Input, Universals, and Transfer in Developing Rhotics: A Sketch in Bilingualism

Abstract

Understanding the role of input in bilingual phonological acquisition is revealing for deciphering the workings of language acquisition processes. Input and usage distributional frequencies guide and differentiate speech sound acquisition patterns cross-linguistically. Such processes are operant in first- and second-language acquisition. There is an under-representation of investigations on how context-specific input in bilingualism influences the early acquisition of rhotics in child developmental speech longitudinally. This study addresses the gap by tracing a Greek/English bilingual girl’s rhotic development between ages 2;7 and 3;11, utilizing naturalistic data during daily interactions with an adult interlocutor. The study reports and schematically illustrates the child’s bilingual usage frequencies, informing language choice in her production variables, which demonstrate, quantitatively and qualitatively, the effects of context-specific input on rhotic accuracy levels and substitution patterns in both languages. Specifically, distributional frequencies in the input govern the child’s phonemic and phonetic tendencies in the languages. Findings are compared with previous reports in the literature and enhance language acquisition theory, revealing the pivotal role of input in the dynamic interplay with developmental universals, language-specific tendencies, transfer, and individual variation.

1. Introduction

Phoneme frequency in a language has an impact on acquisition. Early on, Jakobson ([1941] 1968) stated that the order of acquisition of speech sounds is determined by maturational constraints on perception and production that are subject to available information in the targeted language. Greenberg’s (1963) typology of markedness provides evidence that infrequent forms in cross-linguistic grammars are universally more difficult to master and, thus, slower to acquire. Work comparing speech sound mastery by monolingual children’s babbling and/or speaking of different languages (e.g., de Boysson-Bardies and Vihman 1991; McLeod and Crowe 2018) further indicates universal acquisition patterns influenced by distributional frequencies in the input (e.g., Ingram and Babatsouli 2024 for a review). Such processes are operant in first- and additional-language acquisition by children and adults (e.g., Cummins 2014; Eckman 1984). Sim and Post (2024) review the role of variable phonetic input on speech acquisition, highlighting contextual differences in monolingualism and bilingualism, as well as considering dialectal variation.

Understanding the role of input in bilinguals is both revealing and challenging because of the nature of exposure to each language in terms of input quantity and quality, which is typically context-specific, meaning that it is determined by combinations of factors that vary between children, like the bilingual’s specific pair of languages, the typology of the languages, the linguistic variety (e.g., standard vs. dialect), the onset and length of exposure in each language, the age and educational and socioeconomic status of those modelling language input for the child, the status of the language(s) in these model speakers (e.g., first/additional language), the specific sociolinguistic context of use per language (home vs. societal language), etc. (e.g., Babatsouli 2024a; Flege 2009; Hoff 2018; Sim and Post 2024). High distributional frequencies of a targeted consonant in a bilingual child’s dominant language have a positive effect in the non-dominant language, where a shared phoneme in the languages is acquired at a later age by respective monolingual speakers of a child’s non-dominant language (e.g., Babatsouli 2017). There is a scarcity of studies investigating the interaction between distributional frequencies in the parental input and a bilingual child’s usage frequencies in the less dominant language (e.g., Babatsouli and Nicoladis 2019). Sim and Post (2021) investigated the influence of maternal input quality and “interadult” variation on early phonological acquisition of coda stop release patterns, finding a strong, facilitative correlation between maternal input and child output.

The present study sought to shed more light on the effect of context-specific input in child bilingualism by focusing on the acquisition of rhotic contrasts in a child’s developmental productions, sketching the actual rhotic acquisition path longitudinally. The child’s productions were hypothesized to demonstrate production, perception, and input quality and quantity constraints that lead to the acquisition of the targeted context-specific input. This is addressed in the study quantitatively, by measuring the developmental progress of the rhotic in the child’s languages, and qualitatively, by discussing the evidence in terms of previous reports in the literature.

1.1. Input in the Acquisition of Rhotics by Bilinguals

There is a scarcity of studies investigating input effects on children’s rhotic development early on, in spontaneous speech contexts. Wode (1981) conducted a rare study to investigate children’s acquisition of rhotics in a non-native language based on naturalistic speech. The data covered his four German-speaking children’s L2 English rhotic development over six months and were based on diary notes. The children were between ages 3;11 and 8;11, and their English /ɹ/, with primarily second-language but native-like input in English, showed phonetic variation marked by gradual approximations leading to native-like production. Rhotic acquisition has mostly been investigated in experimental settings. In an acoustic study involving school-aged children in a US two-way immersion program, Menke (2017) compared Spanish tap and trill productions elicited with a word-list task in the speech of thirty-one Spanish/English bilingual children aged 6;8 to 13;5 to examine phonemic contrast differentiation and the effect of age. The findings showed that the bilinguals distinguished the rhotics by the manner of articulation and consonant duration, while grade-level differences in their productions indicated age-dependent progression to adult-like productions of the rhotics; the bilinguals’ rhotics were acquired later than by respective monolinguals. The developmental productions combined sounds from the phonetic inventories of both languages; specifically, English-specific [ɹ], as a substitution of the Spanish rhotics, revealed an influence from the bilinguals’ English.

This topic has been more extensively investigated in adult L2 acquisition in terms of both perception and production. Logan et al. (1991) found that intensive high-variability perceptual training of the English /l-ɹ/ contrast, which is difficult for monolingual Japanese speakers, resulted in improved performance in the eleven 19-to-22-year-old participants. Specifically, the study participants discriminated the contrasts on the perceptual identification task and maintained long-term memory retention, improving the quality, identifiability, and intelligibility of their English /ɹ/-/1/ word productions.

Colantoni and Steele (2018, see also for review) enhanced previous empirical findings on the L2 acquisition of phonetics and the lexicon by looking at the impact that within-category phonetic variability, present in the input, has on non-native speakers’ production accuracy in their L2. They investigated stop and rhotic productions in adult 30–40-year-old English learners of French (twenty participants) and Spanish (thirty-nine participants), and twenty native speakers per language, using a carrier-sentence reading task. The authors found that phonetic variability did not exclusively have a facilitative influence, in that positive effects of input variability may be limited to native language contrasts, suggesting an interplay of other parameters such as challenges in measuring variability, how perception influences input, and differentiating between transfer and response to input. Regarding the effects of input on L2 learning, it was argued that the difficulty lies in predicting the effects of variability by determining the specific input that the learners are exposed to during their learning paths.

Reyes et al. (2017) investigated Spanish rhotic acquisition in twenty-five bilingual/multilingual children (nineteen English monolingual and six Korean–English bilingual children, all learning Spanish as an additional language), alongside nine native Spanish-speaking children using a picture-naming task. They found that all the children quickly mastered native-like pronunciation of the Spanish rhotics, but the Korean–English bilinguals outperformed the English monolinguals, which indicated a bilingualism advantage. They also argued that L1 rhotic skills influenced rhotic acquisition in children’s additional languages, and that transfer effects subsided as universal strategies persisted. In a sociophonetic study of Spanish rhotic acquisition, Lease and Marchesi (2022) concluded that Spanish non-canonical rhotic variants were preferred over canonical ones by bilinguals in the heritage language context in New Mexico, USA, as their Spanish use increased. Furthermore, Spanish rhotics were produced more as English-like approximant variants in the second-generation heritage language speakers. Kehoe (2018) provided evidence that four German/Spanish bilingual children, aged from 1;9 to 3;3, exposed to native input, acquired the German uvular rhotic similarly to the respective monolinguals in the study, but earlier than the Spanish tap and trill, compared to the monolingual Spanish controls. The bilinguals infrequently used substitution [h] for the German uvular rhotic compared to the monolingual German controls, which was explained as having resulted from the absence of /h/ in the Spanish phonemic inventory. Overall, the bilinguals’ substitutions involved more obstruent and fewer sonorant productions, and more variability in the infrequent substitutions compared to the monolinguals in the study. The phonetic variability in these younger bilinguals (Kehoe 2018) matches the phonetic variability reported for the children in Wode’s (1981) study and, more recently, in the Lease and Marchesi 2022 study. Ball et al. (2001) report more variable substitution for the Welch trill /r/ than for English /ɹ/ in the Welsh-English bilinguals of the study, aged 2–6 to 5;0, also reporting more variability in the Welsh-dominant bilinguals than in the English dominant bilinguals. Kehoe’s (2018) direct comparison between bilinguals and monolinguals showed evidence that the substitutions of bilinguals are more variable than those of respective monolinguals. The same was found The results and inferences discussed in this section are important for deciphering phonological/phonetic development in bilingual populations, and they are relevant for comparing the findings of the present study, as shown in the Discussion.

1.2. Rhotics in English and Greek

Rhotic segments are, by default, voiced. English and Greek have a single rhotic phoneme each. The Greek rhotic is a flap /ɾ/, i.e., (+consonantal, coronal, front), while the English rhotic is the approximant /ɹ/, i.e., (−consonantal, coronal, labial (+round), distributed). Greek has no rhotic allophones in complementary distribution, although the trill, [r], is reported as an infrequent allophone in free variation, being speaker-dependent (see reviews in Babatsouli 2023; Okalidou and Babatsouli, forthcoming). In non-rhotic varieties of English, the rhotic singletons in coda position are not produced, while rhotic varieties of English have rhotic vowels in the nucleus (for more detailed comparisons, see, e.g., Babatsouli 2022, 2023). Both Greek and English rhotic segments are coronals (e.g., IPA 2015; Maddieson 1984), but there is also dialectal variation for place of articulation evident on the phonetic level as in, for example, the free variation in the rhotic place of articulation of the General American English (GAE): alveolar, bunched, and retroflex rhotic productions). Flaps/taps and trills are closely related types that result from air-flow obstruction in the oral cavity, frequently serving as allophones (Maddieson 1984). In this study, only word-initial onsets in the two languages are compared, as discussed further in the last section of the ‘Materials and Methods’, below.

1.2.1. Frequency of Rhotic in English and Greek

The rhotic is one of the most frequent sounds across languages. Maddieson (1984) reports that there are both alveolar and retroflex rhotic types in the languages of the world. Among the alveolar rhotic sounds, the trill /r/ and tap /ɾ/ are more frequent (45% each) than approximants (e.g., /ɹ/) (8%) and fricatives (e.g., /ʁ/) (2%). More recently, Everett (2018) employed a database of 6,901 words for different world dialects and languages, to investigate language similarity in terms of speech sound usage frequency. The world lists comprise 100 or a subset of 40 words that are resistant to borrowing (as per Swadesh 2006); the rhotic, [r, ɽ, ɾ, ʀ], is the fifth most frequently used consonant sound at a frequency of 2.98% across consonant and vowel sounds, above the lateral [l] at 2.66%, and below [n] at 5.88%, [k] at 5.46%, [m] at 4.49%, and [t] at 4.45%. In what follows, the results of studies on distributional frequencies of the rhotic are reported first for English, followed by those in Greek; notably, more studies have investigated segmental distributional frequencies in English than in Greek.

Mines et al. (1978) investigated 103,887 speech sound tokens in conversational English, finding that the rhotic is the third most frequently used English consonant and the sixth most frequent sound overall (when consonants and vowels are considered) at 3.87%, following [ə] at 7.3%, [n] at 6.72%, [t] at 5.78%, [ɪ] at 5.15%, and [s] at 4.61%.

Based on Kilgariff’s (n.d.) unlemmatized frequency list for the British National Corpus (www.natcorp.ox.ac.uk, accessed on 1 March 2024) and the Carnegie Melon University Pronouncing Dictionary (http://www.speech.cs.cmu.edu/cgi-bin/cmudict, accessed on 1 March 2024), Doug Blumeyer (2012) found that the rhotic is the second most frequently used consonantal sound and the third most frequently used sound overall (consonants and vowels included) at 6.94%, following [ə] at 11.49%, [n] at 7.11%, and, lastly, [t] at 6.91%. Further, Babatsouli (2024b) examined the phonology of the 1000 most frequent words in English and found that 34.1% of the words comprised one rhotic, while 18 words comprised two rhotic consonants, and 46 words had the rhotic word-initially (WI). Regarding its frequency among all sound tokens, the rhotic has a frequency of 8.8% overall, but of 13.5% compared to other consonants only.

Looking at 102,934 spoken Greek words obtained from 1000 television and radio programs, Trimmis et al. (2006) found that the rhotic has a frequency of 4.18% compared to all consonant and vowel sounds, following [s] at 7.68%, [t] at 7.54%, [n] at 6.17%, and [p] at 4.36%. Protopapas et al. (2012) also computed the following dominant frequencies of consonants to all the consonants and vowels in 35,304 Greek words, as follows: [s] 8.91%, [t] 8.37%, [n] 6.23%, [r] 4.56%, and [p] 4.35%. They found that there are 9.45 phones per word, on average. Multiplying 4.56% by 9.45 results in 43% of Greek words containing a rhotic. More recently, Babatsouli (2024b) examined the phonology of the 1000 most frequent words in Greek and found that 34.5% of the words comprise the rhotic, 15 of which comprise two rhotic tokens, and 14 of which have a WI rhotic. In terms of frequency concerning all sound tokens, the rhotic has a frequency of 6.0%, albeit with a frequency of 11.5% compared to other consonants only. To sum up, rhotic productions are more frequent in English than in Greek, including WI contexts.

1.2.2. Rhotic Norms

Respective Greek and English monolingual norms and crosslinguistic norms show different ages of acquisition for the two rhotic consonants investigated in this study. The alveolar approximant, /ɹ/, is acquired by age 6;0 at the 75% criterion and by 8;0 at the 90% criterion by English monolinguals (Smit 1993; Smit et al. 1990), while cross-linguistic norms show 75–85% acquisition between 3;0 and 3;11, and 90% acquisition between 5;0 and 5;11 (McLeod and Crowe 2018). According to PAL (1995), Greek rhotic norms show an acquisition level of 75% by ages 5;6–6;0 and of 90% after 6;0, while more recent norms indicate 75% rhotic acquisition by age 3;0 (Geronikou and Babatsouli 2024). Crosslinguistic monolingual norms are relevant to and informative for gauging bilingual acquisition in the lack of bilingual norms involving combinations of specific language pairs (Babatsouli 2023). McLeod and Crowe’s (2018) review article on cross-linguistic consonant acquisition involved studies on the following 27 languages: Afrikaans, Arabic, Cantonese, Danish, Dutch, English, French, German, Greek, Haitian Creole, Hebrew, Hungarian, Icelandic, Italian, Jamaican Creole, Japanese, Korean, Malay, Maltese, Mandarin (Putonghua), Portuguese, Setswana (Tswana), Slovenian, Spanish, Swahili, Turkish, and Xhosa. Their results show that the alveolar tap [ɾ], approximant [ɹ], uvular trill [ʀ] and fricative [ʁ] rhotics are 75–85% acquired between ages 3;0 and 3;11, and the alveolar trill [r] is 75–85% acquired between ages 4;0 and 4;11, a time when the tap [ɾ] is found to be 90–100% acquired. The approximant [ɹ] and alveolar trill [r] are the last to be acquired at the 90–100% criterion between the ages of 5;0 and 5;11.

Overall, the differences between the monolingual norms for English /ɹ/ and Greek /ɾ/ are supported by the universal language effects of the distributional frequencies of different rhotic types (Maddieson 1984), in that taps are not only more frequent than approximants in the world’s languages, but also acquired earlier, as was inferred by Ingram and Babatsouli (2024). Ingram and Babatsouli (2024) reviewed the crosslinguistic evidence in McLeod and Crowe’s study (2018), reporting the following: common tendencies across languages regarding the existence of implicational hierarchies, markedness considerations of perceptual/articulatory ease/difficulty, preferences for specific places/manners of articulation and laryngeal distinctions (e.g., laterals precede rhotics), and evidence that increased articulatory complexity leads to slower acquisition for consonants with double places of articulation compared to targets with a single place of articulation (e.g., the labio-alveolar [ɹ] compared to [ɾ]). Comparisons of typically developing monolingual child speech across different languages set the grounds for evidence to establish cross-linguistic phonological universals. This evidence will ultimately advance phonological acquisition theory and guide clinical decision-making during speech assessment and therapy for bilingual children with phonological disorders (e.g., Babatsouli 2024a).

The establishment of developmental norms seeks to identify shared paths in the particularly complex language acquisition process. The universal tendencies that lead to these normative behavior patterns across children, languages, and contexts can be explained by children’s sensitivity to language structures and the processes used to extract these structures from others in the linguistic input; this is referred to as statistical learning (e.g., Romberg and Saffran 2010). Work within the constructivist theoretical stance on language learning provides evidence supporting how children’s specific linguistic experiences impact the outcome of the language acquisition process as children reproduce and manipulate the linguistic input of their caregivers (e.g., Lieven and Brandt 2011). Thus, when considering the child’s “environmental input”, it is argued that “statistical learning operates over the situated learning environment, and not just sequences of sounds in utterances” (Frost and Monaghan 2020, p. 65). In a top-down approach, the authors also emphasize that environmental input (which is referred to in this study as context-specific input) triggers specific linguistic representations in the child that are innate, rather than the other way around, i.e., the acquisition of structure being dependent on computations of distributional properties in the linguistic input (p. 68). We can conclude, therefore, that language acquisition is subject to the combined effects of language typology and children’s context-specific learning environments.

1.2.3. Functional Load

The literature on monolingual and bilingual acquisition indicates the interdependence of lexical and phonological acquisition: as children’s vocabulary becomes bigger and more complex, phonemic representations are also enhanced, leading to better speech production skills (Ferguson and Farwell 1975; Kehoe 2011; Stoel-Gammon 2011). Cross-linguistic research has shown that differences in language typology, which includes the lexical inventory, often lead to variation in children’s developmental phonetic inventories for shared phonemes (e.g., McLeod and Crowe 2018). For instance, /w/ is acquired early by English-speaking monolinguals, but not by Spanish-speaking monolinguals (e.g., Ingram and Babatsouli 2024). The phonemic functional load in a language is determined by the lexical inventory and the distribution of phonemes in its words. High-functional-load phonemes differentiate more words and low-functional-load words differentiate fewer words (e.g., Trubetzkoy [1939] 1958). The effects of functional load on bilingual phonological acquisition may vary, being facilitative when the phoneme is shared between the bilingual’s languages (e.g., Greek/English /θ/ bilingual acquisition, Babatsouli 2017) or inhibitive when it is specific to one language only (e.g., English-specific /w/ bilingual acquisition, Babatsouli 2015).

An explanation for such cross-linguistic disparities in acquisition has been provided by Pye et al. (1987), who investigated the distribution of English and Quiché sound inventories in children’s words in terms of functional load f (or phonemic load), which refers to the degree of prominence a phoneme has language’s within a phonological system. The argument is that a consonant is less likely to be acquired faster than another if it occurs in fewer word types in the language. As an example, English /ð/ has a low functional load and late age of acquisition (>90% by 5;0 in monolinguals) compared to other languages, such as Brazilian Portuguese, Greek, Spanish, and Swahili, for which the sound is shown acquired between 2;0 and 3;0 (McLeod and Crowe 2018). Edwards et al. (2015) investigated phoneme frequency in the speech of 2- and 3-year-old monolinguals in the following languages: English, Greek, Cantonese, and Japanese. They found supporting evidence for the language-specific correlation between consonant type frequency and accuracy level. Similarly, a bilingual child’s English /θ/ was acquired faster than by respective monolinguals due to facilitative interference from the child’s Greek /θ/, which has a higher functional load in the language, compared to English (Babatsouli 2017). An ambiguity in the linguistic input on the phonemic prominence of a sound resulting from misperception, mishearing, or mispronunciation of a sound may confuse the learner and slow acquisition (e.g., Jakobson [1941] 1968). From an exemplar-based model of lexical representation, the relative infrequency of canonical variants in children’s lexical input will lead them to form stronger representations of the non-canonical variants available in the input (Bybee 2001; Bybee and Beckner 2010). This is supported by evidence of a bilingual child’s delayed mastery of /w/ in the child’s non-dominant English at the late age of 3;7, compared to monolingual English norms (Babatsouli 2015).

To conclude this introductory section, it can be stated that disentangling the notorious complexity of bilingual phonological acquisition requires a consideration of the role of linguistic input at variable levels: language learning as human capacity in the context of social interaction, language-specific typologies, the linguistic prominence of words and phonemes in each of the bilinguals’ languages, and the role of context-specific input in the child’s learning environment.

1.3. Goal of the Study

This study investigates the acquisition of rhotics by a Greek/English-speaking bilingual child longitudinally over seventeen months of development with spontaneous speech data; both the study’s longitudinal span of coverage and the naturalistic speech context are under-represented in the literature. This contribution is also original in considering how the acquisition rates and developmental patterns of her rhotic sounds in each language are affected by the context-specific nature of exposure to the languages. The aim is to inform our understanding of early child bilingual speech acquisition. The specific questions are as follows:

• Do language-specific distributional frequencies and phonemic functional load impact the acquisition rates of the bilingual child’s targeted rhotics? The prediction is that the rhotic with the highest phonemic prominence, resulting from high distributional frequencies in the child’s linguistic environment, will be acquired faster.
• What do the child’s productions for the targeted rhotics in the languages during her development longitudinally reveal about the bilingual acquisition path? The prediction is that the developmental path of the child’s rhotics will be characteristic of the context-specific nature of her exposure to her languages.

2. Materials and Methods

2.1. The Participant

This study is based on a bilingual girl’s naturalistic speech in Greek (dominant language) and English (non-dominant language) during daily interactions exclusively in English with her Greek/English bilingual mother. Consistent maternal input in English began at age 1;0, while the child was raised in Greece, where Greek is the language of the ambient society; English was not the language of the community, extended family, or immediate social circle. Both the child’s parents have Greek as their native language, but also speak fluent English as an L2, with several years of residence experience (including higher-education studies) in English-speaking countries. The mother/author had lived in the UK for six years and in the US for two years, before the beginning of the study. Both parents hold a doctoral degree. As a bilingual, the mother spoke Greek and English as life circumstances permitted; however, addressing the child solely in English was not a conscious decision until after the child’s first birthday.

The Grammont rule (Ronjat 1913) was espoused as a regular practice to raise a child bilingually, referring to the “pattern of parental use in bilingual families in which each parent uses only, or primarily, one language with the child” (Genesee et al. 2008, p. 98). The mother only spoke English to the child, and the father only spoke Greek to her; the father hardly ever spoke English in the family context. The child was solely exposed to her mother’s input in their daily interactions. The mother spent the day with the child and would always respond and encourage conversation in English when the child code-switched to Greek. As a result, the child was primarily exposed to English in her daily routines. When she started Greek daycare at age 2;3, the child went through a brief silent period; a silent period is a reported developmental phase in both first-, bilingual, and second-language acquisition (e.g., Genesee et al. 2008; Vihman et al. 1985). Beginning at age 3;0, the child was gradually exposed to native English speech in a bilingual children’s social/instructional setting, where only English was spoken and the adults were native speakers.

The child participant in this study is a simultaneous bilingual exposed to primarily single-person second language (L2) English in an exogenous setting. Greek is her dominant language, or L1, and English is her non-dominant language, or L2 (see e.g., Babatsouli 2020, and Babatsouli and Nicoladis 2019, for grammatical measures that justify these linguistic status assignments). The terms are briefly defined next. Bilingual exposure is exogenous when one of the languages (in this case English) is not the societal language, or language of the ambient environment (e.g., Pienemann and Keßler 2007). Simultaneous bilingualism is assumed when a child’s first exposure to either language occurs before three years of age (e.g., McLaughlin 1978; Meisel 1989; Genesee 1989; Montrul 2008), and is sometimes referred to as bilingual first-language acquisition (Meisel 1989). There is agreement in the literature that neither cross-linguistic interaction (interference/transfer between the child’s languages (e.g., Kehoe 2020; Major 2014)) nor interlanguage ambiguity (foreign accent) in the input disqualifies simultaneous bilingualism status (e.g., Hakuta 1986; Grosjean 1989; Paradis 2000; Meisel 2004). The “interlanguage ambiguity hypothesis” (Paradis 2000, p. 177 and references therein) is defined as cross-linguistic interference resulting from a child’s exposure to foreign- or second-language accent, known as interlanguage. Specifically, interlanguage is the intermediate linguistic system that results when a speaker combines grammatical features of native language (L1), another language (L2), and language universals (U) (Selinker 1972). MacWhinney (2001) argues that simultaneous bilingualism is the point where the barriers between first- and second-language acquisition are brought down for a more unified theoretical framework; this stance continues to be embraced (e.g., Babatsouli and Ball 2019; Babatsouli and Ingram 2018).

2.2. Single-Child Longitudinal Case-Study Design

Approaches to collecting language production data include elicited narrative, word-list reading, and spontaneous speech (Ratner and Menn 2000). Single-child case study research begins with Darwin (1877). Longitudinal single-child case studies inform scientific inquiry, shedding light on the dynamics of developmental paths in minute detail, enhancing language acquisition theory. “A longitudinal study gives a description of linguistic performance at several points of time, analyzing changes which occur between point t1 and point t1 + 1” (Meisel et al. 2008, p. 111). The present study is a single-child case study with longitudinal and uninterrupted data across seventeen months of development. Detailed “longitudinal records”, “sampled at intervals of less than a month”, are needed cross-linguistically (Bernhardt and Stemberger 1998, pp. 8–16), as they allow for “dense longitudinal tracking” (Babatsouli 2017 and references therein). This is especially true when relating usage to input (e.g., Babatsouli and Nicoladis 2019). Furthermore, there is a need for single-child case studies that track the developmental path quantitatively, alongside inferences informing language acquisition theory cross-linguistically. Large-dataset case studies involve arduous data collection and analyses. Phonetic transcription, for instance, is labor-intensive and “a formidable task” MacWhinney (1998, p. 485). The database in this study, from which the data are drawn, is a result of such an arduous methodological regime. Unlike in the present study, the data in the two largest single-child case studies in the language acquisition literature, namely those by parent linguists Neil Smith (1973) and Werner Leopold (1949), were collected in diary notes, are not easily quantifiable, and lack audio files. The current study fills in a methodological gap in the literature of first-language and bilingual child speech development, focusing on Greek–English bilingual speech development.

2.3. The Prosodic Context of Rhotics in the Analyses

As described above, the child’s exposure to English is context-specific, i.e., characterized by the following factors: exogenous, home language, interlanguage ambiguity, and primarily single-person. Because the child was exposed solely to maternal Greek/English interlanguage in the first two years, this determined the choice of prosodic context for the analyses of the child’s rhotics in this study. Specifically, this study focused on word-initial rhotic singleton onsets to control for the maternal English interlanguage ambiguity, notably, mixed rhotic (GAE variant) and non-rhotic (British variants) practice in word-medial and word-final prosodic contexts. The mother’s English pronunciation was judged by a native English phonetician as near-native (Barbara May Bernhardt, pers. comm.), but still an interlanguage, since she has both native-like WI /ɹ/ and occasional transfers of Greek /ɾ/. She does not produce an allophonic trill in free variation in her Greek; neither does the father. A full description of the mother’s Greek/English interlanguage is beyond the scope of the present study, yet the interested reader may find a detailed comparison of phonological and phonetic inventories in Greek and English, as well as the blueprint of Greek/English interlanguage, in Babatsouli’s work (2022, 2023).

2.4. The Dataset

Parental written consent was obtained for the study following the Declaration of Helsinki ethical procedures, although the parents declined public sharing of the dataset. The girl’s spontaneous English/Greek speech from 2;7 to 4;0 was digitally audio-recorded by the mother (also the author) for 1 h daily, 4–5 days a week, using a Sony PCM, 2 Portable Studio Recorder, XLR to 1/4-Inch (PCMD10). Speech data were narrowly transcribed by the author in IPA within 12–24 h of each recording in a CLAN (MacWhinney 2000) database of 511 CHAT files (31,684 child utterances in English, 13,940 child utterances in Greek) totaling 135 h of recordings. Data comprise fewer Greek utterances because the original research focused on the child’s English development and Greek data were mostly obtained during the child’s code-switching. The child’s English and Greek utterances were transcribed orthographically and phonetically (in IPA) by the author (a trained phonetician in Greek and English IPA transcription). The author also created a separate coding tier in the CHAT files for the data analyses. The child’s utterances were time-aligned to respective audio files, as was maternal speech. For a comprehensive description of the management of this database in CLAN, see Babatsouli (2015).

Because the mother/transcriber was the child’s interlocutor in the dataset, she was familiar with the child’s speech and recognized repetitions of specific sounds/productions in the data. The child acquired the sounds [v, l] early on (by age 2;8), and their adult-like realizations were validated acoustically in Praat (Boersma and Weenink 2015) the first time they appeared in the data. The mother/transcriber was also familiar with [w] as a substitution for the child’s English rhotic, and it was therefore recognized as such during the time-aligned-to-audio phonetic transcriptions; this substitution was also validated with acoustic analyses on its first occurrence and upon increasing consistent use in the dataset. Although the mother/investigator was certain of the child’s developmental and adult-like rhotic productions because of her familiarization with the child’s speech, acoustic analyses of a random 10% sample on average per month were also carried out in the dataset. The results of the acoustic analyses confirm the accuracy of the investigator’s phonetic transcriptions. Subsequently, the child’s realizations of the rhotics were tallied using CLAN; they amounted to a total of 242 production tokens for the targeted Greek rhotic, and a total of 1,397 production tokens for the targeted English rhotic. These productions are analyzed and discussed in the sections that follow.

2.5. Criteria for Rhotic Acquisition in the Study

Acquisition or “children’s productions of sounds in word contexts are usually examined in terms of degree of production accuracy and the percentage of children in an age group who reached the level of accuracy in phoneme production” (Dodd et al. 2006, p. 26). Two approaches in phonology have set the 70% (Ingram et al. 1980) and 75% correct use of a structure in obligatory environments (e.g., PAL 1995; Smit 1993; Smit et al. 1990) as criteria for adequate performance. Cazden (1968) established the criterion of 90% correct use as consistent use of a grammatical unit. Different studies use different acquisition criteria, such as the 70% (Ingram et al. 1980), 75%, and 90% criteria (Smit 1993; Smit et al. 1990), the 75–85% criterion (McLeod and Crowe 2018) for English-speaking monolinguals, and the 75% criterion for Greek-speaking monolinguals (e.g., Geronikou and Babatsouli 2024; PAL 1995). The present study utilizes the 70%–75% criterion, meaning that 70%–75% of the targeted sounds are produced in an adult-like manner by the child, i.e., Greek /ɾ/→[ɾ] and English /ɹ/→[ɹ]. Adult-like rhotic production that does not meet the criterion is considered present in the child’s phonetic inventory but not yet acquired in the targeted contexts.

3. Results

The analyses are presented as follows. There is an initial zoom on rhotic productions at age 2;7 to document the child’s language profiles at the beginning of the data collection (Section 3.1). The longitudinal data analyses are discussed next, with targeted Greek /ɾ/ first, followed by targeted English /ɹ/ (Section 3.2).

3.1. Rhotic Acquisition at 2;7

There are 785 English utterances and 688 Greek utterances in the child’s data at age 2;7. The utterances are longer in Greek (tokens-to-utterance ratios (TURs) of 2.02 (English) and 3.68 (Greek). Her mean length of utterance (MLUw) is 2.3 (English) and 4.5 (Greek) respectively. Her English MLU falls within Brown’s (1973) stage III, indicating comparable acquisition levels to English monolinguals.

Rhotic Acquisition Level at 2;7

The rhotic accuracy at 2;7 is at 5% (English) and 1% (Greek), an anticipated finding at this age. The child’s frequent rhotic substitutions are shown in

Table 1. Dominant substitutions of WI rhotic in English and Greek at 2;7.

	WI
	English	Greek
/ɹ, ɾ/→[l]	41% 18/44	40% 4/10
/ɹ/→[v]	43% 19/44	n/a

. The lateral, [l], is common in the languages, but English also shows a /ɹ/→[v]. Examples are restaurant (lɛstant) and ρωτάς /ɾotɐs/→[lotɐs] “ask”. Other substitutions are /ɹ/→[w, b] in read and red, respectively, and /ɾ/→[t] in ρωτάς/ρωτήσουμε [tɐtɐs]/[totisumɛ] “ask”. Greek also has deletions in multisyllabic ρυζάκι [izaci] “rice”.

3.2. Developmental Trajectory of the Child’s Rhotic Longitudinally

This section discusses rhotic development in terms of accuracy and substitution patterns along the developmental path longitudinally, presenting Greek /ɾ/ first, followed by English /ɹ/.

3.2.1. The Child’s Greek /ɾ/

The child’s Greek words with WI /ɾ/ (grapheme <ρ>) and the age of their first production in the data are shown in

Table 2. The child’s new Greek words with word-initial /ɾ/ ¹.

Age	Vocabulary	English Gloss
2;7	ρυζάκι	rice
	ρωτάς/ρωτήσουμε	ask
2;8	ρίξω/ρίξεις	throw
	ρωτάω, ρωτήσω/ρωτήσεις	ask
2;9	ρίχνω/ρίξουμε	throw
	ροζ	pink
	ρουφάμε	suck
	ρώτησα/ρωτήσουμε	ask
	ρόδα/ρόδες	wheel
2;10	ρεύμα	electricity/power
	ρολόι	clock
2;11	ρούχα	clothes
3;0	ράψω	sew
	ρόδακινα	peaches
	ρόδι	pomegranate
3;1	-
3;2	ρίχνει	throw
	ρόκα	arugula
	ρύζι/ρύζια	rice
3;3	ρίσκο	risk
	ρωτήσεις	ask
3;4	-
3;5	ρούχο	clothing item
3;6	-
3;7	ρυάκι	brook
3;8	ρίχνε	throw
3;9	-
3;10	ράμφος	beak
	ρίξει	throw

¹ The terms in bold indicate a list of word inflections separated by the slash (/).

. Because Greek is a highly inflected language (Holton et al. 2015), some word types are inflections of the following: ράμφος “beak”, ράβω “sew”, ρεύμα “power”, ρίσκο “risk”, ρίχνω “throw”, ρόδα “wheel”, ροδάκινο “peaches”, ρόδι “pomegranate”, ροζ “pink”, ρόκα “arugula”, ρολόι “clock”, ρουφάω “suck”, ρούχο “clothing item”, ρυάκι “brook”, ρύζι “rice”, and ρωτάω “ask”.

There are 17 Greek words with WI /ɾ/ for a total of 245 tokens. The monthly tokens are 10 (at 2;7), 5 (2;8), 25 (2;9), 19 (2;10), 5 (2;11), 33 (3;0), 7 (3;1), 29 (3;2), 22 (3;3), 21 (3;4), 14 (3;5), 16 (3;6), 17 (3;7), 8 (3;8), 8 (3;9), 3 (3;10), and 3 (3;11). At age 2;7, Greek WI /ɾ/ is overwhelmingly substituted by [l], except when deleted. Examples of words include ρόζ [∅oz] “pink” (2;9), ρίξουμε [litsumɛ] “throw” (2;10). The monthly tokens of /ɾ/ deletions are: 1 (2;7), 1 (2;8), 2 (2;9), 1 (2;10), 0 (2;11), 2 (3;0), 1 (3;1), 4 (3;2), 3 (3;3), 2 (3;7), 1 (3;8), 1 (3;9). Deletions are evidenced in ρυζάκι [∅izaci] “rice” (diminutive) (2;7), ρωτήσω [∅otiso] “ask” (2;8), ρόδι [∅oði] “pomegranate” (3;0), ρύζι [∅izi] “rice”, ρυάκι [∅iɐci] “brook” (3;7). Figure 1 shows the child’s longitudinal development of Greek /ɾ/→[ɾ] and its main /ɾ/→[l] substitution. We can see that as /ɾ/ accuracy increases, /ɾ/→[l] recedes, creating a mirror effect. The child’s Greek /ɾ/ accuracy reaches the 70% acquisition criterion (e.g., Ingram et al. 1980) around age 3;9, as shown by the solid black line in Figure 1.

3.2.2. The Child’s English /ɹ/

The development of the child’s English rhotic, /ɹ/, is discussed here for the English WI /ɹ/ words in the child’s speech and the age of their first production, shown in

Table 3. The child’s new English words with word-initial /ɹ/ ¹.

Age	Vocabulary
2;7	rabbit, read, reading, red, restaurant, ride, room, run
2;8	rain, raining, reach, ready, repeat, write
2;9	race, real, refrigerator, remember, rice, riding, ring, ringing, rock, running
2;10	rest, roof, round, wrapper
2;11	recorder, remind
3;0	rail, rang, reindeer, rested, rose, rub
3;1	read (past tense), rinse, roller, roses, row
3;2	‘reds’, rooms, rosy, writing
3;3	rails, really, replace
3;4	rainbow, ridden, rollers
3;5	reached, remove, robber
3;6	raincoat, rapping, recognize, remembered, resting, road, runs
3;7	-
3;8	racing, rake, reality, rings, risen, river, rolling
3;9	racket, realize, regular, removed, repeating, return, rocks, “rinser”, rinsing, roll, rusty, wrap, wrapped, written

¹ The terms in single quotation marks reflect child-specific use.

.

There are 85 English word types for a total of 1397 tokens. The monthly tokens are 44 (2;7), 24 (2;8), 110 (2;9), 90 (2;10), 49 (2;11), 69 (3;0), 90 (3;1), 80 (3;2), 100 (3;3), 90 (3;4), 70 (3;5), 94 (3;6), 76 (3;7), 192 (3;8), and 118 (3;9). English /ɹ/ accuracy is very low throughout the child’s development and remains low until the end in the available data, as shown in Figure 2.

In terms of substitutions, WI /ɹ/ is realized mainly as [v, w, l, ɹ, ɾ].

Table 4. The child’s English /ɹ/ substitutions in read, rest, room.

Age	Read	Rest	Room
2;7	3 [v], 1 [w]	-	1 [ɾ], 1 [v], 15 [l], 1 [w]
2;8	1 [ɾ], 1 [v], 1 [l]	-	8 [l]
2;9	33 [v], 1 [l], 2 [Ø]	-	6 [l]
2;10	2 [ɹ], 4 [v], 1 [l]	2 [v]	11 [l], 1 [w]
2;11	-	1 [v]	1 [v], 6 [l], 1 [w]
3;0	2 [ɹ], 7 [v], 1 [l]	2 [v]	2 [l]
3;1	18 [v], 1 [w]	1 [v], 1 [l], 1 [n]	1 [ɹ], 7 [l], 2 [w], 1 [Ø]
3;2	1 [ɹ], 17 [v], 1 [l]	-	2 [ɹ], 2 [v], 11 [l], 2 [w], 1 [Ø]
3;3	2 [ɹ], 29 [v], 1 [l], 1 [w], 1 [Ø]	1 [v], 1 [l]	1 [ɹ], 2 [v], 1 [l], 1 [w], 1 [ð], 1 [m], 1 [Ø]
3;4	8 [v], 1 [w]	1 [v], 2 [l]	3 [v], 2 [l], 1 [w], 2 [ð], 1 [t], 1 [m], 1 [w]
3;5	3 [v], 2 [w]	3 [v], 2 [l]
3;6	10 [v], 4 [w]	2 [v], 1 [w]	1 [l], 6 [w]
3;7	1 [v], 5 [w]	1 [v], 1 [l] 4 [w]	1 [ɹ], 1 [ɾ], 2 [l], 9 [w]
3;8	2 [ɹ], 29 [v], 1 [l], 1 [w], 1 [Ø]		1 [ɹ], 6 [ɾ], 1 [l], 8 [w]
3;9	1 [ɹ], 7 [w]	2 [ɾ], 2 [w]	2 [ɾ], 2 [w]
3;10	1 [ɾ]	1 [ɾ]	1 [ɾ]
3;11	-	1 [ɹ], 4 [ɾ]	3 [ɾ]

Note: The numbers before the phonetic symbols denoting /ɹ/ indicate tokens.

zooms in on English /ɹ/ adult-like and substitution tokens in the words read, rest, and room during development, which demonstrate the variability in production patterns.

Figure 3 shows the child’s developmental production paths for English /ɹ/. Adult-like rhotic productions are shown in black (for /ɹ/→[ɹ]) and red (for /ɹ/→[ɾ]) solid lines, /ɹ/substitutions [l] and [v] are in dashed lines, and substitution /ɹ/→[w] is shown here with a solid blue line to indicate the dominant /ɹ/ substitution of English monolinguals (e.g., Smit 1993). Red is used for /ɹ/→[l], considered to indicate Greek-language specificity, and blue font is used for /ɹ/→[v, w], considered to indicate English-language specificity; see the related elaboration and reasoning in the Discussion section. We can see that the acquisition level is in a plateau/cyclic stage until about ages 3;6–3;7, when it progresses in an S-curve growth pattern towards adult-like use at 80% by 3;11, mainly in the Greek rhotic. Regarding substitutions, it is shown that /ɹ/→[v] is preferred over /ɹ/→[l]. Substitution [w] is evidenced during the progression stage, between 3;5 and 3;9, when the rate of other substitutions declines.

Less-frequent substitutions are also noted. It is interesting that /ɹ/→[t], which first appeared at 2;7, reappears near acquisition without evidence of it in-between during development. Another interesting occurrence is /ɹ/→[m, t, ð] ,which surfaces near acquisition only.

Table 5. Developmental stages in /ɹ/ substitutions.

	During Development (2;7–3;2)	Omissions Interface (2;9–3;3)	Near Acquisition (3;3–3;6)	Rhotic (3;10–3;11)
/ɹ/→[C]	v l w	Ø	m t ð	ɹ ɾ

differentiates substitutions in terms of distinct developmental stages, as follows: (i) in progress (2;7–3;2), (ii) omissions interface (2;9–3;3), (iii) near acquisition (3;3–3;6), and (iv) rhotic production (3;10–3;11).

4. Discussion

This section elaborates on the results, discussing whether the child’s rates of acquisition and the developmental patterns of the rhotic in her languages are influenced by the specific context of her exposure to her languages.

4.1. Understanding the Child’s Rhotic Productions

4.1.1. Accuracy Levels

At the start of the study (age 2;7), the child’s rhotic accuracy was very low in her languages: 5% (English) and 1% (Greek). By age 3;9, her Greek [ɾ] accuracy reached the 70% acquisition criterion (Figure 1), while her English [ɹ] remained low, at 20% (Figure 2). Compared to the monolingual and cross-linguistic norms discussed in the Introduction, the child’s Greek rhotic shows an acquisition level of 70% which is close to the recently reported monolingual norm of 75% acquisition by age 3;0 (Geronikou and Babatsouli 2024) and the cross-linguistic evidence for 75% tap acquisition by 3;0 (McLeod and Crowe 2018). Her English rhotic development does not contradict the monolingual norm of 75% acquisition at age 6;0 (Smit 1993; Smit et al. 1990), although cross-linguistic norms indicate 75% acquisition by 3;0, which is not supported by the evidence in this study.

It is inferred that the bilingual child’s targeted Greek rhotic, which is the rhotic with ambient environment input, was acquired within normative expectations. On the other hand, the targeted English rhotic developed very slowly, not attaining comparable acquisition rates by age 3;0, as per the cross-linguistic evidence on the acquisition of /ɹ/, although it can be inferred that it is still developing, given that English monolinguals acquire it by age 6;0. This outcome is explained as a result of limited exposure to English, compared to monolingual peers in English-speaking communities, since the bilingual child had primarily single-person, exogenous exposure to English in her daily interactions. Data from Smith’s (1973) well-known longitudinal study in English help support this inference; Amahl was raised in England and, despite maternal multilingual use in the home, he showed adult-like /ɹ/ accuracy as early as 2;9, matching both monolingual English and cross-linguistic expectations. These inferences are further supported by the evidence provided by Menke (2017), who reported a delay in articulatory mastery and the acquisition of the Spanish trill by the Spanish/English bilinguals in the study because the trill is not a shared phoneme in the languages of the bilinguals in the study.

Overall, the evidence shows the reverse effect, supporting previous findings that a phoneme’s higher statistical frequency and functional load in one of the languages positively impact the bilingual acquisition of a shared phoneme in the languages compared to the phoneme’s lower statistical frequency and functional load in the other language. This was reported for the Greek–English bilingual acquisition of /θ/, in which the bilingual child’s early acquired interdental in Greek (matching Greek monolingual norms) facilitated faster acquisition of the English /θ/ than in respective English monolingual norms (Babatsouli 2017). This, of course, is not what is found here for English-specific /ɹ/. Although the rhotic is more frequent in the targeted words in the child’s English speech (85 English word types, 1397 tokens) in the database, compared to her Greek rhotic targets (17 English word types, 245 tokens), the child did not achieve mastery of English /ɹ/ by the end of the study (Figure 2). Its slow acquisition reflects the child’s limited exposure to it (single-person and exogenous), i.e., its low statistical frequency in the child’s context-specific learning environment, where Greek [ɾ] is more frequent in the cumulative input the child is exposed to. These results back empirical statements that learning “operates over the situated learning environment, and not just sequences of sounds in utterances” (Frost and Monaghan 2020, p. 65), and that this learning manifests the specific child’s linguistic experiences (e.g., Lieven and Brandt 2011).

Interestingly, the child’s rhotic acquisition rates mirror the evidence of universal markedness effects in the world’s languages, i.e., the frequency of rhotic consonant types, whereby rhotic approximants are less frequent than taps (Maddieson 1984). By inference, the less-frequent sounds in the languages of the world are also more marked in acquisition (Greenberg 1963), and this is also explained in terms of articulatory complexity, since consonants with double articulation, like [ɹ], are cross-linguistically acquired later than less complex consonants, like [ɾ] (Ingram and Babatsouli 2024). Conclusively, a sound’s functional load in bilingualism (shared or not), should be viewed holistically, considering its phonemic prominence in the cumulative distributional frequencies in the child’s linguistic input across both languages and within each child’s specific learning context. This is similar to what is known for bilingual children’s vocabulary: the sum of words in both languages forms the bilingual child’s total vocabulary (e.g., Werker et al. 2009).

As reported previously, Bybee (2001) and Bybee and Beckner (2010) found that the relative infrequency of canonical variants in children’s lexical input will lead them to form stronger representations of the non-canonical variants available in the input. Although /ɹ/ is a canonical variant in the English-specific input, the bilingual child here has formed a strong representation of the non-canonical (as far as English is concerned) variant [ɾ] because it is predominantly available in her context-specific ambient environment input. This also holds for her /ɹ/→[l] developmental pattern, leading us to the discussion on substitutions in the next section.

Concerning the first research question, namely, “Do language-specific distributional frequencies and phonemic functional load impact the acquisition rates of the bilingual child’s targeted rhotics?”, the findings corroborate the prediction that the rhotic with the highest phonemic prominence, when viewed cumulatively across the two languages, was acquired earlier because of its high distributional frequencies in the child’s linguistic environment. In this case, it is the Greek /ɾ/.

4.1.2. Substitutions

Among the substitutions discussed next, v, m, t, l, and ð are shared phonemes/phones between the languages, while w is English-specific. The child’s developmental substitutions for the targeted rhotic sounds in her languages indicate both shared and language-specific patterns. Specifically, the Greek rhotic is overwhelmingly substituted by the lateral, [l], which is also the Greek monolinguals’ predominant substitution (e.g., Babatsouli and Geronikou 2022; PAL 1995). Its developmental path in Figure 1 graphically depicts the mirror effect of the dominant substitution [l] during development longitudinally, which is gradually replaced by adult-like production. The dominant substitutions of English /ɹ/, cumulatively for all the words, involve [v, l, w] (see Figure 3). Among these, /ɹ/→[v, w] are specific to her English rhotic productions, while the [l] is a shared substitution between the languages.

It is interesting to examine English substitution patterns along the developmental path, as was done for the Greek “lateral-to-tap” mirror effect. In the first two months, [l] competed with [v] for dominance, while [l] subsided thereafter. Specifically, the child produced either [v] or [l], depending on the developmental stage she was at and the word she was targeting, as is exemplified by the snapshot zooming in on /ɹ/ development in the frequently occurring words read, rest, and room (Table 4). Before age 3;5, her choice between [v] and [l] was lexically dependent: [v] was preferred in read, and [l] in room, while there was no preference found between [v] and [l] in rest. This is further evidence of lexical–phonological dependence in child developmental monolingual and bilingual speech (e.g., Ferguson and Farwell 1975; Kehoe 2011, 2015). Starting at about 3;5, she substituted /ɹ/ with [v] or [w] instead of [l], independently of the word targeted. Besides these main substitutions, there were also less-frequent substitutions evidenced near acquisition [m, t, ð]. These rare substitutions are attributed to regressive assimilation patterns in the word, as in /ɹ/→[m] in room priming word-final /m/, or in the utterance, priming /ð/ or /t/ earlier on, e.g., this is my room /ɹ/→[ð], there is my room /ɹ/→[ð], come /k/→[t] to your room/ɹ/→[t]. Similar patterns (onset /ɾ/ and /r/→[d, ð, j, l, h, other]) are reported for Spanish monolinguals and bilinguals by Kehoe (2018) and in Ball et al. (2001).

Returning to viewing the substitutions cumulatively for all the words (Figure 3), it is evident that /ɹ/→[v] was preferred over /ɹ/→[l] before 3;5, while a distinct preference for substitution [w] surfaced during the progression stage, between 3;5 and 3;9, when the rate of other substitutions declined. The existence of developmental stages of substitutions is further shown in Table 5, which provides evidence of systemic developmental variation in the child’s English rhotic production. Interestingly, this was not evidenced for her Greek rhotic. The greater phonetic variability of English /ɹ/ substitutions in the child can be explained as resulting from the variable input in the maternal English, the scarcity of exposure to English in the community, and the fact that English was the child’s weaker language in terms of both exposure and use, meaning that the child was processing phonemic representations that surfaced as phonetic variability. The phonetic variability found here matches the phonetic variability or approximations evidenced in other studies engaging bilinguals in different socio-phonetic and language dominance contexts, as reported in the Introduction (e.g., Kehoe 2018; Lease and Marchesi 2022; Menke 2017; Wode 1981), providing evidence that this is typical among bilinguals. The comparison with monolinguals in Kehoe (2018) showed that the bilinguals’ substitutions were more variable than those of the monolinguals in the study. The child in this study also showed more phonetic variability in her English rhotic substitutions compared to her Greek rhotic. Lastly, the child in this study also showed some Greek transfer in her productions of /ɹ/ as [l]. This substitution (i) may be instigated system-internally as interference from Greek, (ii) may be influenced by the interlanguage ambiguity in the mother’s input, where Greek [ɾ] transfers occasionally to her English, and/or (iii) d as the phonemic representation of the rhotic that has the highest cumulative distribution frequency in the child’s environmental input because of its enhanced prominence in the Greek dominant ambient language environment. The dynamics of such strategies invite further elaboration; this is tackled in Section 4.2 below.

4.1.3. Qualitative Remarks on Substitutions

It is known that within the class of liquids, laterals are acquired before centrals (e.g., Dinnsen 1992), so it is not surprising that Greek /ɾ/ is substituted by the earlier-acquired [l]. Regarding the child’s main English /ɹ/ substitutions, they mostly retain the targeted rhotic’s secondary place of articulation, Labial, in [v, w]; glide [w] also satisfies the [−consonantal] feature of /ɹ/, which also explains why [w] (an early acquired sound in English) is the predominant substitution of rhotic in English monolinguals (Smit et al. 1990). The bilingual child’s less preferred [l] also retains the primary place of articulation of /ɹ/ for Coronal. Her less frequent [t, ð] also adheres to Coronal, while [ð] (a later-acquired sound cross-linguistically, e.g., McLeod and Crowe 2018) also adheres to [+distributed] of the targeted /ɹ/. This means that there is nothing out of the norm in the developmental substitution evidenced in the bilingual child. It is worth comparing her developmental substitutions with the longitudinal developmental data of Amahl’s rhotic productions in Smith’s study (1973) (see

Table 6. Amahl’s developmental /ɹ/ substitutions (Smith 1973).

Age	/ɹ/→[C]
2;2	d, g, w, b, d
2;3	d, g, w
2;4	d, g, l, ɹ, w
2;5	d, v, l, g, ɣ, ɹ, j, w
2;6	ɹ, l, w
2;7	ɹ, l, w

). It is worth noting that, even after Amahl’s rhotic is mostly acquired, by 2;7, there are several tokens of /ɹ/→[l] at age 2;9.

We can conclude the following from Amahl’s detailed developmental data longitudinally: (i) while [w] is a common pattern, so are stopping to [b, d, g] and lateralization to [l]. A less frequent pattern is gliding to [j], which can be thought of as an earlier developmental variant, since /l/ is substituted by [j] before its acquisition (e.g., Smit et al. 1990). Substitution [b] early on, at age 2;2, results from regressive assimilation in rubber, as is substitution [ɣ] in ring at 2;5. Similar assimilations were noted in the present study near bilingual /ɹ/ acquisition. Smith (1973) argues that Amahl’s data are monolingual data, but those data show similar patterns of substitutions to those noted in this study.

The /ɹ/→[l] pattern: Transfer is a known psychological and cognitive phenomenon characteristic of all types of learning, in that the learner acquires new knowledge by utilizing previously acquired knowledge (Major 2024 and references therein). Ausubel et al. (1978) argued that even when something new has been learned, future learning also relies on transferred knowledge, which leads to the development of new cognitive structures.

While qualitative evidence of [l] as a rhotic substitution in English and other languages indicates this to be a developmental universal across languages, it is more prevalent, for instance, in languages with taps (like Greek and Spanish) (e.g., Geronikou and Babatsouli 2024; Kehoe 2018) than for targeted English /ɹ/ (e.g., Bernhardt and Stemberger 1998; Smit et al. 1990; Smith 1973). The argument here is that the statistical frequencies evidenced in the detailed longitudinal data provided suggest that the lateral is this bilingual child’s developmental transfer of Greek /ɾ/→[l] in the /ɹ/→[l] context. Thus, the statistical frequencies of the Greek rhotic in the Greek-dominant ambient environment (including her father’s Greek speech), together with the interlanguage ambiguity in the maternal input, where English /ɹ/→[ɾ] is an occasional transfer, explains why [l] is the second dominant substitution of her English /ɹ/. This transfer pattern is also evidenced surfacing in Amahl’s English /ɹ/→[l] pattern from ages 2;4 to 2;9, resulting from L2 rhotic input in his multilingual mother’s speech, which includes productive Standard Indian English trill [r] and Hindi [ɾ]. Similarly to the context discussed for the bilingual child here, Amahl’s frequent /ɹ/→[l], which persists even after his /ɹ/ is acquired, may be explained by interlanguage ambiguity in his maternal input, which will not be prevalent in a strictly monolingual English context. The persistence of [l] even after /ɹ/ was acquired indicates Amahl’s continuing efforts to master the maternal trill/tap to which he had less exposure.

The /ɹ/→[v, w] pattern: Nevertheless, /ɹ/→[l] is not the participant’s most dominant substitution. English /ɹ/ is overwhelmingly substituted by [w] in both monolingual (e.g., Smith 1973) and bilingual English development (e.g., Leopold 1949), keeping faithful to the target’s secondary place articulation, Labial, and also agreeing with [−consonantal]. Notably, the child preferred the substitution [v] over [l]. This is because she adhered to Labial as the secondary place of articulation in English /ɹ/. This indicates that there was enough evidence of English /ɹ/ in the maternal input for the child to create a phonemic representation. The reason why the child did not produce /ɹ/→[w] before age 3;5 is that her English /w/→[w] was not acquired before that age either; English /w/ was substituted mostly by [v], and also by [ʊ] near acquisition (Babatsouli 2015, 2024a).

Again, one should not rush to conclude that the patterns /w/→[v] and /ɹ/→[v], which are indicative of slow [−consonantal] acquisition, result from interlanguage ambiguity in the input. Statistical frequencies in the ambient linguistic input may explain the delay in the acquisition of [−consonantal] in targeted /ɹ/ and /w/ compared to respective monolinguals since, again, the bilingual child had less exposure to these English-specific sounds (Greek does not have such approximants, nor any consonants with double articulation) than a monolingual English peer. For this reason, the child’s preference is for [+consonantal] [v, l, ɾ]: it is more salient in the ambient input compared to [−consonantal] [ɹ, w], which is less salient in the limited English exogenous exposure. The /ɹ/→[v] substitution is also observed as a rare pattern in Amahl’s developmental productions, so it can be considered to be a universal substitution. It is rare in Amahl’s data because he had acquired [w]. On a related note, Kehoe (2018) showed a preference among German–Spanish bilinguals for more obstruent than sonorant substitutions, which resonates in the /ɹ/→[v/ pattern here. Lastly, another explanation for [v] over [w] in the bilingual child here is that the Velar place of articulation (as in [k, g, x, γ]) was acquired in this child; this is reflected in the delay in [w] production that has Velar as its primary place of articulation. This also explains why Amahl’s velar substitutions for targeted /ɹ/→[g, γ] do not surface at all in this bilingual child’s speech. Velar substitutions [γ, x] for the German uvular rhotic are further evidenced in the bilingual German–Spanish children, but not the monolingual children in Kehoe’s (2018) study. It is in a way fascinating that once Velar was acquired by the child here, [w] replaced all other previous substitutions, demonstrating once more that the child acquired the phonemic representation of English /ɹ/. In the final stages of acquisition, when Greek /ɾ/ is acquired, the child also begins producing English /ɹ/ as [ɾ] at 50% by age 3;9 and 75% by age 3;11. This further justifies the argument that her productions are dominated by the prevalence of [+consonantal] in the ambient environment, over weaker [−consonantal]. To sum up, the bilingual girl’s predominance of Labial substitutions for English /ɹ/ provides evidence that /ɹ/→[ɹ] was accurately represented (processed) throughout the bilingual child’s development.

Distributional frequencies: The data provide evidence that /ɹ/→[ɹ] was accurately represented (processed) throughout the bilingual child’s development, but it did not reach acquisition or near-acquisition status by the end of the available dataset affected by the statistically limited exposure to [−consonantal] in the speech around her. However, by the same token, her [+consonantal] rhotic was acquired. The data here also show that while the English rhotic has a high distributional frequency in the language compared to Greek (as reported in the introduction), its phonemic functional load in the bilingual child’s English input was very low, thus leading to minimal progress and eventual failure to meet acquisition criteria by the end of the available data.

As shown in Section 4.1.1, given that /ɹ/ is acquired late in monolingual English and cross-linguistically, it is reasonable to assume that the sound is still developing in this child, and that Greek /ɾ/ is just another developmental substitution for English /ɹ/ that is available to the child, dictated by the following two influences: the distributional frequencies of [+consonantal] in her ambient environment, and universal rhotic distributional frequencies across languages. After all, alveolar trills and taps are more frequent (45% each) than approximant (8%) and fricative (2%) rhotics in the languages of the world (Maddieson 1984). The bilingual evidence in Kehoe (2018) shows earlier acquisition of the German uvular trill compared to the Spanish tap and trill. This does not mean, however, that bilingual children may not show patterns of earlier acquisition (e.g., Ball et al. 2001), as we saw for Amahl’s reported mastery of /ɹ/ by age 2;7 (Smith 1973).

Regarding previous reports in L2: The following findings reported in the introduction on the acquisition of a second language are supported by evidence in this child’s data:

○

L1 rhotic skills influence subsequent language rhotic acquisition, as well as showing transfer effects (Reyes et al. 2017).

○

Variable input is used in experimental settings to train difficult phonemic contrasts in a non-dominant language (e.g., Logan et al. 1991). The findings of the present study do not contradict this experimental result and may serve as an example of how experimental findings may be relevant to “special populations” (p. 977). The English input the child received from her mother was variable and sufficient to permit accurate discrimination of the English rhotic, as shown by the usage distributions of her developmental substitutions in each language.

○

Phonetic input variability does not exclusively have a facilitative influence, in that positive effects of input variability may be limited to native language contrasts dominating (Colantoni and Steele 2018), as observed in the dominance [+consonantal] in the bilingual’s phonetic patterns. This has been explained as resulting from an interplay of factors, including challenges in perception affected by statistical distribution tendencies in the input and challenges in differentiating between transfer and response to input.

○

The relative infrequency of canonical variants (as for L1 English /ɹ/ in this study) is expected to lead to children’s stronger representations of non-canonical variants (as for [ɾ] substituting /ɹ/ in this study), based on an exemplar-based model of lexical representation (Bybee 2001; Bybee and Beckner 2010).

4.2. Understanding the Developmental Path

We have shown in the Results section that the progression path of sounds that are acquired, as shown for [ɾ] and [w] (Figure 1 and Figure 3), follows a logistics curve (or function) of growth that corresponds to Ebbinghaus’ (1913) forgetting and learning S-curve, as it relates to the experimental study of memory. Bills (1934) elaborated on this with a more detailed description of learning curves identifying three properties: negative acceleration, positive acceleration, and plateaus. These distinct stages are evidenced in the bilingual child’s data in the acquisition progress of [ɾ] and [w], although the comparison of the developmental substitutions for English /ɹ/ also shows interesting patterns longitudinally, when viewed both cumulatively (Table 5) and schematically (Figure 3). Wode (1981) speaks of gradual approximations and jumps in the development of the L2 rhotic when the L1 is acquired. The earlier developmental data in this study, showing a developing /ɹ/ in the L2, demonstrate actual developmental stages in the substitutions: (i) in progress with [l, v, w] (2;7–3;2), (ii) omissions interface (2;9–3;3), (iii) near acquisition with [m, t, ð] (3;3–3;6), and rhotic production with [ɾ] and minimal [ɹ] (3;10–3;11).

Dense and uninterrupted developmental data allow the deciphering of developmental processes longitudinally, especially given the sparsity of such studies in either first (child monolinguals/bilinguals)- or second (adult)-language acquisition. Major’s (2014, 2024) Ontogeny Phylogeny Model (OPM) is a theoretical stance on the relationship between L1 and L2 longitudinally. The model was originally advanced to explain L2 speech in adults (meaning ages after 4;0–5;0 when L1 phonological systems are acquired cross-linguistically (Babatsouli 2024c; Ingram 1989; McLeod and Crowe 2018). Nevertheless, given that L1 acquisition (protolanguage) and L2 acquisition in adults (interlanguage) share developmental characteristics, such as speech universals (Us) (e.g., Archibald 1998; Babatsouli and Ingram 2018; Dinnsen 1992; Greenberg 1963; Major 2014, 2024), the OPM has significant implications for understanding bilingual phonological development longitudinally as well. Several L2 theories state that dominant language transfer is strong because the non-dominant language is not mastered yet, especially in the early stages of acquisition (e.g., Flege 1995; Major 2014, 2024).

Before discussing how OPM corollaries help to interpret the child’s rhotic productions longitudinally, a clarification is required. While all substitutions here are, by default, developmental universals (see earlier discussion), usage frequencies indicate language-specific tendencies. Thus, it is argued that [l] is dominant language transfer (hereon an L1 for the following discussion on OPM), [v] is a developmental universal approximating [w] (here in the child’s non-dominant language, or L2 for the purposes of OPM discussion), and [w] is the English-specific developmental pattern.

OPM Chronological Corollary: The concept of transfer is also incorporated in the OPM’s Chronological Corollary, “L2 increases, L1 decreases, and U increases and then decreases”, which states that L1 is so impactful on the L2 production in early stages that it prevents the surfacing of either L2 productions or Universals. This is evidenced here in the substitutions of English /ɹ/. Between 2;7 and 2;9, as the Greek-influenced [l] is very dominant (see Figure 3), it prevents the surfacing of L2 and universal (evident in substitute [v]) in the first stages. As the input in the L2 is processed and disambiguated, the child perceives the “erroneous” pattern [l] and replaces it with a universal, [v], which is shown to be a gradual approximation of both the common monolingual substitution [w] and the targeted /ɹ/. Gradual approximations in the developmental substitution of rhotics are reported to be common in bilinguals, as was inferred earlier, based on findings in the literature.

The evidence here permits the reinterpretation of this chronological corollary to explain the child’s variable substitutions in English. Following the initial bout of transfer, the child’s L2 English substitutions, [v, w], became more prevalent (“L2 increases”), her L1 Greek substitution [l] lost its strength (“L1 decreases”), and U increased and then decreased, as seen in the [v] and [w] before the surfacing of [ɾ]. The OPM assumes that, in the ideal scenario, L2 will attain native-like skill, but this is often stated as the optimal outcome. We have seen that this is not corroborated in the dataset. We know that statistical frequency is often summoned in the OPM, and elsewhere, as influential (Archibald 2021; Eckman 1977; Jakobson [1941] 1968; Trubetzkoy [1939] 1958; also see Introduction). As described earlier, this is also considered to have affected this child’s adult-like rhotic, which ended up being produced as transfer from the dominant Greek by the end of the available data. The sudden bouts from one substitution to another along the developmental path were qualitatively reported by Wode (1981) as jumps in production.

OPM Similarity Corollary: Similarity is known to impede the acquisition of new sounds in an L2 when they are similar to an L1 sound (e.g., Flege and Bohn 2021). The two rhotics in the child’s languages are similar sounds, both being coronal liquids, and this similarity component influences their perceptual salience. The OPM similarity corollary states that “L2 increases slowly, L1 decreases slowly, and U increases slowly and then decreases slowly”. We have evidenced this in the interpretation of the previous, more holistic, corollary, except that here, similarity between sounds slows down the speed of progression from one phase to the next.

While the two rhotics are similar, English /ɹ/ is more complex (phonemically and phonetically), having more distinctive features that are reflected in its articulation: it requires both lip involvement (rounding) and tongue involvement (which varies by dialect) while the airflow is distributed, and there is no point of contact between the articulators. In terms of acquisition level, the highly frequent (across languages) tap is acquired by the child, while the less frequent (in languages) approximant is not. According to the similarity corollary, while the L1 prevails in acquisition early on, when the sounds are similar, L1 persists for longer than for normal phenomena under the chronological corollary, thus showing that similar sounds are more difficult to acquire.

OPM Markedness Corollary: Distributional frequencies are included in the various accounts on markedness; as discussed in the Introduction, infrequent grammatical constructs (Greenberg 1963) in the languages of the world tend to be more marked (meaning more difficult) to acquire. The OPM’s Markedness Corollary states that, for marked sounds, “L2 increases slowly, L1 decreases rapidly and then decreases slowly, and U increases rapidly and then decreases slowly”. English /ɹ/ is more marked than Greek /ɾ/, as also corroborated by the acquisition-level data. The markedness corollary is considered next for the developmental substitutions of /ɹ/. The substitution [v] did not surface clearly until age 2;10 (“L2 increases slowly”), her [l] substitution changed from 60% between 2;8 and 2;9 to 10% at 2;10 (“L1 decreases rapidly”), remaining at about 10% to 20% for several months (“and then decreases slowly), while her [w] (considered a universal for English monolinguals) shot up from 10% at 3;5 to 80% at about 3;9 (“and U increases rapidly”), subsequently falling to 0% at 3;11 and, lastly making a final jump by age 4;0 (“and then decreases slowly”).

With regard to the second research question, namely, “What do the child’s targeted rhotic productions in the languages during her development longitudinally reveal about the bilingual acquisition path?”, the findings corroborate the prediction that the developmental path of the child’s rhotics in the languages is representative of the context-specific nature of her exposure to her languages.

5. Conclusions

Research in developmental bilingualism is relevant to language acquisition theory, carrying the potential to inform both child and adult acquisition patterns in the span of time. Dense longitudinal data framed within specific acquisition contexts in terms of input and usage frequencies also have the potential to shed light on the dynamics of developmental paths longitudinally. Such information is, by and large, under-represented in the literature to date. The present study on a child’s Greek/English bilingual acquisition of rhotics has contributed to this line of work, espousing a cross-disciplinary perspective. The results here provide evidence of context specificity in the effects of the statistical frequencies of speech sounds in the bilingual child’s environmental input. Her Greek /ɾ/ was acquired similarly to monolingual norms due to its cumulatively higher prominence in the child’s overall linguistic environment, that is, due to its higher exposure to the ambient environment input. On the other hand, the child was still developing her English /ɹ/ at the end of the study because of its lower frequency in her cumulative environmental input. Her developmental substitutions for the English rhotic were more variable than for the Greek rhotic, reported to be characteristic of bilingual acquisition in a non-dominant language. This variability indicates phonemic processing of the variable input to the non-dominant language, provides evidence of phonemic functional load at the level of distinctive features, and indicates that the child has acquired the phonemic contrasts of her targeted rhotics, although she still lacks the articulatory skill for /ɹ/, targeted in her non-dominant language. Her delay in the acquisition of the English rhotic relates to her single-person exposure to English, the exogenous setting of acquisition in English, and universal language patterns. Lastly, the findings here illustrate the existence of dynamic paths and distinct developmental stages in the bilingual child’s early phonological development longitudinally that have been used to reinterpret second-language acquisition theory to decipher the early workings of the non-dominant language in child developmental speech. The child’s usage frequencies of rhotic substitutions across seventeen months of development (ages 2;7–3;11) indicate language specificity and help illustrate the dynamic nature of the status of her languages longitudinally. The single-subject longitudinal case study design was an experimentally arduous task, however, future research should explore this approach for the replication of the evidence presented here and further validation of the theoretical interpretations made.