Journal of Research in Reading

The development of reading comprehension, as the ultimate goal of reading acquisition, has attracted many researchers in various languages. In this context, one of the most prevalent theories is the ‘simple view of reading’ (SVR) model, which was first proposed in English by Gough and Tunmer (1986) and by Hoover and Gough (1990). This model posits that reading comprehension can be explained as the product of decoding abilities and listening comprehension. Listening comprehension is typically assessed by using questions about an orally presented text. Decoding refers to the process of translating printed words into speech and is usually assessed by using pseudowords (Hoover & Gough, 1990; Joshi & Aaron, 2000). Additionally, it was emphasised that both decoding and listening comprehension were necessary for reading comprehension but that neither was sufficient alone (Gough & Tunmer, 1986).

Supporters of the SVR in English claim that these two main and dissociated components include many other reading subcomponents (Hoover & Tunmer, 1993) and that their influence varies with the different phases of reading acquisition (Chen & Vellutino, 1997; Florit & Cain, 2011). The contribution of decoding is reported to be stronger in beginning readers and tends to decrease as readers become more skilled (Duke et al., 2004), whereas the contribution of listening comprehension tends to increase (Catts et al., 2005; Chen & Vellutino, 1997). Some authors suggested that this model should be extended to include other factors that influence reading comprehension (Joshi & Aaron, 2000; Kirby & Savage, 2008). The componential model of reading, for instance, is considered an extended version of the SVR model (Joshi & Aaron, 2000) and proposes taking into account the contribution of not only other linguistic and cognitive factors but also environmental and psychological variables (Aaron et al., 2008; Katzir et al., 2009).

The validity of the SVR has been widely replicated in English orthography where the original model was developed, but some researchers have questioned its power to predict reading comprehension in other languages (Chen & Vellutino, 1997). Currently, findings from other languages and orthographies suggest that while the SVR is adequate, on the whole, in explaining reading comprehension, different patterns of correlations could be found between its components, due mainly to the depth of the various orthographic systems (Florit & Cain, 2011). Decoding seems to be a stronger predictor in opaque orthographies, whereas oral language skills explain more variance in reading comprehension in transparent orthographies. A recent study (Joshi et al., 2012) that compared the SVR in Spanish (as a transparent orthography) and in Chinese and English (as opaque orthographies) found that the model explained more variance in reading comprehension in Spanish (~60%) than in English (~50%) and explained more variance in both of these than in Chinese (between 25% and 42%). Additionally, it was reported that decoding explained less variance in Spanish than in Chinese or English. These authors concluded that orthographic transparency plays an important role in the prediction of reading comprehension. Consistent with this conclusion, the meta‐analysis published by Florit and Cain (2011) indicated that, in comparison with deep orthographies, linguistic comprehension in transparent orthographies is a more important predictor of reading comprehension than decoding, even for beginner readers.

Acquiring reading in deep/opaque orthographies is thought to be more challenging than in transparent ones (Seymour et al., 2003; Ziegler & Goswami, 2005). The ambiguity of the grapho‐phonemic relations in deep orthographies makes decoding a process that develops more slowly, and thus its contribution to reading comprehension is longer lasting. The consistency of the grapho‐phonemic relations in transparent orthographies allows for a rapid and easy development of decoding skills (for accurate and efficient reading) by the end of children's first grade, explaining the rapid decrease in its contribution to reading comprehension (Landi, 2010; Leppänen et al., 2008).

In Semitic languages, the first findings from Hebrew (Primor et al., 2011) showed that the SVR explained between 25% and 34% of the variance in children in the fourth grade. This latter study showed that while orthographic knowledge contributed significantly to reading comprehension, both in normal and disabled readers, morphological knowledge contributed only in the reading disabled group. These results indicated that additional components are necessary for reading comprehension in this Semitic language. In a more recent study conducted with second to tenth‐grade children in Hebrew (Joshi et al., 2015), the SVR explained between 37% and 70% of the variance, and the contribution of orthographic knowledge was significant in all grades.

In Arabic, the other Semitic language that shares similarities with Hebrew but has several linguistic particularities that are hypothesised to seriously affect literacy processes (Abu‐Rabia, 1999, 2002; Ibrahim et al., 2002; Mahfoudhi et al., 2011; Saiegh‐Haddad, 2003, 2004, 2007; Saiegh‐Haddad & Henkin‐Roitfarb, 2014), the validity of the SVR has not yet been examined. Hence, this study aimed to assess to what extent the SVR explains reading comprehension in Arabic among elementary school children.

Specificities of the Arabic language

Arabic is a diglossic language (Ferguson, 1959) in which two different varieties of the language are used for different purposes: spoken and literary Arabic (the latter is also referred to as ‘Modern Standard Arabic’, see Saiegh‐Haddad & Joshi, 2014). The differences between spoken and literary Arabic appear in various aspects of the language, including at the phonological, morphological, semantic and syntactic levels (see, for review, Saiegh‐Haddad & Henkin‐Roitfarb, 2014). These differences, in particular at the level of the core components (lexical and sublexical) of literary Arabic, present certain challenges that affect language development itself and reading acquisition and comprehension more specifically (Saiegh‐Haddad, 2003, 2004, 2007). Regarding the patterns of use of spoken and literary Arabic, children first use spoken Arabic only (i.e. their mother tongue) for purposes of oral communication during the preschool period (~5–6 years). Subsequently, they begin acquiring literary Arabic through formal instruction in reading and writing. Of note is the fact that the use of standard literary Arabic for speaking during lessons/instruction hours at school is not systematic at all (except in Arabic language and religion instruction lessons). Regarding language instruction per se, this process generally follows official curricula in which the literary standard vocabulary is slowly but gradually acquired from the first grade together with simple morphological and syntactic knowledge. A recent review by Al Ghanem and Kearns (2015) indicated that the official reading acquisition curricula in several Arab countries put a strong emphasis on orthographic skills, with less emphasis on phonological skills and very little on the morphological features of the words.

The Arabic orthographic system is considered a complex one where challenges in basic perceptual discrimination of letters may significantly affect acquisition of literacy processes (Ibrahim et al., 2002; Saiegh‐Haddad & Henkin‐Roitfarb, 2014), the speed of word recognition (Abdelhadi et al., 2011; Khateb, Khateb‐Abdelgani, Taha, & Ibrahim, 2014) and consequently, reading comprehension. Arabic orthography is characterised by a certain visual density and a great visual similarity between letters. The system comprises 29 consonant letters, of which three also represent long vowels. Arabic letters belong to dyads or triads, in which the letters share the same basic form but differ in minor features represented by the presence or the absence of dots, by the position of these dots (under or above the letter) and by their number (Holes, 2004; Saiegh‐Haddad & Henkin‐Roitfarb, 2014). Short vowels (i.e. when the script is vowelised) provide the necessary phonological information needed for the accurate decoding of the written words. These short vowels are represented by diacritical marks appearing above or below the letters, in which case the orthography is considered transparent. In the beginning of the process of reading acquisition, children learn to read with the vowelised (transparent) orthographic version. At around the fourth grade, children move progressively to reading unvowelised Arabic texts, and then the orthography is considered to be opaque or deep. In this case, the phonological information is partly unavailable because words are presented with consonants and long vowels only, and many words become homographic (Abu‐Rabia, 2001).

The Arabic morphological system is considered rich and dense. Some authors maintain that morphological units in Arabic are an integral part of the orthographic knowledge (Taha & Saiegh‐Haddad, 2016). Additionally, there is some evidence that indicates that morphology in Arabic is one of the organisational principles of the mental lexicon and that morphology significantly contributes to the words' visual recognition processes (Boudelaa & Marslen‐Wilson, 2001, 2005, 2011). In fact, words in Arabic are produced from the combination of roots representing the meaning of the word and patterns that determine their lexical and syntactical categories (see Saiegh‐Haddad, 2013; Saiegh‐Haddad & Henkin‐Roitfarb, 2014). A majority of the words in Arabic are produced through inflectional processes on the roots and patterns and derivational processes on the roots. The words can include a number of morphemes, making them morphologically complex in the sense that one single Arabic word might correspond to one complete sentence in English. For instance the word <سنذهب> /sanaᵭhab/ which comes from the root <ذهب> /ᵭhb/, corresponds in English to the sentence ‘we will go’. In this example, the root is affixed by the morphemes <ن/س> /s, n/, with the first determining the tense of the verb (<س> /s/, for the future) and the second representing the personal pronoun (<ن> /n/, for ‘we’) that determines the number and person of the subject. In early research, it was suggested that morphological density decelerates the reading rate because readers are required to perform segmentation (Shimron & Sivan, 1994). More recent research indicates that the morphological richness of Arabic (and Hebrew) enhances reading and spelling (e.g. Ravid, 2012; Saiegh‐Haddad, 2013). In particular, authors claim that the internal morphological structure of unvowelised Arabic words permits the restoration of the missing (which is otherwise present in the vowelised orthography) phonological information (Abu‐Rabia, 2007; Saiegh‐Haddad, 2013; Saiegh‐Haddad & Geva, 2008). The authors further argue that the unvowelised orthography is viable only in Arabic because of the specific Semitic morphology of the language, an argument that, to a great extent, also holds for reading in unpointed/unvowelised Hebrew (Bar‐On & Ravid, 2011).

In view of the diglossic situation of Arabic and of the particular characteristics of its orthographic and morphological systems, one can assume that the way in which listening comprehension and decoding contribute to reading comprehension might be different in this language. This assumption is based on the fact that listening comprehension relies primarily on oral language which, in the case of diglossic Arabic speakers, is significantly affected by the spoken dialect. In addition, decoding abilities might behave differently because of the complexity and the particularities of the writing system. Up to now, the existing research on reading comprehension has not yet dealt with the question of the validity of the SVR in Arabic. Accordingly, this research aimed to address the following questions:

1. How do decoding and listening comprehension predict reading comprehension in Arabic transparent orthography in elementary school children?
2. How do the unique orthographic and morphological characteristics of Arabic explain reading comprehension beyond decoding and listening comprehension?

On the basis of previous research on other languages and particularly on Hebrew, we hypothesised that the main components of SVR will predict reading comprehension and that they will explain more variance in beginning readers than in more skilled readers. We also predicted that, due to the uniqueness of the Arabic writing system, orthographic knowledge will contribute to reading comprehension beyond the specific contribution of decoding and listening comprehension. Concerning morphology, we predicted that its influence will be marginal, given that all the texts were presented in a vowelised transparent script in which the phonological information is available to the readers.

Material and methods

Statistical analyses

For the linguistic domains examined by using two or more tests, a general measure (index) for the domain was created after the computation of a coefficient of correlation (i.e. when there were two tests) or a coefficient of reliability (Cronbach's α when more than two tests were used). Based either on a significant correlation or on the reliability measure, average performance on the different tests was computed. Thus, for reading comprehension, a large correlation between the two texts was found (r = .66, p < .001). Similarly, for orthographic knowledge, the correlation between the parsing and orthographic choice tests was large (r = .50, p < .001). For morphological knowledge, the result of the Cronbach's α reliability test was α = .77. To determine the contribution of the different measures to reading comprehension, path (saturated) models (Figure 1) were computed separately for each grade by using amos 18.0 software (Arbuckle, 2009). This model is identical to linear regression analysis and gives the same results in terms of standardised coefficients (β) and % explained variance (R²).

Results

To verify to what extent the two basic components of the SVR are dissociated, a factor analysis was conducted by using Varimax rotation. As presented in Table 2, this analysis yielded two factors. The first explained 51.7% of the variance and comprised decoding, orthographic and morphological knowledge with an eigenvalue of 2.59. The second factor comprised listening and reading comprehension skills and explained 26.7% of the variance with an eigenvalue of 1.34. A small (r = .292) but significant correlation (p < 0.01) was found between the two factors.

The descriptive statistics of the participants' scores and the correlation between the different measures are presented by grade in Table 3. The mean (in % and standard deviation) performance across the participants in each measure and each grade demonstrates noticeable developmental changes between the two successive grades (first and second, third and fourth and fifth and sixth), which were examined very frequently using the same tasks (see, in particular, the mean scores of listening comprehension, orthography and morphology in Table 3). The correlations between all the variables were all statistically significant (p < .01). Specifically, the correlation between decoding and reading comprehension was the largest in the first and second grades and then decreased slightly and stabilised from the third grade onwards (Table 3). A relatively similar pattern was found for the correlations between listening and reading comprehension. The medium‐to‐large correlations found between decoding and listening comprehension in the first and second grades decreased gradually and became small in the fifth and sixth grades. The correlations between orthographic knowledge and both decoding and reading comprehension were large and stable across the grades. Similarly, large and consistent correlations were found between orthographic and morphological knowledge. Finally, the correlation between morphological knowledge and reading comprehension was consistently slightly higher than with decoding and even than with listening comprehension.

To address our research questions, two path models were constructed for each grade. The first model addressed the first research question and examined to what extent the SVR explained reading comprehension in Arabic. Table 4 presents the results of this analysis for each grade. The basic version of the SVR model accounted for 56% of variance in the first grade, which diminished gradually until the fourth grade and then stabilised at approximately 40% (of explained variance) between the fourth and sixth grades. In all the grades, the contribution of both decoding and listening comprehension to reading comprehension was significant (p < .001). The standardised coefficients showed that the contribution of decoding was highest in the first grade (β = .43), diminished gradually until the third grade (β = .28), and then remained stable between the fourth and the sixth grades (β = .36). In contrast, the contribution of listening comprehension increased between first grade (β = .45) and third grade (β = .55) and then decreased in the fourth grade (β = .42) and remained stable until the sixth grade.

To address the second research question, an extended version of the basic SVR model was constructed to investigate the additional contribution of orthographic and morphological knowledge to reading comprehension (Figure 1). The results of this extended model (referred to as SVR‐A, for Arabic) are presented for each grade in Table 4. This analysis revealed first that the inclusion of the morphological and orthographic knowledge explained an additional 10% of the variance in the first and second grades, an additional 14% of the variance in the third and fourth grades and an additional 22–16% of the variance in the fifth and sixth grades. Generally speaking, it was observed that the extended Arabic model accounted for a maximum of 66% of the variance in the first grade, and this decreased to 56% in the sixth grade. Moreover, there was a strongly marked decline in the explained variance in the fourth grade relative to the third. The most interesting observation in this extended version of the SVR‐A was the fact that the contribution of decoding to reading comprehension disappeared and was not significant even in the first grade. In addition, the contribution of listening comprehension to reading comprehension was much weaker than in the basic model, and it declined progressively until reaching a minimum in the fourth grade (β = .14; p < .05).

Regarding the contributions of orthographic and morphological knowledge, they were significant (almost always at p < .001) in all grades. The contribution of orthographic knowledge intensified as the children progressed through grades, but this trend was broken (β = .17, p < .01) in the fourth grade and reached its maximum in the sixth. The morphological contribution already found in the first grade (β = .31) decreased in the second grade (β = .19) and strengthened again to reach its peak in the fourth grade (β = .49); it then diminished again in favour of orthographic knowledge.

Discussion

The current study is the first large‐scale cross‐sectional investigation of reading comprehension in Arabic. We examined the validity of the SVR model in Arabic throughout the first six grades of school. In addition, we investigated how the orthographic and morphological characteristics of this language contributed to reading comprehension beyond decoding and listening comprehension. Our results revealed that while the basic SVR model explained 56% of the variance in the first grade, this decreased gradually to 40% in the sixth grade. The Arabic‐extended model tested here, in which orthographic and morphological knowledge were included, explained an additional 10–22% of the variance in the different grades. Both models explained a moderate portion of variance, which tended to weaken as the children became more skilled in reading. The fact that a significant portion of the variance was not explained by the two models may suggest that the complex process of reading comprehension relies on factors other than those examined here (Aaron et al., 2008; Katzir et al., 2009).

The contribution of both decoding and listening comprehension to reading comprehension was consistently significant in the basic model of the SVR. This observation is consistent with findings from other studies of various languages and orthographies, including English, Greek and Hebrew (Johnston & Kirby, 2006; Katzir et al., 2009; Kendeou et al., 2012; Kirby & Savage, 2008; Vellutino et al., 2007). Whereas the contribution of decoding was similar to that of listening comprehension in the first grade, the contribution of listening comprehension became much stronger in the second and third grades and was somewhat stabilised in the fifth and sixth grades. This similarity between the two components in the first grade contrasts with the view that the contribution of listening comprehension to reading comprehension in transparent orthographies should be more dominant and influential than that of decoding (Catts et al., 2005; Duke et al., 2004). The contribution of decoding weakened as the children became more skilled, while the contribution of listening comprehension simultaneously intensified. In both cases, the trend of prediction changed in the fourth grade and stabilised later. These results support other findings, suggesting that the contribution of decoding and listening comprehension varies in the different stages of literacy development (Chen & Vellutino, 1997; Storch & Whitehurst, 2002). Thus, it seems that decoding has a critical role in the very early stage (first grade) of learning to read (Chen & Vellutino, 1997), and then its impact diminishes to give a more prominent place to listening comprehension in the following stages.

The correlation between decoding and listening comprehension suggests that, despite loading in two separate factors, these components are not dissociated, particularly in the first (r = .47) and second (r = .52) grades. This finding, which contrasts with previous findings for English (Cain et al., 2004), indicates that decoding and listening comprehension rely on some common subskills (linguistic or cognitive or both). The difference with English observed here might be explained in terms of differences in orthographic transparency between Arabic and English, hence the necessity to take into account the specificities of each language when assessing reading comprehension and literacy processes in general (Florit & Cain, 2011; Share, 2008). In transparent orthographies, the contribution of decoding in more skilled readers is reported to decrease, whereas the contribution of linguistic comprehension is reported to increase (Florit & Cain, 2011). Here, the consistent contribution of decoding until the sixth grade needs to be interpreted with caution in view of the fact that the vowelised Arabic orthography (used in this study) is considered transparent. A possible explanation for the differences between our results and others from transparent orthographies might be due to the diglossic situation of Arabic‐speaking children who, in this age range, had been only briefly exposed to the auditory literary language. This explanation appears highly plausible when one considers the fact that the contribution of listening comprehension was expected to dominate in the first grade.

In the SVR‐A, in addition to the added variance beyond that explained by the basic SVR, dramatic changes in the contribution of the basic components were observed. In particular, we found that the contribution of decoding was no longer significant, even in the first grade. A similar finding was presented by Landi (2010), who examined reading comprehension in adult German readers. The author reported that while decoding explained 8% of the variance when entered first into the regression, its contribution nearly disappeared (3%) when decoding was entered after vocabulary and other high‐level skills (Landi, 2010). Other similar findings were obtained from Greek‐speaking second, third and fourth‐grade children (Protopapas et al., 2007). These results again support the view that emphasises the differential influence of specific orthographic systems on reading comprehension (Florit & Cain, 2011). Our results are consistent with several studies that documented the contribution of orthographic knowledge to reading comprehension, including English (MacArthur et al., 2010; Mehta et al., 2009; Silverman et al., 2013), Arabic (Elbeheri et al., 2011) and Hebrew (Joshi et al., 2015; Primor et al., 2011). The observation here that the orthographic contribution was significant in all grade levels is similar to recent findings from Hebrew in second to tenth‐grade children (Joshi et al., 2015).

The orthographic contribution might derive from the correlation between this component and fluency, which is critical for reading comprehension (Fuchs et al., 2009; Perfetti, 2007; Tilstra et al., 2009). Researchers have suggested that, in Arabic, the orthographic complexity of the language slows its processing (Ibrahim et al., 2002). Slow processing is hypothesised to reduce fluency and overload working memory at the expense of higher‐order skills such as reading comprehension (Gathercole et al., 2006; Pikulski & Chard, 2005). Indeed, the formation of orthographic patterns allows the reader to abandon relying on basic decoding processes, enabling them to deal with bigger units such as the orthographic patterns that, in turn, enhance fluency and permit the allocation of attentional resources to higher‐order processes. Hence, the absence of the contribution of decoding to the SVR‐A might be explained by the hypothesis of lexical quality (Perfetti & Hart, 2001). This view assumes that good phonological, orthographic and semantic representations can enable the freeing of cognitive resources for the higher‐order comprehension process. In the presence of orthographic contribution, the effects of decoding on reading comprehension might have been mediated by lexical quality. Indeed, previous research showed that the semantic component (as reflected in oral vocabulary) was the strongest predictor of reading comprehension, even after controlling for word reading and phonological awareness (Muter et al., 2004), especially in the higher grades (Ouellette, 2006; Ouellette & Beers, 2009).

The contribution of listening comprehension to the SVR‐A remained significant even after adding the orthographic and morphological knowledge to the model. Here again, this might be explained by the lexical quality hypothesis (Perfetti & Hart, 2001) because listening comprehension contains both phonological and semantic components. However, the trend observed for the contribution of listening comprehension in the SVR (increasing from the first to the third grade) was inverted in the SVR‐A. The increasing contribution of listening comprehension with increases in grade is consistent with findings from English (Catts et al., 2005; Chen & Vellutino, 1997). The inversion of this trend in the SVR‐A might reflect the particular effects of orthographic knowledge in general and of the specificities of the Arabic orthographic system on reading comprehension in particular.

Similar to the contribution of orthographic knowledge, morphological knowledge was consistent across grades, a finding that supports the critical role of morphology in reading comprehension, particularly in Semitic languages (Abu‐Rabia, 2007; Carlisle, 2000; Deacon & Kirby, 2004; Primor et al., 2011; Nagy et al., 2006; Ravid & Mashraki, 2007; Ravid & Schiff, 2006; Schiff et al., 2011; Siegel, 2008). This contribution was already present in the first grade, peaked in the fourth grade (β = .49), and then decreased smoothly towards the sixth grade. The consistent contribution of morphological knowledge might be explained by the role that morphology in Semitic languages seems to play in extracting phonological information from print, even in vowelised words (Bar‐On & Ravid, 2011; Saiegh‐Haddad, 2013; Saiegh‐Haddad & Geva, 2008). Although not tested in the study, the contribution of morphology might also be explained by its strong link with vocabulary, where it is thought to be critical for the development of the mental lexicon (Boudelaa & Marslen‐Wilson, 2011; Carlisle, 2000).

Our results showed that the basic SVR, constructed initially for English and explaining between 70 and 90% of the variance in reading comprehension, could explain ~50% in Arabic. These results support the validity of the SVR model for Arabic but emphasise the need to address the uniqueness of the language at hand (Share, 2008) because the inclusion of orthographic and morphological knowledge explained 10–20% more of the variance. The need to address language specificities appears to be particularly crucial when speaking about ‘major’ differences between languages, such as that of the depth/transparency of the orthographic systems. In fact, the contribution of decoding to reading comprehension was expected to disappear around the second and third grades as in other transparent orthographies (Florit & Cain, 2011), as this is a time when children progress to more efficient and fluent reading. The consistent involvement of decoding up to the sixth grade (in the basic model of SVR) can thus be considered a strong indication that Arabic orthography behaves not as a transparent but rather as a deep orthography. Additionally, this long‐lasting involvement of decoding provides further evidence regarding the difficulty of reading in Arabic (Abu Ahmad et al., 2014) and suggests that children continue to some extent to rely on the primary grapho‐phonemic conversion processes (i.e. decoding) until later and thus are not very efficient in their reading. Thus, it appears that the complexity of the Arabic orthographic system (Ibrahim et al., 2002) obliges the reader to rely on decoding processes that require more cognitive resources, which are necessary for reading comprehension (Perfetti, 2007).

Acknowledgments

This research was supported by the Edmond J. Safra Brain Research Center for the Study of Learning Disabilities in the context of the doctoral work of I. A. and by the Israeli National Science Foundation (grant no. 623/11). The authors declare that there is no conflict of interest related to the work presented in this paper. This article is dedicated to the memory of Prof. Zvia Breznitz, head of the Edmond J. Safra Brain Research Center for the Study of Learning Disabilities, who passed away on 18 May 2014.