In her Keynote, Gathercole (2006) provides a comprehensive review regarding the nature of the nonword repetition (NWR) task and a compelling argument for the utility of the task as a robust index of children's phonological short-term storage capacity. She further argues that temporary phonological storage acts as a primitive learning mechanism that plays an instrumental role in the lexical learning of young children, including children with specific language impairment (SLI). In this Commentary, we focus our remarks on two related themes concerning the relation of NWR and SLI: (a) performance on NWR tasks provides a window into these children's more general cognitive (dis)abilities and (b) the robustness of NWR as a predictor of these children's lexical knowledge and learning abilities.

Gathercole and Baddeley's (1990) paper has become a seminal work for those who investigate the cognitive and language characteristics/associations in children with SLI. Children with SLI are those who are generally developing in a typical manner in all respects except for demonstrable receptive and/or expressive language impairments. These children exhibit language difficulties across the different domains, including lexical, morphological, and syntactic. However, they exhibit particular weakness in the areas of grammatical morphology and syntax (for a review, see Leonard, 1998. Gathercole and Baddeley (1990) were the first investigators to propose that children with SLI have a marked deficit in phonological short-term memory (PSTM) capacity (as indexed by their poor repetition of three-, four-, and five-syllable nonwords). They further argued that this deficit may well be causal to their language impairments.

Gathercole (2006) points out that NWR tasks involve various language-related factors (e.g., knowledge of phonotactic probabilities in the language, knowledge of canonical stress patterns of words) and a range of more general cognitive and memory-related processes (e.g., perceptual and encoding processes, redintegration processes, retrieval, articulatory planning, and execution processes), but argues that it is an especially sensitive index of children's phonological storage capacity. She argues this to be the case for both typically developing (TD) children and children with SLI. Although this may well be true for TD children, we urge caution in making the same claim about children with SLI.

With respect to language-related factors, the data for TD children are consistent in demonstrating that knowledge of constituent syllable structure (Dollaghan, Biber, & Campbell, 1993) and the phonotactic probably of phoneme sequences (Coady & Aslin, 2004; Munson, 2001; Munson, Edwards, & Beckman, 2005) supports NWR. The same also appears to generally hold for children with SLI. For instance, NWR of children with SLI also appears to be affected by the degree of similarity between nonword items and real words (Bishop, North, & Donlan, 1996; Briscoe, Bishop, & Norbury, 2001; Montgomery, 1995; Munson, Kurtz, & Windsor, 2005). Nonwords that more closely approximate the features of real words are repeated more accurately than those that diverge from the properties of real words. Such findings indicate that children with SLI are sensitive to the same lexical properties as TD children and use this knowledge to facilitate NWR. At the same time, however, many children with SLI exhibit lower vocabulary and a reduced network of lexical knowledge (e.g., Kail & Leonard, 1986) compared with their age peers. Such differences would likely lead to lower lexical and phonological neighborhood densities and phonotactic probabilities in the lexicons of children with SLI. These limitations may well make NWR somewhat less a language-related task for children with SLI, at least when compared with many of their TD age peers. However, as the lexicons of children with SLI grow and become more richly and densely connected the same language-related knowledge sources will likely play a greater facilitative role in their NWR.

Comparing the NWR of children with SLI with that of younger, language-matched TD children suggests that the poor NWR of children with SLI is constrained not by poorer language-related knowledge but by a variety of nonlanguage-related factors. Both Gathercole and Baddeley (1990) and Montgomery (1995) reported that, relative to a group of younger, language-matched children, children with SLI showed significantly poorer NWR performance. Because receptive vocabulary was accounted for in both of these studies, the poorer NWR of the children with SLI cannot be easily attributed to differences in receptive language-related knowledge. However, in a more recent study by Munson et al. (2005) no difference in NWR was found between a group of children with phonological disorder and a younger, language-matched group of TD children. These findings notwithstanding, it is safe to assume that at least some of the NWR difficulties of children with SLI are attributable to nonlanguage-related abilities.

Although it is true that various cognitive and memory-related processes mediate TD children's NWR, it may well be that it is these factors that most strongly distinguish children with SLI and TD children in NWR ability. Unlike most TD children, many children with SLI exhibit deficits in one or more of the processes assumed to underlie NWR, including poorer rapid rate acoustic–phonetic processing (Bishop, Bishop, Bright, Delaney, & Tallal, 1999; Tallal, Stark, & Mellits, 1985), perhaps inferior phonological encoding leading to the creation of less precise and/or stable phonological representations (e.g., Edwards & Lahey, 1998; Evans, Viele, Kass, & Tang, 2002), inefficient lexical retrieval operations (Kail & Leonard, 1986), and/or problems with complex articulatory production (e.g., Bishop et al., 1996; Briscoe et al., 2001). It is because NWR entails such a wide range of input and output processing requirements that we suggest that NWR performance represents a more complex psycholinguistic exercise for children with SLI than TD children. That is, the poor NWR by children with SLI reflects more than just reduced PSTM capacity (or a PSTM deficit coupled with weak auditory processing abilities; Bishop et al., 1999). It also reflects these children's broader cognitive deficits. It is precisely because NWR taps so many different cognitive and memory-related skills that it has proved to be a highly reliable discriminator of TD children and children with SLI (Bishop et al., 1996; Conti-Ramsden, Botting, & Farragher, 2001; Dollaghan, & Campbell, 1998; Ellis Weismer et al., 2000). Thus, any conclusion ascribing the poor NWR performance of children with SLI to primarily a phonological storage deficit is premature.

Our second point relates to the robustness of NWR performance in predicting level of vocabulary knowledge and novel lexical learning in SLI. As indicated by Gathercole (2006), the association between NWR and word learning in TD children is far from straightforward. It is important, though, that NWR tends to be a good/stable predictor of TD children's level of vocabulary knowledge and novel word learning. The relation in children with SLI appears to be even less straightforward, especially given the mixed findings reported in several recent studies. Munson et al. (2005), for instance, found NWR and existing receptive vocabulary knowledge to be significantly correlated. Other researchers, however, have failed to report a similar positive correlation (e.g., Botting & Conti-Ramsden, 2001; Edwards & Lahey, 1998). The evidence documenting the relation between NWR and novel word learning (as indexed by receptive measures) in SLI is likewise mixed. Gray (2004), for example, showed a positive association between NWR accuracy and novel receptive word learning in a group of preschoolers with SLI. Ellis Weismer (1996) also reported a similar positive relation in a group of young school-age children with SLI. However, Horohov and Oetting (2004) failed to find a correlation between NWR and receptive word learning in a group of school age children with SLI. The picture regarding an association between NWR and vocabulary in SLI becomes even less clear when one looks for an association with expressive vocabulary. Results from studies by Briscoe et al. (2001) and Ellis Weismer (1996) showed no positive correlation between NWR and level of expressive vocabulary (Briscoe et al.) or between NWR and novel lexical learning as measured by novel word production (Ellis Weismer, 2006).

The lack of consistent correlations between NWR and vocabulary knowledge and lexical learning in SLI may relate to the possibility that NWR tasks do not adequately capture the nature and scope of the complexity of the relation. Gathercole (2006) alludes to this possibility, and we would echo this notion. The SLI literature is clear in demonstrating that children with SLI have verbal memory deficits that extend well beyond phonological storage. They also show difficulty performing a variety of tasks requiring simultaneous verbal processing and storage (e.g., Ellis Weismer, Evans, & Hesketh, 1999; Marton & Schwartz, 2003; Montgomery, 2000). It is far more likely that a greater portion of the variance in SLI vocabulary and word learning would be explained by a matrix of memory measures that captures the simultaneous verbal processing and storage abilities of these children (e.g., Ellis Weismer & Thordardottir, 2002).

There is no debate that children with SLI demonstrate inferior NWR relative to most age-matched TD children and even many younger, language-matched TD children. There is, however, debate as to how best to explain these children's NWR deficit. We argue that the evidence thus far appears to favor a more multicausal explanation than it does a primarily phonological storage deficit account or even a phonological storage deficit plus auditory processing impairment account (Bishop et al., 1999). With respect to the association between NWR and word learning in these children, our understanding is complicated by the fact that (a) the significant correlations between NWR and measures of word knowledge and/or learning are not consistent across studies, and (b) NWR appears to be more predictive of receptive vocabulary than of expressive vocabulary. The NWR task has proven to be a valuable theoretical and clinical heuristic for those researchers investigating the nature of SLI. Future investigations will certainly continue to explore the linguistic and cognitive underpinnings of the NWR task, as well as the ability of the task to explain a meaningful portion of the variance in the language learning and performance of children with SLI. The theoretical and clinical utility of the task as it relates to SLI depends on it.