Individual differences in nonword repetition (NWR) show a particularly strong association with vocabulary acquisition for both first- (L1) and second-language (L2) learners, and they serve as a behavioral marker for specific language impairment (SLI) in children (Gathercole, 2006). However, this association is susceptible to alternative explanations.

Proponents of the so-called phonological processing account (PPA; e.g., Bowey, 1996, 1997, 2001; Chiat, 2001; Snowling, Chiat, & Hulme, 1991; Snowling, Goulandris, Bowlby, & Howell, 1986) view NWR as a complex task involving several components, most involving phonological processing. These include speech perception, the construction, and encoding of a phonological representation in the phonological store, maintenance of this representation, retrieval of the representation from the phonological store, assembly of articulatory instructions, and articulation itself. Difficulties in one or more of these components may impair NWR. Findings that manipulations that increase phonological complexity adversely affect NWR support the PPA. Nevertheless, because these manipulations sometimes place additional demands on phonological store capacity, some require further investigation (Gathercole, 2006).

There is considerable overlap between Gathercole's (2006) position and the PPA. Advocates of the PPA acknowledge that NWR is constrained by phonological store capacity (e.g., Bowey, 1996, 1997). Although phonological store capacity can impair new word learning (e.g., Gathercole, Hitch, Service, & Martin, 1997), Gathercole concedes that NWR is not a pure measure of the capacity of the phonological store, and that poor phonological storage may not itself be sufficient to produce the severe NWR and language deficits associated with SLI.

Although they differ concerning the relative centrality of the role ascribed to phonological store capacity per se in constraining NWR and language learning, the PPA and a phonological store capacity account (PSCA) represent complementary rather than incompatible accounts of NWR. The extent to which NWR reflects phonological processing efficiency or phonological store capacity varies not only between individuals but also with nonword characteristics. Moderately long nonwords with high phonotactic probabilities and familiar prosody are efficiently encoded, maintained, retrieved, and articulated by individuals with a mature and efficient phonological processing system. Thus, NWR is more likely to reflect phonological store capacity when such individuals repeat this type of nonword than when children with immature or inefficient phonological processing systems recall nonwords with unfamiliar phonotactics and prosody.

Several findings support the PPA over the PSCA. For instance, findings that NWR is more predictive of early L1 and L2 vocabulary acquisition (see Gathercole, 2006), where the phonology of the language is less familiar, are consistent with the PPA. However, if NWR primarily reflects phonological store capacity, and if this largely constrains vocabulary acquisition, then NWR should predict vocabulary acquisition just as well beyond the earliest stages, especially in adult L2 learners.

Similarly, the findings of Archibald and Gathercole (2005) pose difficulties for the PSCA but not the PPA. When nonverbal ability effects were covaried, children with SLI differed from language controls in NWR but were no worse in recalling the same syllables presented as a series one-syllable items. Gathercole (2006) views this finding as puzzling, given her view that the phonological content is equivalent in the NWR and serial recall tasks. However, although Archibald and Gathercole claim that all syllables contained tense vowels, nonwords presented as isolated consonant–vowel (CV) syllables are more likely to have fully stressed vowels that are longer in duration, and consonants in isolated CV syllables also contain clearer and more invariant acoustic–phonetic information (see Cole & Jakimik, 1978). These features would make it easier for SLI children to construct higher quality phonological representations of CV nonwords presented as isolated syllables than as a single multisyllabic nonword, and thus reduce the overall reliance of the serial recall task on phonological processing efficiency per se. The serial recall items were presented at rate of 1 per 750 ms, leading to decay of material in the phonological store (Baddeley, 1986; Brown & Hulme, 1995). Although these items could be rehearsed, redintegration of nonwords is minimal (Gathercole, 2006). Because the recall of isolated CV syllables thus comprises a purer test of phonological store capacity, the PSCA wrongly predicts that, relative to language controls, SLI children should show greater deficits in repeating a series of nonword syllables than multisyllabic nonwords.

Gathercole (2006) notes that accounts of SLI must explain the finding that, relative to controls, SLI children show more marked deficits with longer nonwords than shorter nonwords. This finding is readily explicable by the PPA. Item length effects in NWR are unlikely to reflect rehearsal (see Gathercole, Willis, Baddeley, & Emslie, 1994, p. 122). NWR length effects per se can be explained in probabilistic terms, in addition to phonological store capacity, and may also reflect trace decay and possibly interference during retrieval and recall (Brown & Hulme, 1995). If SLI children initially encode lower quality phonological representations of nonwords, which are even more prone to decay and interference, then they will show a more marked length effect (Brown & Hulme, 1995).

Following the early work of Gathercole, Willis, and Baddeley (1991), Bowey (1996, 2001) investigated the extent to which NWR and phonological sensitivity (or phonological awareness¹

) account for largely shared or independent variance in vocabulary in 5-year-olds. NWR and phonological sensitivity (operationalized by phonological identity and rhyme oddity performance) were hypothesized to be alternative surface manifestations of an underlying phonological processing ability. Thus, findings that, with performance IQ effects first controlled, phonological sensitivity and NWR accounted for largely common variance in vocabulary (Bowey, 1996, 2001; see also Metsala, 1999) were interpreted as most consistent with the PPA. However, some aspects of Bowey's (2001) findings suggested that phonological store capacity independently contributes to children's vocabulary acquisition.

Gathercole (2006) suggests that findings that phonological sensitivity and NWR explain largely overlapping variance in vocabulary may reflect the incorporation of a substantial phonological memory component within the phonological sensitivity tasks. A rhyme oddity task, in which children name one of three simple spoken words that does not rhyme with the other two, requires three words to be held in memory while the phonological judgment is made, and thus includes a phonological memory component (see Baddeley, 1986; Bowey, 1994). However, to reduce the memory demands of Bowey's (1996) rhyme oddity task, all spoken words were represented by pictures that remained visible throughout each trial. Furthermore, Bowey (1996, 2001) observed parallel findings from a highly explicit phoneme identity task that minimizes memory requirements (see Bowey, 1994). Here, for each item, children see a series of three pictures, two sharing a final phoneme (e.g., knife, mop, leaf). Children are told the name of the top picture (knife) and the last sound of that word (/f/), and are then asked which of the two spoken words (mop and leaf), whose referents are pictured below, end with /f/. This task is similar to that used to investigate rhyme sensitivity in PV, a patient with an auditory span of only two or three words or digits (Vallar & Badddley, 1984).

Gathercole (2006) distinguishes a phonological sensitivity hypothesis from the more general PPA. Nowhere in Bowey's (1996, 1997, 2001) work is there any proposal that phonological sensitivity per se directly contributes to NWR, vocabulary, or language acquisition.²

Bowey's (1996, 1997) theoretical discussions were framed entirely in terms of the phonological processing demands of the NWR task, with NWR and phonological sensitivity viewed as alternative indicators of a latent phonological processing ability. Bowey (2001) linked the PPA to what may be termed the protracted lexical restructuring hypothesis, according to which the process of establishing fine-grained segmental representations of vocabulary items extends into middle childhood, driven largely by the need to distinguish similar-sounding words (see Charles-Luce & Luce, 1990; Walley, 1993). What Gathercole (2006) refers to as phonological sensitivity should be termed lexical restructuring; thus, the phonological sensitivity hypothesis should be termed the lexical restructuring hypothesis. Researchers who argue that detailed segmental representations are established virtually from the outset (e.g., Ballem & Plunkett, 2005; Fennell & Werker, 2003; Swingley, 2003) challenge this view. Much of the work pertaining to the lexical restructuring hypothesis cannot be definitively interpreted (see Bowey & Hirakis, in press). However, the PPA is not inherently tied to the lexical restructuring hypothesis.

NWR constitutes a behavioral marker for SLI (Gathercole, 2006). However, the extent to which NWR deficits in SLI reflect underlying phonological processing difficulties or deficits in phonological store capacity is not clear. Gathercole and Baddeley (1990) did attempt to eliminate phonological processing deficits as a more basic cause of SLI, although more demanding phonological processing tasks may have revealed differences. SLI children performed as well as controls in articulation speed and in discriminating between minimally contrasting pairs of CVC words and nonwords, although the SLI children appeared to be worse at discriminating between nonword than word pairs. Phonological processing difficulties are common in SLI (see Chiat, 2001; Joanisse & Seidenberg, 1998). Chiat argued that underlying phonological processing difficulties selectively impair the acquisition of aspects of vocabulary and grammar depending on the accurate perception and representation of phonological information, including words whose meaning is underdetermined within any one particular context, phonologically conditioned morphemes (e.g., -ed), and syntactic constructions involving long-term dependencies, particularly when signaled by unstressed morphemes.

Clearly much work remains in testing rival accounts of SLI (Chiat, 2001; Joanisse & Seidenberg, 1998). Given the robustness of NWR as a marker of SLI, further research investigating the processes involved in NWR may assist in understanding not only processes limiting vocabulary acquisition, but also developmental disorders of language acquisition, and possibly in determining subtypes of SLI (Gathercole, 2006).