Pierce, Genesee, Delcenserie, and Morgan have provided an innovative and thought-provoking juxtaposition of the putative role played by phonological working memory (PWM) in explaining individual and group differences among early internationally adopted (IA) children, deaf children with cochlear implants, simultaneous and sequential bilinguals, children who are learning sign languages, and children with otitis media. This novel comparison suggests that adequate exposure to phonology before 12 months of age is key in the development of improved PWM. Consequently, long-term linguistic advantages in vocabulary and learning of morphosyntax arise, but not advantages in other areas of higher cognition. One significant implication of the review is that although many intriguing links exist between language development and PWM, irrefutable conclusions elude the field as to the directionality of a causal relationship between phonological development, other linguistic development, and PWM. In this commentary, I should like to make the somewhat controversial proposal (Gathercole, Reference Gathercole2006, and commentaries) that the evidence presented points to PWM being an epiphenomenon arising out of individual differences in the robustness and richness of phonological representations themselves. The authors hinted several times at the tantalizing relationship between phonology and PWM, but they do not articulate explicitly that PWM could be a redundant construct. I also offer some proposals as to how one might test this suggestion experimentally or in a corpus of child language.
The evidence for PWM being an epiphenomenon derives from the review in the following way: knowledge of phonology precedes PWM; bilingualism or multilingualism from birth leads directly to higher PWM; the effect of PWM disappears by around age 5 when vocabulary knowledge predicts WM; for deaf children with cochlear implants, PWM is reliably lower than for IA children, but otherwise they are developmentally normal, suggesting that early sound exposure is key for PWM, not an independent “capacity” that PWM measures separately; and finally, older children who learn a second language do not have better PWM initially, but they do once the second language and expanded phonological inventory has been established.
The main problem for the construct of PWM independent of language in the literature presented lies in the dangers of circularity and correlation. The “P” in PWM obviously stands for phonological, a temporary store or buffer of (a string of) phonemes. Hence, PWM is in danger of presupposing what it seeks to explain because phonemes are only relevant when considered as part of the pattern of contrasts that a language uses to create a lexicon. The authors note that humans “have access to a powerful information processing system that can analyze, store, and use regularities in the signal to which the child is exposed,” but they do not elaborate on what that system is. Without getting into the debate about innateness and modularity, the studies that the authors cite point strongly to a mechanism that homes in on the phonology of the ambient language, perhaps in utero, but with in utero exposure not shutting off sensitivity before the child is 1 year old. Because this specific sensitivity is so rapidly deployed and set, some limit on the hypothesis space must exist to exclude nonhuman language sounds and other linguistically irrelevant acoustic input. In other words, PWM must have some clue about what P might be even before it starts processing and storing in short-term PWM.
In this context, it is unsurprising that the most reliable tests of PWM are based on nonword repetition tasks that either contain sequences of phonemes in the ambient language or, even more reliably in some cases, contain phonemes completely absent in the ambient language. For example, French and O'Brien's (Reference French and O'Brien2008) used a task containing Arabic phonemes that occur in neither French nor English in their study of elementary French-speaking children learning English.
The authors’ overview points strongly to advantages of simply knowing more phonemes as a predictor of PWM, thus reducing the need for a separate construct that is based on ability to store longer strings of phonemes in short-term memory. This observation explains why, by keeping more phonological options open, early, simultaneous multilingualism leads to improved language skills later. For example, French-speaking children simultaneously learning English may pay heed to the allophonic nature of aspiration in English stops and learn interdental fricatives, whereas English-speaking children learning French will need to keep open phonemic contrasts between oral and nasal vowels and the distribution of lip rounding to include not just (high) back vowels. Moreover, IA Chinese-speaking children who have exposure early enough to French, for example, will learn that voicing contrasts are important in stops, and based on Pierce, Klein, Chen, Delcenserie, and Genesee's (Reference Pierce, Klein, Chen, Delcenserie and Genesee2014) work could retain aspiration in stops as phonemic as well. (If these children learn English, they will need to learn the allophonic nature of aspiration in stops, too!)
The psychological reality of the phoneme or segment-sized units is an important issue in phonology that is not addressed in the text, and should be carefully considered in developing PWM tests. Nathan (Reference Nathan, Dabrowska and Divjak2015) provides a recent and accessible review of the issues. With this caveat, it is clear that when the appropriate grain-size of analysis is considered, for example, as in MacKenzie, Curtin, and Graham (Reference MacKenzie, Curtin and Graham2012), it appears that the ambient language phonotactics alone have a direct impact on vocabulary learning. Mackenzie et al. (Reference MacKenzie, Curtin and Graham2012) demonstrated that 12-month-old infants are open to learning word–object pairings only for pairings that match objects with labels that conform to expected phonological shapes.
The patterns reported by the authors of the review and by MacKenzie et al. suggest some ways to tease apart phonological knowledge from PWM in IA children. If an existing range of knowledge of phonological inventory and phonotactics is key, then carefully designed different tests of PWM could be designed that manipulate both length of nonword (Gathercole, Reference Gathercole2006) and phonemes used. For example, if early exposure to Chinese restricts word label matching, one might expect IA Chinese-speaking children to do less well on PWM tests that include complex onset and especially coda clusters (e.g., Comprehensive Test of Phonological Processing; Wagner, Torgesen, & Rashotte, Reference Wagner, Torgesen and Rashotte1999) due to the restriction on syllable codas in Chinese dialects (Juffs, Reference Juffs1990). However, in a different PWM test that contains phonemes in Chinese but not English (e.g., alveopalatal and velar fricatives and postalveolar retroflex affricates), IA Chinese–English bilinguals might perform better than English-speaking monolingual children on such a test, regardless of length of the nonword. It is worth noting that when such tests as the Comprehensive Test of Phonological Processing are scored on narrow phonetic parameters rather than just number syllables/length, they become indicators of proficiency, further indicating that the PWM could serve as a proxy for other abilities and not an independent construct itself (Juffs & Rodríguez, Reference Juffs, Rodríguez and Young-Scholten2008). Thus, tests of PWM should control both for length of item and the phonemes/clusters included.
In corpus work on IA children, one might expect first language (L1) phonological effects on the robustness of lexical representation. If these effects link to PWM scores, but not phoneme inventory or phonotactics, then this would be evidence for PWM as an independent construct. The lexical quality hypothesis, proposed by Perfetti and Hart (Reference Perfetti, Hart, Verhoeven, Elbro and Reitsma2002) for literacy (i.e., the strength of links between phonological and semantic representation, not including orthography at this point obviously), could be useful here. Thus, supposing that the phonemic inventory that a child has restricts his or her labeling in early stages of acquisition (MacKenzie et al., Reference MacKenzie, Curtin and Graham2012), one prediction might be that the lexical quality of the children's English words would be superior where the phoneme exists in both English and Chinese because earlier and more frequent exposure would lead to more robust instantiation of the phonemic status of a sound that can be deployed in lexical items. IA Chinese-speaking children might also perform better than monolingual English-speaking children in words that had more of a category in Chinese but not English. For example, the range of affricates in Chinese is broader and more frequent; thus, English words with affricates may emerge earlier and more accurately in Chinese IA children. Lexical quality might also be worse than monolinguals with (complex) onset and coda clusters, bearing in mind that clusters are subject to universal sonority constraints (Selkirk, Reference Selkirk, Aronoff and Oerhle1984).
The point I am making is that the authors have opened a door on future research that could manipulate the phonological structure of PWM test items and lexical representation in early child language corpora, enabling researchers to probe whether such a thing as PWM exists as a construct independent of the developing phonology and lexicon. The fact that effects of PWM dissipate rather quickly, relatively speaking, suggests that it is not a separate construct, but across the life span is related to the degree to which sensitivity to new phonemic contrast remains available.
If the construct of an in-born, fixed, psychometrically measurable capacity such as PWM to explain individual differences is problematic (for a variety of reasons), then what could explain individual differences among learners in L1 and special cases of early second language development? Years of exposure are held equal in most studies, for example, 9.5 years on average in Delcenserie and Genesee (2014), but years constitute a coarse measure that perhaps does not capture early differences in the amount of social interaction among parents and infants. As Kuhl (Reference Kuhl2004) and the authors note, differences in interaction and input are key; the authors provide the example that nonnative signing parents sign less with infants or not at all until they themselves learn some sign. One possibility then is that differences in parent–child directed attention have an effect. Variability in the richness of verbal play and shared reference is important in infant development, and shared reference at a very early age has an effect on naming, contrasts, and possibly phonological development (see Tamis-LeMonda, Song, Leavell, Kahana-Kalman, & Yoshikawa, Reference Tamis-LeMonda, Song, Leavell, Kahana-Kalman and Yoshikawa2012, for research that found ethnicity and language related differences in this area). Within-group variation is of course to be expected here too. Furthermore, in adult SLA some researchers have argued that tests of working memory tap into attention ability (Sunderman & Kroll, Reference Sunderman and Kroll2009). Thus, differences in (directed) attention to sounds and word play could be the source of individual differences in “PWM” because the children learn more phonology from more interaction. This third factor explains why phonological development and PWM develop in tandem because the cause is another, separate factor.
In sum, the authors’ review suggested to me the possibility that PWM is an epiphenomenon. To test this idea, more attention could be paid to manipulating the phonemes in nonword PWM tests to identify early L1 influence. As the authors mention from other research (Pierce et al., Reference Pierce, Klein, Chen, Delcenserie and Genesee2014), the L1 remains neurally active among IA children. Thus, the grain-size of future research is important because teasing out L1 effects must account for preexisting, subconscious knowledge of phonology at a level as fine-grained as a linguistic feature, phoneme, and syllable structure. A note of caution here is that MacEnzie et al. (Reference MacKenzie, Curtin and Graham2012) note that phonetic (allophonic) variation seems to be paid no heed by children, perhaps because they need to be tolerant of variation in input and tune into to the sociolinguistic impact of variation later. For corpus studies, a finding some lexical items in vocabulary are less robustly represented based on early L1 phonemes and phonotactics could also be one area to look for alternative explanations of putative PWM effects on development. Finer-grained measures of interaction time could also supplement such research. Finally, whether PWM remains a fundamental construct in explaining development or not, the authors have laid important groundwork for future research in memory and language among bilinguals and other nontypical groups of learners.
You have
Access
Pierce, Genesee, Delcenserie, and Morgan have provided an innovative and thought-provoking juxtaposition of the putative role played by phonological working memory (PWM) in explaining individual and group differences among early internationally adopted (IA) children, deaf children with cochlear implants, simultaneous and sequential bilinguals, children who are learning sign languages, and children with otitis media. This novel comparison suggests that adequate exposure to phonology before 12 months of age is key in the development of improved PWM. Consequently, long-term linguistic advantages in vocabulary and learning of morphosyntax arise, but not advantages in other areas of higher cognition. One significant implication of the review is that although many intriguing links exist between language development and PWM, irrefutable conclusions elude the field as to the directionality of a causal relationship between phonological development, other linguistic development, and PWM. In this commentary, I should like to make the somewhat controversial proposal (Gathercole, Reference Gathercole2006, and commentaries) that the evidence presented points to PWM being an epiphenomenon arising out of individual differences in the robustness and richness of phonological representations themselves. The authors hinted several times at the tantalizing relationship between phonology and PWM, but they do not articulate explicitly that PWM could be a redundant construct. I also offer some proposals as to how one might test this suggestion experimentally or in a corpus of child language.
The evidence for PWM being an epiphenomenon derives from the review in the following way: knowledge of phonology precedes PWM; bilingualism or multilingualism from birth leads directly to higher PWM; the effect of PWM disappears by around age 5 when vocabulary knowledge predicts WM; for deaf children with cochlear implants, PWM is reliably lower than for IA children, but otherwise they are developmentally normal, suggesting that early sound exposure is key for PWM, not an independent “capacity” that PWM measures separately; and finally, older children who learn a second language do not have better PWM initially, but they do once the second language and expanded phonological inventory has been established.
The main problem for the construct of PWM independent of language in the literature presented lies in the dangers of circularity and correlation. The “P” in PWM obviously stands for phonological, a temporary store or buffer of (a string of) phonemes. Hence, PWM is in danger of presupposing what it seeks to explain because phonemes are only relevant when considered as part of the pattern of contrasts that a language uses to create a lexicon. The authors note that humans “have access to a powerful information processing system that can analyze, store, and use regularities in the signal to which the child is exposed,” but they do not elaborate on what that system is. Without getting into the debate about innateness and modularity, the studies that the authors cite point strongly to a mechanism that homes in on the phonology of the ambient language, perhaps in utero, but with in utero exposure not shutting off sensitivity before the child is 1 year old. Because this specific sensitivity is so rapidly deployed and set, some limit on the hypothesis space must exist to exclude nonhuman language sounds and other linguistically irrelevant acoustic input. In other words, PWM must have some clue about what P might be even before it starts processing and storing in short-term PWM.
In this context, it is unsurprising that the most reliable tests of PWM are based on nonword repetition tasks that either contain sequences of phonemes in the ambient language or, even more reliably in some cases, contain phonemes completely absent in the ambient language. For example, French and O'Brien's (Reference French and O'Brien2008) used a task containing Arabic phonemes that occur in neither French nor English in their study of elementary French-speaking children learning English.
The authors’ overview points strongly to advantages of simply knowing more phonemes as a predictor of PWM, thus reducing the need for a separate construct that is based on ability to store longer strings of phonemes in short-term memory. This observation explains why, by keeping more phonological options open, early, simultaneous multilingualism leads to improved language skills later. For example, French-speaking children simultaneously learning English may pay heed to the allophonic nature of aspiration in English stops and learn interdental fricatives, whereas English-speaking children learning French will need to keep open phonemic contrasts between oral and nasal vowels and the distribution of lip rounding to include not just (high) back vowels. Moreover, IA Chinese-speaking children who have exposure early enough to French, for example, will learn that voicing contrasts are important in stops, and based on Pierce, Klein, Chen, Delcenserie, and Genesee's (Reference Pierce, Klein, Chen, Delcenserie and Genesee2014) work could retain aspiration in stops as phonemic as well. (If these children learn English, they will need to learn the allophonic nature of aspiration in stops, too!)
The psychological reality of the phoneme or segment-sized units is an important issue in phonology that is not addressed in the text, and should be carefully considered in developing PWM tests. Nathan (Reference Nathan, Dabrowska and Divjak2015) provides a recent and accessible review of the issues. With this caveat, it is clear that when the appropriate grain-size of analysis is considered, for example, as in MacKenzie, Curtin, and Graham (Reference MacKenzie, Curtin and Graham2012), it appears that the ambient language phonotactics alone have a direct impact on vocabulary learning. Mackenzie et al. (Reference MacKenzie, Curtin and Graham2012) demonstrated that 12-month-old infants are open to learning word–object pairings only for pairings that match objects with labels that conform to expected phonological shapes.
The patterns reported by the authors of the review and by MacKenzie et al. suggest some ways to tease apart phonological knowledge from PWM in IA children. If an existing range of knowledge of phonological inventory and phonotactics is key, then carefully designed different tests of PWM could be designed that manipulate both length of nonword (Gathercole, Reference Gathercole2006) and phonemes used. For example, if early exposure to Chinese restricts word label matching, one might expect IA Chinese-speaking children to do less well on PWM tests that include complex onset and especially coda clusters (e.g., Comprehensive Test of Phonological Processing; Wagner, Torgesen, & Rashotte, Reference Wagner, Torgesen and Rashotte1999) due to the restriction on syllable codas in Chinese dialects (Juffs, Reference Juffs1990). However, in a different PWM test that contains phonemes in Chinese but not English (e.g., alveopalatal and velar fricatives and postalveolar retroflex affricates), IA Chinese–English bilinguals might perform better than English-speaking monolingual children on such a test, regardless of length of the nonword. It is worth noting that when such tests as the Comprehensive Test of Phonological Processing are scored on narrow phonetic parameters rather than just number syllables/length, they become indicators of proficiency, further indicating that the PWM could serve as a proxy for other abilities and not an independent construct itself (Juffs & Rodríguez, Reference Juffs, Rodríguez and Young-Scholten2008). Thus, tests of PWM should control both for length of item and the phonemes/clusters included.
In corpus work on IA children, one might expect first language (L1) phonological effects on the robustness of lexical representation. If these effects link to PWM scores, but not phoneme inventory or phonotactics, then this would be evidence for PWM as an independent construct. The lexical quality hypothesis, proposed by Perfetti and Hart (Reference Perfetti, Hart, Verhoeven, Elbro and Reitsma2002) for literacy (i.e., the strength of links between phonological and semantic representation, not including orthography at this point obviously), could be useful here. Thus, supposing that the phonemic inventory that a child has restricts his or her labeling in early stages of acquisition (MacKenzie et al., Reference MacKenzie, Curtin and Graham2012), one prediction might be that the lexical quality of the children's English words would be superior where the phoneme exists in both English and Chinese because earlier and more frequent exposure would lead to more robust instantiation of the phonemic status of a sound that can be deployed in lexical items. IA Chinese-speaking children might also perform better than monolingual English-speaking children in words that had more of a category in Chinese but not English. For example, the range of affricates in Chinese is broader and more frequent; thus, English words with affricates may emerge earlier and more accurately in Chinese IA children. Lexical quality might also be worse than monolinguals with (complex) onset and coda clusters, bearing in mind that clusters are subject to universal sonority constraints (Selkirk, Reference Selkirk, Aronoff and Oerhle1984).
The point I am making is that the authors have opened a door on future research that could manipulate the phonological structure of PWM test items and lexical representation in early child language corpora, enabling researchers to probe whether such a thing as PWM exists as a construct independent of the developing phonology and lexicon. The fact that effects of PWM dissipate rather quickly, relatively speaking, suggests that it is not a separate construct, but across the life span is related to the degree to which sensitivity to new phonemic contrast remains available.
If the construct of an in-born, fixed, psychometrically measurable capacity such as PWM to explain individual differences is problematic (for a variety of reasons), then what could explain individual differences among learners in L1 and special cases of early second language development? Years of exposure are held equal in most studies, for example, 9.5 years on average in Delcenserie and Genesee (2014), but years constitute a coarse measure that perhaps does not capture early differences in the amount of social interaction among parents and infants. As Kuhl (Reference Kuhl2004) and the authors note, differences in interaction and input are key; the authors provide the example that nonnative signing parents sign less with infants or not at all until they themselves learn some sign. One possibility then is that differences in parent–child directed attention have an effect. Variability in the richness of verbal play and shared reference is important in infant development, and shared reference at a very early age has an effect on naming, contrasts, and possibly phonological development (see Tamis-LeMonda, Song, Leavell, Kahana-Kalman, & Yoshikawa, Reference Tamis-LeMonda, Song, Leavell, Kahana-Kalman and Yoshikawa2012, for research that found ethnicity and language related differences in this area). Within-group variation is of course to be expected here too. Furthermore, in adult SLA some researchers have argued that tests of working memory tap into attention ability (Sunderman & Kroll, Reference Sunderman and Kroll2009). Thus, differences in (directed) attention to sounds and word play could be the source of individual differences in “PWM” because the children learn more phonology from more interaction. This third factor explains why phonological development and PWM develop in tandem because the cause is another, separate factor.
In sum, the authors’ review suggested to me the possibility that PWM is an epiphenomenon. To test this idea, more attention could be paid to manipulating the phonemes in nonword PWM tests to identify early L1 influence. As the authors mention from other research (Pierce et al., Reference Pierce, Klein, Chen, Delcenserie and Genesee2014), the L1 remains neurally active among IA children. Thus, the grain-size of future research is important because teasing out L1 effects must account for preexisting, subconscious knowledge of phonology at a level as fine-grained as a linguistic feature, phoneme, and syllable structure. A note of caution here is that MacEnzie et al. (Reference MacKenzie, Curtin and Graham2012) note that phonetic (allophonic) variation seems to be paid no heed by children, perhaps because they need to be tolerant of variation in input and tune into to the sociolinguistic impact of variation later. For corpus studies, a finding some lexical items in vocabulary are less robustly represented based on early L1 phonemes and phonotactics could also be one area to look for alternative explanations of putative PWM effects on development. Finer-grained measures of interaction time could also supplement such research. Finally, whether PWM remains a fundamental construct in explaining development or not, the authors have laid important groundwork for future research in memory and language among bilinguals and other nontypical groups of learners.