In 1990 Gathercole and Baddeley proposed a strong hypothesis that has generated a wealth of research in the field of language development and disorder. The hypothesis was that phonological memory, as indexed by nonword repetition, is causally related to vocabulary development. Support for the hypothesis came from an impressive range of longitudinal, correlational, and laboratory training studies, and from studies of specific language impairment (SLI). However, more recently, Gathercole, Tiffany, Briscoe, Thorn, and The ALSPAC Team (2005), directly tested the causal hypothesis by following a cohort of children from age 5 to 8 years. Contrary to prediction, children with poor nonword repetition abilities at age 5 had normal vocabulary at the age of 8.

Gathercole (2006) presents a revised version of the hypothesis cast in a framework comprising the key phonological processes contributing to nonword repetition, including input and output processes, the quality of phonological representations, and phonological storage, with a range of factors affecting each stage of the process. Gathercole distinguishes two hypotheses regarding vocabulary growth. According to the phonological storage hypothesis that she favors, the capacity of phonological storage underpins and places constraints on behavioral measures of vocabulary, nonword repetition, and phonological awareness. Within this view, an increase in phonological storage with age leads to increases in nonword repetition, new word learning, and phonological awareness skills.

The phonological sensitivity hypothesis is more explicitly developmental. Within this model, vocabulary growth drives the development of lexical representations, causing progressive segmentalization (Walley, 1993). In turn, the development of segmental representations is a precursor of phoneme awareness, and leads to improvements in a range of phonological processes including nonword repetition and new word learning. Neither hypothesis emphasizes reciprocal causation yet such influences need to be accommodated to fully understand the developmental changes observed in normal and atypical development (e.g., Castro-Caldas, Petersson, Reis, Stone-Elander, & Ingvar 1998).

In our view it is premature to reject the phonological sensitivity hypothesis because there is insufficient evidence to distinguish it from the phonological storage hypothesis. We suggest that future research should be guided by a developmental framework in which phonological representations undergo a continual process of modification as the consequence of lexical development and the constraints that operate on it, and which are critical both to phonological storage and word learning. We contend that the framework should be capable of accounting for the established finding that poor nonword repetition is a risk factor for SLI, but that language learning disorders (including SLI) are unlikely to be caused by a unitary deficit.

An adaptation of the developmental contingency modeling approach of Morton and Frith (1995) is used here to consider the role of different factors in driving nonword repetition, as well as the causation of language disorders. This approach separates cognitive and behavioral levels of explanation, and is a useful way of thinking about contingencies between different processes during development. It also allows for a consideration of the role of the environment, both as a cause of behavior and a mediator of reciprocal influences. The approach does not make any commitment to the model or mechanism that mediates brain–behavior relationships at the cognitive level (Morton, 2004).

The first step in the causal modeling process is to separate processes at the cognitive and behavioral levels. The present concern is with the predictors of two behaviors, nonword repetition, and new word learning, and their relationship to vocabulary knowledge. Gathercole (2006) focuses primarily on nonword repetition that she considers the behavioral marker of a primitive learning mechanism that is impaired in SLI, highly heritable, and relatively impervious to environmental influence. We agree with Gathercole that an exclusive focus on nonword repetition is not adequate to fully explain the impairments of children with language learning disorders. We propose that it is productive to consider the role of semantic representations in vocabulary learning, and to allow for the reciprocal influence of vocabulary growth on the status of both semantic and phonological representations. The proposed causal hypothesis is depicted in Figure 1. At the cognitive level, there is a neural network with bidirectional connections between semantic and phonological representations. The development of segmental phonological representations is fuelled by vocabulary growth, the impetus for which is semantic development (as specified by the phonological sensitivity hypothesis). In addition, input and output processes (not depicted in the framework) influence phonological development, the pacemaker of phonological learning.

The developmental contingency model for the consideration of language learning disorders.

Within this view, the constraints on nonword repetition that Gathercole (2006) discusses have their impact on the development of the representations that underpin nonword repetition, new word learning, vocabulary growth, and other skills such as phoneme awareness. In turn, these behaviors feedback and influence the further development of semantic and phonological representations. One of the strengths of this model is that it can accommodate the dissociations between nonword repetition and vocabulary seen in developmental disorders. Thus, dyslexia is associated with a deficit in phonological representations, whereas semantic representations are unimpaired (Swan & Goswami, 1997); the typical behavioral manifestation is therefore of a phonological learning deficit in the absence of vocabulary impairments (e.g., Aguiar & Brady, 1991). The deficit in phonological representation observed in dyslexia is shared with children with SLI who have decoding deficits in reading (Bishop & Snowling, 2004). However, it is important to emphasize that the deficit in phonological representation (and it follows in nonword repetition) is not seen in all cases of SLI (Catts, Adlof, Hogan, & Ellis Weismer, in press), and it is not sufficient to explain the language or reading profile of others (Bishop, 2006; Nation, Clarke, Marshall, & Durand, 2004). The model leads to the hypothesis that deficits in semantic representation contribute to, and in some cases may be a primary cause of, specific language impairment; furthermore, such deficits may be the sequelae of environmental factors, such as language deprivation that either predates or is a consequence of language difficulties.

In summary, Gathercole's (2006) revision of the nonword repetition hypothesis is to be welcomed. Further research directed at the development of an interactive model of vocabulary growth would be a fruitful next step in understanding the causes of language learning impairments.