Background

People have been interested in the mechanisms that govern the development of language, and have also expressed concerns for the causes of language disorders. A number of assessments have been developed to effectively identify children with language disorders. Nonword repetition (NWR) is one such measure that has been developed for this purpose, and has been found to be a powerful index to not only children’s language development (Adams & Gathercole, 1996; Gathercole & Baddeley, 1989, 1990a; Gathercole & Willis, 1991; Hoff, Core, & Bridges, 2008) but also to children with language disorders (Conti-Ramsden, 2003; D’odorico, Assanelli, Franco,

& Jacob, 2007; Girbau & Schwartz, 2007; Stokes & Klee, 2009; Stokes, Wong, Fletcher,

& Leonard, 2006). In light of its significance to language development, looking into the nature of NWR would provide insights into the mechanisms that support language development.

Over the past 30 years, there have been considerable studies related to the NWR measure. While it has been constantly applied as a measure of either phonological memory or phonological representation in studies, a wealth of research has been carried out in parallel to explore the nature of this task. Consensus has been reached that this task taps the capacity to decode and encode phonological information, and the ability to maintain phonological information in storage. Performances in NWR could also be affected by the perception ability and the ability to organize articulatory gestures (Gathercole, 2006). The complex processing mechanism involved in the task makes it sensitive to any problems in the mechanism that would cause language difficulties or deficits. However, this complexity also makes its interpretation challenging, because a poor performance in this task could result from any of the processes involved, or could be the consequence of several causes. Therefore, it is necessary to understand the

processes involved in this task, and also factors that would affect task performances, so that we can develop more accurate interpretation of children’s performance in this task.

NWR performances develop with age, and demonstrate considerable variation across individuals. Different proposals have been raised regarding the major source of individual variation and developmental changes. Gathercole and colleagues (1989, 1990a) propose that NWR is a measure of phonological short-term memory (see also Gathercole, 2006). They made this proposal based on the working memory model of Baddeley and Hitch (1974). In this model, phonological short-term memory is one of the components specialized in managing verbal information. When verbal information is encountered, it will automatically enter into the phonological store in phonological codes and will be maintained in the store with the rehearsal process. The operation of subvocal rehearsal is most relevant to phonological short-term memory capacity, because a faster rehearsal rate can capture more information in the phonological store within the two-second memory decay period (Hulme, Thomson, Muir, & Lawrence, 1984). However, the driving force of subvocal rehearsal may not be at work in preschool children or younger, because past studies have shown that children do not spontaneously exploit the rehearsal strategy or other active memory strategies until age 7 (Gathercole & Adams, 1994; Gathercole, Adams, & Hitch, 1994; Henry, 1991).

The other line of research proposes that variation and development in NWR are more associated with the development in children’s language knowledge, especially in the young children. For example, Metsala and Chisholm (2010) discovered that preschool children’s NWR accuracy is supported by lexical knowledge. It was found that children had better repetition performances with the syllables that have lexical status, and they tended to change a nonword syllable into a word syllable. Also, their repetition of NWR may be mediated by the density of the lexical neighbors that the constituent syllable of the nonword has. The lexical effect and the neighborhood density effect were most prominent in multisyllabic nonwords.

In addition to mediation at the lexical level, NWR performance is also influenced by children’s phonological analysis ability. Metsala (1999) and Bowey (1996, 2001) found that phonological analysis played an important role in the 4- to 5-year-old English-speaking children’s NWR performances, even when the effect of short-term memory was controlled. The study of Li and Cheung (2014) on 4- to 5-year old Mandarin-speaking children showed that productive phonology was the major predictor of NWR, while digit span made a minor contribution. Moreover, they demonstrated that children’s individual differences in NWR performance may reside in their ability to encode nonwords into appropriate phonological units.

In most of the studies, the effects of lexical knowledge and phonological capacities are considered independently. However, in development they are not two unrelated constructs. In fact, increasing studies have pointed out that phonological capacities develop in conjunction with the increase in vocabulary size. It is proposed that phonological representations are shaped through the dynamics of the production-perception loop in the process of learning the forms of lexical items (Munson, Beckman, & Edwards, 2012). At the beginning, young children’s phonological representations are relatively more holistic, with words or syllables as the basic units (Ferguson & Farwell, 1975; Treiman & Breaux, 1982). The emergence of phonemic representation undergoes a process of gradual reformulation and it is suggested to be propelled by vocabulary growth (Metsala, 1999; Smith, McGregor, &

DeMille, 2006; Walley, 1993). This is also known as the “lexical restructuring account.”

Even though some scholars propose that the developing phonological system affects lexical acquisition to a greater extent than the reverse (Sosa & Stoel-Gammon, 2006;

Stoel-Gammon, 2011), they also admit that this may be limited to the age before 2;6.

From then on, the increase in lexicon size may be the driving force of phonological development. For example, the findings of Sosa and Stoel-Gammon (2006) suggested that phonological reorganization and the emergence of phonemic representation may

take place with the attainment of 150-200 words.

Therefore, regarding the contribution of lexical knowledge to NWR, lexical knowledge does not only mediate NWR at the lexical level, but also could mediate NWR at the sublexical level by affecting phonological abilities. For example, Munson, Kurtz, and Windsor (2005) showed that vocabulary size was the best predictor of the difference in repetition accuracy between high- and low-probability sequences. Also, the study of Edwards, Beckman, and Munson (2004) demonstrated an interaction effect between children’s vocabulary size and the phonotactic probability of nonwords. They discovered that children with smaller vocabulary size showed more prominent influence of the phonotactic probability of nonwords. Based on the findings, they propose that vocabulary size mediates the influence of phonotactic probability on nonword repetition by improving the specificity of phonological categories. Children with smaller vocabulary might have less established knowledge of sublexical units, because this knowledge is formed based on generalization made over lexical items.

As shown in the literature, a well-established model on NWR should not only take into consideration the effect of storage capacity, but also incorporate the influences from long-term lexical knowledge and phonological capacities. Gupta (2009) has incorporated all in a computational model of NWR. In the model, Gupta (2009) has incorporated a serial order mechanism and also linguistic representations at both the lexical and the sub-lexical levels. As suggested by Gupta (2009), the linguistic representations in this model constitute long-term knowledge, and the serial ordering device constitutes the short-term sequence memory. Therefore, when given a nonword, the participant has to decode and encode the nonwords into representations at the lexical level (i.e. word) and the sublexical levels (i.e. syllable and phoneme), and also to maintain the sequential information of the linguistic units. This process could be performed in the phonological buffer, which is subject to time decay (Barrouillet et al., 2009; Towse & Hitch, 1995). Hence, efficient decoding and encoding of the nonwords

is req

However, among the older children, the dependence relationship between the two factors changed. The vocabulary knowledge at age 5 and age 6 predicted children’s performance in NWR at age 6 and 8, respectively. Also, regarding the relationship between vocabulary and phonological development, there may be a change in the direction of influence between them (Stoel-Gammon, 2011). Therefore, what appears to be interesting is the dynamic interaction pattern of these three variables along the course of development.

Particularly, it is of interest how the growth in vocabulary or the growth in phonology influences the performances in NWR. Past studies examine the effect of vocabulary growth by comparing children with large vocabulary size with children with small vocabulary size (Edwards et al., 2004). This approach could allow us to examine how children form different processing strategies and performances when they accumulate different sizes of vocabularies. However, it does not allow us to control for children’s variation in other aspects, such as their phonological analysis capacity, attention span, or other cognitive factors, which might also contribute to variation in NWR performance. Gupta’s (2009) computational model simulates vocabulary growth in the model to examine NWR performance; however, it was done just for the purpose of establishing corpus for the model, and may not be assimilated to the nature vocabulary growth in children.

In addition, the relationship between vocabulary knowledge and phonological development is rarely considered in NWR studies. Even though the studies of Edwards et al. (2004) and Munson et al. (2005) have pointed out the close tie between vocabulary and phonological development, they did not measure children’s phonological capacities independently, but rather manipulated the phonotactic probability of the nonword stimuli. They hold the belief that phonology is an emergent consequence of the mapping between phonetic parameters and lexicons (Munson, Beckman, & Edwards, 2012).

However, it is also likely that phonological capacities have their own independent

contributions, especially in cases where lexical knowledge is only allowed to exert little intervention, such as very young children who has only limited vocabulary size, or nonword stimuli that composed of non-attested syllables.

Most of the current NWR studies are cross-sectional, thus not able to demonstrate the dynamic relationship among the three variables. Longitudinal studies on the relationship between NWR and other measures have been rare (e.g. Gathercole, Willis, Emslie, & Baddeley, 1992; Melby-Lervåg, et al., 2012; Bowey, 2001), and all these studies have focused on children at 4 or above. However, if nonword repetition could potentially be used as an indicator of language abilities, it is necessary to examine its correlation with language abilities at a much younger age.