Allophony in English Language Learners: The Case of Tap in English and Spanish

Abstract

Purpose

The objective of this study was to examine tap production by English language learners (ELLs) in kindergarten whose 1st language is Spanish. The conflicting status of tap in Spanish and English could present challenges for allophonic learning in 2nd language for ELLs. Prior research has evaluated acquisition of other allophone pairs, but none has focused exclusively on tap.

Method

Thirty ELLs, 30 English monolinguals, and 29 Spanish monolinguals participated in the study. Participants completed a single-word repetition task in which numerous opportunities to produce tap were provided. Productions were phonetically transcribed and analyzed.

Results

The great majority of taps were pronounced correctly in both languages. The allophonic status of tap in English and phonemic status in Spanish suggest ELLs could experience negative transfer in learning; however, this was not observed. A significant interaction indicated more t/d substitutions in English and more semivowel/liquid substitutions in Spanish, contradicting the expectation of negative transfer. ELLs were also significantly more accurate at producing tap in Spanish than English.

Conclusion

Findings suggest that, at early kindergarten, ELL children rapidly adapted to English patterns of tap production even though Spanish and English conflict in phonemic/allophonic status of tap. This study was preliminary, and further investigation is warranted.

As of 2017, data from the U.S. Census indicated over 57.5 million Hispanics living in the United States (U.S. Census Bureau, 2017), accounting for 18% of the total U.S. population. In addition, more than 37.5 million Hispanics of ages 5 years and above reported speaking Spanish at home with slightly less (35 million) indicating proficiency in English (Flores, 2017), suggesting that varying degrees of Spanish–English bilingualism are extremely common in the United States and its schools.

The American Community Survey revealed that Hispanic children accounted for over 25% of the total enrollment in U.S. kindergartens in 2016 (U.S. Census Bureau, 2016). Such a large number of Hispanics in schools, a substantial proportion of whom begin kindergarten with Spanish as their only or their primary language, provides an opportunity for scientific investigation of early second language (L2) learning. It also allows for study of bilingual language development in U.S. school children, so that researchers and teachers alike can develop a greater understanding of the potential challenges faced by young bilingual learners. Although bilingualism is common for much of the world's population, most U.S. schools continue to provide English-only instruction and testing despite the growing population of English language learners (ELLs).

Phonological Development in Spanish–English Bilinguals

The study of phonological development in young Spanish–English bilinguals has revealed evidence of both positive and negative transfer. Positive transfer has been said to occur when elements acquired in first language (L1) are similar enough to those of L2 to be utilized in L2 with little or no modification. In contrast, negative transfer or “interference” is said to occur when elements of L1 are not sufficiently similar to transfer and slow down the process of acquisition of L2 elements (Gass & Selinker, 1993; Kroll & Stewart, 1994; MacWhinney, 1997). Durgunoglu, Nagy, and Hancin-Bhatt (1993) and Cobo-Lewis et al. (2002) found evidence of positive transfer in phonological awareness across the languages of bilingual children. Such evidence is consistent with the suggestion that children with a Spanish language background often adapt quickly to English phonology. Fabiano-Smith and Goldstein (2010) found that Spanish–English bilinguals, when tested in English, achieved overall consonant accuracy that was not significantly different from that of monolingual English speakers. However, their productions in Spanish seemed to be negatively impacted by their acquisition of English phonology as overall consonant accuracy was significantly higher for the Spanish monolingual children. These findings suggest positive transfer in the bilinguals' quick adaptation to English but also provide evidence for decreased performance in their native language as compared with monolingual peers.

Gildersleeve-Neumann, Kester, Davis, and Peña (2008), unlike Fabiano-Smith and Goldstein, found evidence of possible negative transfer in the bilingual population they studied. Their study of 3-year-old Spanish–English speaking children found lower rates of speech intelligibility as compared with English monolingual peers, which suggests that these children had more difficulty in navigating the differences between English and Spanish. The bilingual children in their study produced more consonant and vowel errors overall and exhibited more uncommon phonological patterns than their monolingual peers in English. However, the children in the study of Gildersleeve-Neumann et al. (2008) were younger than the children in the study of Fabiano-Smith and Goldstein (2010), so it is possible that age may have contributed to the observed differences.

The studies reviewed above have offered insight into how young bilingual children acquire phonology in comparison to monolingual peers; however, as in a variety of monolingual investigations of phonology, their sole focus has been the production of phonemes. Languages also include allophones, which are context-specific variants of phonemes. Command of these sounds is part of what makes speaker sound “native” in a given dialect or language.

The notions of positive and negative transfer have heuristic merit but, by themselves, represent an oversimplification of the possible interactions between two languages in a bilingual speaker or learner. While recognizing this theoretical limitation, we address a way that the notion of transfer may generate a useful hypothesis to be evaluated about possible problems of learning to correctly produce Spanish and English taps. MacWhinney's (1997) competition model may provide an enhancement in interpreting the learning of differences in taps across languages. This theoretical framework suggests that, for bilingual children, the representations of sounds for each of the two languages are learned as elements of each compete and interact. Both the idea of transfer and that of competition can help inform our interpretation of how young children who are acquiring Spanish and English establish underlying representations for a sound that is phonemic in one language but subject to allophonic variation in the other.

Phonological and Allophonic Differences Between English and Spanish

In spite of the fact that both English and Spanish are Indo-European, their phonologies are sharply distinct, suggesting that L2 learning of English phonology by L1 Spanish-speaking children could be a major task. One aspect of the difference is that English phonology involves substantially more contrastive elements than Spanish. The phonemic vowel inventory of English is more than twice as large as that of Spanish, and the English phonemic consonant inventory is 25% larger than the Spanish phonemic consonant inventory (Stockwell & Bowen, 1965). Adding to its complexity, English phonotactics allows many syllables with final consonants, and its consonant clusters occur in a wide variety of contexts, whereas in Spanish, syllable-final consonants and consonant clusters are quite constrained and comparatively rare in the lexicon. This reasoning does not imply that Spanish is somehow more limited than English in how its phonology can serve to transmit lexical and syntactic material.

These comparative phonological facts are well known, but it is less well recognized that English may also present a challenge to Spanish-speaking ELLs because the allophonic variations of the two languages are also considerably different. Allophony is the property of phonologies by which phonemic elements are phonetically realized differently in different environments. The phonological environment in which a specific phoneme occurs governs the use of a particular allophone. English has extensive allophonic variation, and many English phonemes, including consonants, vowels, and diphthongs, have multiple allophones. English phonemic vowels also vary dramatically under the influence of lexical stress changes, with many appearing as schwa under low stress. The first vowel in parental differs from the first vowel in parent. Similarly, in top the /t/ phoneme is phonetically realized as [t^h], whereas the same phoneme is produced as [t] in faulty and can occur as [ʔ] in mountain. These examples illustrate only a few of the many allophonic variants corresponding to the phonemes of English.

Spanish allophony is comparatively simple in that phonemes tend to have a more limited set of allophones than do English phonemes. One sign of the more limited allophony in Spanish is that the five-vowel system is crisply implemented as five phonetic elements across a wide variety of contexts, whereas English vowels shift dramatically under differing degrees of stress. The more limited types of allophonic variation in Spanish can also be exemplified by relatively few notable changes in consonant phonemes across environments. Particularly notable examples of change are the result of spirantization in which the phoneme /ɡ/ is realized as [ɡ] in gozar (to enjoy), but as the spirant allophone [ɣ] in intervocalic position of bigote (mustache). This pattern is also evident as the phoneme /d/ is realized as [ð] intervocalically, as well, in dedos (fingers). Another example of allophonic variation comes from Mexican Spanish where /s/ becomes aspirated when it occurs in syllable final position. Also in Mexican Spanish, trill /r/ can be realized as [R] or [x] in syllable initial position.

A particularly interesting case of difference between English and Spanish concerns the alveolar “tap” sound [ɾ], which has phonemic status in Spanish and a frequency of occurrence of approximately 5% (Moreno Sandoval, Toledano, de la Torre, Garrote, & Guirao, 2008), but only allophonic status in English. In American English, in medial position, when stress occurs on the preceding syllable, /t/ or /d/ usually becomes a tap as in body [bɑɾi] or party [p^hɑɻɾi]. It is also possible for tap to occur across word boundaries in English, such as in the word “to” in “go to the beach.” In Spanish, tap is often classified as a rhotic sound and is thus treated as related to the trill [r] as in perro (dog; Olsen, 2012); however, it is produced in pero (but) by a single tap of the tongue at the alveolar ridge, much as in the case of allophonic [ɾ] in English. Unlike in English, tap occurs across a wide variety of contexts in Spanish. It occurs intervocalically as in cura (priest) or muros (walls), word finally as in comedor (dining room), and in consonant clusters as in tarde (late) or sartén (skillet). In certain dialects of Spanish, such as Puerto Rican, the tap phoneme as alternates with [l], for example, in carta (letter), pronounced [kalta]. Although not considered a dialect feature, production of [l] for tap has been observed as a substitution pattern in studies of Mexican Spanish speakers (e.g., Anderson & Smith, 1987; Barlow, 2003) to be discussed below.

Given the conflicting status of tap in English and Spanish, it is uncertain in what ways young Spanish-speaking children learning English as L2 treat the element in English and how their underlying representation may influence its production. As previously discussed, research on Spanish–English speaking children has revealed [l] substitution for target tap in Spanish (Barlow, 2003; Goldstein & Washington, 2001). This substitution seems highly plausible given the shared place of articulation, but as tap is presumed to be a rhotic element in Spanish, it seems possible that the English tap would be treated as an r-like element and not be associated with the alveolar stops, /t/ and /d/. If this association is not made, we might expect negative transfer to occur, so that one would predict L2 errors where phonemic /r/s of English words might be replaced by tap or possibly a trill, given that tap and trill can both occur intervocalically in Spanish. Also plausible is that target English taps might alternate with /r/s. And assuming such negative transfer, one would not expect words that include allophonic taps in English to be replaced by [t] or [d], because Spanish does not include allophonic alternations of tap with [t] or [d]. If this pattern were observed in early L2 learning of English, it would suggest that children may begin the task of learning L2 by attempting to apply L1 phonemics rather directly to the superficial phonetic facts presented to them by L2. In addition, it would indicate that they do not recognize the allophonic variants in L2 as alternating members of an L2 phonemic class but rather assign them to the closest L1 phonemic element based on their underlying representation in that specific language.

Alternatively, although positive and negative transfers are often treated as a dichotomous variable, perhaps transfer could be better viewed as a continuum, in which some elements of an L2 are learned faster or slower than others based on degree of similarity rather than absolute differences. For example, although there could be evidence of substitution of rhotics for taps in English based on Spanish phonemics, it may be that as children continually learn more about the intricacies of English, they refine their productions gradually across a range of possible, for example, rhotic-like and stop-like sounds. To assess such gradual changes would require highly detailed transcription or acoustic analysis that may be beyond our capabilities in current linguistic science. But in principle, treating transfer as a continuum appears to be possible, especially if the patterns mentioned above were to be observed.

Allophonic variation has been empirically examined in adults for generations (Bloomfield, 1914; Gleason, 1955) with particular attention in laboratory research to perception studies (e.g., Peperkamp, Pettinato, & Dupoux, 2003; Shea & Curtin, 2010), but an increasing number of studies have explored phonemic and allophonic challenges in both monolingual (e.g., Klein & Altman, 2002) and bilingual children (e.g., Barlow, 2003; Barlow, Branson, & Nip, 2013; Fabiano-Smith & Bunta, 2012; Fabiano-Smith, Oglivie, Maiefski, & Schertz, 2015; MacLeod & Fabiano-Smith, 2014). Although several of these investigations have explored allophonic relationships in terms of production, others have analyzed acoustic measures, such as voice onset time (VOT), to explore bilinguals' knowledge of Spanish and English and the interactions that occur between languages.

Regarding acquisition of /t, d/ allophones in English, Klein and Altman (2002) examined the productions of four English-speaking monolingual children from ages 2 to 5 years. Productions of allophones were elicited using words with either iambic or trochaic contexts. Over the 38-month study duration, the authors found that children varied in their ability to produce tap. However, they did observe that children's accurate productions of flap increased with each additional recording session, and by age of 5 years, flap was produced correctly. Jimenez (1987) examined phoneme production in Spanish monolinguals of ages 3–5 years 7 months. She found that by age 4;7 (years;months) children had reached a 90% level of acquisition for the tap phoneme. Although these studies reveal that tap production accuracy is variable early on in development, by preschool age both Spanish and English-speaking monolingual children have mastered this phoneme and allophone in their respective languages. These findings may offer insight regarding how Spanish–English bilingual children navigate allophony across both languages.

Fabiano-Smith and colleagues examined another allophonic variant by investigating the stop–spirant alternation in Spanish–English bilinguals from ages 2;4 to 8;2 and found that these children were more accurate in their productions of voiced stops [b, d, g] compared with the spirant allophones [β, ð, ɣ]. This result contrasted with their expectation that the children would have higher spirant accuracy, which would have suggested that the spirant was, in fact, the underlying phoneme. These findings further added to the research on the Spanish stop–spirant alternation conducted by Barlow (2003) but also provided preliminary information about phonemic–allophonic relationships across languages regarding underlying representations. In addition, the authors indicated that stop–spirant alternation may not be fully mastered even by 8-year-olds. They had anticipated that bilingual children would understand the stop–spirant alternation by age of 5 years. The outcome thus suggests that bilingual children may have difficulty in navigating allophony.

Barlow's (2003) original work on the stop–spirant alternation utilized acoustic analysis and production patterns to investigate the process of learning spirantization in Spanish. She suggested support of a fortition account to explain what occurs between a voiced stop and its spirant counterpart. This account assumes that the spirants are the underlying form and that their production is closer to that of an approximant than a fricative. In addition to examining stops and spirants, Barlow (2003) investigated error patterns in the Spanish productions of Spanish–English bilingual children. In four selected participants, she found that [l] was the most frequent substitute for tap and trill, which suggests that this pattern could be revealed in the current study as well. It was also observed that [l] was a frequent substitute for English /ɹ/ in the bilingual children's English productions, a pattern consistently observed in previously mentioned research. In addition, two of the bilingual children utilized [l] as a substitute for tap in English.

In a related investigation, Barlow et al. (2013) explored bilingual acquisition of a shared phoneme across Spanish and English. The researchers examined how production of /l/ was influenced by its characterization in each language. To do this, acoustic measures of F1 and F2 were gathered, and their results indicated that Spanish–English bilingual children produced prevocalic [l] with similar frequencies in English and Spanish, closely approximating the values of their Spanish monolingual peers. However, for postvocalic [l] in English, lower F2 values were observed in English than Spanish. Taken together, these findings suggest that Spanish–English bilingual children have a merged allophonic category for prevocalic /l/ in both languages, but distinct categories for postvocalic /l/ in English and Spanish, a fact that may be partly accounted for by the tendency in American English but not in Spanish for postvocalic /l/ to be dark (velarized).

Although Spanish and English share numerous phonemes, production of these phones can be influenced by various factors, including VOT. Fabiano-Smith and Bunta (2012) investigated VOT value differences in 3-year-old Spanish–English bilinguals and their monolingual peers by evaluating production of voiceless bilabial and velar stops. Although VOT is contrastive in both languages, it creates a unique challenge for Spanish–English bilinguals because its values are not the same across languages for comparable phonemic pairs such as /p, b/. The authors discovered that there were no differences in VOT values in English or Spanish for the bilingual children; however, differences were present in English between the monolingual and bilingual children. No differences were observed between these two groups in Spanish. Overall, the findings suggest that VOT may be subject to cross-linguistic influence in bilingual populations and that the phonological systems of these children may continue to be refined over time. If continued refinement is necessary for phonemes as these children did not differentiate VOT for the voiceless stops tested, it is plausible that the acquisition and knowledge of allophones across languages might require similar fine-tuning.

Finally, in a study of allophone acquisition by bilingual Spanish–English and French–English 3-year-olds, MacLeod and Fabiano-Smith (2014) suggested that structural differences across languages and less exposure to each language were likely contributors to bilingual children's challenges with establishing distinct phonological categories. If bilingual children must continuously refine their phonological categories, acquisition of allophones could be particularly difficult. By definition, allophones are constrained to specific environments and structures, which may create a challenge in phoneme categorization.

The current study seeks to add to the research base on production of allophonic variants by ELLs whose L1 is Spanish to determine if tap acquisition in English poses challenges for this population. In addition, investigation of the substitution patterns for alveolar tap and potential evidence of negative transfer from Spanish to English given the complexities of English phonology motivated our inquiry.

Purpose of Current Study

Given the differing allophonic patterns present in English and Spanish and the potential for cross-language phoneme/allophone conflict, we explored the potential interaction between the English allophone [ɾ] and the Spanish phoneme /ɾ/ in ELLs during the early months of kindergarten. Specifically, we investigated the allophony conflict presented by Spanish and English tap in young Spanish-speaking children learning English. To accomplish this goal, a comparison of tap production by ELLs was made with Spanish-speaking monolingual children and with English-speaking monolingual children. The research questions were driven by interest in whether Spanish-speaking ELLs adapt to learning English allophony given the conflicts described above, even in the first semester of kindergarten. Given that tap can occur in multiple environments in Spanish, but its use is limited in English, it seemed plausible that young ELLs could have difficulty achieving high levels of accuracy and might evidence negative transfer in their productions of English tap. Although prior studies of allophonic variation in bilingual populations exist, none has exclusively evaluated the learning of a phonetic element in L2 in which allophonic variation on that element in L2 results in a conflict with the phonemic status of the element in L1.

Our research questions included the following: (a) Are there differences in accuracy of ELLs' tap in Spanish and English? (b) Did the substitution patterns, reflecting assignment of allophones to phonemic categories, exhibited by the ELLs mirror the substitution patterns used by the monolingual children in each language? If children show similar tap accuracy in the two languages and they exhibit substitution patterns resembling those of monolinguals, it would suggest relatively rapid learning of L2 allophony, with little negative transfer from L1 and no need for explicit instruction by educators.

Method

Participants

Eighty-nine children, ranging in age from 57 to 85 months (M = 71.1 months, SD = 6.7 months) participated in this study. The participants included 30 ELLs, 30 English-speaking monolinguals, and 29 Spanish-speaking monolinguals (see Table 1). To gather demographic data, parents of the ELLs and English monolinguals completed questionnaires regarding (a) child's birth country, (b) child's age of arrival in the United States, (c) child's age of first exposure to English, (d) child's preschool attendance, (e) language of preschool instruction, and (f) mother's level of education.

Table 1.

Demographic information.

Participants	Language group
	English language learners	English monolinguals	Spanish monolinguals
Total	30	30	29
Gender	20 F, 10 M	17 F, 13 M	8 F, 21 M
Mean age	67 months	68.4 months	78.1 months
Birth country	11 Mexico, 19 USA	30 USA	29 Mexico
Mean age of first exposure to English	42.8 months (n = 23)	n/a	n/a
Mean mother education level	7.5 years	12 years	NR
Preschool attendance	9 Y, 21 N	15 Y, 13 N, 2 NR	NR
Preschool language	4 E/S, 4 E, 1 NR	English	NR

Note. F = female; M = male; Y = yes; N = no; n/a = not applicable; NR = not reported; E = English; S = Spanish.

ELLs

Children in the ELL group (M = 67 months, SD = 3.7 months) were enrolled in kindergarten at one of two public elementary schools in Memphis, Tennessee. Any child whose parent was willing to allow them to participate in a broader study of language development was accepted for the study. All the ELLs were of Hispanic heritage. Nineteen children were born in the United States, and 11 were born in Mexico with various arrival dates reported. One child each arrived at 6, 12, and 24 months old. Four children arrived at 3 years old. Three children arrived at 4 years old, and one child arrived at 5 years old.

Data for age of first exposure to English was reported for 23 ELLs. Two children were first exposed to English at birth, one at 12 months old, one at 2 years old, two at 3 years old, 10 at 4 years old, and six at 5 years old. Thus, about 70% of the ELLs for whom it was reported were first exposed to English at age of 4 years or later. For the 10 children whose first exposure to English was at 4 years old, three were born in Mexico and seven in the United States. For the six children whose first exposure to English was at 5 years old, four were born in the United States and two were born in Mexico.

Of these same 23 ELLs, nine had attended U.S. preschools. Four children attended Spanish–English bilingual programs, four children attended an English-only program, and for one child, preschool language was not reported. One child, from this subset of 23 children, was repeating kindergarten at the time of our study but was not an outlier in the data we evaluated and therefore was retained for the analysis. Socioeconomic status and dialect information were not available for these children. However, they all attended one of two neighborhood public schools in a middle- to lower middle–class neighborhood.

Monolinguals

English

The 30 English-speaking monolingual children (M = 68.4 months, SD = 4.7 months) were also enrolled in kindergarten at one of the same two public elementary schools in Memphis, Tennessee, as the ELLs. They came from English-speaking homes. Parents also completed a similar questionnaire to gather demographic information (see Table 1). Fifteen children had attended preschool. Thirteen children had no preschool experience, and for two children, preschool attendance was not reported.

Spanish

The Spanish-speaking monolingual children were enrolled in first grade at an elementary school in Guadalajara, Mexico, where kindergarten is not offered in public schools. These children were not reported to be receiving any non-Spanish instruction or input. The Spanish monolingual children (M = 78.1 months, SD = 5.2 months) were significantly older than their bilingual peers (M = 67 months, SD = 3.7 months), t(57)= 9.418, p = .000, d = 2.49; however, this was their first experience with formal schooling. Age was controlled for in comparisons presented below of the ELL and Spanish monolingual groups.

Data Collection

A single-word repetition task was administered to elicit productions from ELLs and English-speaking monolingual children during early kindergarten (October to November). The Spanish-speaking monolingual children completed the single-word repetition task in October of their first-grade year. Repetition could produce more accurate productions than spontaneous speech, but the repetition method has the advantage of affording standardization and comparability of words to evaluate across children. A female native speaker of each language recorded the single-word stimuli used for the repetition task in citation form. The recorded words were then arranged in three randomized sets for each language so that any potential order effects would be minimized. During the single-word repetition task, the recorded words were presented to each child via headphones, and the child was verbally instructed to repeat each word into a microphone. The task was conducted once each with the 30 monolingual English, 29 monolingual Spanish, and twice each (once in each language) for the 30 ELL children, for a total of 119 sessions of testing to be evaluated.

The task included 42 English words and 30 Spanish words. These words were developed to represent common words in each language of three syllables or less. They were selected to facilitate several comparisons of cross-linguistically different consonants and vowels, in which negative transfer was a possibility. For the purposes of this study, only words containing the tap phoneme in Spanish and the tap allophone in English were analyzed. There were 17 opportunities to produce tap in Spanish and 12 opportunities in English. The Appendix shows the stimulus lists.

Dialectal variations were also considered in each language, which altered the total number of tap targets for some children. For example, regarding parecer /paɾeseɾ/, the pronunciation [paɾeseʂ] was treated as an acceptable variation. Production of this variant by an ELL child reduced the total number of tap targets (because [ʂ] is not a tap). For some children, it was the case that more than one accepted dialectal variation for tap was produced in the repetition task. Thus, the total number of tap targets ranged from 12 to 17, dependent on participant.

Data Analyses

Following recording, the children's productions were narrowly transcribed in LIPP (Logical International Phonetics Programs; Delgado & Oller, 2000) using the conventions of the International Phonetic Alphabet. The transcriptions were then analyzed using the same program, which offers user-based specification of analysis strategies for the phonetic features of the International Phonetic Alphabet. The analysis was tailored to seek correct productions of tap and deletions and a wide variety of possible substitutions for tap. Specific attention was given to substitutions of [t] and [d] and semivowel [ʋ] and liquid substitutions ([l], [r] (trill), and English [ɻ]).

Narrow transcription was conducted after intensive training and several passes of reliability evaluation among transcribers. The last author, who had extensive phonetic transcription experience, is a near native speaker of Spanish, and had publications founded on that experience (Oller & Eilers, 1982; Oller & Ramsdell, 2006; Oller, Wieman, Doyle, & Ross, 1976), trained two other transcribers to a high level and trained a group of additional student transcribers through university coursework. These expert transcribers reviewed and adjusted transcriptions provided by less-experienced transcribers, who had also been trained by the last author. If either of the two expert transcribers had doubts about transcription of a word, the last author served as a consensus judge. Prior to this review, overall agreement was 86%. When each transcription was reviewed by the last author, it was done in the presence of the primary expert transcriber, and coding was adjusted accordingly.

The transcriptions for 76% of the sessions (91/119) were conducted “blind” with respect to bilingual or monolingual status of children. That is to say, each utterance was selected without knowledge of who the child was or which group he or she pertained to. Any child speaking in Spanish was known by the coders to be either a monolingual Spanish speaker or an ELL, and any child speaking English was either a monolingual English speaker or an ELL. Both the blind and nonblind transcriptions showed the patterns presented below. Consequently, all the data are collapsed in the report below.

Results

English: ELLs Versus English Monolinguals

The monolingual children achieved higher proportions of tap accuracy in English (M = 0.75, SD = 0.17) than did the ELLs in English (M = 0.58, SD = 0.23), t(58) = –3.253, p = .002, d = –0.85 (see Figure 1) by independent samples t test. No significant difference was observed between the monolinguals (M = 0.14, SD = 0.13) and ELLs (M = 0.15, SD = 0.13) on the proportion of t/d substitutions for tap, t(58) = –0.352, p = .726, d = 0.07 (see Figure 2). A significant difference, however, was observed between the monolinguals (M = 0.003, SD = 0.02) and ELLs (M = 0.06, SD = 0.10) on semivowel/liquid substitutions for tap, t(58) = 2.944, p = .005, d = 0.77 (see Figure 2). The ELLs used significantly more semivowel/liquid substitutions in English than their English monolingual peers.

Figure 1.

Production types across language groups. This figure illustrates the proportions of accurate tap, substitutions, and deletions for the language groups. ELLs = English language learners.

Figure 2.

Key substitutions across language groups. ELLs = English language learners; semiV = semivowel.

The unexpected substitutions for tap in English (semivowel/liquid) were very rare in the monolinguals (as illustrated in Figure 2), but not so rare in the ELLs. As a result, the difference between the proportion of t/d substitutions and semivowel/liquid substitutions in English was far greater for the monolinguals than for the ELLs.

Spanish: ELLs Versus Spanish Monolinguals

The same sorts of effects were evaluated in Spanish, but this required us to take account of the fact that the monolingual Spanish speakers were somewhat older than the ELLs. To control for the age difference, a one-way analysis of covariance (ANCOVA) was used to compare tap accuracy between groups. The independent variable, language status, had two levels: ELL and monolingual. The dependent variable was the proportion of accurate tap productions, and age in months was the covariate. A preliminary analysis evaluating the homogeneity-of-variance assumption indicated no violation. The ANCOVA showed that the relationship between the covariate (age) and the dependent variable (tap accuracy) did not differ significantly as a function of the independent variable (ELL vs. monolingual), F(1, 58) = 0. 37, p = .55. The ANCOVA, however, was significant, F(1, 56) = 4.702, p = 0.034, η² = .077, indicating that the Spanish monolinguals (M = 0.86, SD = 0.19) were more accurate at producing tap than the ELLs (M = 0.72, SD = 0.16; Figure 1).

An ANCOVA was also used to evaluate the differences in key substitutions for tap used in Spanish (see Figure 2) while controlling for the age difference between the ELL and Spanish monolingual children. A preliminary analysis evaluating the homogeneity of regression assumption indicated no violation. For t/d substitutions, there was no significant difference between the monolingual (M = 0.01, SD = 0.03) and ELL children (M = 0.03, SD = 0.04) after controlling for age, F(1, 58) = 1.21, p = .294, and there was no significant age effect, F(1, 56) = 3.50, p = .067, η² = .009. For semivowel/liquid substitutions, there was a nearly significant difference observed between the monolinguals (M = 0.05, SD = 0.09) and ELLs (M = 0.09, SD = 0.11), F(1, 56) = 3.51, p = .066, η² = .059, after controlling for age, and there was no significant age effect, F(1, 58) = 1.278, p = .263, η² = .022.

ELLs: Spanish Versus English

For the ELLs, a paired-samples t test revealed a significant difference between the proportions of accurate tap in English and Spanish. The ELLs were more accurate in Spanish (M = 0.72, SD = 0.16) than English (M = 0.58, SD = 0.23), t(29) = –2.49, p = .02, d = –0.71 (see Figure 1, left half).

Paired-samples t tests were also used to examine the differences in the proportions of key substitutions in each language (see Figure 2, left half). The ELLs used significantly more t/d substitutions in English (M = 0.153, SD = 0.128) than in Spanish (M = 0.031, SD = 0.042), t(29) = 5.092, p < .000, d = 1.27. A significant difference was not observed between semivowel/liquid substitutions in Spanish (M = 0.095, SD = 0.111) and English (M = 0.059; SD = 0.104), t(29) = −1.248, p = .222, d = −0.34.

A two-way repeated-measures analysis of variance was also conducted to examine the interaction of the two types of substitutions used for tap (t/d or semivowel/liquid) across the two languages (English and Spanish). A significant interaction was observed, F(1, 29) = 15.109, p = .001, and revealed that ELLs used a larger proportion of t/d substitutions than semivowel/liquid substitutions in English, whereas in Spanish, they used a larger proportion of semivowel/liquid substitutions than t/d substitutions. Given the possibility of negative transfer, we might have expected the ELL children to associate tap with semivowel/liquids in both languages (because of tap's relation with the Spanish [r]), but they actually associated tap primarily with t/d in their English pronunciations, reflecting the fact that t/d are related to tap in English.

According to parent report, many of the ELLs had been exposed to some English before starting school; therefore, we computed the patterns associated with the substitutions separately for the six ELL children who started English exposure at 5 years of age, presumably at onset of kindergarten. As for the whole group of ELLs, the interaction regarding key substitutions (more t/d substitutions in English, more semivowel/liquid substitutions in Spanish) was also present in the subset of the six whose English exposure began at 5 years of age (see Figure 3).

Figure 3.

Key substitutions exhibited by English language learners (ELLs) with first English exposure at age of 5 years (n = 6). semiV = semivowel.

Transcriber Agreement

Because the great majority of the transcription was done blind with regard to monolingual or ELL status of the children, the results showing significant differences between monolinguals and ELLs should not have been affected by any possible transcriber bias, and thus, we do not view transcriber agreement as a major concern. In contrast, differences within the monolingual and ELL groups for the two languages are not so insulated from potential bias, because the coders could not be blinded to which language they were transcribing. Consequently, we conducted a direct assessment of transcriber agreement to assess this question.

Usually one presents such reliability data in methods, but in this case, it makes sense to present them after results because the most important issues regarding transcriber agreement concern whether the same pattern of results would be obtained with different transcribers. Having presented data in the figures above, we tested whether the same patterns were obtained with a different transcriber.

Fourteen files were used for this purpose. The two expert transcribers produced the standard transcription for this subset of files, whereas the last author carried out the secondary transcription, at a much later point in time, with no reference to prior consensus transcriptions. Files used for coder agreement analysis included data from all the child groups: Spanish monolingual children, English monolingual children, and the ELL group in English and Spanish. Agreement on tap production was calculated in the same way as for the data above. Both transcribers exhibited similar patterns on tap accuracy in that both revealed higher accuracy in Spanish. In fact, the secondary transcription revealed a larger difference between the proportions of accurate tap in Spanish than English than the standard transcription, thus suggesting even sharper differentiation of the languages (see Figure 4).

Figure 4.

Transcriber agreement regarding accuracy of tap productions: Standard and secondary transcription results on tap accuracy.

More importantly, similar patterns were also revealed between the standard and secondary transcriptions regarding key substitutions in each language. More substitutions of t/d were observed in English than Spanish, and more substitutions of semivowel/liquids were revealed in Spanish than English. Again, the secondary transcription demonstrated a larger difference in the proportions of occurrence of these substitution types than the standard transcription, again suggesting even sharper differentiation (see Figure 5).

Figure 5.

Transcriber agreement on key substitutions for tap: standard and secondary transcription results on substitutions. semiV = semivowel.

Discussion

The limited research on allophonic variation in Spanish-speaking children learning English and the conflicting status of tap in Spanish and English prompted this study of tap allophony. Given that tap is a phoneme in Spanish and an allophone in English, we thought it possible that young ELLs might exhibit inconsistency when attempting to produce the tap sound in each language, especially in their L2, English. The association of tap with semivowels and liquids (as is present in Spanish) might encourage transfer from L1 to L2 by producing semivowel/liquid substitutions for tap in English. We sought to explore this potential transfer by examining tap production accuracy in Spanish and English and to draw to comparisons between substitutions for tap by ELL and monolingual speakers of each language. Although prior studies have examined allophone production in bilingual children, none has exclusively examined L2 learning of an element that is in phonemic/allophonic conflict between L1 and L2. This investigation expands research on ELLs whose L1 is Spanish.

Overall, we observed that ELLs were significantly less accurate at producing tap than were monolingual children in each language. It is worthwhile to note that previous research on tap accuracy in monolinguals may also offer some insight into these results as Jimenez (1987) found that Spanish-speaking children had achieved tap mastery at an earlier mean age of 4;7 whereas English-speaking children were observed to accurately produce tap by age of 5 years (Klein & Altman, 2002). Therefore, it was not surprising to find that the ELL group was significantly more accurate at tap production in Spanish than English. In addition, their relatively limited exposure to and experience with English were likely contributors to their less accurate production in English.

The notable surprise was that, in English, the ELLs' most frequent substitutions for tap (t/d) were consistent with the monolingual English speakers' substitutions for tap; this pattern contradicts the expectation of negative transfer that might have yielded more common substitutions of semivowel/liquids for tap in English as tap is associated with the liquid trilled /r/ in Spanish. There was no significant difference between the ELL and English-speaking monolinguals regarding the proportion of t/d substitutions used for tap. Both groups most frequently substituted [d] for tap in English, with several occurrences of [t] for tap. This finding suggests that the ELLs, in English, had quickly acquired underlying knowledge of the tap's allophonic status and association with the /t/ and /d/ phonemes and not with rhotic consonants as it is classified in Spanish. Use of t/d as substitutes may also indicate that the ELL group was continuing to refine their articulation of English tap and, in doing so, erroneously produced [t] or [d] simply because the tap articulation may be more difficult than the stop. It is possible that this was also the case for the English-speaking monolingual children, who also produced examples of t/d substitution for tap. Either way, it suggests the ELLs relatively quickly adopted a pattern like that of English speakers.

Even though the ELLs adapted quickly to English tap as being associated with /t/ and /d/ as reflected in the large number of [t] and [d] substitutions for tap that they produced in English, this does not mean that the children's L1 had no effect on L2. With regard to the other key substitution type, the ELL group used significantly more substitutions of semivowel/liquids for tap in English as compared with their monolingual peers, suggesting a possible influence of L1 on L2. This influence could be viewed as having a possible weak negative effect of transfer, which could be a point on the far end of a transfer continuum. These findings leave open the possibility that transfer played a minor role in the outcomes we observed, suggesting that transfer is not binary, but rather continuous due to the developmental process of language learning.

Another interpretation is suggested by the competition model (MacWhinney, 1997) mentioned above. The ELLs, by establishing direct links in English between tap and the alveolar stops, may have acquired a tendency, when speaking English, to inhibit the direct links in Spanish between tap and the semivowels/liquids. This inhibition can be viewed as a response to competition between tendencies associated with the two languages, a competition that prevents complete segregation of the two languages in the ELL child's mind. In light of the competition model, the languages can interact in potentially complex ways, and in this case, it might be said that inhibition of the contrasting tendencies regarding substitutions for tap across languages appeared to be incomplete.

In Spanish, there was no significant difference between the monolinguals and ELL groups regarding the proportions of t/d substitutions for tap. However, a near significant difference was observed for semivowel/liquid substitutions. The ELL group used a larger proportion of semivowel/liquid substitutions in Spanish than their monolingual peers. In examining this key substitution type, it was observed that the ELLs most frequently used [l] as a substitute for tap, whereas the monolingual Spanish-speaking children substituted [r] (trill) most frequently. Although [l] alternation for tap is a dialectal variation of Puerto Rican Spanish and has been observed in other studies of bilingual children, all of these children were speakers of Mexican dialects of Spanish, and therefore, [l] productions for tap were scored as substitution errors. Although it is possible that these children had previously encountered Spanish speakers using this feature, we have no way of verifying this possibility. However, the results of this study provide further evidence for this pattern of acquisition that has been observed in the productions of Spanish–English bilinguals. This finding also suggests that the ELLs may have had an underlying representation for tap that was not clearly defined but included various sounds that all shared an alveolar place of production. Use of [r] (trill) for tap by the monolingual Spanish speakers is not surprising as previous literature has indicated Spanish-speaking children exhibit errors on rhotics throughout first grade (Goldstein, 1995).

Overall, the findings of this initial investigation of allophony in ELLs revealed that production of phonemic tap in Spanish and allophonic tap in English does not appear to be as problematic for children as we might have expected because high levels of tap accuracy were achieved in both languages during the first few months of kindergarten. We were concerned that there would be evidence of strong negative transfer, but our results did not confirm this concern. It is plausible that the ELLs were influenced by the speech patterns of their monolingual classmates and/or teachers and had heard these native speakers sometimes use /d/ for tap. It would be interesting to study tap production by Spanish language learners whose L1 is English to determine if they displayed similar substitution patterns in each language to those exhibited by ELLs of the same ages. It seems possible that these children may encounter more challenges acquiring tap as it occurs in numerous contexts in Spanish as compared with the limited set in English. Perhaps, the limited environment of tap in English offers an explanation for the generally accurate productions of the ELLs.

Finally, it should be noted that these data were collected in citation form, in isolation, and with words that were possibly unfamiliar to the children. In the current context, the results are best viewed as preliminary. Ideally, the topic could be investigated with more natural, spontaneous productions of individual words or by collecting a connected speech sample, which could allow for observation of tap production across its many different contexts in Spanish as compared with its limited environment in English. In addition, the current study did not incorporate acoustic analysis. Incorporating that type of analysis could be especially helpful in differentiating tap productions from those of [t] or [d] and in assessing subtleties of semivowel and liquid acoustics.

Conclusions

Development of a strong phonological foundation is important for young Spanish-speaking children acquiring English. Although it is plausible that some of their L1 phonological knowledge may aid in the acquisition of L2 phonology, it may also result in inconsistencies, transfer in their L2 productions, or competition and interaction between tendencies acquired for each language.

Understanding phonology and, in turn, allophony might aid in ELLs' acquisition of reading and spelling skills. The tap sound is allophonic in English and phonemic in Spanish. In American English, tap only occurs intervocalically, but in Spanish, it can appear intervocalically, word-finally, and in consonant clusters. Young ELLs must resolve these differences implicitly in order to achieve accurate production of English tap. We deem it highly unlikely that educators teach ELLs or monolingual children explicitly about tap or about allophony in general (Moats, 2009). If we are correct, it suggests children develop implicit understanding of tap and its allophonic status in the absence of explicit instruction, enabling them to accurately produce tap in words such as water or body. As young ELLs transition from being oral language users to readers and spellers, they may become aware of tap in much the same way they become aware of other speech sounds as their phonemic awareness increases. At some point, children must make the phoneme–grapheme and grapheme–phoneme connection between tap and the two different graphemes (t and d).

Our initial investigation of tap and allophony interaction suggests that early kindergarten–age ELLs do not have major difficulty in navigating the tap allophony rule in English, and thus, our expectation of possible negative transfer was not confirmed. This finding is consistent with an empirical suggestion that allophonic variation across languages may not constitute a major concern for development of a phonological foundation in an L2.

Appendix

Stimuli Lists

  English Stimuli

  birdie

  body

  butter

  party

  pretty

  radio

  ready

  recorder

  rider

  sturdy

  tardy

  water

  Spanish Words

  color

  comedor

  cortar

  cura

  diario

  dinero

  gordo

  muros

  parecer

  perdido

  recibir

  sartén

  tarde

  torta

  verdad

Funding Statement

This work was supported by National Institute of Child Health and Human Development Grant R01 HD046947, awarded to D. Kimbrough Oller, PI.