Interaction in Spanish–English bilinguals’ acquisition of syllable structure

Delay

A number of previous studies on interaction in bilinguals have presented evidence of delayed acquisition. However, many of the existent phonological studies have found delay for general categories, comparing consonant or vowel accuracy overall, consonant accuracy for different consonant classes, or error rates in general (Gildersleeve et al., 1996; Gildersleeve-Neumann et al., 2008; Goldstein & Washington, 2001), sometimes in the form of case studies of one or two children (e.g. Holm & Dodd, 1999 on the phonological development of sequential Cantonese–English bilinguals). These findings are valuable, especially given that bilinguals remain underrepresented in the acquisition literature. However, there is still a need for studies that investigate potential causes of delay during bilingual phonological acquisition.

Kehoe (2002) investigated vowel acquisition in Spanish–German bilinguals, and predicted interaction patterns on the basis of systemic differences in complexity. One prediction was that bilinguals’ exposure to German’s more complex vowel system (German employs a length contrast) would result in neither delayed nor accelerated acquisition of the less complex Spanish vowel system (i.e. bilinguals and monolinguals would exhibit comparable rates of acquisition of the Spanish vowel system). A second prediction was that the influence of the less complex Spanish vowel system (framed as a decrease in positive evidence of a vowel length contrast due to its less frequent occurrence in bilingual input overall) would result in bilinguals’ delayed acquisition of the more complex German vowel system compared to German monolinguals. Kehoe (2002) found support for both hypotheses using accuracy measures (i.e. percentages of successful sound production attempts out of all respective production attempts) applied to naturalistic speech data from three German monolingual, two Spanish monolingual, and three German–Spanish bilingual children from the onset of word-production to age 3. As predicted, vowel accuracy in Spanish did not differ between bilinguals and monolinguals. However, bilinguals exhibited lower accuracy scores for both short and long vowels in German at all points when compared to German monolinguals. This difference in accuracy is interpreted as a case of delay where less frequent exposure to the German length contrast (possibly coupled with exposure to a system that does not exhibit a vowel length contrast, i.e. Spanish) caused bilinguals to acquire the target vowel system of German more slowly than monolinguals.

Providing more evidence linking delay with decreased exposure to language-specific phonological phenomena, Fabiano-Smith and Goldstein (2010) compared consonant accuracy measures for Spanish and English bilinguals and monolinguals. They followed Flege (1981, 1987), in hypothesizing that bilingual learners lump phonetically similar sounds (including non-identical sounds) into the same phonemic category accessible in both languages. This implies that bilingual learners would have more frequent experience, in both perception and production, with shared compared to unshared categories across their two systems. If more experience with a category correlates with greater accuracy in the production of that category during acquisition, then bilinguals’ productions of phonetically similar sounds in each language should be more accurate than their own productions of phonetically dissimilar, unshared sounds in each language.

Fabiano-Smith and Goldstein (2010) analyzed single-word and connected speech productions of 24 participants with ages 3–4 years (8 monolingual Spanish, 8 monolingual English, and 8 bilingual Spanish–English). As predicted, analysis showed that bilinguals produced the set of shared sounds in each language with greater accuracy than the set of unshared sounds specific to Spanish or English. No such difference was found for English monolinguals. However, Spanish monolinguals produced the set of shared sounds with greater accuracy than the set of unshared sounds, despite monolinguals not ‘sharing’ sounds with another language. Further analysis suggested that this effect was driven by low accuracy among Spanish monolinguals on the flap and trill, which are marked sounds, and typically acquired later (Acevedo, 1993; Jimenez, 1987). Accordingly, comparisons of shared and unshared sounds should also take the markedness or complexity of those sounds into account. Fabiano-Smith and Goldstein (2010) also found that bilinguals were less accurate in their production of Spanish consonants overall compared to Spanish monolinguals, but comparably accurate in their production of English consonants compared to English monolinguals. In particular, bilinguals exhibited lower accuracy on the trill, fricatives, and glides in Spanish compared to Spanish monolinguals, and on stops and fricatives in English compared to English monolinguals, suggesting that bilinguals’ acquisition of these particular manner classes was delayed. However, for other sounds, bilinguals were performing age-appropriately by monolingual standards in both languages.

Further evidence of delay is presented by Gildersleeve-Neumann et al. (2008). This study analyzed consonant and vowel productions from elicited English single-word productions from 33 participants (10 monolingual English, 20 exposed predominantly to English with 10 or fewer hours of weekly exposure to Spanish, and 3 balanced bilinguals with equivalent exposure to Spanish and English). Participants were recorded twice, once at the beginning and once at the end of the same academic year. The authors found some evidence of delay in bilinguals’ productions of interdentals and affricates in that fewer bilinguals in both groups were producing these sounds at the beginning of the year compared to monolinguals. However, the authors did not find any significant differences between groups on measures of overall vowel and consonant production accuracy. Examination of common consonant error patterns, however, found that both bilingual groups exhibited significantly higher average consonant error rates than the monolingual group, and that the difference between the two bilingual groups on this measure approached significance. This suggests that Spanish–English bilinguals were producing common consonant errors more frequently than English monolinguals, and that the less exposure the child had to English, the higher the consonant error rate was likely to be. For all three groups, word-final consonant devoicing was the most common consonant error, followed by gliding of liquids. Although bilinguals made more of the errors being examined, the types of errors made are not unusual in monolingual acquisition of English.

In terms of structural acquisition, both bilingual groups in Gildersleeve-Neumann et al.’s (2008) study were less likely than monolinguals to produce word-initial, -medial, and -final consonant clusters (i.e. bilinguals were more likely than monolinguals to lack clusters in their phonetic inventories). These differences were most pronounced for the 3 balanced bilinguals. It is possible that bilinguals’ less frequent exposure to structurally complex syllables in English (e.g. complex codas) compared to that of English monolinguals resulted in their delayed acquisition of structural complexity. However, it is difficult to determine whether bilinguals produced different complex syllable types with different levels of accuracy, since structural analysis addressed consonant clusters in general across word-positions. In terms of syllable-level errors, Gildersleeve-Neumann et al. (2008) reported that whole cluster deletion was uncommon in all groups, while final consonant deletion and cluster reduction were more common. While each group improved on these measures between the beginning and end of the year, there was a non-significant trend suggesting that bilingual groups exhibited higher rates of these errors than did monolinguals at the beginning of the year. This indicates that, while complex syllable structures were difficult for all learners, bilinguals’ acquisition of these structures may initially have been delayed.

While Gildersleeve-Neumann et al.’s (2008) study contributes a significant amount of information to the literature on the phonological acquisition of bilinguals, there were some shortcomings in the design that hinder its generalizability, as acknowledged by the authors. First, the bilingual and monolingual participant sets were not balanced, meaning that there was much more information collected about children whose exposure was all English (N=10) or predominantly English (N=20) compared to children whose exposure to both Spanish and English was more balanced (N=3). The authors additionally noted that, because the study was focused on English production, no information was available regarding bilinguals’ phonological development in Spanish, how this compared to their own development in English, or how it compared to monolingual Spanish phonological development. Considered broadly, these findings suggest that investigations of interaction in bilingual phonological acquisition where segmental inventories and syllable type frequencies differ between languages should examine syllable structure, as well as consonant production accuracy in different syllabic positions. Taken in sum, the literature demonstrating delay in bilinguals’ acquisition of phonology suggests that further investigations should focus on cases where languages do not share a linguistic property (e.g. Fabiano-Smith & Goldstein, 2010 on delayed acquisition of unshared sounds; Kehoe, 2002 on delayed acquisition of the vowel length contrast in German–Spanish bilinguals), or more generally where languages use a property with different frequency (e.g. Fabiano-Smith & Goldstein, 2010 on shared and unshared sounds; Gildersleeve-Neumann et al., 2008 on syllabic structure). In either case, delayed acquisition indicated by accuracy measures has been found in cases where bilinguals receive less frequent exposure to a linguistic property compared to monolinguals due to differences in the frequency of occurrence of that property between their languages.

Acceleration

Evidence also shows that the influence of one language can promote accelerated acquisition in a bilingual’s other language. Johnson and Lancaster (1998) presented a case study of the phonological development of Andreas, a 2-year-old learning Norwegian and English. Andreas employed a larger set of sounds in his phonetic inventories compared to monolinguals in either language, including phonetically dissimilar, unshared sounds between the two languages. The authors hypothesized that Andreas developed larger, more varied phonetic inventories as a consequence of trying to distinguish his language-specific productions. Almeida, Rose & Freitas (2012) provided further evidence of acceleration in the case of complex onset structure acquisition by Barbara, a Portuguese–French bilingual child. Barbara’s acquisition of complex onset structure in Portuguese was accelerated compared to Portuguese monolinguals’, and proceeded in both languages with patterns resembling French monolinguals’ acquisition of complex onsets. Almeida et al. (2012) argued that this acceleration was due to Barbara’s exposure to positive evidence of complex onset structure in both languages. This contrasts with Barbara’s typical acquisition of word-medial codas in Portuguese coupled with her delayed acquisition of word-medial codas in French, which the authors argued was at least partially the result of her exposure to comparatively extreme segmental restrictions on this position in Portuguese. However, in each case it is difficult to generalize from the results of case studies that examine the productions of individual children.

Also regarding acquisition of syllable structure, Lleó et al. (2003) investigated bilingual and monolingual acquisition of singleton codas in Spanish and German, measuring structurally accurate coda production rates in the speech of 3 German monolinguals, 3 Spanish monolinguals, and 5 German–Spanish bilinguals in both languages. Structurally accurate coda production referred to producing a coda if the target syllable had a coda, regardless of segmental accuracy. Data (naturalistic speech samples) were collected longitudinally, from the onset of word production (about 1 year of age) until age three. Bilinguals were born in Hamburg, had Spanish-speaking mothers, and were simultaneously acquiring German and Spanish. German and Spanish both allow closed syllables, but codas are more frequent in German (67% of syllables) than Spanish (26.7% of syllables) (Meinhold & Stock, 1980, reported in Lleó et al., 2003, p. 193). German also allows a greater variety of segments in coda position compared to Spanish, which is more restrictive, and allows only some coronal codas word-finally (/θ, ð, n, s, l, ɾ/ for Castilian Spanish, Lleó et al., 2003) and labial and dorsal codas in addition to coronals word-medially. Furthermore, Harris (1983, pp. 17–18) notes that obstruent codas other than /s/ are infrequent in Spanish compared to sonorant codas. German, by contrast, beyond allowing a wide variety of segments in singleton coda, allows coda clusters of multiple segments, which are also highly varied. Coda clusters ending in /s/ are possible in Spanish, but are very infrequent, appear primarily word-internally, and are frequently reduced (Harris, 1983). Singleton coda input to monolingual acquirers of German is therefore more frequent and more varied than singleton coda input to monolingual acquirers of Spanish.

Since frequency of exposure to syllable types influences the rate of acquisition of those syllable types (Kirk & Demuth, 2003; Levelt, Schiller, & Levelt, 2000), we should expect German monolinguals to exhibit higher coda structure production accuracy earlier than Spanish monolinguals. Indeed, this is what Lleó et al. (2003) found; at all time-points, German monolinguals produced a greater proportion of target codas than did Spanish monolinguals. By the end of the second year, the German monolinguals produced on average almost 90% of target codas whereas the production rate for Spanish monolinguals on average was still less than 50%. Lleó et al. (2003) further hypothesized that the difference in frequency of exposure to codas in bilinguals’ input compared to Spanish and German monolinguals’ input could result in acceleration in bilinguals’ acquisition of Spanish codas, delay in bilinguals’ acquisition of German codas, or both. They found that bilinguals produced codas in German at similar proportions to monolingual German speakers, suggesting that bilinguals’ acquisition of German codas was not delayed. However, the percentage of Spanish target codas produced by bilinguals substantially exceeded that of Spanish monolinguals at all points. Bilinguals’ acquisition of Spanish codas was therefore accelerated; bilinguals’ exposure to the greater frequency of occurrence of codas in German promoted their acquisition of codas in Spanish.

Furthermore, bilinguals produced more codas in final stressed syllables in both Spanish and German, similar to German monolinguals but unlike Spanish monolinguals who showed a weak preference for medial codas. Bilinguals also first produced nasal, liquid, and obstruent codas in both languages, unlike Spanish monolingual participants, who tended to produce coda glides first. If it is the case that bilinguals tended to produce more varied codas in Spanish earlier than monolinguals, this could also be seen as a kind of acceleration. This result suggests that bilinguals exposed to languages with differing levels of segmental restrictiveness in a given position might experience accelerated acquisition of different types of segments in this position in the more restrictive language (e.g. Spanish). As noted above, German places fewer segmental restrictions on codas, allowing many different segments to occur in this position, whereas the set of permissible coda segments is more restricted in Spanish. However, as Almeida et al. (2012) observed in the case of Barbara, bilinguals’ acquisition of different types of segments in a given position might also be delayed in the language with fewer segmental restrictions in that position (for example, word-medial codas in French).

Overall, these results suggest that studies investigating acceleration in bilingual phonological acquisition should consider at least two scenarios. First, research should examine bilinguals’ acquisition of languages that use similar properties with different frequency (for example, acquisition of codas in a language where codas occur frequently and in a language where codas occur infrequently). Second, research should examine acquisition of languages with different positional phonotactics (that is, languages with different restrictions on the segments allowed to occur in a given syllabic position).

Current study

The literature discussed above suggests that cross-language differences in systemic frequency of occurrence of linguistic properties are a source of acceleration and delay in bilingual phonological acquisition. Aside from frequency, linguistic complexity, which corresponds to typological markedness (Gierut, 2001), may also influence interaction. Exposure to linguistic complexity in the input has been argued to promote monolinguals’ acquisition of syntactic (Wexler, 1982) and phonological structure (Dinnsen & Elbert, 1984; Gierut, 1999, 2001, 2007; Geirut, Morrisette, Hughes, & Rowland, 1996; Tyler & Figurski, 1994). In terms of syllable structure in particular, Gierut (1999) found that children with delayed onset cluster production exhibited enhanced learning when treated on more linguistically complex, marked clusters (those with small sonority differences: Clements, 1990; Davis, 1990; Steriade, 1982) compared to those treated on less marked clusters (those with larger sonority differences).

Hsin (2012) provided evidence that more exposure to syntactic complexity in the C-domain in Spanish not only supported earlier acquisition of complementizer phrase structure in Spanish monolinguals compared to English monolinguals, but that Spanish–English bilinguals’ exposure to this structural complexity in Spanish promoted their accelerated acquisition of English wh-questions compared to monolinguals. Additionally, Almeida et al. (2012) argued that Barbara’s exposure to complex onsets in both French and Portuguese resulted in accelerated acquisition of this structure in Portuguese compared to Portuguese monolinguals. Furthermore, acceleration has also been found in English–Polish bilingual 7- to 8-year-olds’ acquisition of word-initial English s + obstruent onset clusters based on the results of a non-word repetition task (Tamburelli, Sanoudaki, Jones, & Sowinska, 2015). While s + obstruent consonant clusters are more frequent both word-initially and word-medially in Polish than in English, bilinguals were not more accurate than monolinguals in their repetition of non-words with word-medial s + obstruent clusters. Bilinguals did, however, achieve higher accuracy than monolinguals on their productions of word-initial s + obstruent clusters. Instead of a frequency-based explanation, Tamburelli et al. (2015) argued that bilinguals’ acquisition of word-initial clusters was accelerated due to greater complexity in Polish onset clusters. Polish allows sonority plateaus in onset clusters, whereas English does not. While Polish and English both allow smaller sonority falls in onset clusters (e.g. /sp/), only Polish allows onset clusters with larger sonority falls (e.g. /pt/, /mʃ/), which are more marked (Berent, Steriade, Lennertz, & Vaknin, 2007). English–Polish bilinguals therefore have exposure to greater onset cluster complexity compared to English monolinguals, which supported bilinguals’ acquisition of less marked English onset clusters.

Conversely, Lleó and Cortés (2013) argued that bilinguals may experience delayed acquisition of marked linguistic properties. Like Kehoe (2002), they found that German–Spanish bilinguals’ acquisition of long vowels in German was delayed. The Spanish vowel system does not use a length contrast. As a result, bilinguals are exposed to a vowel system that does not use this dimension of contrast, and are exposed to the vowel length contrast less frequently than are German monolinguals. It is possible that bilinguals’ delayed acquisition of German long vowels arose in part from their less frequent exposure overall to the vowel length contrast, or from the lack of overlap between German and Spanish vowel systems in this respect, rather than from the cross-linguistically marked status of contrastive vowel length. Indeed, Lleó and Cortés (2013) also found that bilinguals exhibited accelerated acquisition of singleton codas in Spanish. Like contrastive vowel length, singleton codas are linguistically marked. Unlike contrastive vowel length, however, singleton codas occur in both languages. Like Lleó et al. (2003), Lleó and Cortés (2013) argue that bilinguals’ accelerated acquisition of singleton codas in Spanish is due to the high frequency of occurrence of this structure in German.

Here, we predict that cross-language differences in both frequency and complexity can influence the appearance of interaction in bilingual phonological acquisition. We test the following hypotheses:

In other words, we expect the frequency of occurrence of a property across bilinguals’ languages to affect their acquisition of that property in each language. We expect interaction to appear as acceleration and delay compared to monolingual acquisition, since cross-language differences in systemic frequency of a property will provide bilinguals with greater overall exposure to that property compared to monolinguals learning one language, and less overall exposure to that property compared to monolinguals learning the other language. Similarly, if a bilingual is exposed to a kind of linguistic complexity in one language, we expect them to benefit from that exposure when acquiring similar structures in both languages. Bilinguals should therefore show accelerated acquisition compared to monolinguals in the language that does not employ the kind of linguistic complexity in question.

In the study presented here, we test these hypotheses for Spanish and English monolinguals’ and bilinguals’ productions of singleton codas and onset clusters in each language. English uses singleton codas with greater frequency compared to Spanish. Onset clusters occur with similar frequency in the two languages, but each language allows different kinds of complexity to occur in this syllable position. We follow Fabiano-Smith and Goldstein (2010), Gildersleeve-Neumann et al. (2008), Kehoe (2002), and Lleó et al. (2003), among others, in treating cross-sectional production accuracy rates as indicative of acceleration or delay. Lower accuracy rates in bilinguals’ productions compared to monolinguals’ productions indicate delayed acquisition of the property being measured, whereas higher accuracy rates in bilinguals’ compared to monolinguals’ productions indicate acceleration.

The frequencies with which codas and onset clusters occur in Spanish and English are listed in Table 1. These frequencies were calculated using written corpora (Mexican Spanish: Guffey, 2002; English: Delattre & Olsen, 1969) meaning they strictly represent neither syllable type frequencies in input to children in general, nor to the participants in this study in particular. However, they are approximations of frequency of occurrence of syllable types in the ambient language, and are therefore used as a general guide for considering differences in input to children between the two languages (see also Gierut & Dale, 2007; Storkel & Hoover, 2010 on the compatibility of frequency values in adult and child lexical corpora ² ).

OPEN IN VIEWER

Table 1.

Syllable type frequency (by tokens); upper case Cs indicate structures relevant to the current study.

Structure	Spanish (Guffey, 2002)	English (Delattre & Olsen, 1969)
All codas (total %)	31.6%	60.3%
Singleton codas: (onset)vC	30.8%	48.7%
Complex codas: (onset)vCC(C)	0.8%	11.6%
All complex onsets (total %)	13.8%	10.5%
2-element complex onsets: CCv(coda)	13.8%	10.3%
3-element complex onsets: CCCv(coda)	0%	0.2%

While singleton codas make up almost a third of syllabic input in Spanish, nearly half of the syllabic input in English uses singleton codas. Extending consideration to syllables with codas in general does not greatly change the proportion of codas in the input in Spanish, since complex codas are infrequent, but coda input in English increases to almost two thirds of total syllable input. Comparatively, the proportions of syllables with onset clusters are similar between Spanish and English.

The frequency of occurrence of a phonological property in the input can influence how early monolinguals acquire it (Levelt et al., 2000; Stites, Demuth, & Kirk, 2004). Consequently, we expect English monolinguals to exhibit higher average accuracy rates than age-matched Spanish monolinguals on productions of singleton codas during acquisition. If cross-language syllable type statistics are a source of interaction in bilingual phonological acquisition, we should expect bilinguals to show accelerated acquisition of singleton codas in Spanish (where bilinguals have more overall exposure to codas than monolinguals), and delayed acquisition of codas in English (where bilinguals have less overall exposure to codas than monolinguals). Given the small difference between languages in the frequency of occurrence of onset clusters, we do not expect frequency to affect interaction in bilingual acquisition of this structure in either language.

Turning to complexity, English and Spanish differ in both codas and onset clusters. English allows greater structural complexity in syllable codas than Spanish does. While Spanish does allow some 2-element complex codas, they are rare, and significantly restricted in terms of permissible segments (Harris, 1983; Trapman, 2007), whereas English allows morphologically simple coda clusters of up to 3 consonants in length (and up to 4 consonants in morphologically complex coda clusters), with many possible segmental combinations (Kreidler, 1989). If exposure to complex linguistic structure facilitates learning, then we should expect to see accelerated acquisition of Spanish singleton codas by bilinguals, since bilinguals are exposed to greater structural complexity in English codas.

Regarding onset clusters, both Spanish and English phonologies exhibit complexity, but do so in different ways. English onset clusters may have two elements or three (always s-initial). Most two-element clusters allow a liquid or glide closest to the vowel, with a stop or voiceless fricative in initial position, as in Figure 1a. However, English also allows s-initial clusters (Figure 1b), which behave differently from other clusters. S-initial clusters may be followed by nasals or voiceless stops, or by most two-element clusters that start with a voiceless stop. Much evidence suggests that /s/ in sC(C) sequences is an adjunct (Davis, 1990; 1992; Giegerich, 1992; Kenstowicz, 1994) or appendix (Selkirk, 1982). In English, s-initial clusters pattern differently from true clusters in acquisition and in the treatment of phonological delays (Barlow, 2001, 2004; Gierut, 1999) further supporting their treatment as adjuncts. S-adjuncts creating clusters with three elements (e.g. in /spɹɪnt/) increase the complexity of onset structure in English. The effects of exposure to this complexity in acquisition are evident in studies such as that of Gierut and Champion (2001), which showed that a child with a speech sound disorder trained on a 3-element cluster (/spl-/) exhibited generalization learning of 2-element true and adjunct onset clusters.

OPEN IN VIEWER

Figure 1.

Two- and three-element onset clusters.

Like English, Spanish also allows two-element clusters with a liquid closest to the vowel and a stop or voiceless fricative in initial position, corresponding to large, rising sonority differences between the two consonants. However, Spanish allows greater complexity in terms of sonority differences between onset cluster consonants. We follow Baković (1994), Barlow (2003a), and Danesi (1982), among others, in treating [b d ɡ] as allophones of underlying approximants /β ð ɣ/. Aside from voiceless stop- and f-initial clusters, Spanish allows approximant-liquid clusters (/βl/ /βɾ/ /ðɾ/ /ɣl/ /ɣɾ/), where the sonority difference between approximants /β ð ɣ/ and a following liquid is smaller than the sonority difference between obstruents and liquids, such as /pl/ or /fl/ (Baković, 1994; Parker, 2002 ³ ). Onset clusters with larger sonority differences are implicationally less marked than onset clusters with smaller sonority differences, both cross-linguistically (Davis, 1990; Steriade, 1982) and in acquisition (Gierut, 1999). In other words, a system that allows onset clusters with smaller sonority differences implies the existence of onset clusters with larger sonority differences in the same system. Literature on the treatment of speech sound disorders has furthermore shown that training on onset clusters with smaller sonority differences results in generalization to onset clusters with larger sonority differences, but not the reverse (Anderson, 2002; Gierut, 1999), as expected by the complexity approach to the treatment of speech sound disorders (Gierut, 2001, 2007, and references therein). While English phonotactics allow onset clusters with greater structural complexity, Spanish phonotactics allow onset clusters with more marked (i.e. smaller) sonority differences. Because bilinguals are exposed to an additional source of increased onset cluster complexity across their languages, we expect them to show accelerated acquisition of onset clusters compared to monolinguals in each language.

Our specific predictions are summarized in Table 2, following from our discussion of cross-language differences in frequency and complexity. We expect bilinguals to exhibit accelerated acquisition of singleton codas in Spanish, due to their more frequent exposure to singleton codas compared to Spanish monolinguals and their exposure to greater coda structure complexity in English. Conversely, we expect bilinguals’ acquisition of singleton codas in English to be delayed, due to their less frequent exposure to singleton codas across their input compared to English monolinguals. Finally, we predict that bilinguals will exhibit accelerated acquisition of onset clusters in each language due to their exposure to different kinds of onset cluster complexity in Spanish and English, even though onset clusters are similarly frequent in each language. While monolinguals learning these languages have exposure to either increased structural complexity in English or increased sonority-based complexity in Spanish, bilinguals have exposure to both.

OPEN IN VIEWER

Table 2.

Predictions for bilingual versus monolingual acquisition separated by language and structure.

	Singleton codas	Onset clusters
Spanish	Bilingual > Monolingual (Acceleration)	Bilingual > Monolingual (Acceleration)
English	Monolingual > Bilingual (Delay)	Bilingual > Monolingual (Acceleration)

Beyond structural accuracy, we also evaluate positional segmental accuracy. Because bilinguals have less frequent exposure than monolinguals to each particular segment produced in codas or onset clusters in either language, it is possible that their segmental accuracy in these positions may be delayed apart from their acquisition of the syllabic structure more generally. In other words, it is possible that exposure to syllable structure across languages supports acquisition of that structure in each language, but that this same exposure may not enable bilinguals to initially “keep up with” monolinguals in terms of the positional acquisition of individual segments. This may be especially true in the case of English singleton codas. While English allows most consonants to appear in singleton coda (Kreidler, 1989), Spanish only allows a small set of segments in this position, which is even more restricted word-finally (Harris, 1983). Since there are many individual segments to master in singleton coda in English and fewer in Spanish, and because bilinguals have less frequent exposure to coda types and tokens in each language compared to monolinguals, we expect delay to occur in terms of bilinguals’ lower segmental accuracy in coda, apart from or in addition to delayed acquisition of English coda structure. Performing separate analyses to evaluate structural and segmental accuracy for each position will allow us to determine whether this is the case.

Participants

Data were collected from 27 child participants, including 12 monolingual speakers of English (mean age (standard deviation (SD)): 3;05 (9.3 months), range: 2;05–4;10), five monolingual speakers of Spanish (mean age (SD): 3;02 (9.4 months), range: 2;04–4;02), and 10 bilingual English–Spanish speakers (mean age (SD): 3;10 (9.0 months), range: 2;01–4;08). Independent one-way analyses of variance revealed no significant differences in age between bilinguals and English monolinguals, F(1,20)=1.431, p=0.246, or between bilinguals and Spanish monolinguals, F(1,13)=2.55, p=0.134. Data from 15 of the 27 participants, including MLE01-05, MLS01-05, and BL01-05, were drawn from the archives of a larger study of the phonological acquisition of Spanish and English by monolingual and bilingual children in the Southern California and Baja California areas. Further data were prospectively collected from seven monolingual English speakers (MLE06-12) and five bilingual speakers (BL06-10) living in the same geographical region. All were determined to be typically developing with normal hearing and normal linguistic, cognitive, and motoric development based on parents’ responses to a child history questionnaire evaluating their development, language input, and language output. Table 3 gives demographic information for monolingual participants. Demographic information for bilingual participants is presented in Table 4.

OPEN IN VIEWER

Table 3.

Demographic data for monolingual participants.

Participant ID	Age (years; months)	Gender	Language
MLE01	2;05	Male	English
MLE02	2;05	Female	English
MLE03	4;01	Male	English
MLE04	4;02	Female	English
MLE05	2;08	Male	English
MLE06	3;01	Male	English
MLE07	4;10	Male	English
MLE08	3;02	Male	English
MLE09	3;05	Female	English
MLE10	3;07	Female	English
MLE11	3;04	Female	English
MLE12	3;06	Female	English
MLS01	2;04	Female	Spanish
MLS02	4;00	Female	Spanish
MLS03	2;09	Female	Spanish
MLS04	4;02	Female	Spanish
MLS05	2;11	Female	Spanish

OPEN IN VIEWER

Table 4.

Demographic data for bilingual participants.

Participant ID	Age (years; months)	Gender	Input (%)		Output (%)
			English	Spanish	English	Spanish
BL01	2;01	Female	80	20	90	10
BL02	3;06	Female	27	73	10	90
BL03	4;06	Male	33	66	33	66
BL04	3;06	Male	33	66	33	66
BL05	4;07	Male	46	46	54	54
BL06	3;11	Male	60	40	60	40
BL07	4;08	Female	40	60	25	75
BL08	3;11	Female	60	40	50	50
BL09	3;11	Female	50	50	50	50
BL10	3;04	Female	20	80	20	80

Bilingual status was decided based on results from an extensive questionnaire, evaluating the child’s language development, input, and output (following Gutiérrez-Clellen & Kreiter, 2003; Pearson, Fernandez, Ledeweg & Oller, 1997; Restrepo, 1998). Children classified as “bilingual”, had a minimum of 20% input in both English and Spanish following findings from Pearson et al. (1997) showing that at least 20% exposure was required for bilinguals to readily produce utterances in the target language. Language input and output percentages were based on parent report. Furthermore, bilingual participants were able to interact with experimenters in each language and to perform both Spanish and English versions of the picture-naming task. We use the term “early bilingual” to categorize the bilingual participants in this study, given that all started acquiring their second language before their first language was fully established (before the age of 5 or 6 years, following Flege, 2007; Flege,Yeni-Komshian, & Liu, 1999; Hamers & Blanc, 2000; McLaughlin, 1978).

Data

Data were transcriptions of participants’ productions of target words with singleton codas or onset clusters, elicited using the Assessment of English Phonology (AEP: Barlow, 2003b) or the Shorter Protocol for the Evaluation of English Phonotactics (Little PEEP: Barlow, 2012), and/or the Assessment of Spanish Phonology (ASP: Barlow, 2003c). These assessments are single-word phonological probes targeting all phonemes of Spanish (the ASP) and English (the AEP and Little PEEP) in all permitted syllable positions. Productions were obtained in isolation in the corresponding language using non-imitation elicitation via a picture-naming task (e.g. “What’s this? It’s a…”/ “¿Qué es esta? Es una…”) with delayed imitation when necessary (e.g. “It’s a flower. What is it? It’s a…” / “Es una flor. ¿Qué es? Es una…”).

The AEP targets 256 words, and provided participants with 77 opportunities to produce onset clusters and 187 opportunities to produce singleton codas in English. The Little PEEP targets 285 words, including 122 onset cluster production opportunities and 229 singleton coda production opportunities. In both English probes, most onset clusters are word-initial and most singleton codas are word-final, while some words contain word-medial onset clusters or singleton codas. The ASP targets 156 words, and provided participants with 36 opportunities to produce onset clusters and 96 opportunities to produce singleton codas in Spanish. Onset clusters in words targeted by the ASP were word-initial or word-medial, and singleton codas were word-final or word-medial. Not all participants attempted to produce all targets of a given probe (though inclusion of a participant as a random effect helped to control for these differences during statistical analysis). These differences were due to constraints on participant availability and attention.

Productions were phonetically transcribed by judges trained in the use of narrow transcription with the International Phonetic Alphabet. Judges were native speakers of English and/or Spanish. English productions were transcribed by native English speakers and Spanish productions were transcribed by native Spanish speakers. Twenty percent of the data were re-transcribed by a second judge for calculation of transcription reliability, with point-to-point inter-judge reliability for each target word at 87% for English and 82% for Spanish.

Analyses

We performed two kinds of accuracy analysis on the cross-sectional data: analysis of structural accuracy; and analysis of positional segmental accuracy. Whereas analysis of structural accuracy counted consonant substitutions as correct (similar to the analysis in Lleó et al., 2003), analysis of positional segmental accuracy did not. Consonant deletions were counted as incorrect in both analyses. The purpose of the structural accuracy analysis was to evaluate participants’ accuracy in producing singleton coda and onset cluster structure, without considering their accuracy in producing the sets of segments allowed in these syllabic positions. The purpose of the positional segmental analysis was to evaluate participants’ accuracy in terms of the segmental phonotactics of singleton codas and onset clusters.

Structural accuracy for each target production for each participant was calculated such that any consonant production in coda for a singleton coda target or any consonant cluster production for an onset cluster target (matching the number of cluster segments present in the target) counted as a hit, regardless of segmental accuracy. Unintelligible productions and unattempted targets were not counted as attempts. Accuracy percentages for each structure were based on the number of successful attempts to produce the structure divided by the total number of attempts to produce the structure (total number of hits and misses). For example, if a child attempted to produce leaf, spill, nose, hug, drum, and glass and produced outputs of [lif], [pɪoʊ], [no], [hʌd] [dʒɹʌm] and nothing for glass, she would have a mean accuracy percentage of 60% for singleton codas (3 hits/5 attempts) and a mean accuracy percentage of 50% for onset clusters (1 hit/2 attempts). This structural analysis is similar to what Lleó et al. (2003) performed in their examination of bilingual and monolingual acquisition of singleton codas in Spanish and German.

Segmental accuracy for each target production was calculated such that a faithful production of the coda or onset was recorded as a hit, whereas consonant deletions or substitutions were recorded as misses. Again, unintelligible productions and unproduced targets were not included as attempts. Accuracy percentages for each structure were based on the number of successful attempts or hits for the structure divided by the total number of attempts for the structure (total number of hits and misses). To compare these measures let us consider the same example used above. Given the child’s productions of leaf, spill, nose, hug, drum, and glass as [lif], [pɪoʊ], [no], [hʌd] [dʒɹʌm] and nothing for glass, she would have a mean accuracy percentage of 40% for singleton codas (2 hits/5 attempts) and a mean accuracy percentage of 50% for onset clusters (1 hit/2 attempts).

Singleton codas

Descriptive statistics for all singleton coda analyses are presented in Table 5. Mixed logit models found that bilinguals’ productions of Spanish singleton codas were more accurate than monolinguals’ Spanish singleton coda productions in both the structural (β = 1.4448, standard error (s.e.) = 0.6214; χ2(1) = 4.5904, p < 0.05) and segmental (β = 1.5989, s.e. = 0.6322; χ2(1) = 5.3327, p < 0.05) analyses, as predicted. Results for English singleton coda production accuracy did not indicate differences between the productions of bilinguals and monolinguals. Figure 2 displays Spanish singleton coda accuracy score means for participants grouped by background (bilinguals versus monolinguals) for structural (a) and segmental (b) analyses. Figure 3 presents these values for English singleton coda productions.

OPEN IN VIEWER

Table 5.

Descriptive statistics for singleton coda analyses.

		Spanish		English
		Bilingual	Monolingual	Bilingual	Monolingual
Structural analysis	Mean	82.5%	62.5%	85.8%	91.7%
	Standard deviation	16.4%	22.3%	10.5%	9.7%
Segmental analysis	Mean	79.6%	55.4%	71.9%	77.2%
	Standard deviation	17.5%	23.8%	15.2%	17.2%

OPEN IN VIEWER

Figure 2.

Spanish singleton codas.

OPEN IN VIEWER

Figure 3.

English singleton codas.

A numerical trend in the predicted direction was found in the structural analysis suggesting that bilinguals may have produced English singleton coda structure less accurately than monolinguals, however this trend failed to reach statistical significance (β = −0.7888, s.e. = 0.5026; χ2(1) = 2.3463, p = 0.1256). Similarly, no difference was indicated between bilinguals and monolinguals in the segmental analysis of English singleton coda productions (β = −0.4702, s.e. = 0.5206; χ2(1) = 0.7996, p = 0.3712).

Onset clusters

We now turn to analyses of onset cluster production. Descriptive statistics for all onset cluster analyses are presented in Table 6. Results indicated that, as predicted, bilinguals were more accurate than monolinguals in their structural (β = 2.3708, s.e. = 1.0344; χ2(1) = 4.3505, p < 0.05) and segmental (β = 2.7911, s.e. = 1.1590; χ2(1) = 4.8148, p < 0.05) productions of Spanish onset clusters. In English, structural and segmental analyses obtained different results. No significant difference in accuracy was found between bilinguals’ and monolinguals’ productions of English onset cluster structure (β = 1.0956, s.e. = 0.8850; χ2(1) = 1.4589, p = 0.2271). However, bilinguals’ productions of English onset cluster segments were found to be more accurate than monolinguals’ productions (β = 1.5431, s.e. = 0.6186; χ2(1) = 5.4825, p < 0.05). Participants’ production accuracy score means grouped by background are presented in Figures 4 and 5 for Spanish and English onset clusters, respectively, and Figures 4 and 5 (a) display data from the structural analyses while Figures 4 and 5 (b) present data from the segmental analysis.

OPEN IN VIEWER

Table 6.

Descriptive statistics for onset cluster analyses.

		Spanish		English
		Bilingual	Monolingual	Bilingual	Monolingual
Structural analysis	Mean	70.1%	37.0%	86.1%	72.3%
	Standard deviation	33.4%	34.7%	14.0%	30.9%
Segmental analysis	Mean	64.9%	30.1%	82.0%	57.8%
	Standard deviation	31.6%	32.2%	14.2%	28.6%

OPEN IN VIEWER

Figure 4.

Spanish onset clusters.

OPEN IN VIEWER

Figure 5.

English onset clusters.

In sum, statistical analyses indicated that bilinguals were more accurate than monolinguals in their productions of Spanish singleton codas, for both structural and segmental measures. Bilinguals were also more accurate than monolinguals in their productions of Spanish onset cluster structure, and in their productions of Spanish and English onset cluster segments. Numerical differences in the predicted direction were found between bilinguals’ and monolinguals’ production accuracy rates for English singleton coda structure, but these differences failed to reach significance. Similarly, no differences were found between bilinguals and monolinguals in terms of their production accuracy rates for English singleton coda segments or onset cluster structure.