we could see how their reading was coming along. They were told that they would be recorded, and would need to speak loudly and clearly. Each child's reading responses were recorded for later error analysis. After a gap of 10 min, the same procedure was followed to assess English word reading.
RESULTS
The study examined the mean reading accuracy on words in Hindi and in English obtained for normally progressing and dyslexic readers in a 2(Group) × 2(Language) × 3(Spoken frequency of usage) analysis of variance, with group (normally progressing and dyslexic readers) as a between-subjects variable and language (Hindi and English) and spoken frequency of usage (high, medium, and low) as within-subjects variables (Table 3).
The analysis showed a significant main effect of group, F (1, 58) = 1519.86, p<.001, a significant main effect of language, F (1, 58) = 35.59, p<.001, a significant main effect of spoken frequency, F (2, 116) = 414.75, p<.001. Although there were no significant interactions between group and language and between language and frequency, significant interactions were found between group and spoken frequency, F (2, 116) = 203.29, p<.001, as well as between group, language, and spoken frequency, F (2, 116) = 10.64, p<.01.
The main effect of group showed that normally progressing readers read the words more accurately than dyslexic readers in Hindi as well as in English. The main effect of language revealed that both groups read the words with significantly greater accuracy in Hindi than in English. The main effect of spoken frequency showed that in the case of both groups, words of high spoken frequency were read with significantly greater accuracy than words of medium spoken frequency, which in turn, were read with significantly greater accuracy than words of low spoken frequency.
The interaction among group, language, and spoken frequency showed that in comparison to dyslexic readers, the normally progressing readers were significantly better at reading words of all the three spoken frequencies in Hindi as well as in English. Further, in the case of Hindi, normally progressing readers showed no significant difference in reading accuracy on words of high, medium, and low spoken frequency, whereas dyslexic readers showed significant difference in reading words of all the three spoken frequencies, F (2, 58) = 231.69, p<.001. Post hoc inspection (Newman–Keuls) revealed that words of high spoken frequency were read with significantly greater accuracy than words of medium spoken frequency, which in turn, were read with significantly greater accuracy than those of low spoken frequency.
In English, both normally progressing, F (2, 58) = 44.93, p<.001, as well as dyslexic readers, F (2, 58) = 181.85, p<.001, showed significant differences in reading words of all the three spoken frequencies. Post hoc inspection (Newman–Keuls) showed that normally progressing readers read words of high spoken frequency with significantly greater accuracy than those of medium and low spoken frequency, whereas words of medium spoken frequency were read with significantly greater accuracy than words of low spoken frequency. Dyslexic readers showed significantly greater reading accuracy on words of high spoken frequency than on those of medium and low spoken frequency; however, no significant difference was found between reading accuracy on words of medium and low spoken frequency. The pattern of reading as a function of spoken frequency of usage shown by the two groups in Hindi and in English is represented in Figure 1.
The mean reading accuracy as a function of group, spoken frequencyof usage, and orthography for ([bull ]) Hindi and ([squf ]) English.
Reading errors
From Table 4 it is evident that in Hindi, although normally progressing readers produced only nonword errors, dyslexic readers generated a far greater percentage of nonword than word errors. In English, normally progressing readers produced a greater percentage of word than nonword errors, whereas the dyslexic readers showed a greater percentage of nonword than word errors.
It is common in reading error classification to try to distinguish between what might be visually driven errors and what are more clearly mispronunciations. However, such distinctions are more difficult to apply to errors made when reading orthography is highly consistent in its GPCs, where the error transcriptions concomitantly tend to look like the stimulus that prompted them to the same degree as they tend to sound like it, and where there is no clear-cut boundary between mispronunciations and look-alike errors (Ellis & Hooper, 2001). For this reason, a slightly different track was taken in the further analysis of these errors. These errors were analyzed in terms of absolute proportion of letters and in terms of serial position of letters of the target word stimuli correctly represented in them. In the case of the former, the number of letters correctly represented in each error was counted and their proportion was calculated. For example, in case of English, for the target word “spell” read as “sapil,” out of five letters (s, p, e, l, and l) only three letters (s, p, and l) are correctly represented; hence, the proportion of letters would be 3/5 = .60. For the latter, the number of letters represented in the correct serial order in each error was counted, and their proportion was calculated. For example, in the case of the same English target word “spell” read as “sapil,” out of five letters (s, p, e, l, and l), only two letters (s and l) are correctly represented; hence, the proportion of letters would be 2/5 = .40. Similarly, in the case of Hindi, for the target word “
” (jaha:j) (ship) read as “
” (ja:j) (nonword), out of four elements (
,
,
, and
) (j, h, a: and j), three (
,
, and
) (j, a: and j) are correctly represented; hence, the absolute proportion would be 3/4 = .75, whereas only one element (
) (j) is represented in the correct serial order; hence, the proportion of the letters represented in the correct serial order would be 1/4 = .25. All the errors in Hindi as well as in English were analyzed in this manner.
Means and standard deviations of absolute proportion and absolute serial position of letters correctly represented in reading errors are presented in Table 5. In the case of normally progressing as well as dyslexic readers, no significant differences were found between Hindi and English in terms of absolute proportion of letters in errors. A more stringent measure of this notion is the degree of overlap between the transcribed error and the stimulus in terms of letters correct in absolute serial position. In the case of normally progressing as well as dyslexic readers, no significant differences were found between Hindi and English in terms of absolute serial position of letters in errors.
DISCUSSION
In our study, normally progressing readers showed a significantly greater accuracy in Hindi word reading than in English word reading. Further, in Hindi, these readers showed no significant difference in reading accuracy on words of all the three spoken frequencies. There were no errors in the case of words of high spoken frequency, and whatever errors were committed in the case of words of medium and low spoken frequency, those were all nonwords. On the other hand, in English, words of high spoken frequency were read with significantly greater accuracy than those of medium and low spoken frequency. Although there were no errors in the case of words of high spoken frequency, a few errors were committed in the case of words of medium and low spoken frequency. On analysis of these errors for words of medium spoken frequency, a far greater percentage of word (85.71%) than nonword errors (14.29%) was found, whereas in the case of words of low spoken frequency, percentages of word and nonword errors were 48.21 and 51.79%, respectively. Overall error analysis revealed that the normally progressing readers produced only nonword errors in Hindi, whereas in English, they produced a greater percentage of word than nonword errors. This suggests that in Hindi, despite different levels of familiarity, these children seemed to be relying on sublexical processing to read words, whereas in English, they may be applying logographic strategies to read words of high and medium spoken frequency and alphabetic strategies to read words of low spoken frequency. In the case of English words of high and medium spoken frequency, partial cues might lead to successful reading (Ehri, 1997; Stuart & Coltheart, 1988), whereas in the case of words of low spoken frequency, there might be a tendency to combine the logographic strategy with alphabetic strategy.
Dyslexic readers showed a significantly greater accuracy in Hindi word reading than in English word reading, with words of high spoken frequency being read with significantly greater accuracy than those of medium and low spoken frequency. Error pattern revealed a far greater percentage of nonword (78.69, 93.27, and 97.98%) than word (21.31, 6.73, and 2.02%) errors in the case of words of high, medium, and low spoken frequency, respectively. In English, words of high spoken frequency were read with significantly greater accuracy than words of medium and low spoken frequency, with no significant difference between reading accuracy on words of medium and low spoken frequency of usage. Error analysis revealed a greater percentage of nonword (67.86, 67.08, and 54%) than word (32.14, 32.92, and 46%) errors in the case of words of high, medium, and low spoken frequency, respectively. Overall error analysis revealed that in Hindi, dyslexic readers produced a far greater percentage of nonword than word errors, whereas in English, they produced a greater percentage of nonword than word errors. This suggests that in Hindi, irrespective of familiarity of words, dyslexic readers used sublexical processing, whereas in English, these children might be applying a combination of sublexical as well as lexical strategies to read words. During the process of word reading, dyslexic readers may start reading a word by assembling phonemes following the GPC rules. If the beginning of a word seems familiar to them, they would perhaps guess the rest of the word. Such a strategy may lead to several correct responses in the case of familiar or highly frequent words, whereas in the case of unfamiliar or less frequent words, it may lead to a greater number of errors.
For both groups, greater accuracy in Hindi than in English word reading might be attributed to former being the first language of these children. However, this argument cannot be accepted, as the words chosen for the English and Hindi reading tasks were matched for the psycholinguistic properties, such as spoken frequency of usage, imageability, AoA, and word length. Further, children's level of familiarity for both the languages was comparable as the two languages were being used simultaneously in their day-to-day interaction in school and at home. Hence, this finding implicates the orthographic transparency of Hindi and the orthographic ambiguities of English for the reported difference in the reading accuracy for the two languages. Hindi is a transparent orthography in which one grapheme represents only one phoneme, whereas English is an opaque orthography in which individual graphemes represent a number of different phonemes in different words, and there are many exceptions to GPC rules. Thus, in a transparent orthography children have to learn only the unambiguous GPC rules that can be used to elicit correct pronunciation of any word (familiar as well as unfamiliar words) and even of the nonwords. On the other hand, in an opaque orthography, children have to learn not only the grapheme–phoneme conversion rules but their exceptions as well. Mere knowledge of GPC rules will not lead them to correct pronunciation of words, especially unfamiliar words and nonwords. This implies that a greater effort is required to learn to read in English than in Hindi. Presumably, this might be the reason for greater word reading accuracy in Hindi than in English shown by both groups. In a recent study, Gupta and Jamal (2004) have reported that both normally progressing as well as dyslexic readers showed a greater word reading accuracy in Hindi than in English.
In Hindi, for normally progressing as well as the dyslexic readers, preponderance of the nonword errors compared to a small percentage of whole-word errors indicates that both the groups were reading words mainly by assembling pronunciations from GPCs. The present finding is supported by Gupta (2003), who reported that children with dyslexia as well as RA- and CA-matched controls generated a greater percentage of nonword than word errors in Hindi word reading. In our study, despite using the same reading strategies as the normally progressing readers, the dyslexic readers showed accuracy deficits apparently due to their incomplete knowledge of GPCs.
In English, in the case of normally progressing readers, although a greater percentage of word than nonword errors shows their reliance on lexical processing route, a substantial percentage of nonword errors indicates that in addition to lexical processing, these children were also using sublexical route for reading. Conversely, in the case of dyslexic readers, a greater percentage of nonword than word errors implies that they were reading words indirectly by assembling pronunciations from GPCs. However, a substantial percentage of word errors also indicates a deviance from a purely synthetic assembly strategy. It appears that children also relied on partial visual analysis of the target words for reading. Appendices C and D list the examples of errors in Hindi and English generated by normally progressing and the dyslexic readers.
In our study, in the case of Hindi, both normally progressing as well as dyslexic readers seemed to follow the sublexical route for processing words, whereas in the case of English, both the groups seemingly used a combination of lexical and sublexical processes. However, normally progressing readers reflected a greater reliance on the lexical route whereas dyslexic readers showed a greater reliance on the sublexical route for processing words. Although in the case of normally progressing readers this finding is in line with the existing studies (e.g., Ellis & Hooper, 2001; Spencer & Hanley, 2003; Stuart & Coltheart, 1988; Wimmer & Hummer, 1990; Wimmer & Goswami, 1994), in the case of dyslexic readers, it provides an interesting contrast to previous findings. The study by D'Anguilli, Siegel, and Serra (2001), who investigated the development of reading in English and Italian in bilingual children, provides a plausible explanation for our findings in the case of dyslexic readers. They found that exposure to a language with more predictable GPCs, such as Italian, may enhance phonological skills in English. According to them, it is possible, that Italian phonology disambiguates the choice of the segments more easily than English phonology does. Less skilled readers who are exposed to Italian may realize this feature of Italian, and they may overgeneralize the regularity of Italian words and transfer it to English words.
Overall, in our study, we found that the normally progressing readers used different strategies (sublexical or lexical) of reading according to the demands of the different orthographies. The dyslexic readers, however, seemed to have failed to recognize the differences in the orthographic depth of the two languages, and perhaps adhered to the alphabetic decoding strategy despite the failures it led to in the case of orthographically opaque English. The present study has made certain important observations regarding the nature of reading strategies followed by the bilingual children in the case of orthographically different languages. More studies need to be carried out in this direction.
APPENDIX A
APPENDIX B
APPENDIX C
APPENDIX D
