Computerized training of the correspondences between phonological and orthographic units

Transcription

1 Computerized training of the correspondences between phonological and orthographic units Sini Hintikka, Mikko Aro, and Heikki Lyytinen University of Jyväskylä, Finland The outcomes of computerized training in the correspondences between phonological and orthographic units are reported. Forty-four Finnishspeaking first-graders with poor pre-reading skills were assigned to one of two groups, intervention or control. The children in the intervention group received computerized training over a 6-week period (mean 170 minutes in total) while the children in the control group received regular reading instruction only. Although the short intervention program produced accelerated growth in letter naming, no differential outcomes emerged between the groups in terms of reading acquisition. The outcomes for the poorest performers on six cognitive-linguistic disadvantages were analysed to identify the factors mediating responsiveness to the training. In terms of reading acquisition, the intervention was more effective than ordinary instruction for children with low phoneme awareness skills and attention difficulties as defined by teacher ratings. 1. Introduction Despite the many varying theoretical accounts of dyslexia, the strongest empirical evidence suggests that dyslexia is based on an underlying deficit in phonological skills. More specifically, Vellutino, Fletcher, Snowling, and Scanlon (004) suggest that word identification problems in dyslexia may be the result of deficiencies in phonological awareness, alphabetic mapping, and phonological decoding, which in turn lead to difficulties in establishing the connections between the spoken and written counterparts of words. Because of the general consensus concerning the relationship between phonological skills and reading development, most of the preventive/intervention research in dyslexia has Written Language & Literacy 8: (005), issn / e-issn John Benjamins Publishing Company

2 156 Sini Hintikka, Mikko Aro, and Heikki Lyytinen focused on the training of phonological skills. Hatcher, Hulme, and Ellis (1994) conducted one of the first studies in which the effects of phonological training were compared with the outcomes of training using both phonology and reading (with an emphasis on letter sound correspondences). They found that the combined training of phonology and reading produced the strongest gains in subsequent reading growth, thus substantiating their phonological linkage hypothesis. Meta-analyses of phonological training studies have shown that phonological training enhances both phonological and reading skills, and that the most long-lasting effects on reading skills can be achieved when the training of phonological skills is combined with the use of letters (Bus & IJzendoorn, 1999; Ehri, Nunes, Willows et al., 001). Although most of the studies geared towards the improvement of phonological skills have been carried out with English-speaking participants, there is some evidence that similar interventions may apply in the case of more consistent orthographies. Schneider, Roth, and Ennemoser (000) compared three kindergarten intervention programs on the reading and spelling skills of German at-risk children with low phonological processing skills. They found that the combined training of phonological awareness and letter sound training yielded the strongest effects on reading and spelling in Grades 1 and when compared either with phonological awareness (PA) training or letter sound training alone. However, comparisons between the three groups were rendered difficult by the fact that in this study the letter sound training group received less training (10 weeks) than the other two groups (0 weeks). The indications from the comparison between the PA group and combined group are consistent with the results of meta-analyses: although the PA group scored higher on the phonological awareness tests, the increased amount of PA training did not yield similarly positive results regarding subsequent reading and spelling performance. One possible explanation for the advantage of using letters has been proposed by Adams, Treiman, and Pressley (1998): the use of letters might draw the child s attention to the sounds of spoken words, and as visual symbols, also anchor the phonemes perceptually. It is important to note that in the highly consistent Finnish orthography, 1 letter knowledge and phoneme awareness show a high degree of overlap (Lyytinen, Ronimus, Alanko, Taanila, & Poikkeus, 005). Aro et al. (1999) propose that in Finnish the skills of letter knowledge and phoneme awareness develop simultaneously. Additional evidence for the importance of letter-related processing in reading development stems from the results of the Jyväskylä Longitudinal Study of Dyslexia (for recent reviews see Lyytinen, Ahonen, Eklund et al. 004; Lyytinen, Aro, Eklund et al. 004). Of the 00 children

3 Computerized training 157 followed, each child who failed to achieve reading skills at the end of the first grade (at age 7 or 8) had lower letter knowledge at age 5 or 6 years than any other child in the sample. This finding supports the interpretation that children who fail to acquire adequate reading skills seem to have difficulties in reliably storing letter names in the memory or retrieving them fluently or, possibly, the difficulty lies in forming associations between phonological and orthographic representations. The notion that children with dyslexia have a specific difficulty in grapheme phoneme conversion and especially in increasing the conversion efficiency was already proposed by Snowling (1980) and is consistent with the recent findings reported by Vellutino et al. (004). The question addressed in the present study is whether it is possible to improve the growth of literacy in children with low pre-reading skills by an intervention emphasizing the associations between phonological representations and their written counterparts while also aiming at increasing the efficiency of grapheme phoneme conversion. A few earlier studies have been published in which letter sound correspondence has been a major focus of a preventive intervention and which offer comparisons between letter sound interventions and other training methods. Evidence from these studies suggests that training, with a major focus on letter sound correspondences, can have positive effects on letter knowledge and phoneme awareness (Defior & Tudela 1994; Hohn & Ehri 1983; Rvachew, Nowak, & Cloutier 004). However, the impact on reading skills is inconclusive (e.g., Hohn & Ehri 1983) or at least delayed. For example, in the studies by Defior and Tudela (1994) and Elbro and Petersen (004) the effects on reading were not manifested until later during the follow-up tests. One motive for the present study derives from interest in the issues of the heterogeneity of the children with risk for reading delay and individual variation in responsiveness to training. In recent years, several phonological awareness training studies have raised the question as to whether all children profit from training (e.g., Lovett, Borden, Lacerenza, Benson, & Brackstone, 1994; Poskiparta, Niemi, & Vauras, 1999; Torgesen et al., 1999, 001). In many of these studies the initial level of phoneme awareness has been a predictor of subsequent reading growth. On the other hand, there are studies that have sought to evaluate whether more explicit phonological awareness instruction compared to implicit training helps children who have low levels of phonemic skills. Some studies have provided a positive answer to this question (Foorman, Francis, Fletcher, Schatschneider, & Mehta, 1998; Torgesen et al., 1999), although others have shown that both types of intervention produce equally good outcomes (Wise, Ring, & Olson, 1999). Furthermore, there are findings

4 158 Sini Hintikka, Mikko Aro, and Heikki Lyytinen suggesting that the characteristics of children at risk for reading delay extend beyond the domain of phonological weaknesses. For example, classroom teacher ratings of attention and behaviour, socioeconomic status, and home environment (Torgesen et al., 001); word span, alliteration, and rapid naming of German-speaking kindergarten children (Schneider, Ennermoser, Roth, & Küspert, 1999); verbal intelligence, working memory, and the counting skills of Finnish first graders (Poskiparta et al. 1999) have also predicted reading development in children receiving training in phonological skills. Here, we are interested in evaluating whether the low initial levels of cognitive-linguistic abilities (letter knowledge, phoneme awareness, rapid serial naming, and short-term memory), nonverbal intelligence or attention problems reported by classroom teachers have an effect upon efficacy of treatment. In the computer program used in the present study, the task is to associate the relevant orthographic unit with the corresponding phonological unit. The connections between single phonemes and graphemes are the major focus of the training, but syllables and words are also introduced. In Finnish the single letters map onto phonemes in a regular manner. After the basic letter sounds are learned, phonemic blending skill is required for mastering basic recoding. In Finnish this phonological assembly is a fairly simple serial process of combining the letter sounds in sequence (Lyytinen et al., 006). The additional aim of the computerized training is implicitly to demonstrate the blending of phonemes into syllables and syllables into words, and provide children the opportunity to discover and practice the assembly strategies. To ascertain that the connections between the printed items and their phonological counterparts are learned, the children practice these correspondences until they succeed in choosing the correct item three consecutive times. A play-like element is also created by time-limiting the exposure of the options, which aims to accelerate the speed of making the correct connections between phonological and orthographic representations. In the present study, we hypothesize that a game-like computer-based training program that focuses on linking the phonological and orthographic codes can help children with low pre-reading skills to improve their letter knowledge and efficiency in grapheme phoneme conversion. Second, we wish to investigate whether this type of training has a further effect on reading skills during the first grade after achieving an improvement in letter sound correspondences. The outcomes of this computerized training program are compared with those of a control group receiving regular reading and language support which, in Finland, typically consists of systematic phonics instruction. In the intervention group, the intensity and amount of letter sound correspondence practice

5 Computerized training 159 is greater than in the control group. Third, we anticipate that the program will be relatively more effective among children whose pre-reading skills are low and who therefore are less ready to benefit from the ordinary reading instruction provided in school.. Method.1 Participants A group of Grade 1 non-readers from 13 elementary school classes located in an urban district in central Finland was included in the study. One month after school entry, teachers selected and nominated those children in their classes who had poor pre-reading skills, such as letter knowledge. This initial evaluation was based on observations and tests conducted in school by the teachers. This was followed by individually administered testing of the children (N = 59), whereby the children s inability to decode (sounding-out) syllables or words was verified using the reading material described below. At this point, 11 children were excluded from the study, either because decoding skills were present or they were not native speakers of Finnish. One classroom with 4 participants was unable to follow the intervention schedule and this classroom was not included in the analysis. The final participant group consisted of 44 non-readers (8 boys and 16 girls) selected from a group of 61 first-grade children. The age range was between 81 to 104 months and the mean age was 87.1 months (SD 4. months). Detailed characteristics of the groups are presented in Table 1. Table 1. Descriptive characteristics of children in the two groups Variable Intervention Group Control Group N (Boys/Girls) (14/8) (14/8) Age (in months) 87.7 (5.) 86.5 (.8) Raven (max. 36) 4.3 (3.9) 3.4 (4.) Digit span (max. 16) 5.4 (1.0) 5.3 (1.5) Pseudoword repetition (max. 18) 10.5 (.8) 10.6 (.9) Auditory discrimination (max. ) 14.3 (3.0) 14.0 (3.0) Phoneme awareness (max. 30) 9.0 (4.3) 9.3 (5.9) Rapid naming (37.5) a (.8) a Teacher ratings: Attention problems (max. 16) 4.9 (5.1) 6.1 (4.6) Note. a Significant difference between the intervention and control group (p <.05).

6 160 Sini Hintikka, Mikko Aro, and Heikki Lyytinen We were interested in controlling for the possible influence of the teacher s instructional methods, hence the children from the same classroom were assigned to two groups (intervention or control group): Children were first ordered in rank with respect to their performance in letter knowledge and tests of reading and then the subsequent pairs of children within one classroom were randomly assigned to intervention or control group. Both the intervention and the control group ultimately consisted of participants. Statistical analyses of the pre-test measures revealed that, on average, the assignment to the two groups was successful (see Table 1). With the exception of rapid naming speed of colours and objects, in which the intervention group was slower, t(34.7) =.19, p <.05, there were no differences between the groups in any of the measures.. Design and Materials The study consisted of a pre-test, a training period, a post-test and a follow-up assessment. The pre-test was carried out at the beginning of October, approximately 6 weeks after the beginning of school. The training period lasted for 6 weeks, starting at the end of October, and post-tests were conducted after the training period at the beginning of December. The follow-up tests were administered 14 weeks after the training period...1 The pre-intervention assessments Tests of non-verbal abilities, short-term memory and phonological tasks were chosen as screening measures to ensure that there were no significant group differences in terms of cognitive level. Non-verbal abilities. Non-verbal IQ was assessed with the Raven s Coloured Progressive Matrices (Raven 1956). Short-term memory. The test items were chosen from the Finnish version of the Wechsler Intelligence Scale for Children Revised (Wechsler 1999) digit span sub-test, in which the child was required to repeat a series of spoken digits of increasing length. The score was the number of series repeated correctly. The WISC-R sub-test items requiring backward repetition were not included in the assessment. Pseudoword repetition task. Children were asked to repeat 6 verbally-presented bi-syllabic and 1 tri-syllabic pseudowords (e.g., tiippa, mittuupakke). The score was the number of correctly repeated items. Auditory discrimination. In this task the child was required to say whether two successively presented pseudowords (e.g., aradas, aaradas) were exactly the

7 Computerized training 161 same or not. Only the phonemic length distinguished the words. Twenty-two pairs of words were presented and the total number of correct answers was used in the analyses. Phoneme awareness skills. The three phonological tasks were taken from the Finnish Diagnostic Tests 1: Reading and Writing (Poskiparta, Niemi, & Lepola 1994). In the Single Phoneme Identification Task (10 items), the words were presented orally to the child and he or she was required to say aloud the first phoneme of a given word. Correct letter name responses were also accepted. This task was followed by the Phoneme Deletion Task (10 items), in which the same words as in the identification task were used, with the requirement to delete the initial phoneme and to say aloud the remaining part, which forms a word. In the Phoneme Blending Task (10 items), words were presented by the experimenter phoneme by phoneme and the child was asked to say aloud the resulting word. The words contained two to four phonemes. For the analyses, the three tasks were combined into a score for phoneme awareness (maximum score 30; Cronbach s alpha for internal consistency =.57). Rapid naming tasks. Two subtests (colours and objects) were selected from the Test of Rapid Serial Naming (Ahonen, Tuovinen, & Leppäsaari, 003). Stimuli were arranged in five rows in random order (altogether 50 presentations of 5 different items) with no successive presentations of the same stimulus item. The object-naming task consisted of pictures (car, house, fish, pencil, and ball) and in the color-naming task the items used were red, black, yellow, green, and blue. Prior to the test proper, the child was issued with practice items, including all 5 items from each of the color and object stimuli, to ensure that he or she was familiar with the names of the items. The children were instructed to perform the task as quickly and accurately as they could. In the analyses the two subtests were combined and total task-completion time in seconds was used as the measure (the Pearson correlation coefficient for consistency =.69)... Teacher ratings Classroom and special education teachers completed a rating scale separately for each participating child. The 8-item questionnaire consisted of items related to hyperactivity (e.g., has difficulty remaining seated when that is expected, is always on the go ), 4 items related to inattention (e.g., has difficulty following through instructions, is easily distracted by extraneous stimuli) and items related to executive functions (e.g., has difficulty controlling his/her behavior). A 3-point rating scale (0 = never, 1 = sometimes, = often) was used. The Pearson correlation coefficient between the classroom teacher and special education teacher evaluations was r(35) =.79, p <.001. Only the classroom

8 16 Sini Hintikka, Mikko Aro, and Heikki Lyytinen teacher s ratings were used in the analyses (Cronbach s alpha for internal consistency =.93)...3 Outcome measures Tests of letter knowledge and reading, used in earlier studies (Seymour, Aro, & Erskine, 003; Aro, Tolvanen, Poikkeus, & Lyytinen, 004) were administered. The lists were preceded by the presentation of practice items (in the reading lists three items and in letter naming six items). Letter naming and spelling. A list of 3 letters of the Finnish alphabet was used for letter/phoneme naming (the child gave either the phoneme or letter name and both were considered correct). Letters (c, q, w, x, z, å) which do not have a distinct corresponding phoneme in Finnish were excluded. Cronbach s alpha for internal consistency was.91. The list-completion time was measured to elicit a measure of letter naming speed (Cronbach s alpha for test retest =.77). The same 3 letters were used for letter spelling, in which the stimuli were read aloud by the examiner and the children were asked to write the letters on designated lines on a sheet of paper (the Pearson correlation coefficient for test retest =.79). Reading lists. Children were presented with 3 lists of items for reading. Each list was presented on a sheet with the items arranged vertically and two parallel versions of the list were used during the study (the same list was presented at the pre- and post-test and the parallel version at the follow-up test). The materials included (a) a list of 9 familiar content words (high imageability; e.g., house, run), (b) a list of 1 syllables (with the structures VV, CV, VC, and CVC), and (c) a list of 9 bisyllabic pseudowords (with the structures VCV, CVCV, VCVC). For familiar words and syllables, children were asked to read out each item on the list. In the case of pseudowords, they were told that the items are made-up words which they might, nonetheless, be able to pronounce. For the statistical analyses, the lists of content words and syllables were combined (the maximum score was 1; the Pearson correlation coefficient for consistency =.85). For pseudowords, the test retest reliability coefficient was.59. Reading fluency. Assessment included a standardized time-limited reading test Lukilasse (Häyrinen, Serenius-Sirve, & Korkman, 1999), in which 90 words of increasing length are presented in lower-case letters in a list arranged vertically on a sheet of paper in three columns. The task was to read the words as accurately and rapidly as possible within a -minute time-limit. The score was the number of correctly read items (Cronbach s alpha for internal consistency =.98).

9 Computerized training Testing procedures Testing was conducted individually in one session and in a fixed task order. Participants responses in letter naming, reading and rapid naming tests were recorded for subsequent scoring. The examiner presented the lists to the child, simultaneously started a stop watch and stopped it when the last item had been attempted. If a child was blocked by a particular item he or she was encouraged to move on and complete the list...5 Training program The Literate computerized training program was developed at the University of Jyväskylä. The goal of the program was to enhance the accuracy of processing for phonemic sounds and to learn to connect phonemes fluently with their orthographic equivalent. A single auditory stimulus was delivered (with high quality headphones) concurrently with a number of orthographic items (target and distractors) that appeared at the top of the screen embedded within balls. The balls immediately began to drop downwards on the computer screen and the player s task was to hone in on the relevant orthographic item and to catch it by clicking the mouse. If the player did not catch the correct spelling prior to the ball hitting the ground or erroneously clicked on the incorrect spelling, the target item was repeated in the next trial and the correct response was color-highlighted. The version of the program used in this study contained five levels, starting with a grapheme phoneme level (in which the auditory stimuli were letter sounds, not letter names) and proceeding to syllables and words. Each level could be played in either upper case or lower case letter formats. To ascertain that the items were learned, the children practiced the items until they succeeded in choosing the correct answer three consecutive times. As the emphasis was on adaptation, the number of orthographic alternatives (distractors) and the speed at which the balls fell was initially set very low. However, as the game proceeded, the number of distractors and speed were adjusted in keeping with the developing level of the individual player, ensuring thereby that the game was always challenging without being so difficult that the child became frustrated by thwarted attempts. The program also saved data on the progress of each child, thus allowing for continuity of subsequent levels of difficulty...6 Training procedures Intervention group. Computerized training for the intervention group was included as part of the reading and language support that is typically provided

10 164 Sini Hintikka, Mikko Aro, and Heikki Lyytinen for struggling readers within the Finnish school curriculum. Ordinary classroom instruction is based on synthetic phonics while the special education procedures resemble the classroom instruction (see Poskiparta et al., 1999), except that the children are taught in small groups. The intervention and control group received the same amount of practice, but the methods varied. In the intervention group, the intensity and amount of letter sound correspondence practice was greater than in the control group. The computerized training was conducted in four classes by university staff, in four classes by the classroom teachers and in five classes by the special education teachers. The teachers were informed about the purpose of the study and were instructed to arrange a tutoring session with the players to explain the functioning of the program. After the first tutoring session, players were able to work independently with the program. However, in all the classrooms the instructor kept a record of the children s playing time, arranged the sessions and ensured that the children could practice undisturbed. The instruction to the teachers was to arrange short (10 0 minutes) training sessions at least three times per week. The total playing time over the 6 weeks of training varied between 111 and 36 minutes with an average of minutes (SD 35.6 minutes). The number of playing sessions varied between 11 and 1, and averaged 14.3 (SD.3). The mean duration of one session was 10.7 minutes. Control group. The control group was exposed to reading and language support as determined by the teachers, but without the specific computerized training. The methods and content of this instruction included a variety of pen-and-paper tasks and other types of computer-assisted practice in phonological awareness, letter knowledge, and reading. Typically speeded practice was not included in the regular reading instruction given to the controls. 3. Results Scores for the letter knowledge and reading measures were subjected to analysis of variance with group (intervention, control) as the between-subjects factor and test session (pre-test, post-test, follow-up) as the within-subjects factor. The test session effects were analyzed first for the change between pre-test and follow-up and then the contrasts between the pre- and post-tests and post- and follow-up tests were specified. For letter spelling, the analysis was performed for only two test sessions (pre-test vs. post-test). The means and standard deviations for the groups in the pre-, post- and follow-up tests are presented in Table.

11 Computerized training 165 Table. Means (and standard deviations) by group in literacy measures at three assessment points in Grade 1. Test Maximum possible Session Intervention Group (n = ) Control Group (n = ) Letter knowledge Letter naming 3 Pre-test 15.0 (4.7) 16.6 (6.4) Post-test 0.0 (3.0) 19.3 (4.7) Follow-up.1 (1.0) 1.6 (1.8) Letter naming speed a Pre-test 5.5 (4.4) 41.0 (17.3) Post-test 3.6 (14.4) 9.3 (1.8) Follow-up 4.4 (6.5) 4.6 (8.1) Letter spelling 3 Pre-test 14.3 (4.9) 15.6 (5.9) Post-test 19.0 (.8) 18.4 (4.7) Reading measures Syllables and words 1 Pre-test 1.5 (1.7).0 (.0) Post-test 1.9 (5.) 11.9 (7.6) Follow-up 18.4 (.3) 17.7 (4.6) Pseudowords 9 Pre-test 0.1 (0.4) 0.1 (0.3) Post-test 3.4 (.9) 3.4 (3.1) Follow-up 6.5 (.5) 6.5 (3.0) Reading fluency 90 Pre-test 0. (0.7) 0.5 (0.8) Post-test 10.4 (7.4) 11.1 (9.7) Follow-up 5. (10.7) 7.1 (13.5) Note. The groups did not differ significantly (p >.05) at these assessment points. For test-session and interaction effects see the Results section. a In the analyses, a logarithmic transformation was used for speed in letter naming. 3.1 Growth in letter knowledge Letter naming. There was a significant main effect of test session for letter naming, F(1.4, 58.0) = 58.07, p <.001, η p =.58, and the contrasts showed that the difference between the mean score on the pre-test and post-test was significant, F(1, 4) = 71., p <.001, η p =.63, as was the difference between the posttest and follow-up, F(1, 4) = 1.87, p <.001, η p =.34. There was no significant main effect of group, F < 1, and the group by test session interaction across all three test sessions failed to reach significance, F(1.4, 58.0) =.53, p >.10, η p =.06. However, the contrasts showed that the growth of letter naming differed reliably between the groups from pre-test to post-test, F(1, 4) = 6.49, p <.05, η p =.13. During the training period the intervention group showed greater gains (+5 letters) in their letter knowledge than the control group (+.7

12 166 Sini Hintikka, Mikko Aro, and Heikki Lyytinen letters). In the follow-up tests, the intervention group performed slightly better than the control group, although the difference failed to reach significance as also did the interaction, F < 1. However, it must be noted that, by the time of the post-test and especially the follow-up test, scores for letter naming were close to the ceiling level (at the follow-up test the accuracy level for the intervention group was 96% and for the control group, 94%). Letter naming speed. A logarithmic transformation was computed for the speed variables, as they were not normally distributed. The effect of test session was highly significant, F(1.8, 74.3) = 56.8, p <.001, η p =.58, with the speed of letter naming rising from pre-test to follow-up test. Neither main effect of group nor the group by test session interaction effect, Fs = 1, were significant. The contrasts showed that the test session effect was significant between the pre-test and post-test, F(1, 4) = 5.69, p <.001, η p =.56, and the post-test and follow-up, F(1, 4) = 16.56, p <.001, η p =.8. However, the interactions failed to reach significance, Fs 1. Letter spelling. Again, for letter spelling, the effect of test session was highly significant, F(1, 4) = 57.00, p <.001, η p =.58. There was no significant main effect of group, F < 1. For the intervention group the gain (+4.7 letters) was slightly greater than for the control group (+.8), the difference falling short of conventional levels of statistical significance, F(1, 4) = 3.61, p =.06, η p = Gains in reading measures The results for all the reading accuracy measures were similar: there were highly significant gains in all measures showing a rapid improvement in reading ability for both groups. Accuracy in the reading of syllables/words and pseudowords showed a significant improvement across the test sessions, F(1.6, 67.0) = 51.13, p <.001, η p =.86; F(, 41) = 119.7, p <.001, η p =.85, respectively. A significant effect of test session was also obtained for the reading fluency test, F(1.4, 61.1) = , p <.001, η p =.80. The contrasts showed that this improvement in all the tests was significant from pre-test to posttest (syllables and words: F(1, 4) = 151.9, p <.001, η p =.78; pseudowords: F(1, 4) = 56.51, p <.001, η p =.57; reading fluency: F(1, 4) = 69.93, p <.001, η p =.63) and from post-test to follow-up (syllables and words: F(1, 4) = 5.48, p <.001, η p =.56; pseudowords: F(1, 4) = 64.07, p <.001, η p =.60; reading fluency: F(1, 4) = 186.1, p <.001, η p =.8). However, in all the reading measures, neither the group effect nor the group by test session interaction reached significance (all Fs < 1). This was also true for the contrasts between the preand post-test and post- and follow-up test.

13 Computerized training Did children with low pre-reading skills profit? Since we were interested in the reading development of children who may be at-risk for reading delay, the poorest performers within both the intervention and control group were selected on the basis of their performance on each of the six pre-test measures (non-verbal IQ, attention difficulties, letter naming, phoneme awareness, naming speed, and short-term memory). For each of these six sub-groups, we included only those participants whose scores fell below the median on the respective measure. Thus, data from 11 children from the intervention and control groups were included. It should be noted that, in the groups of children with low non-verbal abilities, phoneme awareness, and short-term memory, several children attained the median value and, in these cases, the size of the groups differed. Because some children showed problems in several abilities, they were included in more than one group. For economy of presentation and because the training period was so short that the interesting results were more observable during the intervention period, we report here only the improvement during the intervention period and only with respect to letter naming and reading development. In the first instance, we determined that there were no differences between the intervention and control group in these poorly performing groups in terms of the pre-test letter naming and reading measures. For letter knowledge, analyses of variances were performed separately for each sub-group with training condition (intervention, control) as the between-subjects factor and test session (pre-test, post-test) as the within-subjects factor. The effects of test session were highly significant across all of the six poorly performing groups. The F-values ranged from (in the group of children with low letter naming) to (in the group of children with lower non-verbal abilities). The main effects of training condition (intervention, control) within each group were absent. The interaction effects are reported below separately for each low performing group. For reading measures, the pre-test reading means and variances were in many cases close to zero (see Table 3) and therefore ANOVAs are performed with post-test data only. In pseudoword reading and reading fluency the post-test scores tended to favor the children receiving computerized training (especially in reading fluency in the group of participants with attention difficulties, F(1, 1) = 3.90, p =.06, η p =.16), but no statistically significant differences between the control and intervention children emerged, all other Fs 3, ps.10. The results of syllable/word reading are reported below. Low non-verbal abilities group. All members of the lower ability group obtained a score below 4 on the Raven s Matrices. The interaction effect in letter

14 168 Sini Hintikka, Mikko Aro, and Heikki Lyytinen Table 3. Means (and standard deviations) in literacy measures at pre- and post-test in Grade 1 by group for poorly performing children. Test Maximum possible Session Intervention Group Control Group 1. Low non-verbal IQ Letter naming 3 Pre-test 16.5 (4.6) 14.6 (7.3) Post-test 19.8 (3.8) 17.9 (5.6) Reading measures Syllables and words 1 Pre-test 1.1 (1.1).0 (1.9) Post-test 11.8 (5.) 1.0 (8.) Pseudowords 9 Pre-test 0.0 (0.0) 0.0 (0.0) Post-test 1.6 (1.6) 3.7 (3.4) Reading fluency 90 Pre-test 0.1 (0.3) 0.4 (1.0) Post-test 7.4 (5.1) 10.7 (9.9). Attention difficulties Letter naming 3 Pre-test 14.5 (4.6) 15.1 (7.5) Post-test 19. (3.4) 17.8 (5.3) Reading measures Syllables and words 1 Pre-test 1.7 (1.4) 1.3 (.) Post-test 1.4 (5.1) a 7.3 (6.6) a Pseudowords 9 Pre-test 0.0 (0.0) 0.0 (0.0) Post-test 3.5 (3.0) 1.6 (.3) Reading fluency 90 Pre-test 0. (0.4) 0. (0.4) Post-test 10.1 (8.3) 4.5 (4.7) 3. Low letter naming Letter naming 3 Pre-test 11.0 (.9) 11.7 (5.6) Post-test 18.3 (3.3) 16.3 (5.0) Reading measures Syllables and words 1 Pre-test 1.1 (0.9) 0.8 (1.3) Post-test 11.5 (5.0) 7.5 (7.9) Pseudowords 9 Pre-test 0.0 (0.0) 0.0 (0.0) Post-test 3.1 (.8) 1.6 (.5) Reading fluency 90 Pre-test 0.1 (0.3) 0.3 (0.6) Post-test 9.1 (7.) 6.1 (8.1) 4. Low PA Letter naming 3 Pre-test 13.8 (5.6) 13.1 (7.3) Post-test 19.6 (3.9) 16.3 (5.5) Reading measures Syllables and words 1 Pre-test 0.8 (0.8) 0.4 (1.0) Post-test 13.1 (5.3) a 6.3 (7.5) a

15 Computerized training 169 Table 3. (Continued) Test Maximum possible Session Intervention Group Control Group Pseudowords 9 Pre-test 0.0 (0.0) 0.1 (0.3) Post-test. (.8) 1.3 (.5) Reading fluency 90 Pre-test 0.0 (0.0) 0. (0.4) Post-test 10.4 (7.6) 5.9 (9.7) 5. Slow naming Letter naming 3 Pre-test 15.0 (4.6) 16. (4.) Post-test 19. (3.3) 19.6 (3.6) Reading measures Syllables and words 1 Pre-test 1.6 (1.5) 1.6 (1.7) Post-test 11.4 (5.3) 1.5 (7.4) Pseudowords 9 Pre-test 0.0 (0.0) 0.1 (0.3) Post-test 3.6 (.9) 3.5 (3.0) Reading fluency 90 Pre-test 0.1 (8.1) 0. (0.6) Post-test 10.0 (8.1) 9.8 (8.4) 6. Poor STM Letter naming 3 Pre-test 13.3 (4.4) 15.3 (7.6) Post-test 19.4 (3.3) 17.8 (5.5) Reading measures Syllables and words 1 Pre-test 1.5 (1.9) 1.8 (.3) Post-test 13.4 (5.3) 9.3 (7.6) Pseudowords 9 Pre-test 0.6 (0.) 0.0 (0.0) Post-test 4.5 (.8).8 (3.1) Reading fluency 90 Pre-test 0. (0.8) 0.4 (0.9) Post-test 11.3 (7.7) 9.6 (10.3) Note. a Significant difference between the intervention and control group (p <.05). 1. Low non-verbal IQ = Group of children with low non-verbal abilities, n =,. Attention difficulties = Group of children with attention difficulties, n =, 3. Low letter naming = Group of children with low letter naming at beginning of school, n =, 4. Low PA = Group of children with low phoneme awareness, n = 0, 5. Slow naming = Group of children with slow naming speed, n =, 6. Poor STM = Group of children with poor short-term memory span, n = 5. naming was not significant, F < 1, indicating that the intervention and control children developed similarly in letter naming. In syllable and word reading at the post-test no differences emerged between the groups, F < 1. Attention difficulties group. The analyses revealed no significant differences between the intervention and control group in letter naming, F(1, 0) =.41, p >.10, η p =.11 but group differences were observed in reading development. The children in the intervention condition scored higher than the children

16 170 Sini Hintikka, Mikko Aro, and Heikki Lyytinen in the control condition in syllable and word reading at the post-test, F(1, 1) = 4.1, p.05, η p =.17. Low letter naming at school outset group. The intervention group improved more during the training period in letter naming: the gain was 7.3 letters in the intervention group and 4.6 letters in the control group, F(1, 0) = 5.0, p <.05, η p =.0. In terms of reading syllables and words at the post-test the analyses revealed no significant difference between the groups, F < 1. Low phoneme awareness group. The interaction effect in letter naming approached significance, F(1, 18) = 3.15, p =.09, η p =.15, with the trained group tending to show higher gains. In the post-test reading of syllables and words the children receiving computerized training performed better than the control group, F(1, 19) = 5.51, p <.05, η p =.3. Slow naming speed group. In the group of children with slow naming skills, there were no differences between the intervention and control group in the development of letter naming (interaction was not significant, F < 1) or in the reading of syllables and words, F < 1. Poor short-term memory span group. There was a significant interaction for letter naming, F(1, 3) = 8.60, p <.01, η p =.7, with the computer-trained group showing more gains. In syllable and word reading no significant group difference at the post-test emerged, F < Discussion The main purpose of this study was to determine whether struggling beginning readers could profit from training of correspondences between phonological and orthographic units. The results indicated that a brief period of computerized training, averaging 170 minutes, was associated with improvements in letter sound correspondences, especially among children with compromised pre-reading skills. In the computer program, the task of the child was to select the orthographic item that corresponded to the spoken item. After practicing with this program for less than 4 hours, the intervention group showed benefits in letter naming. For children who were at the beginning of reading development, this type of training offered an intensive way in which to rehearse the memorizing of letter sound correspondences. During the follow-up period, the control group children, who received only regular reading instruction, caught up with the intervention group children and, after 7 months of formal reading instruction, the scores for both groups on the letter naming task approached the ceiling level (accuracy for the intervention group was 96% and

17 Computerized training 171 for the control group 94%). These results are consistent with earlier findings of children achieving good letter knowledge during the first school year (Seymour et al. 003). The contribution of the intervention was that, in comparison to regular reading instruction, the development of letter knowledge was accelerated. A central aim of the training was to increase the efficiency of grapheme phoneme conversion, as it appears that the difficulty in letter sound associations is manifested as a slow retrieval rate rather than inaccuracy (Lyytinen et al., 005; de Jong & van der Leij, 1999; Wimmer, 1993). The results showed that both groups became faster in the naming of letters. The intervention group had more room for improvement because they started from a lower level. At the follow-up tests the groups performed similarly, but there were no reliable group or interaction effects. A similar finding was reported by de Jong & Vrielink (004) who found that, even after intensive training (130 repetitions), rapid letter sound naming did not improve in comparison to the results of a group that practiced serial addition training. Apparently, more long-lasting practice is necessary, and the type of computer program used here may offer the possibility for numerous repetitions. The results of the reading acquisition analyses showed that in both groups rapid growth of reading skills was achieved. Accuracy levels during the followup tests in reading syllables and words were 88% for the intervention group and 84% for the control group. One important finding of this study was that a combination of a highly regular orthography and systematic phonics instruction seemed to provide a good basis for the rapid acquisition of reading ability, which is consistent with the results of earlier studies (Aro et al., 004). It must be noted that, in spite of the rapid growth of reading, the results were not comparable to those of children with typical pre-reading skills. Aro et al. studied the acquisition of literacy of 63 first-graders and found that, within 4 months of beginning of school, the children could read words with an accuracy level of 87%. In the present study, the same materials were used and assessed at the same time point and the accuracy level was 60% for the intervention group and 54% for the control group. In reading acquisition, no significant differences between the intervention and control group emerged. There may be several reasons for this finding. First, the intervention was extremely short in duration (on average less than 3 hours) compared to many phonologically based intervention studies, where up to 70 hours of training may be provided. The second explanation for this lack of differential outcomes may relate to the activities of the control group. The control group received regular reading and language support which, in Finland,

18 17 Sini Hintikka, Mikko Aro, and Heikki Lyytinen typically consist of synthetic phonics instruction. In addition, many teachers supplement the regular reading instruction with tasks of letter knowledge and phonological awareness. It could be that there was greater similarity in instruction in the two groups than we expected, which means that the effectiveness of the intervention program was in fact subjected to a stringent test. In spite of the similarities in instruction in the two groups, we consider that in the intervention group the intensity of the letter sound correspondence practice was greater than in the control group. In addition, for the intervention group the practice was targeted at developing speed in associating the phonological units with orthographic items, which is not a typical method in regular reading instruction. The benefit of the program was that it provided for the easy implementation of individually-targeted training. After the teachers had arranged a tutoring session with the player, each child was then able to work with the program individually thus teacher time did not need to be allocated for tasks that required drilling, such as repetitive exercises for memorizing the letter sound associations. Were we able to identify the risk factors which mediated/moderated responsiveness to the training program? To examine this question, we selected groups of children whose performance was below the median, separately for each of the six cognitive-linguistic abilities (non-verbal IQ, attention difficulties, letter naming, phoneme awareness, naming speed, and short-term memory), which according to the literature are associated with reading acquisition. This kind of analysis is not optimal for detecting individual bottlenecks in the development and predictors of reading acquisition, but our sample size set limits to conducting a more fine-tuned analysis. We decided to pursue this type of analysis because the correlations showed that these cognitive abilities were not completely overlapping (at the whole group level, only phoneme awareness and letter knowledge showed significant correlation, r s (44) =.48, p <.01) and because we found that the groups with different disadvantages did indeed have different results in literacy development. First of all, the results showed that, in all the groups with poor pre-reading skills, the children improved their reading skills and learned to decode words during the intervention period. In the groups of children with low non-verbal reasoning skills and slow naming, no differences between the participants in the intervention or control group were found. However, the intervention was more effective than ordinary instruction for children with low letter knowledge and who had a poor short-term memory span, in terms of improving their letter knowledge, and for children with attention problems and low phoneme awareness, in syllable and word reading.

19 Computerized training 173 It has been proposed that phoneme awareness problems can be conceived as access to phonemes per se (Landerl & Wimmer, 000) and that this access might be relatively fragile (de Jong & van der Leij, 003). The present study suggests that it may be possible to improve the access to phonemes of children who have initial difficulties with this process. To facilitate this process, it is important to emphasize the connection between print and phonology in instruction and to provide a sufficient number of exposures to the phonemic representations. The present results are consistent with the findings of Hatcher, Hulme, and Snowling (004), showing that the children with risk for reading delay benefit from additional training in phoneme awareness and linking phonemes with letters, and those of Magnan, Ecalle, Veuillet, and Collet (004), indicating that phonological representations can be specified by training involving both phonological and orthographic units. Although this type of training had a positive effect for children who had low entry levels of letter knowledge upon letter naming, no specific effects on reading acquisition were found. It may be the case that this group required more repetitions in order to learn the connections between print and phonology (especially in letter sound correspondences) and that an intervention of such a short duration was insufficient to produce effects on reading. Examination of the training times showed that, in comparison to the group with low phoneme awareness, the group with low letter knowledge spent more time not only at the smaller unit level but also at the higher levels of syllables and words. An important finding was that children with attention difficulties benefited from the program. This is an interesting observation that relates to the findings by Torgesen et al. (1999). They concluded that the attention and behaviour control problems of children made it very difficult for them to profit from a one-to-one teaching situation. When working in groups or even in one-toone tutoring sessions, these children may be less able to direct their attention towards the practice at hand. A game-like method of instruction, even during short training sessions (average 10 minutes), offers a sufficient number of repetitions for learning a letter sound conversion to help children with attention difficulties master the alphabetic decoding strategy. The characteristics of this program can be described as attention catching. First, the player is constantly alerted to maintain concentration in order not to miss points. Second, because the task is similar at every level, the playing principle is easy to master and the child is constantly aware of what he or she is required to do. Third, the child is able to obtain immediate visual feedback from his or her performance, which supports the maintenance of sustained attention in both the auditory and visual domains.

20 174 Sini Hintikka, Mikko Aro, and Heikki Lyytinen The group of children with slow naming speed did not specifically profit from the intervention. However, the children with slow naming speed, in general, showed their best gains in the reading measures. In the Finnish language, naming rate has been shown to be one of the most important predictors of reading speed for older students, at age 8 (Holopainen, Ahonen, & Lyytinen, 001), and continues to be predictive for more than 10 years in the development of reading skills (Korhonen, 1995). It is possible that the correlation between rapid naming and reading fluency will be observed later when there is no longer such a strong emphasis on accurate assembly in reading instruction (Aro et al., 004). To conclude, by means of the training program, contrasted with regular reading instruction, we were able to enhance the letter knowledge of the participants with low pre-reading skills, but not the reading acquisition of the whole garden-variety risk group. However, the results were promising for children with poor access to phonemic representations: their early reading acquisition was accelerated by short-term training in the correspondences between phonological and orthographic units. With this short training program we were unable to obtain differential outcomes between the intervention and control group either in terms of increasing the speed of retrieving phonemic representations or in terms of affecting the reading acquisition of children with low initial letter knowledge or rapid serial naming. Such an outcome can be interpreted in two ways. First, there is a difference between the speed of retrieving a phonological representation from the long-term memory and the quality of this representation, as suggested by de Jong and van der Leij (1999) or, second, that it is difficult to quickly improve speed-related processes in rapid naming (de Jong & Vrielink, 004) or word recognition (Thaler, Ebner, Wimmer, & Landerl, 004). One surprising finding was that children teacher-rated as having low attention abilities, also profited from this game-like computerized training. Game-like instruction with an immediate feedback and short training sessions may offer children with attention problems a sufficient number of repetitions to assist them in learning to use letter sound conversion in the alphabetic decoding strategy. Acknowledgements This paper is part of the first author s doctoral thesis. This research belongs to the Finnish Center of Excellence Program ( ), which has been supported by the Academy of Finland, Niilo Mäki Institute and University of Jyväskylä, and the Finnish National