WHAT DO WE MEAN BY WM?

There is much current debate concerning how best to characterize WM within a set of higher level cognitive processes, termed executive functions. We do not attempt to resolve this debate in the current article but instead take WM to be a cognitive system that is involved in language acquisition, among other abilities. Although WM does not function in isolation, in this paper we consider it, for the most part, without discussing other executive functions. WM is a process involving the temporary storage and manipulation of incoming information, be it visual, auditory, or otherwise. WM processes are deemed necessary in order to translate incoming information into long-term knowledge. Baddeley and Hitch (Reference Baddeley, Hitch and Bower1974) first proposed that WM is a multicomponent subsystem, composed of the following: (a) the visuospatial sketchpad, a subsystem responsible for the storage and manipulation of visual information, and (b) the phonological loop that allows for the temporary storage and processing of verbal and/or acoustic information. Responsible for controlling these subsystems is (c) a central executive that relies on attentional mechanisms to control the functioning of each of these “slave” subsystems within working and episodic memory (see Baddeley, Reference Baddeley2000, Reference Baddeley2007). The phonological loop can be broken down further into the phonological store, which can hold phonological content for a few seconds before it fades, and an articulatory rehearsal process where phonological information may be rehearsed in order to refresh the memory trace (Baddeley, Reference Baddeley2003, Reference Baddeley2007; Repovš & Baddeley, Reference Repovš and Baddeley2006). It has been demonstrated that the different subcomponents of the phonological loop function together to support different aspects of language learning and processing (for reviews, see Baddeley, Reference Baddeley2003; Baddeley, Gathercole, & Papagno, Reference Baddeley, Gathercole and Papagno1998; Gathercole, Reference Gathercole2006; Juffs & Harrington, Reference Juffs and Harrington2011).

It is has been proposed that the phonological loop supports language acquisition by allowing incoming (unfamiliar) language-relevant information to be maintained in memory long enough that it can be processed and, over time, transformed into long-term stored knowledge about language, such as vocabulary or grammar. It has been suggested that the phonological store in particular, as opposed to subvocal rehearsal, may be involved in the process of language acquisition (e.g., Baddeley et al., Reference Baddeley, Gathercole and Papagno1998) as there is robust evidence for a relationship between PWM capacity and language acquisition as early as 2 years of age; however, subvocal rehearsal does not appear to emerge until much later, around roughly age 7 (for reviews, see Cowan & Kail, Reference Cowan, Kail and Gathercole1996; Gathercole & Hitch, Reference Gathercole, Hitch, Collins, Gathercole, Conway and Morris1993). Moreover, the ability to perform well on a nonword repetition task, which requires that individuals hear and repeat back an item well within the temporal capacity of the phonological store, and before rehearsal processes are possible, is associated with improved language acquisition. That such a relationship exists further suggests that language acquisition is supported by the creation of adequate phonological representations within the phonological store (Baddeley et al., Reference Baddeley, Gathercole and Papagno1998).

If PWM is necessary for language development, then individual differences in PWM abilities, and/or in the quality of phonological representations supporting PWM, should be linked to variation in language outcomes. This has been found to be the case in typical language development contexts. PWM capacity predicts children's acquisition of both a first and a second language (for reviews, see Baddeley et al., Reference Baddeley, Gathercole and Papagno1998; Gathercole, Reference Gathercole2006; Juffs & Harrington, Reference Juffs and Harrington2011). In children with typical development, PWM is positively associated with vocabulary learning (Baddeley, Reference Baddeley2003, Reference Baddeley2007; Gathercole, Reference Gathercole2006; Repovš & Baddeley, Reference Repovš and Baddeley2006), grammatical development (Adams & Gathercole, Reference Adams and Gathercole1995), reading skills (e.g., Goff, Pratt, & Ong, Reference Goff, Pratt and Ong2005; Nikolopoulos, Goulandris, Hulme, & Snowling, Reference Nikolopoulos, Goulandris, Hulme and Snowling2006), letter learning (Torppa, Poikkeus, Laakso, Eklund, & Lyytinen, Reference Torppa, Poikkeus, Laakso, Eklund and Lyytinen2006), and phonological skills (Alloway, Gathercole, Adams, & Willis, Reference Alloway, Gathercole, Adams and Willis2005; Durand, Hulme, Larkin, & Snowling, Reference Durand, Hulme, Larkin and Snowling2005; van Daal, Verhoeven, van Leeuwe, & van Balkom, Reference van Daal, Verhoeven, van Leeuwe and van Balkom2008). Moreover, this relationship has been found in both naturalistic studies of child second language (L2) learners (e.g., Cheung, Reference Cheung1996; Service, Reference Service1992; Service & Kohonen, Reference Service and Kohonen1995) and carefully controlled experimental studies examining children's acquisition of new vocabulary in a laboratory setting. For example, in a study of school-aged children (12-years-old, on average) who were learning English as an L2 (Cheung, Reference Cheung1996), PWM capacity predicted the number of trials necessary to learn new words in English. However, this was only the case when the children's English vocabulary was low, indicating that once adequate vocabulary is acquired, PWM capacity plays a less dominant role in further vocabulary acquisition (e.g., Gathercole, Willis, Enslie, & Baddeley, Reference Gathercole, Willis, Emslie and Baddeley1992).

THE ROLE OF EARLY EXPERIENCE IN PWM DEVELOPMENT

While there is robust evidence that PWM capacity is positively associated with both early language learning and later language outcomes, there is less evidence of how initial differences in PWM capacity might emerge. Children's PWM capacity develops rapidly, and with much variability, over the first few years of life (e.g., Gathercole & Adams, Reference Gathercole and Adams1993). There is relatively little known about developmental changes that occur to support the development of phonological memory (Gathercole, Reference Gathercole1999), although limited evidence suggests that WM develops within the first 6–12 months of life (Nelson, Reference Nelson1995), with evidence for the beginnings of early PWM development typically being reported around 2 years, 6 months (2;6) to 4 years of age, once children are able to repeat words and sentences (Gathercole & Baddeley, Reference Gathercole and Baddeley1989; Gathercole & Hitch, Reference Gathercole, Hitch, Collins, Gathercole, Conway and Morris1993). Improvements in WM capacity are then observed during the next few years. As reviewed in the previous section, research that has focused on the role of WM in language acquisition suggests that individual differences in WM capacity are related to various aspects of both first language (L1) and L2 acquisition and processing even from an early age. However, exactly how these individual differences emerge is an open question. While some accounts view variability in WM as due to inborn capacity differences (e.g., Baddeley et al., Reference Baddeley, Gathercole and Papagno1998), here we argue that one critical factor in determining PWM abilities may be early language learning experience (e.g., Gathercole, Reference Gathercole2006). Specifically, both timing of language exposure and quality and quantity of language input during an early sensitive period for phonological development might influence children's short- and long-term language learning as a result of variations in PWM.

Sensitive periods are common in the brain development of organisms, including humans. They are defined as a period during development when the acquisition of certain abilities or behaviours is facilitated. Following the closure of this period, these abilities can no longer be acquired, or are acquired to a lesser degree and much less efficiently (i.e., through compensatory mechanisms). A great deal of research has delineated sensitive periods for several aspects of visual development (for a review, see Hensch, Reference Hensch2005; Werker & Hensch, Reference Werker and Hensch2015), with research on language development revealing less consistent results (e.g., Birdsong, Reference Birdsong1999; Johnson & Newport, Reference Johnson and Newport1989; Krashen, Reference Krashen1973; Lenneberg, Reference Lenneberg1967; Morgan, Reference Morgan, Brookes and Kempe2014). However, there is considerable evidence that a sensitive period, or important age-related constraints, exists for phonological development. This period lasts until roughly the end of the first year of life, at which point children's language abilities become fine-tuned such that they are better able to distinguish phonemic contrasts in their developing native language (or languages) than those in foreign languages (Kuhl, Williams, Lacerda, Stevens, & Linblom, Reference Kuhl, Williams, Lacerda, Stevens and Linblom1992; Werker & Tees, Reference Werker and Tees1984). During this period, it is suggested that neural architecture is altered so that circuits become specialized to the acoustic and statistical properties of phonetic units in the native language. This facilitates native language learning, while making L2 learning more difficult as the processing of any alternate (i.e., nonnative) phonetic units is made less efficient (Kuhl, Reference Kuhl2000, Reference Kuhl2004). Some evidence suggests that molecular processes are engaged at the end of sensitive periods to prevent specialized circuits from being altered by subsequent experience with similar stimuli (for a review, see Werker & Hensch, Reference Werker and Hensch2015). Thus, specific distributions of native language phonetic units, in place by 8 to 10 months of age, become increasingly difficult to change and persist into adulthood (Baker, Idsardi, Golinkoff, & Petitto, Reference Baker, Idsardi, Golinkoff and Petitto2005; Mattock, Molnar, Polka, & Burnham, Reference Mattock, Molnar, Polka and Burnham2008; Maurer & Werker, Reference Mauer and Werker2014; Maye, Werker, & Gerken, Reference Maye, Werker and Gerken2002; Polka & Werker, Reference Polka and Werker1994; Werker, Gilbert, Humphrey, & Tees, Reference Werker, Gilbert, Humphrey and Tees1981; Yoshida, Pons, Maye, & Werker, Reference Yoshida, Pons, Maye and Werker2010).

In light of this, we focus on the relationship between the sensitive period for phonological development and the establishment of the PWM system. In particular, early experiences might influence phonological development and in turn PWM through the timing, as well as through the quality and quantity of language input experienced during or after a sensitive period. Input encountered while a sensitive period is open might be particularly influential in shaping the quality of phonological representations that will later be used by the PWM system, which arguably requires the establishment of adequate and stable phonological representations (Gathercole, Reference Gathercole2006) acquired during the earliest stages of language learning.

While evidence for the specific role of children's earliest language experiences on PWM development is limited, there is some evidence from monolingual language learners that experience in general does influence PWM outcomes. For example, several studies report that language familiarity leads to enhanced PWM processing. Adults generally perform faster and more accurately when they are required to repeat words as opposed to nonwords (e.g., Chiat & Roy, Reference Chiat and Roy2007), and both adults (Coady & Aslin, Reference Coady and Aslin2004) and 2;6- to 3;6-year-old children (Thorn & Gathercole, Reference Thorn and Gathercole1999) have been found to repeat nonwords more accurately if they are composed of highly familiar, frequent, or nativelike phonemes, as opposed to nonwords constructed of phonemes with low phonotactic probability, demonstrating a relationship between learned phonemic knowledge and PWM capacity. There is also evidence that PWM capacity in children depends on language-specific representations. For example, Messer, Leseman, Boom, and Mayo (Reference Messer, Leseman, Boom and Mayo2010) found phonotactic probability effects in 4-year-olds that were similar to those observed by Coady and Aslin. Messer et al. also found that monolingual Dutch-speaking children showed greater phonotactic probability effects in Dutch than did Turkish–Dutch bilinguals. The reverse was also true: bilingual children for whom Turkish was their L1 showed greater phonotactic probability effects in Turkish than did Dutch monolinguals. Familiarity effects have also been observed in production. Keren-Portnoy, Vihman, DePaolis, Whitaker, & Williams (Reference Keren-Portnoy, Vihman, DePaolis, Whitaker and Williams2010) found that 26-month-old children who had been producing consonants for longer also performed better on nonword repetition tasks using stimuli that conformed to the regularities of the native language, particularly for consonants that they had individual experience producing compared to consonants not yet in their repertoires.

Evidence of experiential effects on other components of PWM has also been reported. For example, factors affecting subvocal rehearsal processes, such as articulation rates, also influence PWM capacity. In a sample of first, third, and fifth graders, the faster children were able to articulate words and digits, the better their PWM (Cowan et al., Reference Cowan, Wood, Wood, Keller, Nugent and Keller1998). This relationship was interpreted as more efficient use of the phonological loop in that the more information that can be rehearsed within a shorter time span, the more information can be maintained at a given time. It is important to note that although effects on subvocal rehearsal are not evident until later stages of development, they may be influenced by experiences that occurred earlier. For example, early exposure to rich linguistic stimuli may affect general processing speed, which may in turn affect articulation rates and, thus, PWM outcomes as well (e.g., Tallal et al., Reference Tallal, Miller, Bedi, Byma, Wang, Nagarajan and Merzenich1996).

Similarly, experiences that influence the acquisition of long-term stored vocabulary might also influence PWM abilities. By the age of 5 or 6, once children have acquired some proficiency in their native language, vocabulary becomes a much stronger predictor of PWM than the reverse (although there is much variability in this age range; see Gathercole et al., Reference Gathercole, Willis, Emslie and Baddeley1992; Jones, Gobet, & Pine, Reference Jones, Gobet and Pine2007). This is important to note because factors that lead to differences in vocabulary size (including early differences in phonological representations underlying PWM) might in turn influence PWM processing at later stages of development. Children who are slower to develop vocabulary due to underspecified phonological representations might show increasingly worse PWM due to the lack of adequate vocabulary to support subsequent PWM processing. At the same time, the development of sufficient vocabulary might mask initial differences in PWM abilities owing to increased support from stored vocabulary knowledge, a point we return to later.

While much of the research reported here focuses on PWM in monolingual language learners, the evidence from this group is limited. Moreover, the influences of early experiences are difficult to tease apart in typically developing monolinguals because by the time PWM can be measured in these children (i.e., once they are able to repeat words and sentences), the sensitive period for phonological development is likely closed. However, one way to shed light on the influence of the earliest experiences on PWM development is by looking at groups of language learners with more variation in their early language experiences than is usually evident in groups of monolingual learners, particularly variation that occurs during the earliest stages of language acquisition. In the following sections, we review research on different kinds of learners, which we believe provides a rich source of evidence for uncovering the relationship between early experience and PWM development.

A SPECTRUM OF EARLY EXPERIENCES IN LANGUAGE ACQUISITION

While much research uses monolingual and uninterrupted language acquisition as the default model for studying language development, this is not the only, nor necessarily the most frequently occurring, environment in which children acquire language. There is a wide variety of language acquiring populations who experience different learning environments compared to those reported in studies of native monolingual language learners, including bilinguals and internationally adopted children as well as children with perceptual deficits. Not only has there been a growing interest in how different contexts affect the rate, pattern, and ultimate attainment of language, we propose that examining populations with particular variations in early language experience allows us to delineate the influence of early experience on PWM outcomes in greater depth. This also allows us to draw implications for language learning outside those populations to language learning mechanisms more generally. Specifically, in the following sections, we examine the effects of delay in language input (i.e., internationally adopted children and oral deaf children with cochlear implants) as well as the effects of relative language enrichment or deprivation (simultaneous bilinguals and L2 learners versus children with otitis media and deaf signing children born to nonsigning parents) on the development of PWM. Based on this review, we argue that language input that is delayed, disrupted, and/or impoverished during an early sensitive period for phonological development will lead to specific disturbances and weaknesses in PWM that may in turn influence the development of more complex aspects of language. We speculate that disturbances in PWM stem from underspecified phonological representations for the language that experienced the disruption. In contrast, learners who experience early language input that is enriched in some manner, as in bilinguals who learn more than one language, will show selective advantages in PWM owing to stronger, more stable, and/or a greater repertoire of phonological representations. It may also be that selective advantages are a result of greater availability of executive function mechanisms that arise from exposure to and use of multiple languages. Throughout the following sections, we highlight the general patterns of development that arise in these different populations as well as the differences that can be predicted to arise from each language learning context.

PUTTING IT ALL TOGETHER

In the preceding sections, we reviewed studies that indicate that very early language experiences shape children's later language learning. In particular, we proposed that early cognitive underpinnings, namely, the development of PWM, play a key role in this relationship. Here, we consolidate that evidence in order to discuss how variations in the timing, quality, and amount of language exposure after birth might modulate effects on later language development as a result of their effects, short and long term, on phonological processing and, in particular, PWM.

The first thing we note about the language development patterns of the specific groups we have reviewed is that, while early experiences can affect all aspects of language development, phonological aspects of language (including PWM) appear to be particularly vulnerable. As reviewed, deaf adults who experience delayed exposure to L1 signed language as children exhibit particular difficulties in phonological processing in sentence repetition tasks in comparison to adults with early L1 sign exposure (Mayberry, Reference Mayberry2007), and children who experience delay in L1 sign language exposure exhibit greater difficulties acquiring and processing phonological information than late L2 sign language learners who become deaf after early acquisition of spoken language. Similar patterns have also been found in oral deaf children's rapid processing of English phonotactics (Geers et al., Reference Geers, Nicholas, Tobey and Davidson2015) and in later reading ability, which has been argued to be related to earlier PWM abilities (Kyle & Cain, Reference Kyle and Cain2015). Similarly, while IA children often score within the typical range on standardized tests of language such as vocabulary and grammar, they score significantly lower in comparison to both test norms and matched control groups on tests of sentence recall and WM involving phonological processing (Delcenserie & Genesee, Reference Delcenserie and Genesee2014a). These outcomes arguably occur due to disruption in phonological input during an early sensitive period for phonological development, and in the case of signing deaf children, these effects might also reflect the impoverished nature of the input that delayed sign language exposure entails (Lu et al., Reference Lu, Jones and Morgan2016), a point we return to later. Other groups who experience impoverished language input during the sensitive period for phonological development also exhibit reduced performance on tasks dependent on phonology. For example, children with otitis media early in life have been shown to exhibit lags on tasks assessing PWM while often scoring in the typical range on other tests of language including expressive and receptive vocabulary, although subtle effects on these measures have also been reported (see Roberts et al., Reference Roberts, Long, Malkin, Barnes, Skinner and Hennon2005, for a review).

Similar patterns also appear in children who learn an L2. L2 learners who, by definition, exhibit lags in exposure to an L2 but normal exposure to their L1 at birth, in some cases perform more poorly on phonological processing tasks in the language that they acquire later (e.g., Abrahamsson & Hyltenstam, Reference Abrahamsson and Hyltenstam2009; Parra et al., Reference Parra, Hoff and Core2010); these effects can persist into adulthood, though lags may be masked with increasing proficiency in that language, potentially through the recruitment of additional cognitive abilities (Delcenserie & Genesee, Reference Delcenserie and Genesee2016). Such effects are more evident in L2 learners with less proficiency in the L2 than in highly proficient L2 learners (Delcenserie & Genesee, Reference Delcenserie and Genesee2016), although they can be evident even in L2 learners who can pass for native speakers during oral conversations if sufficiently sensitive language tests are used (Abrahamsson & Hyltenstam, Reference Abrahamsson and Hyltenstam2009). In contrast, L2 learners and simultaneous bilinguals who achieve high levels of L2 proficiency, on par with monolinguals, do not necessarily exhibit such lags and may even exhibit advantages in PWM in comparison to monolingual learners (Delcenserie & Genesee, Reference Delcenserie and Genesee2016). Of particular note, simultaneous bilinguals who have exposure to two languages from birth score significantly higher than L2 learners with equally high levels of proficiency in the target language on tests of PWM, arguing again that it is early experience with language that is especially consequential for phonological processing and language learning.

At the core of our hypothesis is the speculation that the development of PWM is tied to the establishment of language-specific phonological representations insofar as there is evidence that phonological representations form the basis for what is most efficiently processed by the PWM system (Gathercole, Reference Gathercole2006). We argue more specifically that the maturation of the WM system itself is modulated by the same types of language experience that facilitate the brain's specialization for its developing native language. In this argument, the sensitive period for language is therefore tied into a shared time frame for the activation and instantiation of the PWM component. Moreover, because native-language specific phonological representations are established early and remain robust over time, we propose that subsequent PWM processes that are active beyond the initial stages of language acquisition continue to be influenced by that early experience.

Evidence from the language learning groups we have reviewed supports this claim and suggests how these effects might be substantiated in the brain and, thereby, how they might influence language development. Beginning with IA children, perceptual attunement typically occurs during a time when they are exposed to a language they will subsequently discontinue learning. Thus, by the time they are adopted (often between 12 and 24 months of age in the case of children from China, the object of much research) IA children have become at least partially, if not completely, specialized at discriminating the speech sounds of their birth language but will not have acquired phonological representations and full maturation of PWM for specific sounds in the adoption language. If early acquired phonological representations for the birth language persist, as suggested by Pierce, Klein, Chen, Delcenserie, and Genessee (Reference Pierce, Klein, Chen, Delcenserie and Genesee2014), they might interfere with or otherwise impact acquisition of phonological representations for sounds in the adoption language (Pierce et al., Reference Pierce, Chen, Delcenserie, Genesee and Klein2015).

The findings from Pierce et al. (Reference Pierce, Chen, Delcenserie, Genesee and Klein2015) support this argument and suggest further that this is the case for typically developing L2 learners as well. Of particular relevance to the current discussion, IA children who began acquiring their “second first language” between 6 and 25 months of age exhibited similar patterns of brain activation during the PWM task as same-age Chinese–French bilinguals who had acquired French as an L2 at approximately the same ages as the adoptees, but who continued to learn and use Chinese, their birth language. While all groups activated regions associated with PWM, the IA children and the Chinese–French bilinguals activated additional regions that are typically associated with nonverbal memory and cognitive control processes. Pierce et al. interpreted these results to indicate that the adoptees and the bilingual participants recruited regions of the brain that differed from those activated by the French monolinguals in order to compensate for phonological representations in the brain being fine-tuned to Chinese but the current language environment was French. As a result, it may be that subsequently acquired phonological representations for the adoption language were weaker or less precise than those acquired by L1 learners. In turn, acquisition of the adoption language after the sensitive period for phonological development, along with maintenance of phonological representations from the birth language, may account for the persistent PWM lags exhibited by IA children. These factors may in turn contribute to specific behavioral differences observable in IA children's acquisition of the adoption language, as evidenced by their performance on tasks of verbal memory and vocabulary that implicate those representations (for reviews, see Delcenserie, Reference Delcenserie, Genesee and Delcenserie2016; Scott & Roberts, Reference Scott, Roberts, Genesee and Delcenserie2016).

Turning now to deaf children of hearing parents, weaker PWM for spoken language has been found in profoundly deaf children born of hearing parents who receive CIs. However, in contrast to IA children, these children receive no or greatly reduced sensory input during the sensitive period for phonological development before and during the CI process. Thus, influences from a previously acquired language and, in particular, early acquired phonological categories and relationships, cannot account for the language and verbal memory outcomes of deaf children with CIs. How then can we account for their patterns of language development? There are at least two possibilities. One possibility is that they acquire unstable or underspecified phonological representations of spoken language because they are acquired after the sensitive period for phonological development. In other words, the phonological representations of the input language that congenitally deaf children establish and use for language learning following implantation may be unstable or weak because they were not laid down during the sensitive period for phonological development.

A second possibility is that, because they acquire spoken language after the sensitive period for phonological development, children with CIs, like IA children, draw on alternative, more general cognitive systems to acquire the phonological representations that support their developing PWM system. In support of this, an increasing number of studies suggest that individual differences in speech–language outcomes after cochlear implantation in deaf children may be due to underlying differences in core neurocognitive processes (e.g., Harris et al., Reference Harris, Kronenberger, Gao, Hoen, Miyamoto and Pisoni2013). That reliance on alternative cognitive systems results in different outcomes in comparison to CI children's hearing peers and deaf native signers comes from evidence that a significant number of deaf children with CIs continue to exhibit weaknesses in language development, particularly phonological processing skills, following implantation. This suggests that phonological processing and PWM that rely on alternative neurocognitive processes may be less than optimal (Geers & Hayes, Reference Geers and Hayes2011; Miller, Lederberg, & Easterbrooks, Reference Miller, Lederberg and Easterbrooks2013; Spencer & Tomblin, Reference Spencer and Tomblin2009).

Additional evidence supports the hypothesis that alternative cognitive systems (i.e., those not typically active during early language acquisition) might be activated for processing phonological information in a language that is acquired after the sensitive period for phonological development, with or without prior sensory input. For example, Barone, Lecassagne, and Kral (Reference Barone, Lecassagne and Kral2013) have found that the auditory cortex of congenitally deaf, but not hearing, cats received unique inputs from visual and somatosensory brain regions, which the authors viewed as compensatory (for a review of issues surrounding cortical reorganization during and after deafness in humans, see Campbell, Macsweeney, & Woll, Reference Campbell, Macsweeney and Woll2014). In humans, this may imply that alternative brain networks take over, or supplement, phonological processing for a language that was not acquired during the earliest stages of development. Supporting this, Logan, Lively, and Pisoni (Reference Logan, Lively and Pisoni1991) report that adult L2 learners are able to learn nonnative phonemic contrasts through extensive training (see Bradlow, Pisoni, Akahane-Yamada, & Tohkura, Reference Bradlow, Pisoni, Akahane-Yamada and Tohkura1997; Ingvalson, Holt, & McClelland, Reference Ingvalson, Holt and McClelland2012; Lively, Pisoni, Yamada, Tohkura, & Yamada, Reference Lively, Pisoni, Yamada, Tohkura and Yamada1994; McCandliss, Fiez, Protopapas, Conway, & McClelland, Reference McCandliss, Fiez, Protopapas, Conway and McClelland2002) and that this learning occurs through the use of compensatory mechanisms that are highly task dependent (Gervain & Werker, Reference Gervain and Werker2008; Werker & Tees, Reference Werker and Tees2005). Despite such improvements, however, it is rare for the phoneme discrimination abilities of late L2 learners to reach the level of native speakers or to generalize outside the training context; nor is there evidence that newly acquired phonemes are used during word recognition (e.g., Bradlow et al., Reference Bradlow, Pisoni, Akahane-Yamada and Tohkura1997; Ingvalson, McClelland, & Holt, Reference Ingvalson, McClelland and Holt2011; Pallier, Bosch, & Sebastián-Gallés, Reference Pallier, Bosch and Sebastián-Gallés1997). These findings, along with results from research on children with CIs, are compatible with our hypothesis that language exposure that happens during the sensitive period sets up a reciprocal relationship between the building blocks of language and PWM. If both systems are set up from very early in development, the ensuing language skills may be most efficient. However, if development is disrupted, as illustrated in the populations reviewed previously, compensatory mechanisms might be used to enable language development to progress, although this may come at some cost.

Evidence that early language exposure in any form is better than no language exposure at all comes from studies of native sign language learners (DCDP who are exposed to a natural signed language during an early sensitive period). In this case, a WM system with full processing capacities can be developed (Marshall et al., Reference Marshall, Jones, Denmark, Mason, Atkinson, Botting and Morgan2015). While native signers are extremely rare, their language development profiles point to the importance of early language exposure in some modality (even if it is not speech) as a crucial factor in PWM development. In contrast, the results we reviewed suggest that if initial language exposure is in a language that is ultimately discontinued (the case of IA children) or if early input is impoverished (DCHPs or children with otitis media, to be discussed next), phonological processing and PWM are affected to some degree later in development.

While IA children likely develop phonological representations for a language they will ultimately discontinue and CI children may miss the early opportunity to establish optimal phonological representations, children with impoverished language input during the first years of life fall somewhere in between. Specifically, children who experience multiple bouts of otitis media early in life and deaf children who rely on sign language input but are born to hearing parents just learning a signed language both receive language input from birth or, often, by the time they are 6 months of age. However, their early exposure to language is often degraded or impoverished. Despite early exposure to language, the language outcomes observed in these groups and, in particular, their lags in tasks requiring phonological processing and/or PWM (e.g., otitis media: Mody et al., Reference Mody, Schwartz, Gravel and Ruben1999; Pearce et al., Reference Pearce, Saunders, Creighton and Sauve1988; DCHP's sign language skills as adults: Emmorey, Bellugi, Friederici, & Horn, Reference Emmorey, Bellugi, Friederici and Horn1995; MacSweeney, Waters, Brammer, Woll, & Goswami, Reference MacSweeney, Waters, Brammer, Woll and Goswami2008; Orfanidou, Adam, Morgan, & McQueen, Reference Orfanidou, Adam, Morgan and McQueen2010) resemble those of both CI and IA children. Although beyond the scope of this article, it is also interesting to note that similar weakened PWM abilities are thought to underlie the particular deficits observed in children with specific language impairment (e.g., Archibald & Gathercole, Reference Archibald and Gathercole2006). While PWM deficits in specific language impairment are proposed to stem from neurocognitive impairments (e.g., Ullman & Pierpont, Reference Ullman and Pierpont2005), in the present populations we propose that similar lags can be caused by environmental disruption. Specifically, the pattern of reduced PWM abilities observed in children with otitis media and DCHPs may be due to degraded language input during the earliest stages of development when phonological representations are typically acquired. Thus, the input that DCHPs and children with otitis media are exposed to may not be consistent or precise enough to establish phonological representations that are on par with those of children who have continuous and nondegraded language exposure from birth. As a result, children with impoverished early language experiences may have underrepresented or weak phonological representations that are less efficient for supporting PWM processes.

In contrast to the preceding cases of delayed and impoverished input, bilinguals acquire language under conditions that might be considered enriched insofar as they are exposed to and acquire two language systems, including two different phonological systems. That dual language exposure constitutes a form of enrichment that can enhance phonological processing and, in particular, PWM is attested to by a number of studies, to be reviewed shortly, but these effects are conditional on proficiency and AoA. With respect to proficiency, research that has examined the development of PWM in bilinguals, simultaneous and successive, suggests that bilinguals exhibit greater WM in the language in which they are more proficient (see, for example, Parra et al., Reference Parra, Hoff and Core2010, for a case of simultaneous bilinguals; and Gathercole, Reference Gathercole1999, for a case of successive L2 learners). That proficiency is important also comes from research demonstrating that successive bilinguals perform significantly better on PWM tasks than monolinguals once they have had several years exposure to the L2 and, thus arguably, some minimal level of proficiency in that language (Blom et al., Reference Blom, Küntay, Messer, Verhagen and Leseman2014; Delcenserie & Genesee, Reference Delcenserie and Genesee2016; Kaushaskaya et al., Reference Kaushanskaya, Gross and Buac2014). In a related vein, simultaneous bilinguals who are matched on proficiency with native speakers have been found to actually outperform monolinguals on tasks of PWM (e.g., Blom et al., Reference Blom, Küntay, Messer, Verhagen and Leseman2014; Delcenserie & Genesee, Reference Delcenserie and Genesee2016). Delcenserie and Genesee's finding that the advantages of bilingualism with respect to PWM are most evident in simultaneous bilinguals in comparison to L2 learners whose exposure to the L2 begins after the sensitive period for phonological development once again attests to the importance of early experience.

While there is no clear explanation for the bilingual advantage with respect to PWM, one possibility that has gained some empirical support, although it is strongly contested by others (e.g., Morton, Reference Morton2014), is that WM is advantaged in bilinguals owing to the positive effects of bilingualism on executive functions. More specifically, there is growing evidence that complex mental activities that “promote mental stimulation” have positive cognitive outcomes (Kramer & Mota, Reference Kramer and Mota2015, p. 311). It has been argued that bilingualism entails complex cognitive processes that can enhance executive functions and, of particular relevance to the present discussion, WM, a component of executive functions, through the effects of cognitive control. It has been argued further that the acquisition and management of two languages during communication may enhances bilinguals’ WM abilities because they need to pay careful attention to and process information about context, the interlocutor, and the discourse in memory in order to use their languages appropriately and effectively, all tasks that implicate executive functioning. That cognitive processes related to attention and the control of information might play a role in accounting for WM advantages in bilinguals is compatible with the findings we discussed earlier that such additional cognitive processes might be activated to compensate for delays in language exposure in IA children or for impoverished linguistic input in DCHP and cases of repeated bouts of otitis media.

Up to this point, we have focused on the role of early experience on phonological development and PWM. The acquisition and processing of complex language might also be influenced by early language experience owing to the foundational nature of native-language phonology and its cascading effects on more complex language skills that are built on phonological representations. In this regard, there is evidence that once sound categories for the native language have been formed, infants can use these categories to help with word learning and segmentation (Jusczyk & Aslin, Reference Jusczyk and Aslin1995; Stager & Werker, Reference Stager and Werker1997; Swingley & Aslin, Reference Swingley and Aslin2002; Werker, Fennell, Corcoran, & Stager, Reference Werker, Fennell, Corcoran and Stager2002). Similarly, more accurate native language speech sound discrimination in infancy (from newborn to 20 months) has been positively associated not only with early vocabulary skills at 24 months of age (Tsao, Liu, & Kuhl, Reference Tsao, Liu and Kuhl2004) but also with word knowledge at 3 years of age (Molfese & Molfese, Reference Molfese and Molfese1985), verbal fluency, language comprehension, and production abilities at 4 years of age (Bernhardt, Kemp, & Werker, Reference Bernhardt, Kemp and Werker2007), and reading abilities at 5 years of age (Kuhl, Reference Kuhl2010). Thus, there is considerable evidence that the nature of the input received during the earliest stages of language acquisition, and arguably during an early sensitive period for phonological development, may establish a long-lasting foundation for phonological processing that has cascading effects on the acquisition of higher order language skills.

Before concluding, it is important to note that children in all of the groups whom we have discussed, even those who experience delayed and/or impoverished early language input, can develop high levels of language proficiency. In some cases, they even may perform at the same level as children who were exposed to language continuously from birth when tested using general measures of language, such as vocabulary and grammar (e.g., Delcenserie, Reference Delcenserie, Genesee and Delcenserie2016). It is difficult to establish the exact extent to which this is true at present owing to a lack of sufficient and appropriate evidence; differences often emerge only when fine-grained analysis of vocabulary and grammar are examined, and at present, this kind of testing is not common (e.g., Abrahamsson & Hyltenstam, Reference Abrahamsson and Hyltenstam2009; Chilosi et al., Reference Chilosi, Comparini, Scusa, Orazini, Forli, Cipriani and Berrettini2013; Delcenserie & Genesee, Reference Delcenserie and Genesee2014b, Reference Delcenserie, Genesee, Orman and Valleau2014c).

Nevertheless, we would argue that achievement of high levels of language proficiency, although perhaps not “like that of monolingual native speakers,” is compatible with the view that early phonological/PWM development is uniquely and differentially affected by early language exposure because proficiency can be achieved in other ways. First, and as discussed earlier, individuals may be able to recruit additional or alternative neural processes in order to support the acquisition and processing of phonological elements that were not acquired during early language acquisition (e.g., Pierce et al., Reference Pierce, Chen, Delcenserie, Genesee and Klein2015). Second, it may be possible to reopen phonological sensitive periods at a later-than-usual point in development. In partial support of this possibility, there is some evidence that sensitive periods can be reinstated in the auditory (Engineer et al., Reference Engineer, Riley, Seale, Vrana, Shetake, Sudanagunta and Kilgard2011) and visual cortices (Maya Vetencourt et al., Reference Maya Vetencourt, Sale, Viegi, Baroncelli, De Pasquale, O'Leary and Maffei2008, Reference Maya Vetencourt, Tiraboschi, Spolidoro, Castrén and Maffei2011; Silingardi, Scali, Belluomini, & Pizzorusso, Reference Silingardi, Scali, Belluomini and Pizzorusso2010) of adult rats. With respect to humans and language acquisition, Werker and Hensch (Reference Werker and Hensch2015) have suggested that specific training paradigms might lead to the reopening of a sensitive period, possibly due to the deployment of focused attention (Kaliman et al., Reference Kaliman, Alvarez-López, Cosín-Tomás, Rosenkranz, Lutz and Davidson2014; Slagter, Davidson, & Lutz, Reference Slagter, Davidson and Lutz2011). This in turn, they speculate, might activate biological changes and allow for the creation of neural circuits that subserve phonological acquisition to once again become plastic. Extant evidence suggests that if a sensitive period can be reopened as a result of exposure to critical acoustic input (e.g., deaf children with CI) or enriched input (late L2 learners), there may nevertheless be limits on the level of proficiency that can be attained subsequently.

In a related vein, early language experience itself (or lack thereof) may play a role in determining the onset, duration, and offset of the sensitive period for phonological acquisition. For example, IA and CI children might experience extension of the sensitive period for phonological development as a result of delayed exposure to a new or first language, respectively. In the case of IA children, because they are exposed to language from birth, the onset of the sensitive period for phonological development should be opened as expected. If onset of the adoption language does not occur early enough, the period might also close as expected because, as far as their language systems are concerned, they are following a typical trajectory for a monolingual language learner. However, if IA children get input in the adoption language before the period closes, the added linguistic variability or the change in repertoire of input sounds entailed by the adoption language might keep the window open longer, as has been argued to occur in bilingual language learners (e.g., Petitto et al., Reference Petitto, Berens, Kovelman, Dubins, Jasinska and Shalinsky2012; Werker, Byers-Heinlein, & Fennell, Reference Werker, Byers-Heinlein and Fennell2009). As reported earlier, there is a significant negative correlation between age of exposure to the adoption language and PWM capacity and language outcomes, at least during the first several years following adoption (Scott et al., Reference Scott, Roberts and Krakow2008). Thus, an extended sensitive period for phonological development might explain later high levels of language proficiency in some cases.

Differences in sensitive period timing might lead to different PWM outcomes at different points in development. For example, if the sensitive period for phonological development is extended, but PWM is measured early in development (shortly after IA children switch to the adoption language), it may be the case that observed lags are due to phonological acquisition starting at a later point in development, and/or that IA children are still in the process of acquiring phonological representations for their new language. In this case, however, phonological acquisition would utilize learning mechanisms that are typically active during the sensitive period, and thus initial lags may resolve at some later point in development. In contrast, if acquisition of the adoption language truly begins outside of a sensitive period for phonological development, then those IA children might rely on alternative mechanisms to support phonological development and PWM processing. In that case, lags in PWM processing may be observed indefinitely, except in cases where the compensatory mechanisms or adaptations are able to mask underlying differences in phonological processing.

The case for children with CIs is quite different. Because input is required to begin the cascade of processes involved in opening and holding a sensitive period open (e.g., see Cyander & Mitchell, Reference Cynader and Mitchell1980; Fagiolini, Pizzorusso, Berardi, Domenici, & Maffei, Reference Fagiolini, Pizzorusso, Berardi, Domenici and Maffei1994; Mower, Reference Mower1991; Philpot, Sekhar, Shouval, & Bear, Reference Philpot, Sekhar, Shouval and Bear2001, for examples from the visual system), the very reduced phonological input experienced by deaf children with CIs might delay the opening of the sensitive period and then ultimately hold it open later in development than expected, up to a point, following implantation (e.g., Kral & Sharma, Reference Kral and Sharma2012; for a review, see Werker & Hensch, Reference Werker and Hensch2015). Thus, CI children may be able to acquire phonological representations once their CI is functioning via similar means as native language learners, if they receive input early enough. However, even the newest CI technology cannot reproduce the quality of input experienced by children with normal hearing. Moreover, given their later stage of general maturation at the time of cochlear implantation, it is possible that other factors might influence phonological acquisition that occurs later than expected. At present, we lack sufficient evidence to explicate these possibilities further, but they remain theoretically viable and interesting.