Working memory—the ability to store information while simultaneously carrying out processing operations—is a well-established predictor of individual variation in reading comprehension performance in both adults (Daneman & Merikle, 1996) and children (Cain, Oakhill, & Bryant, 2004a). In the literature, it is debated whether individual differences in reading comprehension are best explained by processing or storage capacities of working memory. Various studies support the view that processing capacities tapped by working memory tasks in both the phonological and the semantic domain are important in explaining variance in reading comprehension (Daneman & Merikle, 1996). The role of storage has been investigated in the phonological domain but is less clear in the semantic domain since studies have typically used storage measures that tap into storage of phonological information rather than into semantic information (Haarmann, Davelaar, & Usher, 2003). Although some studies with adults (Haarmann et al., 2003) and children with difficulties in reading comprehension (Nation, Adams, Bowyer-Crane, & Snowling, 1999; Nation & Snowling, 1998) have suggested a link between reading comprehension and semantic storage, it is currently unknown if semantic storage contributes to reading comprehension in typically developing children. Moreover, it is by no means clear what the relative contribution is of phonological and semantic storage, on one hand, and phonological and semantic working memory, on the other hand, to children’s reading comprehension. Furthermore, it is unclear how semantic storage, semantic working memory and reading comprehension are related. Therefore, in the present study, children’s reading comprehension were related to their storage and processing capacity, in both the phonological and semantic domain.
Reading comprehension is the product of a complex integration of knowledge and skills such as decoding (Lyon, 1995; Torgesen, 2000), vocabulary (Verhoeven & van Leeuwe, 2008), and syntactic (Cutting & Scarborough, 2006; Oakhill & Cain, 2011) and semantic processing (Nation et al., 1999; Torgesen, 2000). In addition, reading comprehension depends on higher-level control functions (Cain, 2006; Christopher et al., 2012), among which working memory is the most well-established predictor in both adults (Daneman & Merikle, 1996) and children (Cain et al., 2004a, b). A commonly applied working memory model in the reading comprehension literature, is the model of Baddeley and Hitch (1974; see also Baddeley, 2000). According to the original model, working memory is composed of a central executive and two storage components, namely the visuospatial sketchpath and the phonological loop, encoding visuospatial and verbal information, respectively. More specifically, the phonological loop temporarily preserves verbatim representations of presented words and keeps this information active and accessible during the performance of complex cognitive tasks, which is controlled by the central executive. Various memory tasks have been designed based on Baddeley’s (2000) model, including tasks measuring the storage of information only, and working memory tasks reflecting the processing component of the central executive, in addition to storage of information. Working memory measures have a higher predictive value of reading comprehension performance than measures that assess storage only in adults (Daneman & Merikle, 1996), children (Cain, 2006) and children with reading comprehension difficulties (Carretti, Borella, Cornoldi, & De Beni, 2009). These results have been taken to suggest that it is the general processing capacities tapped by working memory tasks that are important in explaining variance in reading comprehension, rather than the storage component (Cain et al., 2004a, b; Daneman & Merikle, 1996).
Indeed, working memory tasks explain variance in reading comprehension regardless of whether they mainly involve non-verbal processing (recall visual patterns and/or spatial traces) or verbal processing (Carretti et al., 2009; Daneman & Merikle, 1996). There is, however, substantial evidence that the linguistic information tapped by working memory tasks is of primary importance with regard to explaining variance in reading comprehension (Daneman & Merikle, 1996). There is considerable variation in the kind of language processing involved among the different types of verbal working memory tasks, ranging from tasks that tap mainly into phonological processing (e.g., backward digit span tasks) to tasks that tap mainly into semantic processing (e.g., listening span tasks), and tasks that lie somewhere in between. During a backward digit span task, participants are asked to recall verbally presented digits in reverse order. Hence, the task requires storage and processing of verbatim information that contains a minimal amount of syntactic and semantic relations between items. During a listening span task, participants listen to a set of unrelated sentences and judge if sentences are semantically correct or incorrect. After the set of sentences has been presented, participants are asked to recall the sentence-final words (Daneman & Carpenter, 1980). In addition to verbatim encoding, the listening span task requires participants to integrate the presented items based on syntactic and semantic information. In other words, the listening span task relies on processes that serve language comprehension (Hulme et al., 1997; Knott, Patterson, & Hodges, 1997; Walker & Hulme, 1999). Semantic working memory tasks have been shown to be better predictors of reading comprehension than working memory tasks that mainly tap phonological processing (Cain et al., 2004a, b; Daneman & Merikle, 1996; Oakhill, Cain, & Bryant, 2003; Seigneuric, Ehrlich, Oakhill, & Yuill, 2000) and non-verbal working memory tasks (Shah & Miyake, 1996) in both typically developing children and adults. Similarly, children with difficulties in reading comprehension have shown deficits solely in verbal working memory, with the most profound deficits on tasks mainly tapping into semantic processing (Cain, 2006; Cain, Oakhill, & Lemmon, 2004b; Carretti et al., 2009; De Beni, Palladino, Pazzaglia, & Cornoldi, 1998; Nation et al., 1999). This has lead to the claim that not all variation in working memory can be explained by general processing capacity, but that linguistic information tapped by memory tasks, must play an important role as well (Daneman & Merikle, 1996).
In a similar way, the degree to which tasks that measure storage only rely on semantic rather than phonological aspects of stored representations may influence the extent to which performance on storage tasks explains variation in reading comprehension. Studies investigating the role of storage in reading comprehension have commonly used measures such as the forward digit span task, which requires immediate verbatim recall of a number of items (digits, letters or words) in exact serial order, thought to take place in Baddeley’s phonological loop (Baddeley, 2000). Correlations between performance on phonological storage tasks and reading comprehension in children were not significant (Leather & Henry, 1994; Swanson & Berninger, 1995; Yuill, Oakhill, & Parkin, 1989) or very low (Daneman & Carpenter, 1980; LaPointe & Engle, 1990; Turner & Engle, 1989). Additionally, children with difficulties in reading comprehension performed similarly to controls on these types of storage tasks (Nation et al., 1999; Oakhill, Yuill, & Parkin, 1986; Stothard & Hulme, 1992). To summarize, phonological storage has been found to be a poor indicator of reading comprehension performance.
There are, however, several indications that the ability to store semantic information may contribute to individual variation in reading comprehension. Children with difficulties in reading comprehension do not appear to benefit from the availability of long-term semantic representations to the same extent as controls: children with reading comprehension difficulties show a poorer performance on the recall of abstract and low frequency words compared to the control children, but perform similarly on the recall of concrete and high frequency words, suggesting that the deficiencies lie in the recall of semantic information (Nation et al., 1999; Nation & Snowling, 1998). Moreover, Haarmann et al. (2003) have shown that the conceptual span task designed to tap mainly into semantic storage explained unique variance in adult reading comprehension over and above a word span task. Based on these results it can be hypothesize that semantic, rather than phonological information tapped in storage tasks may explain variation in reading comprehension. Moreover, these results question the assumption that it is mainly the general processing component tapped by working memory tasks, rather than storage of the items involved, that is important in explaining variance in reading comprehension.
However, to our knowledge, research into the contribution of semantic storage and inherently, the relative contribution of phonological and semantic storage, on one hand, and phonological and semantic processing, on the other hand, to reading comprehension, has not yet been reported. Additionally, although Daneman and Carpenter’s listening span task (1980), is assumed to reflect simultaneous storage and processing of semantic information, the contribution of semantic storage to performance on the listening span task has not been explicitly investigated. Insight in this matter would be useful as the listening task is frequently used to assess working memory in the reading comprehension literature.
In the present study we aimed to examine the relation between phonological and semantic storage and processing capacities and reading comprehension in Dutch fifth grade children, after controlling for their vocabulary and word decoding. More specifically, we posed four research questions. The first question relates to the contribution of the phonological and semantic storage measures to reading comprehension:
-
1.
Is semantic, but not phonological, storage a direct predictor of reading comprehension?
The other three questions speak to a model in which semantic and phonological working memory tasks, which assess processing in addition to storage, were added to the model:
-
2.
Is processing, but not storage, a direct predictor of reading comprehension?
-
3.
Is semantic, but not phonological, processing a direct predictor of reading comprehension?
-
4.
If so, does semantic storage indirectly predict reading comprehension via semantic working memory?