In the 1970s and 1980s, infant speech perception and word learning were relatively new fields. Eimas, Siqeland, Jusczyk, and Vigorito’s (Reference Eimas, Siqueland, Jusczyk and Vigorito1971) study demonstrating that 1-month-olds could distinguish /b/ and /d/ defied widely held preconceptions of the limited linguistic potential of young babies. It was in this context that Janet Werker (Werker & Tees, Reference Werker and Tees1984) made her now classic discovery that phonetic perception changes significantly over the first year of life in support of native language acquisition. Since this time, Werker has not only developed and refined a toolbox of methods widely used to study infants’ linguistic sensitivities but also broken new ground with a wealth of studies on how phonetic sensitivity relates to later word learning. Her work is methodologically stunning and theoretically rich. In her Keynote article, she offers a compilation of her results as she describes infants’ journey from sounds to words.
Distinguishing this article from previous reviews are two significant contributions. The first is that word learning affects perception just as perception affects early word learning. The second is that speech perception, itself, is a multisensory process that includes both visual and auditory information at a minimum, and that infants’ perception of speech is inextricably linked to accompanying cross-modal experiences. Here we add to the narrative that unfolds in Werker’s article, raising two points for consideration. First, we consider aspects of infants’ language learning environments that may be minimized in laboratory approaches and possible consequences of this for our appreciation of the factors at play in the “real” world. Second, we discuss the bounds of perceptual narrowing and suggest that a focus on intermediate developments between perceptual narrowing and word learning may be informative.
With respect to the first point, Werker identifies a powerful and testable hypothesis to account for the origins of phonetic categories: distributional learning. She reports findings from two highly compelling studies suggesting that infants’ sensitivities to frequency peaks within a distribution of sounds provides a viable route to phonetic categories. However, the conditions under which these abilities were demonstrated differ quite substantially from those found in a natural language environment. Specifically, infants’ sensitivity to distributional cues to phonetic categories was tested in noninteractive, screen-based tasks where infants were presented with a single distribution of sounds, varying in relative frequency along a single dimension. Yet these tasks are designed to measure linguistic developments that occur in a social world, most likely via interactions, which can produce very different results as noted by Werker (Kuhl, Reference Kuhl2007; Kuhl, Tsao, & Liu, Reference Kuhl, Tsao and Liu2003). In addition, natural language environments contain multiple, concurrent phonetic distributions that exhibit complex distributional profiles. These distributions vary multidimensionally and contain widespread “cue sharing” where a single cue to phoneme identity is influenced by several interdependent phonological features. Furthermore, input analyses suggest that natural phonetic distributions may overlap such that the statistical peaks differentiating native phonetic categories are not as pronounced as those produced for laboratory experiments (Swingley, Reference Swingley2009). While distributional learning provides a strong and convincing route to phonetic category acquisition, how do we know that infants make use of this route to organize the sounds they hear in their vastly more complex and variable language environments? While issues of ecological validity of laboratory testing are not unique to infant speech perception, we suggest that simulated distributional learning in the laboratory could be complemented by the use of real life language statistics to further validate distributional learning as a route to a native phonological inventory.
Specific cross-linguistic comparisons may reveal infants’ use of distributional learning in their natural environments. For example, Korean maintains a three-way verb onset time (VOT) distinction with one clearly marked VOT distinction (aspirated vs. lenis sounds) and one murkier VOT distinction (lenis vs. fortis sounds; Abramson & Lisker, Reference Abramson and Lisker1972). Do Korean infants demonstrate earlier and/or better discrimination of sounds at the higher end of the VOT continuum where distributions are better formed and nicely separated compared to sounds on the lower end where distributions overlap? Moreover, English and Korean each partition the same VOT continuum of stop consonants in different and complex ways potentially leading to overlapping statistical peaks in a mixed English–Korean environment. What about bilingual infants of English and Korean? Do they demonstrate delayed discrimination of English and Korean VOT-contrastive stops in relation to monolingual speakers of either language or do they show evidence of a tag-and-sort mechanism (e.g., Green, Reference Green1998) that permits language-selective VOT distributions?
Another interesting comparison comes from bilingual infants learning Cantonese and English. Unlike Korean, Cantonese and English both exhibit a two-way VOT distinction, resulting in a “pha-ba/pa”-like distinction in both languages. Given that both languages maintain similar bimodal distributions (albeit with some within-category variation across languages), does the distributional similarity between Cantonese and English result in earlier sensitivity to VOT contrasts in each language in Cantonese–English bilinguals compared with Korean–English bilinguals who must maintain both a bimodal and a trimodal distribution?
Using the distributions found in the ambient language is analogous to the approach adopted in laboratory studies of semantic narrowing (e.g., Göksun, Hirsh-Pasek, & Golinkoff, Reference Göksun, Hirsh-Pasek and Golinkoff2010; Göksun et al., Reference Göksun, Hirsh-Pasek, Golinkoff, Imai, Konishi and Okada2011). These studies explore how infants from different language communities perceive motion events based on how these events are described in their everyday social experience. This approach is also adopted in studies of phonological perceptual narrowing (e.g., Polka & Werker, Reference Polka and Werker1994; Werker & Tees, Reference Werker and Tees1984) that examine the effects of infants’ life experience on perception. However, we suggest that support for distributional learning could be strengthened by appealing to real-life statistics as a driver of perceptual narrowing. Such an approach would further confirm the use of statistics learning as a viable explanation for the genesis of phonetic categories.
Related to this is the issue of variability. Many of the studies reported in the article rely on paradigms that minimize sources of variability (e.g., talker variability), yet infants encounter rampant variability in their environments. Variability has been shown to exert strong effects on the specific developmental processes described in the target article, such as phonetic category acquisition and emergent word learning. For example, variability appears to sharpen sensitivity to phonetic contrasts in phoneme categorization tasks (Feldman, Myers, White, Griffiths, & Morgan, Reference Feldman, Myers, White, Griffiths and Morgan2013), word segmentation tasks (Singh, Reference Singh2008), and in learning similar sounding words (Rost & McMurray, Reference Rost and McMurray2009). Variability may even bolster distributional learning by distinguishing high-quality tokens from lower quality tokens (see Adriaans & Swingley, Reference Adriaans and Swingley2012). In addition to being an integral part of language input, Newman (Reference Newman2008) claims that variability supports the infant’s search for native language structure by providing a basis for linguistic generalization and discrimination. Without it, infants might continue to store unnecessary details from the linguistic stream. In this way, variability effects may be relevant to the comparison–contrast mechanism articulated within PRIMIR (Curtin, Byers-Heinlein, & Werker, Reference Curtin, Byers-Heinlein and Werker2011).
Differences in phonetic variability in infants’ natural language environments may influence the outcomes of distributional learning. Fundamental tenets of statistical learning, such as the central limit theorem, predict that the degree of variability shapes the quality of distributional information. High-variability environments, for example, containing a broader range of exemplars, give rise to more pronounced distributional peaks and highlight invariant phonetic properties more prominently than a narrower range of exemplars (e.g., Apfelbaum & McMurray, Reference Apfelbaum and McMurray2011). As with the previous point on laboratory-based versus real-world experience, it is possible to consider effects of real-world variability on infants’ uptake of native language structure. It would be potentially informative to investigate whether individual differences in phonetic variability in infants’ language environments (e.g., number of caregivers, number of accents, and close cross-language phonological correspondences in mixed bilingual environments) strengthen statistical learning. In doing so, studying variability in infants’ natural environments and its effects of phoneme categorization may narrow the distance between laboratory environments and natural language input when investigating the acquisition of phonetic categories.
We turn now to our second point. Werker describes the maintenance and attenuation of phonological sensitivity as a chief precursor to word learning, demonstrating compelling links between these developments. We suggest that this link may be further elaborated by investigating intermediate developments. As noted by Schvachkin (Reference Schvachkin1973/1948), demonstrated by Stager and Werker (Reference Stager and Werker1997) and others (e.g., Swingley, Reference Swingley2016; Swingley & Aslin, Reference Swingley and Aslin2002; White & Morgan, Reference White and Morgan2008), infants are sometimes seemingly insensitive to distinctions they have acquired. Moreover, after infants lose nonnative phonetic distinctions, they occasionally resurrect these sensitivities when learning new words. For example, English-learning infants demonstrate reduced sensitivity to lexical tone by 9 to 10 months (Mattock & Burnham, Reference Mattock and Burnham2006; Yeung, Chen, & Werker, Reference Yeung, Chen and Werker2013, but see Liu & Kager, Reference Liu and Kager2014). Yet, when learning new words, English-learning infants are willing to use lexical tone contrasts 5–6 months later (Hay, Graf Estes, Wang, & Saffran, Reference Hay, Graf Estes, Wang and Saffran2015; Singh, Tam, Chan, & Golinkoff, Reference Singh, Tam, Chan and Golinkoff2014). Perhaps more surprising, at 17 months (many months after native phonetic categories are evident) infants map words onto beeps and colors that are not part of any language’s phonological system (Namy, Reference Namy2001). This suggests that while perceptual narrowing may set the stage for deriving phonological determinants of meaning, there may be further critical developments that lead up to the final act of faithfully linking phonology to words. We suggest that investigating the intermediate transition(s) where native phonological knowledge infiltrates lexical processes may further our understanding of precursors to word learning.
The notion that infants relax their phonetic category boundaries when learning words is quite widely acknowledged in child language research (e.g., Garnica, Reference Garnica1973; Werker, Reference Werker2002) and may have to do with the demands of word learning as suggested by Werker (Reference Werker2002) and further supported by findings that even adults demonstrate similar effects when learning new words (White, Yee, Blumstein, & Morgan, Reference White, Yee, Blumstein and Morgan2013). However, at some point, word learning must become constrained such that native phonological variation is preserved and nonnative variation is either assimilated to native variation or recognized as nonnative. We suggest future examination of developmental “bridges” that link perceptual narrowing and word learning. Such bridges may explain how and when infants go from accepting a broad range of labels such as nonlinguistic mouth sounds as labels for words (Hollich, Hirsh-Pasek, & Golinkoff, Reference Hollich, Hirsh-Pasek, Golinkoff, Brand, Brown, Chung and … Rocroi2000), to a highly sophisticated ability to distinguish words based on subtle phonemic variation, to an even more sophisticated ability to collapse across phonemic variation when hearing a new accent (Buckler, Oczak-Arsic, Siddiqui, & Johnson, Reference Buckler, Oczak-Arsic, Siddiqui and Johnson2017). We propose some possible research questions below that may inform our understanding of how this second development unfolds, which for purposes of simplicity, we will call “phonological–lexical narrowing” here.
First, it would be interesting to investigate how closely coupled perceptual narrowing and phonological–lexical narrowing may be. This bears on the broader theoretical question of whether perceptual narrowing and phonological narrowing exist on a continuum resulting in an extended time table of narrowing or whether they represent distinct developmental processes, pointing to a discontinuous, dual-process account of narrowing. An initial foray into this question could be to investigate whether the timing of perceptual narrowing predicts the timing of phonological–lexical narrowing within individuals.
Second, research by Werker and others has revealed that the course of perceptual narrowing varies for different phonetic categories and is constrained by stimulus properties as well as by how sounds are represented in the learner’s environment (e.g., Anderson, Morgan, & White, Reference Anderson, Morgan and White2003; Narayan, Werker, & Beddor, Reference Narayan, Werker and Beddor2010; Yeung et al., Reference Yeung, Chen and Werker2013). It remains unclear whether the same constraints apply to phonological–lexical narrowing. It is possible, for example, that when engaged in mapping sounds to words, infants rely more heavily on leading social cues to accurately interpret phonetic variation (Hollich, Hirsch-Pasek, et al., Reference Hollich, Hirsh-Pasek, Golinkoff, Brand, Brown, Chung and … Rocroi2000) and/or that phonological–lexical narrowing is driven by vocabulary growth (May & Werker, Reference May and Werker2014). Related to this question, distributional learning provides a strong mechanism by which perceptual narrowing can occur. It remains unclear whether distributional learning (of the same variety) can account for phonological–lexical narrowing or whether different mechanisms might prevail in this stage of development. We suggest that further investigation of phonological–lexical narrowing, as well as of relationships between perceptual narrowing, phonological–lexical narrowing, vocabulary size, and use of social cues may reveal a more detailed and expansive picture of the phonological precursors to word learning.
In conclusion, the Keynote article presents a wealth of powerful evidence attesting to early phonetic sensitivities as precursors to later language acquisition. In this Commentary, we have suggested two research directions that may complement the extensive research discussed in the target article. Werker’s studies reveal how statistically structured learning mechanisms available to infants may be elegantly calibrated to a statistically structured phonological environment. It is this relationship that paves the way to sounds and words and possibly to the successful negotiation of variability. We have suggested that companion studies that investigate infants’ responsiveness to phonetic contrast using statistics computed over their lifetime may confirm the role of distributional learning in infants’ natural environments. In addition, we have suggested that investigating phonological–lexical narrowing may further our understanding of how infants develop a lexicon that is truly well specified.
You have
Access
In the 1970s and 1980s, infant speech perception and word learning were relatively new fields. Eimas, Siqeland, Jusczyk, and Vigorito’s (Reference Eimas, Siqueland, Jusczyk and Vigorito1971) study demonstrating that 1-month-olds could distinguish /b/ and /d/ defied widely held preconceptions of the limited linguistic potential of young babies. It was in this context that Janet Werker (Werker & Tees, Reference Werker and Tees1984) made her now classic discovery that phonetic perception changes significantly over the first year of life in support of native language acquisition. Since this time, Werker has not only developed and refined a toolbox of methods widely used to study infants’ linguistic sensitivities but also broken new ground with a wealth of studies on how phonetic sensitivity relates to later word learning. Her work is methodologically stunning and theoretically rich. In her Keynote article, she offers a compilation of her results as she describes infants’ journey from sounds to words.
Distinguishing this article from previous reviews are two significant contributions. The first is that word learning affects perception just as perception affects early word learning. The second is that speech perception, itself, is a multisensory process that includes both visual and auditory information at a minimum, and that infants’ perception of speech is inextricably linked to accompanying cross-modal experiences. Here we add to the narrative that unfolds in Werker’s article, raising two points for consideration. First, we consider aspects of infants’ language learning environments that may be minimized in laboratory approaches and possible consequences of this for our appreciation of the factors at play in the “real” world. Second, we discuss the bounds of perceptual narrowing and suggest that a focus on intermediate developments between perceptual narrowing and word learning may be informative.
With respect to the first point, Werker identifies a powerful and testable hypothesis to account for the origins of phonetic categories: distributional learning. She reports findings from two highly compelling studies suggesting that infants’ sensitivities to frequency peaks within a distribution of sounds provides a viable route to phonetic categories. However, the conditions under which these abilities were demonstrated differ quite substantially from those found in a natural language environment. Specifically, infants’ sensitivity to distributional cues to phonetic categories was tested in noninteractive, screen-based tasks where infants were presented with a single distribution of sounds, varying in relative frequency along a single dimension. Yet these tasks are designed to measure linguistic developments that occur in a social world, most likely via interactions, which can produce very different results as noted by Werker (Kuhl, Reference Kuhl2007; Kuhl, Tsao, & Liu, Reference Kuhl, Tsao and Liu2003). In addition, natural language environments contain multiple, concurrent phonetic distributions that exhibit complex distributional profiles. These distributions vary multidimensionally and contain widespread “cue sharing” where a single cue to phoneme identity is influenced by several interdependent phonological features. Furthermore, input analyses suggest that natural phonetic distributions may overlap such that the statistical peaks differentiating native phonetic categories are not as pronounced as those produced for laboratory experiments (Swingley, Reference Swingley2009). While distributional learning provides a strong and convincing route to phonetic category acquisition, how do we know that infants make use of this route to organize the sounds they hear in their vastly more complex and variable language environments? While issues of ecological validity of laboratory testing are not unique to infant speech perception, we suggest that simulated distributional learning in the laboratory could be complemented by the use of real life language statistics to further validate distributional learning as a route to a native phonological inventory.
Specific cross-linguistic comparisons may reveal infants’ use of distributional learning in their natural environments. For example, Korean maintains a three-way verb onset time (VOT) distinction with one clearly marked VOT distinction (aspirated vs. lenis sounds) and one murkier VOT distinction (lenis vs. fortis sounds; Abramson & Lisker, Reference Abramson and Lisker1972). Do Korean infants demonstrate earlier and/or better discrimination of sounds at the higher end of the VOT continuum where distributions are better formed and nicely separated compared to sounds on the lower end where distributions overlap? Moreover, English and Korean each partition the same VOT continuum of stop consonants in different and complex ways potentially leading to overlapping statistical peaks in a mixed English–Korean environment. What about bilingual infants of English and Korean? Do they demonstrate delayed discrimination of English and Korean VOT-contrastive stops in relation to monolingual speakers of either language or do they show evidence of a tag-and-sort mechanism (e.g., Green, Reference Green1998) that permits language-selective VOT distributions?
Another interesting comparison comes from bilingual infants learning Cantonese and English. Unlike Korean, Cantonese and English both exhibit a two-way VOT distinction, resulting in a “pha-ba/pa”-like distinction in both languages. Given that both languages maintain similar bimodal distributions (albeit with some within-category variation across languages), does the distributional similarity between Cantonese and English result in earlier sensitivity to VOT contrasts in each language in Cantonese–English bilinguals compared with Korean–English bilinguals who must maintain both a bimodal and a trimodal distribution?
Using the distributions found in the ambient language is analogous to the approach adopted in laboratory studies of semantic narrowing (e.g., Göksun, Hirsh-Pasek, & Golinkoff, Reference Göksun, Hirsh-Pasek and Golinkoff2010; Göksun et al., Reference Göksun, Hirsh-Pasek, Golinkoff, Imai, Konishi and Okada2011). These studies explore how infants from different language communities perceive motion events based on how these events are described in their everyday social experience. This approach is also adopted in studies of phonological perceptual narrowing (e.g., Polka & Werker, Reference Polka and Werker1994; Werker & Tees, Reference Werker and Tees1984) that examine the effects of infants’ life experience on perception. However, we suggest that support for distributional learning could be strengthened by appealing to real-life statistics as a driver of perceptual narrowing. Such an approach would further confirm the use of statistics learning as a viable explanation for the genesis of phonetic categories.
Related to this is the issue of variability. Many of the studies reported in the article rely on paradigms that minimize sources of variability (e.g., talker variability), yet infants encounter rampant variability in their environments. Variability has been shown to exert strong effects on the specific developmental processes described in the target article, such as phonetic category acquisition and emergent word learning. For example, variability appears to sharpen sensitivity to phonetic contrasts in phoneme categorization tasks (Feldman, Myers, White, Griffiths, & Morgan, Reference Feldman, Myers, White, Griffiths and Morgan2013), word segmentation tasks (Singh, Reference Singh2008), and in learning similar sounding words (Rost & McMurray, Reference Rost and McMurray2009). Variability may even bolster distributional learning by distinguishing high-quality tokens from lower quality tokens (see Adriaans & Swingley, Reference Adriaans and Swingley2012). In addition to being an integral part of language input, Newman (Reference Newman2008) claims that variability supports the infant’s search for native language structure by providing a basis for linguistic generalization and discrimination. Without it, infants might continue to store unnecessary details from the linguistic stream. In this way, variability effects may be relevant to the comparison–contrast mechanism articulated within PRIMIR (Curtin, Byers-Heinlein, & Werker, Reference Curtin, Byers-Heinlein and Werker2011).
Differences in phonetic variability in infants’ natural language environments may influence the outcomes of distributional learning. Fundamental tenets of statistical learning, such as the central limit theorem, predict that the degree of variability shapes the quality of distributional information. High-variability environments, for example, containing a broader range of exemplars, give rise to more pronounced distributional peaks and highlight invariant phonetic properties more prominently than a narrower range of exemplars (e.g., Apfelbaum & McMurray, Reference Apfelbaum and McMurray2011). As with the previous point on laboratory-based versus real-world experience, it is possible to consider effects of real-world variability on infants’ uptake of native language structure. It would be potentially informative to investigate whether individual differences in phonetic variability in infants’ language environments (e.g., number of caregivers, number of accents, and close cross-language phonological correspondences in mixed bilingual environments) strengthen statistical learning. In doing so, studying variability in infants’ natural environments and its effects of phoneme categorization may narrow the distance between laboratory environments and natural language input when investigating the acquisition of phonetic categories.
We turn now to our second point. Werker describes the maintenance and attenuation of phonological sensitivity as a chief precursor to word learning, demonstrating compelling links between these developments. We suggest that this link may be further elaborated by investigating intermediate developments. As noted by Schvachkin (Reference Schvachkin1973/1948), demonstrated by Stager and Werker (Reference Stager and Werker1997) and others (e.g., Swingley, Reference Swingley2016; Swingley & Aslin, Reference Swingley and Aslin2002; White & Morgan, Reference White and Morgan2008), infants are sometimes seemingly insensitive to distinctions they have acquired. Moreover, after infants lose nonnative phonetic distinctions, they occasionally resurrect these sensitivities when learning new words. For example, English-learning infants demonstrate reduced sensitivity to lexical tone by 9 to 10 months (Mattock & Burnham, Reference Mattock and Burnham2006; Yeung, Chen, & Werker, Reference Yeung, Chen and Werker2013, but see Liu & Kager, Reference Liu and Kager2014). Yet, when learning new words, English-learning infants are willing to use lexical tone contrasts 5–6 months later (Hay, Graf Estes, Wang, & Saffran, Reference Hay, Graf Estes, Wang and Saffran2015; Singh, Tam, Chan, & Golinkoff, Reference Singh, Tam, Chan and Golinkoff2014). Perhaps more surprising, at 17 months (many months after native phonetic categories are evident) infants map words onto beeps and colors that are not part of any language’s phonological system (Namy, Reference Namy2001). This suggests that while perceptual narrowing may set the stage for deriving phonological determinants of meaning, there may be further critical developments that lead up to the final act of faithfully linking phonology to words. We suggest that investigating the intermediate transition(s) where native phonological knowledge infiltrates lexical processes may further our understanding of precursors to word learning.
The notion that infants relax their phonetic category boundaries when learning words is quite widely acknowledged in child language research (e.g., Garnica, Reference Garnica1973; Werker, Reference Werker2002) and may have to do with the demands of word learning as suggested by Werker (Reference Werker2002) and further supported by findings that even adults demonstrate similar effects when learning new words (White, Yee, Blumstein, & Morgan, Reference White, Yee, Blumstein and Morgan2013). However, at some point, word learning must become constrained such that native phonological variation is preserved and nonnative variation is either assimilated to native variation or recognized as nonnative. We suggest future examination of developmental “bridges” that link perceptual narrowing and word learning. Such bridges may explain how and when infants go from accepting a broad range of labels such as nonlinguistic mouth sounds as labels for words (Hollich, Hirsh-Pasek, & Golinkoff, Reference Hollich, Hirsh-Pasek, Golinkoff, Brand, Brown, Chung and … Rocroi2000), to a highly sophisticated ability to distinguish words based on subtle phonemic variation, to an even more sophisticated ability to collapse across phonemic variation when hearing a new accent (Buckler, Oczak-Arsic, Siddiqui, & Johnson, Reference Buckler, Oczak-Arsic, Siddiqui and Johnson2017). We propose some possible research questions below that may inform our understanding of how this second development unfolds, which for purposes of simplicity, we will call “phonological–lexical narrowing” here.
First, it would be interesting to investigate how closely coupled perceptual narrowing and phonological–lexical narrowing may be. This bears on the broader theoretical question of whether perceptual narrowing and phonological narrowing exist on a continuum resulting in an extended time table of narrowing or whether they represent distinct developmental processes, pointing to a discontinuous, dual-process account of narrowing. An initial foray into this question could be to investigate whether the timing of perceptual narrowing predicts the timing of phonological–lexical narrowing within individuals.
Second, research by Werker and others has revealed that the course of perceptual narrowing varies for different phonetic categories and is constrained by stimulus properties as well as by how sounds are represented in the learner’s environment (e.g., Anderson, Morgan, & White, Reference Anderson, Morgan and White2003; Narayan, Werker, & Beddor, Reference Narayan, Werker and Beddor2010; Yeung et al., Reference Yeung, Chen and Werker2013). It remains unclear whether the same constraints apply to phonological–lexical narrowing. It is possible, for example, that when engaged in mapping sounds to words, infants rely more heavily on leading social cues to accurately interpret phonetic variation (Hollich, Hirsch-Pasek, et al., Reference Hollich, Hirsh-Pasek, Golinkoff, Brand, Brown, Chung and … Rocroi2000) and/or that phonological–lexical narrowing is driven by vocabulary growth (May & Werker, Reference May and Werker2014). Related to this question, distributional learning provides a strong mechanism by which perceptual narrowing can occur. It remains unclear whether distributional learning (of the same variety) can account for phonological–lexical narrowing or whether different mechanisms might prevail in this stage of development. We suggest that further investigation of phonological–lexical narrowing, as well as of relationships between perceptual narrowing, phonological–lexical narrowing, vocabulary size, and use of social cues may reveal a more detailed and expansive picture of the phonological precursors to word learning.
In conclusion, the Keynote article presents a wealth of powerful evidence attesting to early phonetic sensitivities as precursors to later language acquisition. In this Commentary, we have suggested two research directions that may complement the extensive research discussed in the target article. Werker’s studies reveal how statistically structured learning mechanisms available to infants may be elegantly calibrated to a statistically structured phonological environment. It is this relationship that paves the way to sounds and words and possibly to the successful negotiation of variability. We have suggested that companion studies that investigate infants’ responsiveness to phonetic contrast using statistics computed over their lifetime may confirm the role of distributional learning in infants’ natural environments. In addition, we have suggested that investigating phonological–lexical narrowing may further our understanding of how infants develop a lexicon that is truly well specified.
ACKNOWLEDGMENTS
The writing of this Commentary was supported by Ministry of Education Tier 1 Grant FY2013-FRC2-009 and an HSS seed grant from the National University of Singapore to the first author, Grant SBE-SMA-1514493 from the National Science Foundation to the second author, and Institute of Education Sciences Grant R305A110284 to the third and fourth authors. Correspondence concerning this article should be addressed to Singh [leher.singh.nus@gmail.com] or Golinkoff [Roberta@udel.edu].