Research on the role of working memory in language disorders has stemmed mainly from the phonological loop model (e.g., Baddeley, 1986; Gathercole and Baddeley, 1993) or the capacity theory of comprehension (e.g., Just and Carpenter, 1992). These models differ in their conception of working memory and in the paradigms typically used to assess this construct (cf. Montgomery, 2000a); however, a central premise of both frameworks is that there is a limited pool of operational resources available to perform computations, such that processing and storage of linguistic information is degraded when demands exceed available resources. Numerous investigations based on these two approaches have demonstrated an association between working memory capacity and normal language functioning in children and adults. In young children, individual differences in phonological working memory predict vocabulary development and are related to differences in word repertoire, utterance length, and grammatical construction use (e.g., Gathercole and Baddelely, 1990b; Adams and Gathercole, 2000). School-age children's performance on working memory measures is significantly correlated with spoken language comprehension as well as with reading recognition and comprehension (e.g., Gaulin and Campbell, 1994; Swanson, 1996). Working memory capacity predicts a number of verbal abilities in adults, including reading comprehension levels, understanding of ambiguous passages and syntactically complex sentences, and the ability to make inferences (e.g., King and Just, 1991; Carpenter, Miyake, and Just, 1994).

Investigators have examined short-term or working memory abilities in children with varying profiles of language and cognitive deficits, including children with Down syndrome, Williams' syndrome, Landau-Kleffner syndrome, learning disabilities, and specific language impairment (SLI). Of special interest are children with SLI, who demonstrate significant language deficits in the absence of any clearly identifiable cause such as mental retardation or hearing loss. One theoretical camp views SLI in terms of limited processing capacity. There are various formulations of limited capacity accounts of SLI, including hypotheses about specific deficits in phonological working memory and hypotheses regarding more generalized difficulties in information processing and storage that affect performance across modalities (cf. Leonard, 1998). Difficulties discussed here are limited to poor nonword repetition, reduced listening span, and poor serial recall.

Children with SLI exhibit deficits in nonword repetition, a paradigm that has been used extensively by Baddeley and colleagues (and others as well) as a measure of phonological working memory (Gathercole and Baddeley, 1990a; Montgomery, 1995; Dollaghan and Campbell, 1998; Edwards and Lahey, 1998; Ellis Weismer et al., 2000; Briscoe, Bishop, and Norbury, 2001). Nonword repetition has proved to be useful clinically as a culturally nonbiased measure for distinguishing between children with and without language disorders. In one of the initial investigations of nonword repetition in SLI, Gathercole and Baddeley (1990a) concluded that children with SLI demonstrate significantly poorer phonological working memory than controls matched on nonverbal cognition or language level (however, see van der Lely and Howard, 1993; Howard and van der Lely, 1995). The findings of Gathercole and Baddeley (1990a) were replicated by Montgomery (1995), who similarly interpreted his results as indicating that children with SLI have reduced phonological memory capacity.

Other studies have sought to determine whether difficulties with nonword repetition reflect cognitive processes other than working memory deficits (Edwards and Lahey, 1998; Briscoe, Bishop, and Norbury, 2001). After a thorough investigation of possible explanations for nonword repetition deficits in SLI, Edwards and Lahey (1998) concluded that neither auditory discrimination nor response processes could account for the difficulties. Instead, they attributed the deficits to problems in the formation or storage of phonological representations in working memory. Children with SLI usually do not differ from normal language peers in their ability to repeat short, simple nonwords; rather, breakdowns on nonword repetition tasks typically occur on the most complex stimuli (Ellis Weismer et al., 2000; Briscoe, Bishop, and Norbury, 2001). When children with SLI were compared with children with mild to moderate hearing loss, both groups showed similar difficulty with longer nonwords, but children with SLI also displayed deficits on digit recall and were more negatively affected by phonological complexity (Briscoe, Bishop, and Norbury, 2001). These investigators concluded that auditory perceptual deficits are not sufficient to explain the range of language and literacy difficulties observed in children with SLI and suggested that some kind of processing capacity limitation underlay their language deficits.

Several genetic investigations of developmental language disorder have examined phonological memory as indexed by nonword repetition. Bishop, North, and Donlan (1996) administered a nonword repetition task to participants in a study of twins with language impairment. Children with persistent language impairment as well as those with resolved language impairment exhibited significant deficits in nonword repetition. Comparison of nonword repetition performance in monozygotic and dizygotic twin pairs revealed a significant heritability component. Based on these results, Bishop et al. suggested that deficits in nonword repetition provide a phenotypic marker of heritable forms of developmental language disorder. Bishop and colleagues (1999) replicated the earlier results and found that nonword repetition gave high estimates of group heritability. This measure was a better predictor of low language scores than was a measure of auditory processing (Tallal's Auditory Repetition Test). Tomblin and colleagues (2002) recently investigated candidate genes associated with developmental language disorder, testing for associations between candidate loci in a sample of 476 children and their parents. A two-stage approach was used to search for loci associated with language disorder, as diagnosed by standardized tests of listening and speaking or by a measure of phonological memory (nonword repetition). Preliminary results were suggestive of an association of CFTR (a marker on chromosome 7) with both the phonological memory and spoken language phenotypes.

Another paradigm widely employed in research on the association between language and working memory abilities uses a listening/reading span task (e.g., Daneman and Carpenter, 1980). The person is required to perform two tasks concurrently (involving processing and storage), such as making true/false judgments about sentences and recalling the last word in each sentence following the presentation of all sentences in a set. The number of sentences within a set increases throughout the task in order to assess memory span. Ellis Weismer, Evans, and Hesketh (1999) found that children with SLI evidenced limitations in verbal working memory compared to age-matched controls, based on their performance on a listening span task developed by Gaulin and Campbell (1994). Findings primarily pointed to quantitative differences between the groups involving reduced capacity for the children with SLI; however, there were some indications of qualitative differences in terms of distinct patterns of word-recall errors and different patterns of associations between working memory and performance on language and nonverbal cognitive measures. Montgomery (2000a, 2000b) examined the relation between working memory and sentence processing in children with SLI. Using an adaptation of Daneman and Carpenter's listening span task, he demonstrated that children with SLI exhibit reduced capacity under dual-load conditions. Performance on the listening span measure was significantly correlated with performance on an off-line sentence comprehension task but not with on-line sentence processing. Montgomery concluded that the slower real-time sentence processing in children with SLI was primarily a function of inefficient lexical retrieval operations rather than limitations in working memory; however, he posited that their difficulties with off-line sentence comprehension tasks were related to difficulties coordinating the requisite processing and storage functions, revealing limitations in functional working memory capacity.

Serial memory deficits in children with SLI have been documented by Gillam and colleagues (Gillam, Cowan, and Day, 1995; Gillam, Cowan, and Marler, 1998). The initial study employed a suffix effect procedure in which a spoken list to be recalled was followed by a “suffix” (nonword item) that was not to be recalled. The suffix had a disproportionately negative effect on recency recall for the children with SLI when strict serial position criteria for scoring were imposed. The subsequent investigation by Gillam et al. sought to determine the nature of working memory deficiencies in children with SLI, using a modality effect paradigm in which input modality, rate of input, and response modality were manipulated. To control for differences in capacity across the groups, trials were administered at a level consistent with each child's working memory span. Children with SLI and controls demonstrated traditional primacy, recency, and modality effects and similar performance when audiovisual stimuli were paired with spoken responses. However, children with SLI exhibited reduced recency effects and poor recall when visually presented items were paired with pointing responses. The investigators concluded that neither output processes nor auditory temporal processing could account for the working memory deficits in children with SLI. They suggested instead that children with SLI have problems retaining or transforming phonological codes, particularly on tasks requiring multiple mental operations. They further speculated that these capacity limitations in working memory may be due to rapid decay of phonological representations or to performance limitations involving the use of less demanding coding and retrieval strategies.

In conclusion, there is considerable evidence that children with SLI have limitations in working memory, yet there are a number of unresolved issues. In light of the known heterogeneity of the SLI population, it seems unlikely that any single factor can account for the language difficulties of all children. Additional research is warranted to examine individual variation within this population and to explore whether limitations in working memory are differentially implicated in various subtypes of SLI. Another important issue pertains to whether deficits in working memory capacity are restricted to processing of verbal material or extend to nonverbal information as well. That is, it is important to determine whether the evidence supports a domain-specific model or a generalized capacity deficit model. Finally, future studies should determine whether memory limitations are actually a causal factor in SLI, an outgrowth of the language problems, or an independent area of difficulty for children with language disorder.