11.2.1 Interlocutor Sensitivity

Reference Quay, Cenoz and GeneseeQuay (2001) studied the child Freddy, born and raised in Japan, with a German-speaking father and an English-speaking mother. Regular exposure to Japanese started at 0;11 when Freddy began attending a Japanese childcare center. The parents spoke German to each other and were proficient in Japanese. Freddy’s perception and production skills in English were better than in the other languages. He did not differentiate languages in either the German or English context, using Japanese freely instead. However, the parents also accepted Japanese language mixing and mixed Japanese elements into their own language, which might have affected the child’s behavior. Freddy’s case is quite different from that of Xiaoxiao (Reference QuayQuay, 2008), who also grew up in Japan with two minority languages at home: English from her father, Mandarin Chinese from her mother. She was exposed to Japanese at day care from five months of age onward. Quay analyzed thirty-minute video recordings during dinner table conversations, covering the ages between 1;10 and 2;4, and focusing on the two minority languages spoken in the home. Xiaoxiao selected her language not only based on the languages spoken to her but also based on her interlocutor’s proficiencies, thus showing interlocutor sensitivity. She used more Chinese than other languages with her mother, more English than other languages with her father, and mostly Japanese when both parents were present, because Japanese was the language (other than English) that both parents shared.

Reference MontanariMontanari (2009a) investigated language choice in a Tagalog–Spanish–English trilingual child, Kathryn, longitudinally between 1;10 and 2;4. The author wanted to know whether there is evidence for interlocutor sensitivity when users of different languages are present, and whether the choice of the non-target language can be explained by lexical gaps or the interlocutors’ bilingual strategies (e.g., whether they encouraged mixing). Kathryn was born in Los Angeles to a Filipino-American mother and a Chilean-American father who spoke to her in Tagalog and Spanish, respectively. Input in English was provided through Kathryn’s older sister and outside the home. Until 2;2 Kathryn was exposed mostly to Tagalog (48%), and to a lesser extent to Spanish (29%) and English (23%). From 2;3 she started attending an English day care and, as a consequence, relative exposure to English increased to 50%. Her total number of words in each language matched relative exposure. At 1;9, 40.3% of Kathryn’s words were Tagalog, 36.1% were Spanish, and 23.6% English. The relative share of English vocabulary increased over time, as did exposure to this language. According to MLU measures, too, Tagalog was initially her dominant language. Measuring which language the girl selected when confronted with different language users showed that she differentiated the languages from the earliest sessions: she used more Tagalog with her mother, she increased her English use with the English-speaking interlocutor, and used more Spanish in the Spanish context. Kathryn did not strictly adhere to the interlocutor’s language, however, and this was related to her relative proficiency in the three languages. For instance, during the earlier stages, when Tagalog was very strong, there was more mixing in the non-Tagalog contexts. Toward the later recordings, mixing in English decreased as her English proficiency increased. While the interlocutors mixed rarely (the Tagalog-speaking mother slightly more than the English and Spanish interlocutors), the child showed mixing in all three languages but less in Tagalog, indicating that proficiency was a stronger determinant of her mixing behavior than parental mixing behavior. Overall, the study shows early language differentiation, but also the ability to do so in a context where speakers with three different languages are present at the same time.

Reference Nibun and WigglesworthNibun and Wigglesworth (2014) looked at language choice from as early as between 1;4 and 1;9. The study follows Michael, who grew up in Australia with a German-speaking father and a Japanese-speaking mother. Michael’s estimated exposure to Japanese (61%) was higher than to German (33%). Exposure to English started at 1;7 twice weekly but was estimated to be minimal (and was not investigated explicitly). The analysis, based on audio and video recordings, showed that at 1;4 Michael already differentiated language contexts, predominantly using the language of the interlocutor. He abided by the context language 60–75% of the time, despite an obvious dominance in Japanese due to more exposure to this language. Over time, mixing increased in the German context, consisting mostly in lexical borrowings, while decreasing in the Japanese context.

Reference EinfeldtEinfeldt (2022) investigated the acquisition of a trilingual child, Tommaso, growing up in South Germany with Italian (language of the mother), Swiss German (language of the father), and German (language of the community). The child has been observed on a regular basis between 1;4 and 2;8 in each of the three languages. The Italian recordings were done with the mother, the Swiss German ones with the father, and the Standard German ones with the researcher. Figure 11.1 shows the number of utterances in each recording by language context (dots represent German utterances; dark grey, Italian; medium grey, Swiss German; light grey, utterances containing elements from more than one language). This study shows an impressive case of trilingual language separation from an early age. Most mixing occurred from both home languages into the community language (CL), which was the language developing most slowly during the second year.

Figure 11.1 Language choice in the trilingual child Tommaso (German, Italian, Swiss German).

(Reference EinfeldtEinfeldt, 2022)

Overall, all case studies reviewed in this section demonstrate that trilingual children show interlocutor sensitivity from early on, just like bilingual children do. However, the degree to which trilinguals use their interlocutor’s language may be compromised by their proficiency in each language (Reference Quay, Cenoz and GeneseeQuay, 2001; Reference MontanariMontanari 2009a; Reference EinfeldtEinfeldt, 2022), the interlocutor’s behavior (Reference Quay, Cenoz and GeneseeQuay, 2001; Reference ChevalierChevalier, 2013), and possibly even the interlocutors’ proficiency (Reference QuayQuay, 2008).

11.2.2 Code-Mixing

Analyses of language choice, as illustrated in Figure 11.1, typically include quantifications of mixed utterances, and a distinction is drawn between inter-utterance (intersentential) mixing and intra-utterance (intrasentential) mixing. The former means that entire utterances (including one-word utterances) are produced in a language that is not the one spoken in the context or by the interlocutor. Intra-utterance mixing, by contrast, refers to utterances containing elements from more than one language (represented in light grey in Figure 11.1). Percentages of intra-utterance mixing vary across children, language contexts, and age range (as mixing tend to decrease with growing age).Footnote ³ Examples are provided in Table 11.2.

Some studies on trilingual children were exclusively dedicated to the analysis of language mixing. Reference Stavans and SwisherStavans and Swisher (2006) investigated mixing in two trilingual children, M and E, who acquired Spanish (father), Hebrew (mother), and English (CL) simultaneously from birth. English was the children’s dominant language. Data were collected over twenty months from age 2;6 to 4;2 (M) and 5;5 (E), and mixed utterances were drawn from thirty-two hours of spontaneous conversation. The children mostly responded in English when being addressed in their home languages, possible because English was the children’s dominant language and because both parents were fluent in English. Most intrasentential mixes occurred between two languages only, but some cases also involved three languages and even violation of the morphosyntactic boundaries; for example, the arones (the closets) with an English determiner, a Hebrew stem (aron), and a Spanish plural suffix (-es). According to the authors, these are indications of a “unique trilingual competence development that capitalizes on multiple languages” (217). The children exploit the formal properties of language, and thereby achieve “economic efficiency” in communication.

Reference Poeste, Müller and GilPoeste, Müller, and Arnaus Gil (2019) wanted to know whether code-switching is related to language dominance. The authors investigated data from 122 children with more than one language from birth, including fifty-one bilinguals (mean age 4;9), sixty-two trilinguals (mean age 4;7), and nine quadrilinguals (mean age 4;5). The data were originally collected to elicit properties unrelated to code-switching (e.g., adjective placement, subject placement, finite verb placement in German). It seems that not all children were tested in all of their (minority) languages, but dominance was assessed for as many as 107 children based on vocabulary and accordingly children were classified as balanced, unbalanced, or unknown. Mixing rates were low overall (7% total); trilinguals (8.8% mixing) and multilinguals (9.3% mixing) showed higher rates than bilinguals (4.3% mixing). Interestingly, only 45% of all children showed mixing at all, and among these, half were balanced and half unbalanced. This led the authors to the conclusion that language dominance does not determine mixing, contrary to earlier observations by Reference Quay, Cenoz and GeneseeQuay (2001) and Reference MontanariMontanari (2009a). Intersentential mixing was more frequent between typologically close languages and decreased noticeably once Catalan–Spanish speaking children were excluded. Overall, the study shows comparatively low mixing rates for trilinguals (as for bilinguals). However, the authors’ conclusion that dominance and mixing are unrelated might have been compromised by the fact that not all trilingual children were examined in all of their minority languages. Moreover, while the study makes a strong case for the role of typological proximity, it will be interesting to investigate in future studies whether this finding is generalizable to typologically close languages that are not spoken within the same community, unlike Catalan and Spanish. An example would be trilinguals, who speak two related dialects as home languages and a typologically different CL (Profile 5, Table 11.1) − a so far understudied scenario in 3L1 acquisition. Another case in point is the following study by Reference Sivakumar-Thiyagarajah, Müller, Arnaus Gil, Suárez-Gómez and Guijarro FuentesSivakumar, Müller, and Arnaus Gil (2020).

Reference Sivakumar-Thiyagarajah, Müller, Arnaus Gil, Suárez-Gómez and Guijarro FuentesSivakumar and colleagues (2020) examined language mixing in a trilingual boy (Diego), who grew up speaking Italian and Spanish as minority languages in France. The child spoke French and Italian as his stronger languages (measured based on vocabulary) and Spanish as his weaker language. Like all previous studies, Reference Sivakumar-Thiyagarajah, Müller, Arnaus Gil, Suárez-Gómez and Guijarro FuentesSivakumar and colleagues (2020) found evidence for language separation according to context. Mixing was most frequent in the child’s weaker language (Spanish) and happened primarily between Italian and Spanish. At first brush, the results seem to underline the role of typological proximity in language mixing. However, since French was the CL, an alternative explanation is that the child rarely heard mixing into French.

In summary, studies of language choice and code-switching in children acquiring two minority languages in a country where a third language is spoken show that (1) children differentiate their languages by discourse even before the age of two years, (2) children vary in terms of interlocutor sensitivity and are more likely to use their parents’ minority languages if the parents refrain from using the community language in the home, and (3) the amount of exposure to each language (and, by implication, dominance and relative proficiency), the caregiver’s abilities and language strategies, and typological distance additionally determine the amount and directionality of mixing.

11.3.1 Phonology

Reference Yang and HuaYang and Hua (2010) studied the inventory of phonemes in child D, who grew up in Paraguay with Paraguayan Spanish, Mandarin, and Taiwanese. His mother addressed D mostly in Mandarin (and sometimes in Taiwanese), Spanish was D’s father’s language and the family language, while the Taiwanese input was provided by the grandmother. Audio and video data were collected during playtime, covering the ages between 1;3 and 2;0. This resulted in 184 Spanish words, 163 Mandarin words, and 72 Taiwanese words. The authors were interested in language interaction, potential effects of relatedness between the languages, and differential input. Although D’s language production was predominantly in Spanish (54% of his utterances), by 2;0 D was able to produce all sounds but /r/ in Spanish and all sounds but /p^h/ and /g/ in Taiwanese. Given the child’s divided input space, this is impressive, even if five sounds were still missing from Mandarin and not all sounds in all languages were produced regularly yet (113). There were similar error patterns across the three languages. For instance, final consonant deletion and deaspiration, or replacement of /f/ by /x/, were found in more than one language, but the attested patterns were comparable to those of monolinguals reported in the literature. These similarities to monolingual children suggest that limited amounts of input in each language do not impede phonological development. However, more input does not necessarily lead to faster acquisition, because the Taiwanese sound inventory developed faster than the Mandarin sound inventory although D had the least input in Taiwanese. The authors speculated that the development of Taiwanese may have been helped by the typological similarity between Mandarin and Taiwanese. The study shows that the child differentiated the phonologies of three languages from early on, while there was also interaction between the languages.

Reference MontanariMontanari (2011) studied the phonological systems of Kathryn at the age of 1;10. Kathryn grew up with a Chilean Spanish-speaking father and a Tagalog-speaking mother in an American English-speaking environment. Kathryn heard English regularly from her older sister and from family conversations. At 1;10 her estimated relative exposure to Spanish was 29%, to Tagalog 48%, and to English 23%, which roughly matched her word types at that age (see Section 11.2.1). Data were collected in a setting where all languages were present. The produced consonants were coded for whether they matched the target sounds in terms of organ (oral articulator involved in producing the segment), location (place of articulation), degree (manner of articulation), and glottis gesture (voicing). The data show different developments across languages. In Tagalog and English (compared to Spanish), words were produced more faithfully with regard to the organ and location involved in the production of the target sounds. In Spanish and English (compared to Tagalog), the target sounds were produced more faithfully with regard to manner of articulation. In terms of voicing, Kathryn performed better in English than in the other two languages. Finally, the type of organ errors differed across languages. In English and Spanish, they concerned exclusively coronal targets, while in Tagalog, they involved velar targets (although coronal and velar sounds are present in all three languages). Thus, the types of errors and the phonetic inventories suggested that the child was at distinct stages of phonological development in each language. Kathryn further produced some language-specific phonemes that went beyond the average monolingual inventory, suggesting heightened attention to phonological properties.

Reference Mayr and MontanariMayr and Montanari (2015) studied two sisters, aged 6;8 and 8;1, who grew up in California with an English-speaking father, an Italian-speaking mother, and a Mexican Spanish-speaking nanny (who did not learn any English or Italian). The estimated input distribution was 24% of English, 33% of Spanish, and 43% of Italian during the girls’ first four years. From age 6;0 onward, as the girls attended an Italian-English dual language program, estimated exposure shifted to 46%, 16%, and 38% respectively. The study focused on word-initial voiceless and voiced stops. The stop systems of Italian and Spanish overlap, as both are voicing languages with prevoiced /b d g/ and short lag voice onset time (VOT) for /p t k/. However, in Mexican Spanish, word-initial /d/ can be lenited, yielding an approximant [ð̞]. English, by contrast, is an aspiration language, with /p t k/ produced in the long-lag VOT range (with aspiration) and /b d g/ in the short-lag VOT range (although prevoicing may occur). Stop sounds were elicited in three recordings per language over a period of two months. The child’s English stops were target-like, but in the girls’ two minority languages they differed in spite of the overlapping properties. While Italian stops were subject to crosslinguistic influence (CLI) from English, Spanish stops were unaffected by English: voiceless stops were generally produced with short-lag VOT and voiced stops were lenited. Thus, the children developed different stop systems for Italian and Spanish despite the similarities between these two languages. The authors speculated that these differences between Italian and Spanish can be explained based on the contexts in which the languages were acquired. The children heard a lot of Italian in an English context and they had a lot of contact with Italian speakers who also spoke English, while their only input source for Spanish was the nanny, who did not speak any English.

Reference EinfeldtEinfeldt (2022) has provided the most detailed case study of trilingual phonological development, covering the ages between 1;4 and 2;6 in Tommaso, who is growing up with Standard German, Swiss German, and Italian (see Section 11.2.1). One part of the study focused on consonant inventories, which differ substantially between the three languages. Here, the child showed mostly monolingual-like developments with some evidence for acceleration and transfer. Another focus was VOT. While all three languages have binary stop systems, they are based on different parameters: German distinguishes long-lag VOT (fortis stops) from short-lag VOT (lenis stops), Italian distinguishes short-lag VOT (fortis stops) from prevoicing (lenis stops), while the Swiss German stop system is based on closure duration (CD) (all stops have short-lag VOT but fortis stops have a longer CD). At 2;4 Tommaso made monolingual-like distinctions in German and Italian. In Swiss German, all stops were produced within the short-lag range, although the CD contrast had not been acquired. This is typical for Swiss German children, who do not acquire this contrast before age 4;0. The third focus were geminate consonants in Italian. The contrast between short consonants (singletons) and long consonants (geminates) is absent in German, which makes this phenomenon potentially susceptible to CLI from German to Italian. However, Tommaso distinguished geminates from singletons already at 1;4. Overall, all analyses showed evidence for language separation from early on. The development was largely monolingual-like, with some evidence for acceleration and transfer, but no delays.

Overall, the studies on trilingual phonology conform to the idea of early differentiation (Reference Yang and HuaYang & Hua, 2010; Reference MontanariMontanari, 2011), albeit with some interaction (Reference Yang and HuaYang & Hua, 2010; Reference Mayr and MontanariMayr & Montanari, 2015; Reference EinfeldtEinfeldt, 2022). There seem to be no delays compared to monolingual children, suggesting that acquiring three languages is not necessarily more costly than acquiring two. Cases of interaction may be triggered by typological proximity (Yang & Hua, 2019), and cases of separation may be triggered by different contexts of use (Reference Mayr and MontanariMayr & Montanari, 2015).

11.3.2 Vocabulary

Multilingual children have to acquire two or more labels for the same referent. This goes against Reference Clark and Mac WhinneyClark’s (1987) Principle of Contrast, which says that any difference in form goes along with a difference in meaning. Linked to this principle, research in the 1970s (e.g., Reference 294Volterra and TaeschnerVolterra & Taechner, 1978) proposed that bilingual children initially have a fused lexicon, which includes words from both languages but no translation equivalents (TEs). Contrary to this idea, however, research on bilingual children has convincingly shown that children do learn translation equivalents from early on. Trilingual children face the additional challenge that they have to acquire three labels for the same referent.

Reference MontanariMontanari (2010b) studied lexical differentiation in Kathryn, who has grown up with Spanish, Tagalog, and American English. The author reconstructed the child’s vocabulary between 1;4 and 2;0 by means of diary data and audio recordings of spontaneous speech, collected in a setting where all three languages were spoken. She analysed the cumulative appearance of words (all words in all languages that Kathryn knew at each monthly interval) and conceptual vocabulary (TEs counted only once, as the conceptual representation for a TEs is assumed to be the same across languages). Moreover, she distinguished doublets (concepts for which the child has TEs in two languages; e.g., Ta. susi−Sp. llaves “keys”), triplets (concepts for which the child has TEs in three languages; e.g., Ta. aso−Sp. perro−En. dog), and neutral terms (words like okay, existent in all three languages). Results showed that the child’s cumulative vocabulary in each language grew more or less at the same rate and neutral terms grew somewhat faster, suggesting that words that are identical or similar in a multilingual child’s languages are acquired earlier. The conceptual vocabulary, which is best suited to compare lexical knowledge in multilinguals and monolinguals, showed that Kathryn’s overall lexical development was a bit slower in the beginning but otherwise comparable to that of bilinguals and monolinguals in other studies. TEs were produced from early on and the percentage increased with the gradual increase of new words. The number and types of TEs seemed to depend on the amount of input heard, as doublets with Tagalog, to which the child had more exposure, were most frequent. Triplets were less frequent than doublets, possibly due to a higher number of cognates between English and Spanish. Kathryn could also use equivalents with different language users from 1;6 on, providing robust evidence for lexical differentiation.

In summary, there is evidence for language separation from vocabulary acquisition, and the number of doublets grows faster than that of triplets. With that said, the number of systematic studies comparable to that of Reference MontanariMontanari (2010b) are still scarce.

11.3.3 Syntax

There is evidence from bilingual children for differentiated syntactic development from the moment that children start to produce their first two-word utterances (see Reference De Houwer, Fletcher and MacWhinneyDe Houwer, 1995; and Reference Meisel, Hyltenstam and OblerMeisel, 1989, Reference Meisel2011 for overviews). Comparable studies of 3L1 children are rare.

Reference MontanariMontanari (2009b) studied Kathryn’s early word combinations between 1;7 and 2;1 (MLU less than 1.5). Kathryn is the aforementioned child who acquired Spanish and Tagalog in the home and English outside the home. The frequencies of particular word order sequences in her speech were calculated in terms of whether Arguments (Arg) precede or follow Predicates (Pred). English is an SVO language (i.e., Arg+Pred). In Tagalog, word order is relatively free, but in the unmarked case, the verb comes first, resulting in either VSO or VOS (i.e., Pred+Arg). In Spanish, word order is determined by both grammatical functions and information structure, resulting in SVO Arg+Pred or VOS (Pred+Arg). Thus, all languages differ from each other in their word order patterns, which is an ideal scenario to investigate language differentiation. Kathryn’s earliest constituent combinations indeed mirror the language-specific patterns described. In Tagalog, Pred+Arg prevails over Arg+Pred (73% Pred+Arg/27% Arg+Pred). In English, the reverse pattern was found (20% Pred+Arg/80% Arg+Pred). In Spanish, the ratio of the two orders is more balanced (40% Pred+Arg/60%Pred+Arg). These ordering preferences are also mirrored in the interlocutors’ speech (Pred+Arg order: Tagalog 76%, English 5%, Spanish 41%). Thus, not only did the child show language differentiated patterns, but also the exact patters of the adult language.

Reference Devlin, Folli, Henry and SevdaliDevlin, Folli, Henry, and Sevdali (2015) studied a trilingual child, S, acquiring English, Italian, and Scottish Gaelic from birth. The mother interacted with S in Italian and the father in Scottish Gaelic. English was the language of the family, of the community, and at the nursery (attended by S from the age of 0;5). English was also S’s dominant language according to MLU measures.Footnote ⁴ Data from between 2;3 and 3;7 showed that the child “doubled” the pronoun it by a full DP object, as in I like it, the carrot, and she dislocated subjects to the right and left periphery of the clause, as in He’s cross, daddy (he and daddy being co-referent). These constructions are ungrammatical in English, but the corresponding equivalents in Italian are grammatical. The authors argued in favour of CLI from Italian, where dislocations are frequent (e.g., L’ho letto, il libro [literally: It I have read, the book]). Data from the other languages were not presented, so it remains unclear whether the child produced the corresponding structures in Italian. Reference Devlin, Folli, Sevdali, De Angelis, Jessner and KresicDevlin and colleagues (2017) studied the same child aged 5;4–5;7 with respect to the unaccusative/unergative distinction in Italian and in English. Italian differs from English as the unaccusative/unergative distinction is marked by means of auxiliary selection. In English, the distinction is not marked on auxiliaries, but it is visible in past participle and resultative constructions. Results show that S had acquired the distinction in both English and Italian. However, her Italian developed rather slowly and the ability to select auxiliaries according to the aforementioned distinction lagged behind monolinguals. Some miscategorization occurred and there was some overlap between the lexical verbs that S miscategorised in English and in Italian. The authors argued for CLI from Italian, the child’s weaker language, into English. The role of Scottish Gaelic was not discussed.

Reference KazzaziKazzazi (2011) provided a qualitative analysis from two trilingual children (Anunsheh and Irman) who grew up in Germany with Farsi and English as their home languages. German was the children’s dominant language and English the weakest. Data for this study are based on written notes (covering about four years for one child and ten years for the other). One of the structural properties the author investigated were modified NPs, including determiner-noun sequences and compounds. English and German are premodifying languages, while Farsi is postmodifying. The data showed occasional influence from German to Farsi resulting in premodification in Farsi, but cases where the Farsi structure “won out” over the German structure were far more common. This resulted in compounds, such as Ge. Schuhehaus (Hausschuhe “slippers”), Ge. Keksebutter (Butterkekse “shortbread”), Ge. Feuerlager (Lagerfeuer “campfire”) or En. mix-caugh (“cough-mixture”). The author argued that the Farsi structures “cater in a particular way to certain general cognitive tendencies such as iconicity and transparency,” which is why they are transferred more easily, not only to the weaker language (English) but also to the stronger language (German).

Reference Arnaus Gil and MüllerArnaus Gil and Müller (2018b) ran a cross-sectional study with sixty-three children who acquired French as one of their L1s. There were nineteen bilingual children (aged 3;11–6;4) and 44 trilingual and multilingual children (aged 2;4–9;3). While monolingual French children go through a stage (before age 2;7) during which they produce postverbal subjects (e.g., est tombé Philippe “is fallen Philippe” instead of Philippe est tombé), the multilingual children in this study seemed to skip this stage, which the authors interpreted as acceleration.Footnote ⁵ In another cross-sectional study, Reference Arnaus Gil and MüllerArnaus Gil and Müller (2018a) studied verb placement in German main clauses in bilingual and trilingual children (ages 2;0–9;3). German is a V2 language, which requires that the verb is in second position, no matter whether the sentence starts with a subject or whether −for stylistic purposes− a prepositional phrase (PP) or object is fronted. Monolingual German children pass through a stage (roughly between 1;8 and 3;6) during which they produce target-deviant verb-final structures. By contrast, the trilingual children in this study (German + a Romance language + another language) seemed to skip the verb-final stage that is typical for monolingual German children. The authors argued for positive CLI from the Romance language.

In summary, the outcomes of studies on the syntactic development of trilingual children parallel those from bilingual studies, showing early separation of languages (Reference MontanariMontanari, 2009b) and crosslinguistic influence regardless of language dominance (Reference KazzaziKazzazi, 2011; Reference Devlin, Folli, Henry and SevdaliDevlin et al., 2015, Reference Devlin, Folli, Sevdali, De Angelis, Jessner and Kresic2017; Reference Arnaus Gil and MüllerArnaus Gil & Müller, 2018a, b). Compared to the acquisition of phonology, it is perhaps less obvious that children follow the same patterns as monolingual children. Studies that provide quantified data from all three languages and that capture the emergence of syntax are the exception (Reference MontanariMontanari, 2009b being the only one).

11.3.4 Pragmatics

Trilingual pragmatics is understudied. Reference Safont-JordàSafont-Jordà (2011, Reference Safont-Jordà2013) investigated data from a consecutive trilingual boy, Pau, who grew up in Valencia, Spain, where Spanish is spoken as the majority language. Pau acquired Catalan as his L1; exposure to Spanish started at the age of 2;0 and to English at 2;11. The mother used Catalan and Spanish with Pau, the father Catalan only, while English was restricted to playtime, TV, and instruction at school. The study focused on the pragma-linguistic routines of requesting in the three languages, examining direct and indirect request strategies. Data involved audio and video recordings during play at home with the mother-researcher. Indirect request strategies are vague, opaque expressions that do not show the speakers intention explicitly (e.g., It’s hot in here, isn’t it? – as a request to open the window). Direct requests are realized (e.g., by imperative verbs, such as Look!). There are cross-linguistic differences in politeness orientations, Spanish and Catalan being positive politeness-oriented languages where direct requests are not considered as offensive or rude, unlike in English. The first study (Reference Safont-JordàSafont-Jordà, 2011) covered the age between 2;6 and 3;6. It showed an increase in indirect forms and a decrease in direct requests in Spanish and Catalan, coinciding with the introduction of the third language English. Thus, since indirectness in English requests affected Catalan and Spanish request forms, the pragmatic systems were closely linked. Pau’s production therefore involved a wider variety of forms than found with monolingual Catalan and Spanish children at comparable ages, suggesting qualitative differences vis-à-vis monolinguals. Reference Safont-JordàSafont-Jordà (2013) showed a decrease of indirect request forms in Catalan and Spanish starting at 3;7, along with a steady increase in English during the same age. This means that the amount and type of request forms varied for each language and developed in line with appropriateness criteria of language use. Direct requesting was closely linked in Catalan and Spanish but completely different in English. Thus, the pragmatic systems of trilingual children interact too, and the influencing language can even be one that is neither the home language nor the CL.

11.3.5 Summary

While there is fairly good coverage of studies on interlocutor sensitivity, studies on core linguistic areas are still scarce, especially quantitative longitudinal studies that capture the onset of speech production in all three languages. Table 11.3 lists studies that more or less meet these criteria.

Use of the Community Language at Home.

Two independent studies have provided evidence that using the CL at home might jeopardize the development of minority languages. Reference De Houwer, Hoffmann and YtsmaDe Houwer (2004) collected survey data in Flanders, the Dutch-speaking region of Belgium, asking for language use in bilingual and trilingual families with at least one child aged six to nine. All children in the sample heard Dutch at school but differed in terms of home language use. Among the 18,049 questionnaires, De Houwer identified 308 trilingual families (1.7% of the families surveyed). Families were considered trilingual if, within the family, at least two languages X and Y were spoken in addition to or instead of Dutch. She found that in less than half of these families the children actually spoke three languages. There were two main conditions that favored active trilingualism: (1) neither one of the parents spoke Dutch at home, and (2) both parents did not speak the same two languages X and Y. This means that, if parents used Dutch at home, it became a strong competitor for the other home languages. Reference Braun and ClineBraun and Clyne (2010) conducted semi-structured interviews with seventy trilingual families in Germany and England to see whether parents used their native languages with their children. They distinguished three types of families: In Type I families, each parent speaks one different language but neither parent speaks the CL; in Type II families, each parent speaks two native languages; in Type III families, each parent speaks three NLs. In Type II and III families, one of the languages spoken by the parents is the CL. For the parents in Type I families, it was easiest to implement the one person–one language strategy, and these parents were also more successful in raising their children trilingually than parents who spoke two or three native languages. Additionally, Type I children often had grandparents who did not speak the CL, which was a factor that motivated the parents to pass on their native languages to their children (see also Reference BraunBraun [2012] for a more detailed account of the grandparents’ role). Type II and III families encountered problems in maintaining their second or third language once children started nursery or primary school, and the grandparents typically did not provide native language support, as they were often bilingual or multilingual with the CL themselves. In summary, the studies strongly indicated that parental use of the majority language at home is likely to jeopardize active trilingualism.

Language at Day Care.

Reference QuayQuay (2011b) studied language use at the day care of the two trilingual children Freddy and Xiaoxiao. Both children acquired two minority languages at home that were different from the CL, German/English (Freddy) and Mandarin/English (Xiaxiao) (Reference Quay, Cenoz and GeneseeQuay, 2001, Reference Quay2008; see Section 11.2.1 for details). Both children spent 6–8.5 hours per day at day care where they were exposed to Japanese starting at the age of 0;11 (Freddy) and 0;5 (Xiaoxiao). A quantitative analysis of language choice showed that both Freddy and Xiaoxiao produced predominantly Japanese utterances: 95% in the case of Freddy and 98.7% in the case of Xiaoxiao. The data are very robust, based on thirty recorded sessions for Freddy (age 1;1–1;10) and fourteen for Xiaoxiao (2;0–2,4). Neither child ever produced German (Freddy) or Mandarin (Xiaxiao), and they produced only a very small amount of English or Japanese–English mixed utterances. The study showed clear separation of home languages and day care language. Furthermore, in a qualitative analysis, Quay (2011) showed particular recurring discourse patterns, namely gesture and onomatopoetic expressions, especially in the case of Freddy, which – so the author argues – make input more transparent for language-learning children. Both children acquired Japanese effortlessly and even became dominant in Japanese.

Caregivers’ Discourse Styles.

Reference ChevalierChevalier (2013) investigated the influence of caregiver discourse styles, specifically the way in which parents react when their children mix, based on a case study of Lina, who was growing up in the German-speaking part of Switzerland. Lina’s mother is Swiss German, she has a French-speaking father from Belgium, and her English input source is her aunt, who came to visit twice a week; English is also the language used between the parents. Overall, Lina spent considerably more time with her mother and less time with her aunt. Monthly recordings were done between two and three years in all three languages. The data showed that the parents used their languages consistently when addressing the child (mother 98%, father 89%, aunt 99.7%), but Lina’s speech with each of her conversation partners showed different amounts of mixing: with her mother, 97% of her utterances are in Swiss German, with her aunt, 54% are in English, but with her father, only 13% are in French. The analysis of caregiver speech was restricted to Lina’s father and her aunt (i.e., the main interlocutors in the two non-CLs), and how they responded if the child did not use their respective languages. Responses were classified on a continuum of five categories with responses constraining a monolingual mode at one extreme (e.g., expressing incomprehension) and responses allowing or encouraging multilingualism at the other extreme (e.g., switching to the child’s language). The responses by Lisa’s aunt clustered on the monolingual end of the scale, while those of Lisa’s father clustered on the multilingual end. Based on these data, Chevalier argued that the explanation of the differential amount of mixing in the child’s speech lied in the caregiver’s different conversational styles.

Metalinguistic Awareness and the Role of Translanguaging.

Reference ChoiChoi (2019) describes language practises from a translanguaging perspective, asking what language practices a trilingual child engages in to communicate meaning in a multilingual household. This is an ethnographical case study of the child Gyuan based on 300 audio and 300 video recordings covering the period between birth and age 6. Gyuan grew up in the US. His main caregiver was his grandmother, who was living with them and spoke Farsi to the child. The father also used Farsi, while his mother spoke to him in Korean. Thus, Farsi and Korean were provided almost equally on a daily basis at home. At 4;6 Gyuan started English at pre-kindergarten. The study nicely illustrates the child’s heightened awareness of the interlocutors’ languages and needs as well as his own metalinguistic knowledge as he engaged in complex, nuanced translation (requesting and offering translation as well as paraphrasing).Footnote ⁷ Code-switching happened flexibly and purposefully regardless of proficiency. However, the child refrained from code-switching in English contexts and, over time, increasingly code-switched English into the two HLs. While the author emphasized the translanguaging theories’ intent to prevent monolingual ideology in schools and the strategic inclusion of students’ HLs in instruction, she also warned that if translanguaging practices are introduced in minority language settings, they might have the effect that the CL will eventually take over and jeopardize the maintenance of the HLs, as indicated by the increasing tendency to use English (the CL) in HL contexts.

In summary, many studies have emphasized that introducing the CL at home will have implications for the home languages (Reference De Houwer, Hoffmann and YtsmaDe Houwer, 2004; Reference Braun and ClineBraun & Clyne, 2010; Reference BraunBraun, 2012; Reference ChoiChoi, 2019), while the CL is acquired without efforts, regardless of whether it is used at home or not (Reference QuayQuay, 2011b). Discourse styles may play an additional role (Reference Quay, Cenoz and GeneseeQuay, 2001; Reference ChevalierChevalier, 2013). Given that domains of language use are divided by three rather than two, the impact of the CL may be stronger in trilinguals than in bilinguals.

13.2.1 L1 Lexical Attrition

Lexical attrition has been the most widely researched aspect of attrition thus far. The lexicon has been posited to be the first domain of grammar to attrite (Reference EckeEcke, 2004; Reference Paradis and KöpkeParadis, 2007; Reference Schmid, Köpke, Schmid, Keijzer and DostertSchmid, 2007). Observed changes are thought to be due to changes in L1 lexical access as a result of co-activation with the L2 and resulting competition, and changes can present in the form of increased disfluency in production (e.g., reduced speech rate, self-correction, repetitions) and comprehension (in the form of longer reaction times), reduced lexical accuracy, and reduced lexical diversity (repetition, variety, and dispersion of tokens across word types), including sophistication (i.e., usage rate of rare/infrequent words) (for a comprehensive review of these phenomena, see Reference Jarvis, Schmid and KöpkeJarvis, 2019).

Reference Paradis and KöpkeParadis (2007) reports that the linguistic variables most commonly associated with lexical attrition are frequency of lexical items in the L1 and L2 and similarity between L1 items and their L2 counterparts. Reference Schmid and JarvisSchmid and Jarvis (2014) found heavy reliance on frequent items in L1 speech production and Reference Schmid, Köpke and PavlenkoSchmid and Köpke (2009) have found that these items incur lower processing costs. Frequency aside, L1 items with true cognate L2 counterparts are more resilient to attrition; the opposite is true of “false” cognates. As we will see throughout this review of the literature, the degree of these effects can be modulated by extralinguistic factors, including L1 and L2 usage and proficiency. For example, the use of the L1 for professional purposes (Reference Schmid and JarvisSchmid & Jarvis, 2014) at the onset of reduced L1 input can mitigate L2 lexical interference, although mitigation can also depend on L2 proficiency and experience (Reference Schmid and JarvisSchmid & Jarvis, 2014). Most recently, neuroimaging data have shown L2 effects (1) modulated by recency of L2 use on neural activation and connectivity associated with L1 lexical processing (Reference Brice, Frost and BickBrice et al., 2021) and (2) modulated by L1 proficiency on ERP responses to lexico-semantic anomalies associated with word retrieval and conflict monitoring (Reference Kasparian and SteinhauerKasparian & Steinhauer, 2016).

In addition to being the most widely studied aspect of attrition (and perhaps multilingualism, more generally), the lexical domain is inextricably linked to the other domains of grammar discussed herein, affecting “underlying syntactic knowledge as well as syntactic processing, and any lexical … aspects of phonology” (Reference Bardovi-Harlig, Stringer, Schmid and KöpkeBardovi-Harlig & Stringer, 2019: 376). Specifically, it is thought that lexical attrition can trigger a domino effect, yielding potential changes in syntactic complexity and variety as well as phonological information stored in lexical representations (for examples, see Reference Bardovi-Harlig, Stringer, Schmid and KöpkeBardovi-Harlig & Stringer, 2019). The following sections address attrition in each of these two domains.

13.2.2 L1 (Morpho)Syntactic Attrition

While lexical attrition is largely accepted as a natural consequence of bilingualism, the vulnerability of L1 morphosyntactic properties has been (and continues to be) widely debated. As illustrated in recent keynote articles on attrition by Reference Schmid and KöpkeSchmid and Köpke (2017a) and Reference Hicks and DomínguezHicks and Domínguez (2020b) and subsequent commentaries, there seems to be general agreement that language-internal L1 syntactic properties that interface with language-external domains (such as discourse) are vulnerable to L2 influence (e.g., overt subject pronoun use conditioned by discourse constraints; Tsimpli et al., 2004; Iverson, 2012; Reference DomínguezDomínguez, 2013). Sorace’s Interface Hypothesis (IH; applied to L1 attrition in Sorace, 2000; Reference Tsimpli, Sorace, Heycock and FiliaciTsimpli et al., 2004) posits that these interface properties’ vulnerabilities result from limited processing resources that are needed for online integration across internal and external domains. However, there is less consensus on the status of language-internal syntactic knowledge,Footnote ³ and specifically whether L2 influence can yield L1 representations that take on L2 feature specifications, or optionality between L1 and L2 representations. Some authors have maintained that language-internal morphosyntactic properties are not vulnerable to L2 influence (e.g., Reference SchmidSchmid, 2002; Reference Tsimpli, Sorace, Heycock and FiliaciTsimpli et al., 2004; Reference PerpiñánPerpiñán, 2011)Footnote ⁴ with Sorace (2020) citing the L1 reversibility attested upon re-exposure in Reference Chamorro, Sorace and SturtChamorro and colleagues’ (2016) study as an indication that any observed instability is behavioral. These behavioral examples are supported by eye-tracking data indicative of stable L1 syntactic processing (Reference Grüter and HoppGrüter & Hopp, 2021). In contrast, several behavioral studies have pointed to instability of language-internal syntactic properties, including properties of binding (e.g., Reference Gurel and YilmazGurel & Yilmaz, 2011; Domínguez & Hicks, 2016), infinitival constructions (e.g., Reference Ribbert and KuikenRibbert & Kuiken, 2010), and subject-verb inversion phenomena (Reference IversonIverson, 2012). Reference Kasparian, Vespignani and SteinhauerKasparian and colleagues (2017) have also reported evidence from Event Related Potentials (ERPs) of L1 vulnerability to change in number disagreement processing strategies associated with reduced L1 use. Regardless of any disagreement over whether these data sets constitute true language-internal attrition, several researchers (e.g., Reference LohndalLohndal, 2020; Reference PutnamPutnam, 2020; Reference WestergaardWestergaard, 2020) back Reference Hicks and DomínguezHicks and Domínguez’s (2020a) notion that syntactic attrition is plausible and should thus be incorporated in any complete model of bilingualism.

If language-internal syntactic attrition is possible, why is it so rare? Reference Hicks and DomínguezHicks and Domínguez (2020a) and Reference Schmid and KöpkeSchmid and Köpke (2017a, Reference Schmid and Köpke2017b) have both addressed this paradox; the takeaways are that this highly constrained type of attrition is potentially limited to contexts in which (1) the L1 and L2 are structurally similar languages, (2) an L1 property has an L2 analogue with shared (but not identical) feature configurations, and (3) there is consistent co-activation of the L1 and L2. Finally, there is the question of why syntactic attrition is realized differently than lexical attrition or, as we will see in the following section, phonetic/phonological attrition. Following Reference Schmid and KöpkeSchmid and Köpke (2017a), this difference likely boils down to the discrete nature of morphosyntactic phenomena, contrasted with lexical and phonetic/phonological variance that can often occur without incurring “an outright violation of the target norm” (654).

13.2.3 L1 Phonetic and Phonological Attrition

Similarly to the case of morphosyntactic attrition, the primary questions related to phonetic and phonological attrition underline what is affected and to which extent. These questions are reflected in the differential consideration of phonetic drift, phonetic attrition, and phonetic drift and phonetic attrition. Herein, I follow Reference ChangChang’s (2019a) definition of phonetic drift and phonetic attrition as subphonemic changes that are short term due to recent L2 experience or long term, respectively. Phonological attrition, on the other hand, is a long-term change that “crosses a phonemic boundary” (Reference ChangChang, 2019a: 203) or yields a difference in the realization of allophonic alternations (e.g., Reference Celata, Schmid and KöpkeCelata, 2019). This area of research has come to encompass production, perception, and processing of segmental and suprasegmental phenomena. In what follows, I will first review the research on phonetic drift and attrition, followed by phonological attrition. Despite reports of L1 phonological resilience (e.g., Reference Guion, Flege and LoftinGuion et al., 2000; Reference Oh, Guion-Anderson and AoyamaOh et al., 2011), all three phenomena have been documented at the segmental and suprasegmental level in both production and perception.

13.2.4 Summary: L1 Attrition

To summarize, the occurrence and degree of attrition appear to vary depending on (1) the module of grammar, (2) linguistic variables that are common across modules as well as module-specific, (3) extralinguistic variables related to language experience and usage, and (4) whether data are elicited via measures of production or comprehension. Moreover, there are conflicting data when it comes to variables such as L1 usage, whereby some researchers posit that greater usage is associated with L1 stability and others posit that greater usage leads to increased vulnerability due to a high degree of L1/L2 co-activation.

Starting with a comparison across lexical, morphosyntactic, and phonetic/phonological attrition (1), morphosyntactic attrition is the least commonly attested, with evidence largely limited to phenomena that require online integration of language-external elements. Phonetic drift and phonetic attrition have been commonly observed, with comparably fewer accounts of lesser-studied phonological attrition as measured in production and perception alike. Finally, there is ample evidence of lexical attrition, and, in particular, lexical access. When it comes to linguistic variables (2), frequency and crosslinguistic similarity (including cognate status) are commonly reported to predict attrition across modules of grammar, while variables such as functional load, acoustic distance, and psychoacoustic salience have been linked specifically to phonetic/phonological attrition. Extralinguistic variables associated with attrition (3) – beyond age of L2 acquisition, given that this chapter focuses on attrition in sequential bilinguals – encompass attitudinal/affective, socioeconomic/educational, and sociolinguistic factors, as well as quantity, timing, and context of L1 and L2 input and usage, L2 proficiency, and cognitive factors (including inhibitory and biological/aging factors). To complicate matters, these factors do not always predict outcomes in isolation; rather, they act in tandem with other variables and can be analyzed as “cluster” factors (see, e.g., Reference Köpke, Köpke, Schmid, Keijzer and DostertKöpke, 2007). Further, it is not clear whether these variables affect all modules to a similar degree. Finally, if we consider the whole of the body of research available, there is a noticeable connection between evidence of attrition and whether data are based on speech production patterns. That is, there is a need for additional comprehension/perception data to better understand the nature of attested L1 effects and ultimately glean whether and how attrition affects representation.

13.2.5 Predictions for L1 and L2 Attrition in L3 Acquisition

There are a number of critical questions that are left to answer when it comes to attrition and its selective nature. While attrition research has “reached adulthood” (Reference De Bot, Schmid and Köpkede Bot, 2019: 502), we are in the early stages of understanding several critical questions, including those related to the reversibility of attrition, how L1 modules of grammar might be differentially affected within a single learner (Reference Schmid, De Leeuw, Schmid and KöpkeSchmid & De Leeuw, 2019: 187), whether degree of attrition depends on directionality and/or L1/L2 structural similarity (Reference Schmid and KöpkeSchmid & Köpke, 2017a), and whether epiphenomenal effects precede structural/representational changes. What we do know at this juncture, however, is that a sequential bilingual’s L1 is not impervious to L2 influence. Thus, the logical prediction that follows is that an L3 learner’s L1 and L2 will not be impervious to an L3, either. In fact, there is ample evidence of L2 change as a result of reduced L2 input and usage, exclusive of influence from a subsequently acquired language (for an overview of the L2 attrition literature, see Mehotcheva & Köpke, 2019). Within this scenario of L1 and L2 attrition, we can also predict that many of the variables that modulate L1 attrition will also be at play, and likely at an increased level of complexity given the additional system in the mix. As we will see in Section 13.3, L3 researchers have begun to test these predictions with the aim of further informing L1 versus L2 maturational differences as well as the benefits that L3 acquisition can afford L2 development.

13.3.1 Introduction

While the observation of L3 effects on existing linguistic systems has occupied a much smaller space in L3 acquisition research than that of initial transfer and acquisition, a growing body of work has established the epistemological value of this type of research. Two primary questions have driven this work: The first is effectively the mirror-image of the most researched question in the field, which asks which variables drive the reliance on one existing system or another at the initial stages of L3 acquisition. The inverse question asks: when a learner has two existing systems, which variables determine any differential regressive effects of an L3 on these systems? In addition to informing the cognitive organization of the multilingual mind, this question has important implications for our understanding of the constitution of an (early-acquired) L1 versus a (late-acquired) L2 system. The primary debate in L1–L2 differences has centered on maturational constraints, specifically the relationship between reduced brain plasticity and the ability to acquire a target-like representation in adulthood. Applying this notion of maturational constraints to the maintenance of a linguistic system, comparing L3 effects on an L1 versus L2 can uniquely reveal constitutional differences in terms of a critical period for stability.Footnote ⁷

In addition to AoA-related differences, we can home in on other factors that predict differential attrition outcomes. The second question focuses on the interaction between an L3 and late-acquired L2 and considers trickle-down effects in L3 acquisition. Specifically, if a learner acquires a phenomenon in the L3 that presents similarly in the L2 but had not been acquired prior to L3 acquisition, is this knowledge transferred to the L2? This question has important implications in a classroom setting and can further inform the benefits of multilingualism. I address each of these questions in turn, followed by relevant theoretical and methodological considerations.

13.3.2 Research Question 1: L3 Effects on L1 versus L2

The question of how an L3 might affect an L2 differently than an L1 has been examined primarily via phonological effects. Reference Cabrelli Amaro and RothmanCabrelli Amaro and Rothman (2010) first formalized a hypothesis that favored a relationship between age of acquisition and stability in the form of the Phonological Permeability Hypothesis (PPH), which states that a late-acquired phonological system is more vulnerable to L3 effects than an early-acquired system. In studies by Reference Cabrelli AmaroCabrelli Amaro (2013, Reference Cabrelli Amaro2017b), the strong version of the hypothesis operationalized these effects as changes to representation, while the weak version operationalized them as epiphenomenal (i.e., effects due to differences in L1 versus L2 processing routines and inhibitory ability). Spanish data from advanced mirror-image English/Spanish learners of L3 Brazilian Portuguese (Reference Cabrelli AmaroCabrelli Amaro, 2017b) were compatible with the weak version, such that L2 Spanish production of word-final mid vowels pointed to greater L3 influence than L1 Spanish production while no effects were found in Spanish perception regardless of AoA (i.e., whether Spanish was the L1 or L2). Longitudinal case study data in Reference Cabrelli AmaroCabrelli Amaro’s (2013) study was collected at the same time as the group data and tracked an L2 Spanish speaker’s Spanish from the onset of L3 acquisition. In line with the group data, after twelve weeks of instruction, production of Spanish word-final mid vowels had shifted toward L3 Brazilian Portuguese and accent ratings of his Spanish speech dropped significantly, with Spanish perception remaining stable (for additional evidence of L2 perceptual stability, see Reference 353Wrembel, Gut, Kopečková and BalasWrembel et al., 2020).

Of the studies that have followed, a clear pattern has yet to emerge, although none has replicated the effect found by Reference Cabrelli AmaroCabrelli Amaro (2017b). In a study of L1 Polish/L2 Danish/L3 English, Reference SypiańskaSypiańska (2013) found L3 VOT effects on both L1 Polish and L2 Danish but L3 vowel formant measures in Reference SypiańskaSypiańska (2016) signaled effects only on L1 Polish. Without mirror-image data, it is difficult to gauge whether the L1 Polish data are a consequence of AoA, particularly when we consider the robust evidence of L2 influence on L1 phonology (Section 13.2), L3 influence on L1 phonology (Reference SypiańskaSypiańska, 2013; Tordini et al., 2018; Reference LiuLiu, 2019; Tordini, 2019), or another variable (or variables) such as global structural similarity. In a recent pilot study, Reference Brown and ChangBrown and Chang (2022) examined this role of similarity in L3 Spanish effects on speech rhythm in the L1 and L2 of two groups of mirror-image English/German bilinguals and found L3 effects on English speech rhythm independent of its status as an L1 or L2. Based on the assumption that English is more similar to Spanish than German is, Brown and Chang concluded that typology is positively associated with instability and proposed the Similarity Convergence Hypothesis (SCH), whereby

While AoA was not a significant predictor of attrition as measured by the learners’ production, the data indicated a trend toward greater effects on the L1, whether it was German or English. This finding led to the acknowledgement that there might be a role for an interaction of more than one variable, and underlines the importance of further research to understand the relationships of AoA and typology (what I refer to as global structural similarity).

Differential L3 effects have been the focus of just a few lexical and morphosyntactic studies. To my knowledge, the only morphosyntactic study is based on judgment data in Reference Cabrelli Amaro, Angelovska and HahnCabrelli Amaro’s (2017a) study from the mirror-image sample from Reference Cabrelli AmaroCabrelli Amaro (2013, Reference Cabrelli Amaro, Angelovska and Hahn2017a) and Reference Cabrelli Amaro, Amaro, Rothman and PeukertCabrelli Amaro and colleagues (2015). The study replicated the finding that, among advanced learners with target-like L3 knowledge of subject-to-subject raising across a dative experiencer, L2 Spanish was more susceptible to L3 Portuguese influence than L1 Spanish. Specifically, L2 Spanish speakers rated ungrammatical sentences significantly higher than L1 Spanish speakers, in contrast to L3 initial stages data showing rejection of these sentences by L1 and L2 Spanish speakers alike. As with the phonological production data from Cabrelli Amaro (2013, 2017a), the participant from the case study reported by Reference Cabrelli AmaroCabrelli Amaro (2013) patterned with his peer group of L1 English/L2 Spanish speakers (Reference Cabrelli Amaro, Amaro, Rothman and PeukertCabrelli Amaro, 2015). In contrast to the phonological data, which are compatible with the weak version of the PPH, these morphosyntactic data suggest that differential stability might indeed extend to representation. While judgment tasks, typically intended to tap representation, have been posited to reflect metalinguistic performance (see, e.g., Reference Schütze, Schindler, Drożdżowicz and BrøckerSchütze, 2020), the author contended that the results were indicative of a representational change because the structure examined is one (1) that is not explicitly taught in an L1 or Ln classroom and (2) for which negative input is not available. Taking the phonology and morphosyntax data together, Reference Cabrelli Amaro, Angelovska and HahnCabrelli Amaro (2017a) amended the scope of the PPH to include other modules of grammar and presented the Differential Stability Hypothesis (DSH) to reflect this change.

While the DSH has not yet been tested for lexical attrition, the results of a ten-year case study detailed by Reference Ecke and HallEcke and Hall (2013) are compatible with its tenets. The authors analyzed lexical retrieval failures in the first author’s L1 German, L3 English, and L4 Spanish (the speaker’s L2 Russian and L5 Portuguese were not in regular use). In contrast with the DSH’s predictions, Ecke and Hall predicted that L1 lexical retrieval would be more vulnerable to attrition than L3 English or L4 Spanish due to less frequent L1 use while in the L3 or L4 environment. However, data showed that regardless of decreased use, L1 German stabilized after an initial period of adjustment spurred by a temporary increase in maintenance costs, in line with research that shows phonetic drift. German was also the overall greatest source of crosslinguistic influence regardless of the relative typological relationship with English and Spanish.Footnote ⁸ Unlike the L1, although L3 English and L4 Spanish were maintained at a level of near-native proficiency, their stability fluctuated as a result of lessened usage. These data bolster the idea that degree of stability is negatively associated with age of acquisition.Footnote ⁹

Two more recent studies have also compared lexical attrition in early- versus late-acquired languages.

Reference LarsonLarson (2018) reported error analyses from elicited production tasks in the L1, L2, and L3 from four L1 English/L2 Spanish learners of L3 Brazilian Portuguese. The data recorded after thirteen weeks of instruction revealed that 6 percent of lexical items produced in L2 Spanish were instances of form-based transfer (as opposed to meaning-based transfer such as the use of semantic extensions and calques) from Portuguese, while no Portuguese items were observed in L1 English. While these data are compatible with the DSH, additional data would be needed to support the hypothesis: First, a mirror-image comparison group is necessary to determine whether the observed effects are a result of crosslinguistic structural similarity or AoA. Second, to adequately isolate AoA, learners must have target-like representations in the L1 and L2 prior to L3 acquisition and, ideally, global native-like proficiency. In this case, while proficiency was not measured, high proportions of form-based lexical transfer have been associated with lower levels of proficiency (e.g., Reference LindqvistLindqvist, 2010) and the participants had completed Spanish coursework at a third-year undergraduate level, which typically correlates with intermediate to high-intermediate proficiency.

The third study, from Reference SuhonenSuhonen (2020), carried out three L3 translation ambiguity experiments with L2 English speakers of either L1 German or L1 Swedish. The experiments spanned (1) initial L3 learning of an artificial language by L1 Swedish/L2 English speakers measured after one hour of exposure (masked priming), (2) longitudinal development of intensive L3 Swedish (L1 German/L2 English) over ten months in Sweden (word-pair similarity ratings), and (3) a cross-section of L3 Swedish (L1 German/L2 English) learners that resided in either Sweden or Germany (word-pair similarity ratings). In the artificial language study, a nonsignificant effect of translation ambiguity was found in L1 and L2 reaction times, with cognitive control as a significant predictor of L1 reaction times. In the longitudinal and cross-sectional studies, L3 effects were reported for time on task in the L1 and the L2 and for similarity ratings (modulated by time on task) in the L2 data, although these effects were significant only in the longitudinal study. In this case, decreased time on task in L1 German was modulated by cognitive control and perceived similarity.

However, time on task increased in L2 English when perceived L2–L3 similarity was high. Suhonen posited that the L2 outcome was due to hypercorrection or avoidance (Reference SchachterSchacter, 1974) and further suggested that the distinct outcomes in the longitudinal (significant L3 effects) and cross-sectional (no significant L3 effects) data sets were due to language experience. That is, the longitudinal sample was in an intensive immersion learning context for which there is documentation of temporarily increased L1 inhibition (e.g., Reference Linck, Kroll and SundermanLinck et al., 2009; see also Section 13.2.3.1) that is “separate from attrition” (167). On the other hand, the effects found in the cross-sectional sample tested in Sweden, whose duration of exposure was longer and whose learning context was less intensive, were proposed to be representative of lexical attrition. Interestingly, L3 effects on the L1 and L2 were global rather than item-based, which contrasted with item-based L1 and L2 effects on the L3. This finding led the author to posit that this difference reflects the lower processing burden that comes with wholesale inhibition of a less proficient language compared to an L1 or advanced L2.

Whether an L1 or L2 differ in degree of susceptibility to L3 influence, we know that neither system is impervious to change. Is it possible that the durability and reversibility of L3 effects, particularly those that are posited to reflect changes in representation, are qualitatively or quantitatively different in an L1 versus an L2? For example, take the data from Reference Hyltenstam, Bylund, Abrahamsson and ParkHyltenstam and colleagues (2009), who tested phonemic discrimination of Korean stops by Korean adoptees that arrived in Sweden at the age of three months. Despite a lack of a full L1 stabilization period, these adoptees were able to re-activate their L1 perceptual grammar in a re-exposure scenario an average of twenty-two years after the onset of L1 input loss. Could we predict the same outcome for L1 Swedish speakers that had had three months’ exposure to L2 Korean as adults?

Thus far, the only data available that tracks reversibility comes from the L1 English/L2 Spanish/L3 Portuguese case study participant introduced by Reference Cabrelli AmaroCabrelli Amaro (2013), whose initial testing session in Brazil showed attrition of L2 Spanish phonological production and morphosyntactic representation after twelve weeks of L3 Portuguese exposure. Shortly after this testing session, the participant returned to the United States, where he was working as a Spanish instructor with minimal Portuguese input and use. Reference CabrelliCabrelli (2020) collected data from the participant nine months (phonology) and two-and-a-half years (morphosyntax) after the participant’s return to the United States. The morphosyntax judgments in both Spanish and Portuguese had shifted from Portuguese-like to Spanish-like and his Spanish vowel production data had returned to baseline (Portuguese production data were not collected). While generalizations should not be made based on a single case study, these data suggest that, similar to an L1, changes to a near-native L2 are reversible. This is true even at the level of representation and at least under optimal conditions in which the L2–L3 input ratio is reversed (i.e., the learner is re-immersed in the L2 and experiences reduced input in the L3). The findings line up with the assertion that “attrition effects … decrease as a result of … re-exposure” (Reference Chamorro, Sorace, Schmid and KöpkeChamorro & Sorace, 2019: 35), which is also supported in an L1 attrition case study by Reference Sancier and FowlerSancier and Fowler (1997). To determine whether reversibility differs in an L1 versus L2, we need to compare the L2 of learners such as Cabrelli Amaro’s case study participant with the L1 of mirror-image participants in an L3-dominant environment, followed by re-immersion in the environment of the L1/L2. If an early-acquired system is in fact more stable than a late-acquired system, one could then predict that an L2 representation will undergo reversal more easily than an L1. We can also expand this question of reversibility by comparing closely related language pairs such as Portuguese and Spanish with dissimilar pairs to determine how crosslinguistic structural similarity modulates any AoA effects.

With just a handful of studies that vary broadly along a number of dimensions, L3 effects on early versus late-acquired systems is still an exploratory question. To move this question forward, we will need to develop research that examines all of the variables that have been identified as critical in L1 and L2 attrition research, including – but not limited to – those discussed in the studies just mentioned. Consider the data sets that indicate a greater degree of L2 attrition versus L1 attrition: What variables are responsible for these differential outcomes? Are they linguistic (e.g., crosslinguistic proximity, the specific phonological units measured), experimental (e.g., elicitation of production versus perception data, instrument choice), extralinguistic (e.g., AoA, proficiency, context of acquisition as well as testing, usage, affective factors), cognitive (e.g., inhibitory ability, relative activation), metacognitive (e.g., crosslinguistic awareness), or a combination of several of these? For example, take the L3 Portuguese effects on L2 Spanish documented by Reference Cabrelli AmaroCabrelli Amaro (2013, Reference Cabrelli Amaro, Angelovska and Hahn2017a, Reference Cabrelli Amaro2017b): Could they be attributed to an interaction of crosslinguistic structural similarity and AoA, whereby an L2 effect is dependent on a high degree of crosslinguistic similarity between the L3 and L2? And, are similarity effects driven by a grammar’s parser, metacognitive factors, or both? Reference Jessner, Oberhofer and MegensJessner and colleagues (2021) discuss the notion that acquisition of an L3 that shares a cognate-rich relationship with a learner’s existing system enhances crosslinguistic awareness, particularly in experienced learners (Reference MegensMegens, 2020). What might happen in a scenario absent of such a relationship, such as English/Spanish bilinguals learning L3 Mandarin?

Additional factors may be at play that have been shown in L1 attrition and L2 attrition research to affect the degree of change. These include between-group differences in Spanish usage (e.g., Reference Mayr, Sánchez and MennenMayr et al., 2020) and other variables often associated with the construct of dominance, the age of reduced Spanish contact (e.g., Reference BylundBylund, 2009), attained proficiency in the L2 (e.g., Reference MehotchevaMehotcheva, 2010), affective factors such as attitude and motivation (e.g., Reference XuXu, 2010 for L2 attrition), or, most likely, a combination thereof (Reference De Bot, Lowie and Verspoorde Bot et al., 2007; Reference Schmid and DusseldorpSchmid & Dusseldorp, 2010). We will need to consider each of the variables mentioned and their interaction to fully understand whether and under which conditions an L2 is more vulnerable than an L1 to L3 influence.

13.3.3 Research Question 2: L3 Facilitative Effects on the L2

Knowing what we do about variability in L2 outcomes, most L2 learners will not have fully acquired a target-like system prior to the L3 onset of acquisition for reasons related to individual differences, learning conditions, and/or linguistic features including complexity, frequency, and salience (see, e.g., Reference Housen and SimoensHousen & Simoens, 2016 for a discussion). In this section, I explore the question of whether (and under which conditions) L3 acquisition can facilitate L2 development. The research in Section 13.3.2 demonstrates that an L3 can affect an L2 in a nonfacilitative manner; it is therefore plausible to hypothesize that an L3 can have facilitative effects on a non-target-like L2. If this hypothesis is supported, such an outcome has practical implications adjacent to our understanding of multilingual crosslinguistic influence. Most importantly, it would bolster the promotion of the study of multiple languages, lending support to multilingual initiatives and objectives such as the one set forth at the European Commission Social Summit in 2017 for all citizens to speak their mother tongue and two other languages by 2025 (COM(2017)0673),

While most L3 development studies have been designed primarily to observe nonfacilitative transfer, a number of studies have set out to examine potential facilitative L3 effects on the L2 at linguistic and cognitive levels. Starting with lexical facilitation, two studies have examined L2/L3 language pairs with substantial cognate relationships, each with distinct outcomes. Reference GriesslerGriessler (2001) documented L2 English lexical development in L1 German speakers learning French as an L3. This same development was not attested in a comparison group without L3 French, leading Griessler to attribute development to what she called a “French add-on effect” (58). The author considered the effect to be potentially due to English–French cognate relationships, while also citing possible “mnemonic advantages.” Such metacognitive advantages include enhanced “experienced-based indicators of future recall, including memory of previous recall attempts, accessibility of target information, and cue familiarity” (Reference KelemenKelemen, 2000: 800). These and other metacognitive advantages such as enhanced metalinguistic and crosslinguistic awareness have been documented in L3 learners (e.g., Reference MegensMegens, 2019) and could serve to supplement gaps in linguistic knowledge in the L2. With that said, the only other study to examine L3 lexical effects specifically on an L2 (Reference Forecelini and SundermanForcelini & Sunderman, 2020) reported a nonfacilitative effect of L3 (Portuguese) on L2 (Spanish) lexical similarity and suggested that decreased lexical processing speed in L2 Spanish was the result of increased demands on cognitive control due to lexical crosslinguistic similarity. Further research is therefore needed to determine the level(s) (linguistic, (meta)cognitive) at which a lexically similar L3 can fill L2 gaps, and whether any observed effects are limited to language pairs with substantial lexical overlap.

As with lexical knowledge, Reference GriesslerGriessler’s (2001) study was the first to show facilitative regressive transfer of the L3 in the morphosyntactic domain (specifically, verbal morphology). Beyond Griessler’s findings, most L3–L2 regressive transfer research on morphosyntax has originated in East Asia. Several studies in Hong Kong have tested L1 Cantonese speakers of advanced L2 English learning French or German. These studies have demonstrated L3 French effects on L2 English plural morphology (in free writing but not in a grammaticality judgment-correction task; Reference TsangTsang, 2016) and subject relative clauses (Reference HuiHui, 2010), as well as L3 German effects on L2 English tense/aspect (Reference Cheung, Matthews, Tsang, De Angelis and DewaeleCheung et al., 2011; Reference 349Matthews, Cheung and TsangMatthews et al., 2014). Reference Ahn and JangAhn and Jang (2019) also reported facilitative effects specific to generic NPs in the L2 English of L1 Korean speakers learning French as an L3 with low-intermediate proficiency. Similarly, Reference Ahn and MaoAhn and Mao (2019) found that L2 English interpretation of reflexive binding was (descriptively) more target-like in a group of L1 Korean/L2 English/L3 Chinese participants compared to an L1 Korean/L2 English control group, and attributed the between-group difference to L3 Chinese influence. Finally, in the only study to my knowledge to examine L4 regressive transfer, data from Reference Llinàs-Grau, Mayenco, Ibarrola-Armendariz and Ortiz De Urbina ArruabarrenaLlinàs-Grau and Puig-Mayenco (2016) revealed a positive effect of L4 German on that-deletion in L3 English in simultaneous Catalan/Spanish bilinguals. In contrast to these studies, the only null effect reported has come from L1 Chinese speakers in Taiwan (Reference ChuangChuang, 2002) whose L2 English did not evidence facilitative effects of L3 German relative clause acquisition.Footnote ¹⁰

Moving on to phonology, two recent studies have investigated L3 enhancement of L2 representations. A study of L1 Mandarin/L2 English/L3 Cantonese speakers by Reference Luo, Li and MokLuo and colleagues (2020) examined L3 effects on L2 perception and processing speed. Two groups of L3 Cantonese participants with (1) less than a year of immersion or (2) more than five years’ immersion in a Cantonese-dominant environment evidenced faster L2 English reaction times than a comparison L1 Mandarin/L2 English group. Processing speed was modulated by proficiency, whereby the strength of the positive L2–L3 correlation was greatest in the experienced group. In contrast to Luo and colleagues’ findings, Lewandowska and Wrembel (2019) did not find a facilitative effect of L3 German or French on the upper-intermediate L2 English vowel perception of L1 Polish speakers after approximately six weeks of classroom-based L3 exposure. Since the German vowel inventory patterns more similarly with English than French does, the authors predicted that L3 German would have a greater facilitative effect on L2 English perceptual acuity than L3 French. Contrary to the prediction, the groups did not differ in L2 accuracy, and the L3 French learners, whose L3 categorizations were more accurate than those of the L3 German learners, categorized L2 vowels faster than the L3 German learners. While it is of course possible that the data reflect a true null effect, it is equally possible that the L3 had not yet reached a point in development in which influence on the L2 could obtain. In fact, L3 German accuracy averaged only 66 percent, which indicates that the target vowel representations had not been fully established. Another point to consider is the role of explicit instruction, which is unclear based on the studies cited in this section, since none state whether the phenomena under investigation were part of the curriculum.

Taken together, these studies suggest that faciliative regressive effects might be more robust for morphosyntax and lexical knowledge than phonology but can occur across modules of grammar. This is true even when (1) the L3 is as different from the L2 as Chinese and English (Reference Ahn and MaoAhn & Mao, 2019) and (2) L3 proficiency is low-intermediate (Reference Ahn and JangAhn & Jang, 2019; cf. Lewandowska & Wrembel, 2019; as well as Reference TsangTsang, 2016, who found that exertion of L3 influence was limited to high-proficiency learners). At the same time, while L2 representations might benefit, crosslinguistic similarities might lead L2 processing to operate at a deficit (Reference Forecelini and SundermanForcelini & Sunderman, 2020). To dig deeper into the question of which factor(s) constrain facilitative L3 influence on an L2, first steps will be to replicate existing research across levels of proficiency with language triads that vary in similarity and to determine how facilitative outcomes differ depending on explicit instruction and context more generally (classroom versus naturalistic).

13.4.1 Modeling Attrition in Multilingualism

Building this line of investigation requires not just a description of what happens, but also how and why. The optimal starting point is, once again, existing models that have provided accounts of (or that can account for) L1 attrition, which vary widely in their theoretical orientation and coverage. These models, some of which were developed for specific modules of grammar, all address the interaction of representation and processing at some level, while varying in their conceptualization of the primacy of the role of processing, the mechanisms that underlie attrition processes, and their specification of a formal component. Those with a more explicit formal specification tend to center on morphosyntactic structures and include Attrition via acquisition (Reference Hicks and DomínguezHicks & Domínguez, 2020b), Multiple Grammars (Reference Amaral and RoeperAmaral & Roeper, 2014; see Slabakova for its direct application to L1 attrition), the Feature Reassembly Hypothesis (Reference Putnam, Perez-Cortes, Sánchez, Schmid and KöpkePutnam et al., 2019), Modular Online Growth and Use of Language (MOGUL; Reference Sharwood Smith, Schmid and KöpkeSharwood Smith, 2019), the Interface Hypothesis (Reference Chamorro, Sorace, Schmid and KöpkeChamorro & Sorace, 2019), and, specific to phonetics and phonology, the revised Speech Learning Model (Reference Flege, Bohn and WaylandFlege & Bohn, 2021). Other models such as the Activation Threshold Hypothesis (Reference ParadisParadis, 2004) and the Savings paradigm (e.g., Reference De Bot, Martens and Stoesselde Bot et al., 2004), as well as connectionist models of the bilingual lexicon, are more explicit in their definition of the processing mechanisms involved. Several accounts, including MOGUL, the Competition Model (Reference MacWhinney, Schmid and KöpkeMacWhinney, 2019), and Complex Dynamic Systems Theory (CDST; Reference Opitz, Schmid and KöpkeOpitz, 2019; see also Chapter 4 of this volume, for its application to L3/Ln acquisition) also specify a role for individual differences related to social and affective factors and/or metacognitive factors. This last set of models in particular aligns with the notion that Reference MacWhinney, Schmid and KöpkeMacWhinney (2019) puts forth in his extension of the Competition Model to L1 attrition, which assumes that

Looking to existing L1 attrition accounts, while a range of factors undoubtedly extend to L2 attrition and are sufficient to explain the L3 facilitation data in Section 13.3.3, we still need to stipulate the factors that contribute to any observed differences in an L1 versus L2 and the mechanisms that yield these differences. The accounts of L3 regressive influence mentioned thus far (PPH/DSH, TCH) are essentially formal accounts, with the former distinguishing between representational versus processing effects. Both require further elaboration; they explain why a particular outcome will obtain, but do not fully explain how it will obtain and need to specify the mechanisms that underlie the outcomes. Additionally, if we follow MacWhinney’s position – and it is reasonable to assume, that, regardless of one’s theoretical assumptions, we all acknowledge a role for neuro-cognitive and social-affective factors – the end goal will be to broaden the scope of these accounts of L3 regressive influence. Currently, the Dynamic Model of Multilingualism (DMM; Reference Herdina and JessnerHerdina & Jessner, 2002) is the only proposal that accounts for attrition in multilingualism and encompasses linguistic, cognitive (here, (meta)linguistic awareness), and social-affective factors. The DMM applies CDST (see Chapter 4 of this volume) to multilingualism and assumes that a multilingual’s systems are interconnected and consistently fluctuating, with interactions that are unpredictable but systematic. Attrition and acquisition processes are driven by a multilingual’s current communicative needs. As an example, if a multilingual is acquiring an L3 in an immersion setting, language maintenance effort (LME) will be reduced in the L1 and L2 as a result of the effort dedicated to the L3, which will lead to L1 and L2 attrition.Footnote ¹¹ In addition to elaboration of the mechanisms that underpin these changes, what remains to be stipulated is how and why changes to an L1 may be qualitatively and quantitatively different than changes to an L2 (Research Question 1). That is, while the DMM describes the context in which L1 and L2 attrition will be likely to occur and can cover L3 facilitative effects on an L2 (Research Question 2), it “does not make any prediction with respect to the relative strengths of language attrition in L1 and L2 and how these might fluctuate over time, instead treating L1 and L2 similarly” (Reference Ecke and HallEcke & Hall, 2013).

To advance theoretically informed modelling of attrition in L3 acquisition, it is important to follow Reference Schmid, Köpke, Schmid and KöpkeSchmid and Köpke’s (2019) call for theoretically motivated research that includes linguistic, psycholinguistic, and sociolinguistic approaches. With the need to consider a multitude of factors, the question then becomes whether to do so by (1) opting for so-called “simple” models that generate falsifiable predictions, which CDST proponents have traditionally viewed as reductionist and oversimplified (e.g., Reference De Bot, Larsen-Freeman, Verspoor, De Bot and LowieDe Bot & Larsen-Freeman, 2011), or (2) prioritizing the elaboration of fine-grained descriptions (often of single cases) of the complexities of attrition phenomena. In the absence of falsifiable predictions, generalizations are typically limited to post hoc evaluations of outcomes against existing theoretical proposals (e.g., Reference Chan, Dörnyei, Henry, Henry, MacIntyre and DörnyeiChan et al., 2014). In his evaluation of this difference in approaches, Reference PallottiPallotti (2021) considers alternatives to post hoc evaluations, which he does not consider to be a viable resolution due to their lack of falsifiability. His proposed solution to describe as well as explain language development (including attrition) is essentially a compromise between CDST and traditional approaches (and methodologies) that presupposes that complexity and testable hypotheses are not incompatible. CDST approaches and their associated methods of visualization and computational modeling can be optimal for zooming in on variability at a fine-grained level, together with traditional methods that necessarily simplify reality, zooming out in an effort to work practically toward explaining parts of the whole. I continue the discussion of methodological considerations in multilingual attrition research in the following section.

13.4.2 Methodological Considerations

While a one-size-fits-all methodological framework is implausible given that design decisions will often vary depending on elements ranging from the theoretical framework to the participant sample, there are a number of best practices that are broadly applicable.

14.3.1 Support for Typological Primacy

The three following studies examined L3 English article realization in Turkish, Vietnamese, and Russian HL populations in Germany, finding evidence for transfer from German (i.e., the typologically closer language). Note, however, that all three studies are also compatible with the idea that transfer comes from the dominant language (i.e., the language a speaker is more proficient in).

The first study by Reference Kupisch, Snape, Stangen, Duarte and GogolinKupisch and colleagues (2013) analyzed article use in the spoken production of fifteen HSs of Turkish, who grew up in Turkish-speaking families (mean age = 24). From a total of 1,528 instances of NPs and DPs, only 5 percent contained target-deviant productions. These were usually due to articles being oversupplied, which points to transfer from German, since transfer from Turkish would have surfaced as article omission. Further, many of the ungrammatical productions translated into grammatical sentences in German.

Further evidence for transfer from German was found in a second study by Reference Siemund, Lechner and PeukertSiemund and Lechner (2015), who exhibited target-like production of English articles in HSs of Vietnamese and Russian despite a lack of comparable article system in the HLs. While the twelve-year-old HSs made fewer article-related errors than their monolingual peers, this effect was no longer apparent for the sixteen-year-old HSs. The authors suggested this to reflect a higher degree of metalinguistic awareness in younger years, which may potentially disappear as early bilingual children grow older.

Third, Reference HoppHopp (2019) studied a group of thirty-one Turkish-German early bilinguals and a group 31 L1 German controls acquiring English. Using sentence repetition and oral production tasks, Hopp investigated article and subject realization and V2 word order. In the production of articles and subjects, both groups performed similarly, close to ceiling level. At a numerical level, the HSs omitted a larger number of subjects and articles, as would be typical for Turkish speakers, though the difference from the L1 German control group was not significant. Both groups had difficulty with verb raising, which seemed to indicate negative transfer from German and an overall role for typological proximity.

Four further studies examined the extent to which transfer patterns differ between early bilingual L3ers and late sequential L3ers, finding unanimous support for the Typological Primacy Model (TPM), which predicts wholesale transfer from the structurally closer language at the initial stages of L3 acquisition (e.g., Reference RothmanRothman, 2011, Reference Rothman2015; Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019).

The first was a study by Reference Iverson, Pires and RothmanIverson (2009). Iverson’s study, in addition to examining patterns of transfer into the L3, aimed to challenge the proposal that L2 learning relies on a combination of rote learning and L1 transfer rather than on inborn mechanisms of language acquisition (e.g., Reference Bley-Vroman, Gass and SchachterBley-Vroman, 1989). Iverson proposed that any transfer occurring from a late-acquired L2 could be taken as evidence for a mental representation in that language and, in turn, for continued access to Universal Grammar (e.g., Reference Schwartz and SprouseSchwartz & Sprouse, 1996; Reference WhiteWhite, 2003).Footnote ⁵ To test this hypothesis, Iverson compared early and late English–Spanish bilinguals in L3 Brazilian Portuguese at the initial stages of acquisition. The phenomena of investigation were gender agreement (between nouns, determiners, and adjectives) on the one hand, and noun drop (e.g., “I don’t want the white car, I want the red (*one)”) on the other. These two phenomena are found in Spanish and Brazilian Portuguese, but not in English. The participants included simultaneous bilinguals (N = 13) and advanced L2 Spanish speakers (N = 22; AoO in Spanish > 15 years), who were learning BP in a study abroad program in Brazil. Data was collected using a grammaticality judgment task (GJT) to test gender agreement and noun drop, and a translation task to elicit noun drop. The early bilingual and late sequential L3 participants exhibited target-like knowledge of both phenomena, suggesting facilitative transfer from Spanish, since these properties could not have been transferred from English. Accordingly, Iverson argued that linguistic representations can be acquired beyond childhood; further, transfer stemmed from the typologically closer language, irrespective of whether Spanish was the HL or late-acquired L2.

This finding was confirmed by Reference Giancaspro, Halloran and IversonGiancaspro and colleagues (2015) for the acquisition of Differential Object Marking (DOM) in L3 Brazilian Portuguese. As with Reference Iverson, Pires and RothmanIverson (2009), both HSs of Spanish and late L2 Spanish learners were included in the study. Using a GJT, the authors tested the participants’ acceptance of the preposition a in Brazilian Portuguese, which appears before certain direct objects (with the properties [+human/+specific]) in Spanish, but not in Brazilian Portuguese. Their results indicated that both experimental groups tended to accept DOM in Brazilian Portuguese, suggesting nonfacilitative transfer from Spanish, once again regardless of whether this was the HL or the L2.

Reference Puig-Mayenco and MarsdenPuig-Mayenco and Marsden (2018) also contrasted L3 acquisition in early bilinguals and late sequential L3ers (Catalan–Spanish–English), and showed no effect of form of bilingualism. The phenomenon of investigation was knowledge of the position of the polarity item “anything” in English negation, questions, and conditionals. While “anything” is grammatical after negation and ungrammatical in affirmative contexts in all three languages, English and Catalan additionally allow “anything”/res in questions and conditionals, which is not permitted with the Spanish form nada. Catalan also differs from Spanish and English in that res can appear in pre-verbal position, for example in the sentence Res (no) els fa por “nothing frightens them.” Accordingly, nonfacilitative transfer from Spanish was predicted to lead to difficulties accepting “any” in questions or conditionals, and facilitative transfer from Catalan to a higher acceptance thereof. Nonfacilitative transfer from Catalan was expected to lead to the acceptance of “any” outside the scope of negation. Puig-Mayenco and Marsden's results point to transfer from Catalan in both groups, who were significantly more likely than L1 Spanish controls to accept the use of “any”/ “anything” in conditionals and before negation. The authors proposed that the relative closeness of Catalan and English (as opposed to Spanish and English) at the level of phonology might explain the occurrence of transfer from Catalan (for illustration of the relevant phonetic-phonological properties that overlap, see Reference González Alonso, Puig-Mayenco, Fábregas, Chaouch-Orozco and RothmanGonzález Alonso et al., 2021: 4). This interpretation would be in line with the TPM’s cue hierarchy, posited by Reference RothmanRothman (2015: 185) to guide the parser in its selection of the typologically closer language (see also Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019: 163). However, the authors point out that this result also aligns with the LPM and Scalpel Models, which predict transfer from the structurally more similar source.

In an attempt to tease apart the predictions of the TPM and the LPM, Reference González Alonso, Puig-Mayenco, Fábregas, Chaouch-Orozco and RothmanGonzález Alonso and colleagues (2021) tested four properties that pattern differently in the three languages of early sequential bilinguals (Catalan/Spanish) acquiring L3 English during eight weeks of initial exposure to English. The phenomena tested included (1) Direct Object Marking (DOM), which is grammatical only in Spanish; (2) Articles before proper nouns (grammatical only in Catalan); (3) VSO word order (grammatical only in Spanish); and (4) causative structures with a lexical DP, such as “John made Mary go out,” which pattern similarly between English and Spanish, but differ in Catalan (the DP appears in dative case after the verb). Unlike in the study by Reference Giancaspro, Halloran and IversonGiancaspro and colleagues (2015), no negative transfer from Spanish was found for DOM. Instead, participants accepted the use of articles before proper nouns in L3 English, suggesting negative transfer from Catalan. Unexpectedly, VSO constructions were given low ratings across all three languages, suggesting that the participants’ Spanish displayed effects of crosslinguistic influence (CLI) from Catalan; the results for the fourth property, causative constructions, were hard to interpret (for discussion, see Reference Giancaspro, Halloran and IversonGiancaspro et al., 2015: 19–20). Overall, the results suggest a predominant role for transfer from Catalan, which, as in the study by Reference Puig-Mayenco and MarsdenPuig-Mayenco and Marsden (2018), is argued to be more similar to English at the level of phonology.

In summary, the studies in this section all point to a strong role for typological primacy. The three Germany-based studies on article realization suggest transfer from the typologically closer and dominant language, German. The four remaining studies, which focused on various aspects of grammar in initial state L3 learners, showed no differences in transfer patterns between HSs and late sequential L3ers with the same L1s (both transferred from the typologically closer language). Since the typologically closer language in most studies coincides with the L2 rather than the L1, these results would seem to suggest that for transfer into the L3, age of onset in the L2 is less deterministic than typological similarity.

14.3.2 Support for Selective Transfer

The following studies have found evidence for property-by-property transfer in early bilingual L3ers.

Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard and colleagues (2017) investigated 22 Norwegian-Russian early bilingual school students (aged 11 to 14), and two age-matched groups of L1 Norwegian and L1 Russian controls in L3 English. They tested two phenomena, including subject-auxiliary-inversion in questions (where English patterns with Norwegian), and adverb placement in declaratives (where English resembles Russian). In line with the expectation that transfer of a property occurs if there is structural similarity (independently of overall genealogical/typological proximity), the L1 Russians and the Norwegian–Russian early bilinguals more frequently accepted L3 English stimuli with the correct adverb placement (Adv-V), while L1 Norwegian speakers did so less often. Similarly, L1 Norwegians and Norwegian–Russian bilinguals more frequently accepted residual V2 in English (found in English in wh-questions), while L1 Russians did so significantly less often. These results suggest a cumulative effect of the background languages on the L3, and the equal availability of both background languages for transfer. Based on this result, the authors proposed the Linguistic Proximity Model (LPM; Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017), which predicts selective transfer from either background language based on similarities between individual structures (or micro-cues) in the languages.

The Scalpel Model (Reference SlabakovaSlabakova, 2017) expands on the notion of structurally driven transfer, positing that it can be driven by additional variables. In two studies, Reference Slabakova and Garcia MayoSlabakova and García Mayo (2015) and Reference García Mayo and SlabakovaGarcía Mayo and Slabakova (2015) investigated L3 English in Basque–Spanish early bilinguals and L1 Spanish/L2 English learners. The first study (Reference Slabakova and Garcia MayoSlabakova & García Mayo, 2015) examined focus fronting and topicalization. The primary aim was to extend and test the predictions made by the Interface Hypothesis (e.g., Reference SoraceSorace, 2011) to L3 acquisition, according to which syntactic constructions that are dependent on discourse information are particularly difficult for learners due to their high demand on processing abilities. The authors suggested that this difficulty should be even greater for L3 learners, who need to suppress two background languages. Should this expectation be correct, then the trilingual participants in this study were predicted to have difficulty incorporating discourse information in focus-fronted constructions in English. However, this expectation was not borne out, and performance for focus-fronting was target-like. Some difficulties were found in distinguishing between acceptable and unacceptable topicalizations in English, which was interpreted as an effect of the relatively higher construction frequency of clitic left dislocations in Spanish. A second study (Reference García Mayo and SlabakovaGarcía Mayo & Slabakova, 2015) likewise found evidence for nonfacilitative transfer from Spanish in the acquisition of L3 English null objects. These findings led the authors to propose a role for construction frequency in determining patterns of transfer into the L3, in line with the Scalpel Model, which proposes property-by-property transfer modulated variables such as construction frequency, availability of unambiguous input, prevalent use, and structural linguistic complexity.

Reference Jensen, Mitrofanova and AnderssenJensen and colleagues (2021) also found evidence for the LPM and Scalpel Models in thirty-one of school-aged HSs of Russian in Norway acquiring seven different semantic and morphosyntactic properties of L3 English. Their result points to a combination of facilitative (and to a lesser extent nonfacilitative) CLI from both Russian and Norwegian, driven partially by structural similarity, but modulated by the complexity and saliency of the properties being acquired.

14.3.3 Language Dominance and Language Use

Another proposal has been that language dominance might play a role in predicting the source of transfer in early bilingual L3ers. In the relevant literature, language dominance has been operationalized as relative use, and/or as proficiency (see Reference Silva-Corvalán, Treffers-Daller, Treffers-Daller and Silva-CorvalánSilva-Corvalán & Treffers-Daller, 2015). Although the two might coincide, they are often measured differently: language use measures tend to be based on questionnaires and are more subjective in their nature, while proficiency measures generally rely on data from language tasks. As we know from HL research, HSs’ dominance profiles are not clear-cut in themselves, meaning that HSs are not simply dominant in one language over the other. For example, early bilinguals often become dominant in the majority language once they reach adulthood (Reference Benmamoun, Montrul and PolinskyBenmamoun et al., 2013; Reference Montrul, Treffers-Daller and Silva-CorvalánMontrul, 2015, Reference Montrul2016; Reference Stangen, Kupisch, Proietti Erguen, Zielke and PeukertStangen et al., 2015; Reference Kupisch and RothmanKupisch & Rothman, 2018), while they might be balanced bilinguals or dominant in the HL during childhood with a dominance shift often happening during primary school (Reference Kupisch, Montrul and PolinskyKupisch, 2021; Reference Kupisch, Kolb, Rodina and UrekKupisch et al., 2021). Related to this, CLI can be bidirectional during childhood, while being largely unidirectional into the HL during adulthood (for overviews, see Reference Kupisch, Schmid and KöpckeKupisch, 2019, Reference Kupisch, Montrul and Polinsky2021). It seems plausible that such patterns can affect the extent to which transfer occurs from one language or the other. A related proposal has been that language use might determine the source of transfer.

Evidence for a predictive role of language dominance (operationalized as the language of communication) was attested in two studies that examined early bilingual Mazandarani/Persian speakers by Reference Fallah, Jabbari and FazilatfarFallah and colleagues (2016, Reference Fallah and Jabbari2018). The first study investigated prenominal possessives in thirteen- to fourteen-year-olds learning English at school. Participants were divided into three groups based on their language of communication: (1) L1 Mazandarani/early L2 Persian speakers living in a Mazandarani-speaking area (= Mazandarani (dominant) users); (2) Speakers with the same language profile but living in a Persian-speaking area (= Persian (dominant) users), and (3) L1 Persian/early L2 Mazandarani speakers (= Persian (dominant) users). Since the position of possessives is the same in Mazandarani and English (prenominal) but differs in Persian, the authors predicted facilitative transfer in the case of the Mazandarani (dominant) users, and nonfacilitative transfer in the Persian (dominant) users. In line with these predictions, the Mazandarani users judged English possessors mostly correctly (80 percent), while the Persian users tended to prefer postnominal possessors. The second (2018) study tested the same group of speakers for their use of attributive adjectives (prenominal in English and Mazandarani, but postnominal in Persian). As in the 2016 study, Mazandarani users achieved higher scores on all tests, indicative of facilitative transfer from Mazandarani, while the Persian users transferred the Persian word order. These results are taken as evidence for transfer from the most frequently used language, irrespective of whether this was the L1 or early acquired L2.

The role of language dominance was also investigated by Reference Lloyd-Smith, Gyllstad, Kupisch and QuagliaLloyd-Smith and colleagues (2021), who examined the acceptability judgments of English embedded wh-questions following the word order of German and Italian in advanced learners of English. The participants were twenty-one Italian HSs in Germany (mean age = 23) who were dominant in German based on results from a vocabulary measure in German and Italian, and based on their overall language use. At the same time, they differed from one another with regard to how much Italian they used across the lifespan, which was measured with a detailed background questionnaire. The study aimed to ascertain (1) whether transfer would occur holistically from the overall dominant and typologically closer language, German, or (2) whether a higher amount of Italian use would lead to a higher amount of transfer from Italian. Their results showed that, while performance was close to ceiling at a group level, certain wh-items following the Italian word order (e.g., ^*Mary doesn’t know where is John) were accepted by the majority of HSs and L1 Italians, but not by L1 German controls. This was interpreted as nonfacilitative transfer from Italian. However, no role for overall dominance in German was found, nor for Italian use. The result suggests evidence for structurally driven morphosyntactic transfer, in line with the LPM and SM models, irrespective of dominance patterns.

A somewhat clearer role for dominance has been found in L3 developmental data (i.e., when L3 performance measured across different time-points). Reference Puig-MayencoPuig-Mayenco (2019) examined Catalan–Spanish early bilinguals learning English in a classroom setting (see also Reference Puig-Mayenco, Rothman and TubauPuig-Mayenco et al., 2020). He investigated the interpretation of negative quantifiers at two different time points, namely at the initial state after just a few hours of input, and again after eleven months of classes. Language dominance in Catalan and Spanish was measured as a continuous variable on a scale from “completely Catalan-dominant” to “completely Spanish-dominant” using an adapted version of Birdsong’s BLP (e.g., Reference Birdsong, Gertken and AmengualBirdsong et al., 2012). At Time 1, all learners negatively transferred their interpretations of negative polarity items and negative quantifiers from Catalan, irrespective of their dominance profiles. At Time 2, negative transfer was more likely to be overcome by speakers who were dominant in Spanish, but was still an issue for those who were Catalan dominant. This result suggests that negative transfer is harder to overcome if it occurs from a dominant language. It also led the author to propose that the effects of individual variables, such as dominance, become more visible at the later stages of L3 acquisition (see also Reference González Alonso and RothmanGonzález Alonso & Rothman, 2017: 6 with the claim that processes of linguistically driven transfer become clouded by individual variation beyond the initial state).

A similar result was obtained by Reference Cabrelli, Iverson, Giancaspro, González, Molsing, Becker Lopes Perna and Tramunt IbañosCabrelli and colleagues (2020), who compared L1 Spanish/(early) L2 English with L1 English/(late) L2 Spanish speakers acquiring DOM in L3 Brazilian Portuguese, and found that nonfacilitative transfer from Spanish was overcome faster by the group who acquired Spanish as a late L2, whereas the L1 Spanish group had more difficulties in overcoming nonfacilitative transfer from Spanish. The authors suggested this could be because Spanish was acquired from birth (AoO), and/or because Spanish was dominant. To tease these explanations apart, the authors suggested that a study with early bilinguals would be needed: If Spanish HSs overcame transfer from Spanish faster than L1 Spanish speakers, this could be explained by their dominance in the majority language. In contrast, if both groups performed similarly, this would suggest early exposure to Spanish (i.e., AoO) was the central factor. This hypothesis was put to the test by Reference Cabrelli and IversonCabrelli and Iverson (2023), who examined the called-for group of English-dominant HSs of Spanish that were advanced learners of L3 Brazilian Portuguese. Their results revealed that the HSs patterned with the L1 Spanish speakers: both groups failed to recover from nonfacilitative transfer from Spanish, accepting DOM constructions. This result was interpreted as an effect of having learned Spanish at an early age. Rather than being a function of AoO per se, the authors suggested that it is the lifetime quantity of exposure to Spanish that matters. Based on this assumption, the authors put forward the Cumulative Input Threshold Hypothesis, which postulates that the rate of recovery from nonfacilitative transfer is proportional to the input quantity in the transferred language. In other words, the more exposure a speaker has had to a construction in the transferred language, the longer it will take to overcome nonfacilitative transfer from that language in the L3. However, as the authors point out, a continuous measure of input would be needed to more accurately test the role of cumulative input.

We wish to argue that HS L3 learners are the ideal population to further test Cabrelli and Iverson’s Cumulative Input Threshold Hypothesis, since input quantity in the HL tends to be highly variable between individuals and can also be quantified relatively easily. For example, Reference Kupisch, Lloyd-Smith, Stangen and BayramKupisch and colleagues (2020) presented a questionnaire that can be used with HS adults to measure self-reported exposure to the HL across the lifespan, rendering a continuous score on a scale from high exposure to low exposure. A similar measure of HL use was also shown by Reference Lloyd-SmithLloyd-Smith (2021) to predict phonetic-phonological transfer from the HL into the L3 (see Section 14.4.3). Other, more extensive, questionnaires measuring cumulative input have also been developed for early bilinguals and could be adapted for use with older speakers (Reference Unsworth, Nicoladis and MontanariUnsworth, 2016). Using such a measure, one could test for correlations between a higher HL use/exposure score and a slower recovery from nonfacilitative transfer. Such a correlation, if found, would provide strong evidence for their hypothesis.

14.4.1 Support for Typological Primacy

Transfer from the typologically closer language was attested by two studies that examined voice onset time (VOT) in HS populations in Germany. Reference Gabriel, Kupisch and SeoudyGabriel and colleagues (2016) reported German-like performance in the perception and production of voiceless stops in L3 French in Mandarin HSs, even though they theoretically could have transferred their knowledge of shorter VOT values from Mandarin. Although all three languages belong to different families, German and French are typologically closer to each other than either of the two languages is to Mandarin, which means that typological primacy might have played a role (for typological primacy, family membership is not relevant, since the parser does not know anything about family membership). However, these results could also be explained in terms of dominant language transfer, since all learners were dominant in German. To tease apart dominance and similarity, another HS group that is dominant in Mandarin would have to be tested.

In a further study on VOT, this time in L3 English in the German context, Reference Geiss, Gumbsheimer, Svenja, Lloyd-Smith and KupischGeiss and colleagues (2021) investigated twenty-one HSs of Italian and found evidence for transfer from German. This study used two control groups of L2 English, one with L1 Italian and one with L1 German. Both the HSs and L1 German/L2 English controls were on target in English, in contrast to L1 Italian/L2 English controls, who produced VOTs with shorter-than-target values. Once more, the results would also be compatible with dominant-language transfer besides typological proximity. To tease apart dominance and similarity, another HS group dominant in Italian would have to be tested.

In a study that aimed to tease apart the role of dominance from typological proximity (along with several other variables), Reference Cabrelli and PichanCabrelli and Pichan (2021) tested the production of intervocalic stops in L3 Brazilian Portuguese and L3 Italian at the initial stages. While postvocalic voiced stops /b d g/ surface as [−continuant] in English, Brazilian Portuguese, and Italian, they surface as [+continuant] in Spanish (e.g., todo [toðo] “all”). Therefore, transfer from English into Italian or Brazilian Portuguese would be facilitative, while transfer from Spanish into Italian or Brazilian Portuguese would be nonfacilitative. Cabrelli and Pichan examined three different groups: (1) L1 English/L2 Spanish, (2) L1 Spanish/L2 English, and (3) HSs of Spanish/L1 English. If dominance was predictive, transfer patterns should differ between the L1 Spanish and HS Spanish groups, since only the former were dominant in Spanish. In contrast, transfer from Spanish in all three groups would provide evidence for typological proximity. Cabrelli and Pichan found that most speakers produced stops in Italian and Brazilian Portuguese in a Spanish-like manner. This lines up with predictions made by the TPM and also to some extent with the LPM. However, a number of participants produced target-like [−continuant] stops, or stops with values that lay between English and Spanish, especially the L3 Italian learners, and this was unrelated to their background (i.e., not an effect of dominance). A follow-up questionnaire suggested that, unlike the L3 Brazilian Portuguese learners, the L3 Italian learners had been explicitly taught how to produce stops in Italian, which might be why they acquired this property faster. The authors argue for a prominent role of typological proximity across all three learner groups, irrespective of whether Spanish was the HL, a monolingual L1, or an L2, in line with Reference Iverson, Pires and RothmanIverson (2009) and others for morphosyntax.

14.4.2 Support for Selective Transfer

Transfer from typologically distant HLs has also been found by some studies that investigate multiple sound segments.

For example, two studies on school-aged HSs of Turkish in Germany acquiring L3 English found evidence for transfer from both phonological systems for L3 English speech rhythm and vowel production (Reference Domene MorenoDomene Moreno, 2021). This led Reference Domene MorenoDomene Moreno (2021) to propose that phonological transfer occurs in a cumulative, “bit-by-bit” fashion, based on similarities of individuals pieces in the phonetic-phonological systems, similar to the notion of structural transfer proposed by the LPM and Scalpel Models for morphosyntax. Following a similar logic, Reference ArchibaldArchibald (2021) argued that a detailed analysis of metrical phonology across language pairings would be the most effective basis for explaining and predicting phonetic-phonological CLI.

However, structural comparisons across languages do not account for all findings, since structurally motivated transfer seems to be modulated by other variables. For example, Reference Gabriel, Krause and DittmersGabriel and colleagues (2018) found that HSs of Turkish and Russian in Germany produced shorter, more target-like values for the fortis stops /p, t, k/ in L3 French as compared to German monolingual controls. This was interpreted as facilitative transfer from Turkish and Russian, in which fortis stops are produced with short-lag VOT (closer to French). However, the HSs produced voiced stops in French (i.e., /b d g/) in a German-like fashion (with short lag), and this time did not make use of the target property (i.e., prevoicing) in their HLs. To explain this result, the authors proposed that perceptual salience might play a role, since the (non)aspiration of voiceless stops is more salient than prevoicing is (68). In other words, nonfacilitative transfer from the typologically closer or dominant language can be overridden by facilitative transfer from the more distant language, especially if the feature in question is particularly salient.Footnote ⁶ This finding could be explained by the LPM and the Scalpel Models, which specify that structurally based transfer is possible, but may not always occur.

Another factor proposed to influence L3 phonetic-phonological development is universal difficulty, at least when it comes to articulating sounds in a new foreign language. Reference KopečkováKopečková (2016) investigated the production of various rhotic sounds in nineteen (2)L1 German/L2 English speakers learning Spanish. Five of the participants also spoke an additional HL, four of which contained a rhotic /r/, and one of which did not. The results showed that the five HSs were more accurate in producing rhotic sounds as compared to the monolingual learners. However, their degree of ability to produce the notoriously difficult Spanish trill /r/ depended on whether or not this sound was present in the HL.

14.4.3 Language Dominance and Language Use

The roles of dominance and language use have also been investigated for L3 phonology in HSs. For example, Reference Llama and López-MorelosLlama and López-Morelos (2016, Reference Llama, López-Morelos, Babatsouli and Ball2020) investigated adolescent Spanish HSs in Canada in their production of VOT in L3 Canadian French. Spanish and French both have short lag VOT for the voiceless stops /p,t,k/, while English has long lag VOT. The authors found that most speakers produced voiceless French stops in line with English, their dominant language, even though they had acquired monolingual-like stops in L1 Spanish, which they arguably could have transferred. This finding was confirmed in their 2020 study. As the authors point out, this result might also be explained by the English-like input received from their peers, since exposure to naturalistic French (or French as it is spoken in France) would have been seldom.

A further area of interest in HS L3ers has been global foreign accent, which has been examined in two different studies on L3 English in Germany. In the first study, Reference Lloyd-Smith, Gyllstad and KupischLloyd-Smith and colleagues (2017) examined “perceived accent strength” (on a Likert-scale) and “origin of accent” in 19 adult-aged HSs of Turkish and in three control groups (L1 speakers of German, Turkish, and English) by means of an accent rating study with twenty English teacher raters. The accents of the HSs were perceived as moderate, in line with the ratings for the L1 German controls, but as significantly weaker than the accents of the L1 Turkish controls. As a group, the HSs were rated as German-sounding in 60 percent and as Turkish-sounding in 20 percent of ratings, indicating more pronounced influence from German, but some influence from Turkish. In regression analyses, the number of “Turkish-like” ratings in English was significantly predicted by “phonological proficiency in Turkish,” measured in a separate rating experiment with Turkish raters (see Reference Kupisch, Lloyd-Smith, Stangen and BayramKupisch et al., 2020). This suggests that phonetic-phonological transfer from a typologically distant language is more likely if phonetic-phonological proficiency in that language is high.

In the second study on global foreign accent, Reference Lloyd-SmithLloyd-Smith (2021) this time investigated Italian HSs (n = 23) in Germany, using a similar design. As with the (2017) study, the HSs were rated as mostly German sounding (66 percent of ratings), and an Italian-like accent was detected in only 11 percent of ratings. In regression analyses, the probability of sounding Italian-like (classified binarily) was predicted by the “Italian Use Score,” measured using a HL questionnaire, suggesting a role of HL use for transfer from that language. When taken together, these accent studies suggest that, while L3 accent is mostly affected by the dominant and typologically closer language, phonetic-phonological influence from the typologically more-distant HL can co-occur in speakers who use their HL regularly, or who have a high phonological proficiency in that language.

14.5.1 Do HS L3ers Differ?

As this chapter has illustrated, one central question in research with HS L3ers has been the extent to which they differ from late sequential L3ers, the traditionally studied population of L3 learners. Differences between these populations might be expected, for example, due to the largely declarative nature of L2 knowledge in late L2ers, which is often lacking in HSs, or based on HSs’ increased experienced dealing with two linguistic systems from early on (i.e., prior to (potential) sensitive periods for language acquisition). In fact, it was also proposed by Reference RothmanRothman (2015) that, if the cognitive advantages found for bilinguals as compared to monolinguals in psycholinguistic research (e.g., Reference Bialystok, Gogolin and NeumannBialystok 2009) can be extended to language acquisition, then it is thinkable that early bilinguals might be able to more effectively inhibit their background languages during the early stages of L3 acquisition (as compared to sequential bilinguals), and thereby avoid nonfacilitative transfer.

However, in the few existing studies that explicitly compare HS with late sequential L3ers (i.e., Reference Iverson, Pires and RothmanIverson, 2009; Reference Giancaspro, Halloran and IversonGiancaspro et al., 2015; Reference Puig-Mayenco and MarsdenPuig-Mayenco & Marsden, 2018; Reference Cabrelli and PichanCabrelli & Pichan, 2021; Reference González Alonso, Puig-Mayenco, Fábregas, Chaouch-Orozco and RothmanGonzález Alonso et al., 2021), it seems that transfer patterns are similar across both groups of bilinguals. These studies evidenced transfer from the typologically closer language, in line with the TPM or with other models that accommodate data consistent with transfer driven by similarity. Importantly, it made no difference to the outcome whether the second background language had been acquired as an L1 (heritage and non-heritage alike), or as an L2. If these studies are on the right track, the implication, as far as we understand, would be that there is no single cut-off point that would allow us to divide L3ers into two groups (early vs. late) in terms of how morphosyntactic transfer is expected to pattern.

14.5.2 Morphosyntax versus Phonology

A second point of interest in this chapter has been the extent to which patterns of transfer differ across the domains of morphosyntax and phonology. On the one hand, it was suggested early on that it is “fairly obvious” that phonological transfer is conditioned by different explanatory principles from those guiding syntax (Reference Sharwood Smith and KellermanSharwood Smith & Kellerman, 1986: 7). On the other hand, following the notion of full transfer as proposed by, for example, the TPM, one would not make different predictions for transfer across different modules of grammar. One reason to assume that phonetics-phonology is different from other linguistic domains are the neuro-motor routines of the L1(s) that constrain the acquisition of subsequent languages. Though this may not be (as) relevant to L3 perception studies, it is relevant to L3 production. In the same vein, in research on heritage languages, syntactic representations have been suggested to be more stable in their nature (e.g., Reference Benmamoun, Montrul and PolinskyBenmamoun et al., 2013), whereas pronunciation has been shown to be more susceptible to changes in a leaner’s environment. As pointed out by Reference KupischKupisch (2020) in relation to HL research, though phonology is often claimed to be one of the most robust domains of grammar, accent in fact tends to be one of the least stable. As we will argue, L3 phonetics-phonology (or, at least the produc-tion thereof) seems more likely to be affected by variables such as language use than other areas of the grammar.

With regard to L3 morphosyntax, a strong role for typological primacy (and for the TPM) has been found by many of studies discussed in this chapter (Reference Iverson, Pires and RothmanIverson, 2009; Reference Kupisch, Snape, Stangen, Duarte and GogolinKupisch et al., 2013; Reference Giancaspro, Halloran and IversonGiancaspro et al., 2015; Reference Puig-Mayenco and MarsdenPuig-Mayenco & Marsden, 2018; Reference HoppHopp, 2019; Reference González Alonso, Puig-Mayenco, Fábregas, Chaouch-Orozco and RothmanGonzález Alonso et al., 2021). In contrast, others have found evidence for morphosyntactic transfer from the typologically more-distant language (e.g., Reference García Mayo and SlabakovaGarcía-Mayo & Slabakova, 2015; Reference Slabakova and Garcia MayoSlabakova & García-Mayo, 2015; Reference Westergaard, Mitrofanova, Mykhaylyk and RodinaWestergaard et al., 2017; Reference Lloyd-Smith, Gyllstad, Kupisch and QuagliaLloyd-Smith et al., 2021), often dependent on variables such as construction frequency (Reference SlabakovaSlabakova, 2017), dominance (Reference González Alonso, Puig-Mayenco, Fábregas, Chaouch-Orozco and RothmanGonzález Alonso et al., 2021), and language use (Reference Fallah, Jabbari and FazilatfarFallah et al., 2016, 2017). More work is needed to identify the extent to which these variables are important across different populations; for example, whether they apply in the same manner to late successive L3ers as they do to HS L3ers.

The L3 phonology studies reported herein seem to have found more support for transfer from multiple sources. While evidence for typologically based transfer from a single language does exist (e.g., Reference Gabriel, Kupisch and SeoudyGabriel et al., 2016; Reference Cabrelli and PichanCabrelli & Pichan, 2021; Reference Geiss, Gumbsheimer, Svenja, Lloyd-Smith and KupischGeiss et al., 2021), combined phonetic-phonological transfer seems more common, modulated by factors such as similarity of phonetic-phonological segments (e.g., Reference Domene MorenoDomene Moreno, 2021), perceptual salience (Reference Gabriel, Krause and DittmersGabriel et al., 2018), universal difficulty (Reference KopečkováKopečková, 2016), HL proficiency (Reference Lloyd-Smith, Gyllstad and KupischLloyd-Smith et al., 2017), and HL use (Reference Lloyd-SmithLloyd-Smith, 2021).

The asymmetry between the domains of phonetics-phonology and morphosyntax in L3 acquisition was also discussed in the PhD thesis by Reference Lloyd-SmithLloyd-Smith (2020). While Italian use was a predictor of an Italian-sounding accent in their L3 English accent data, it did not predict morphosyntactic transfer from Italian. Reflecting on this, Lloyd-Smith proposed that “global accent” is simply a more dynamic area of language than the area of morphosyntax they examined and, accordingly, is more likely to be affected by individual variables such as language use. This is interesting considering the fact that syntax and phonology are sometimes assumed to be subserved by the same memory system, procedural memory (Reference ParadisParadis, 2004). At the same time, such differences have only been observed when the phonology data comes from speech production, rather than perception, suggesting that this may also be explained by the difference in modalities.

Phonetics-phonology studies also seem more likely to find advantages for HS L3ers when compared to L2 learners of the same target language (e.g., Reference KopečkováKopečková, 2016; Reference Gabriel, Krause and DittmersGabriel et al., 2018). This is perhaps unsurprising, since HS L3ers have an overall larger pool of phonetic-phonological properties at their disposal than L2 learners (which, of course, applies equally to late sequential L3 learners). However, Reference KopečkováKopečková (2016) points out that, when such advantages occur, they tend to be linked to similar properties already available in the speakers’ repertoire, rather than pointing to any kind of general advantages (see also Reference Geiss, Gumbsheimer, Svenja, Lloyd-Smith and KupischGeiss et al., 2021). For future studies, it would be desirable to further tease apart the often-assumed general language learning advantages from advantages due to facilitative transfer. According to the general advantage explanation, early bilinguals would have advantages in linguistic learning because they have experienced more structural variety during their early childhood, with the effect that they are more flexible and cognitively apt in foreign language learning. The alternative scenario is that early bilinguals – or any L3 learners for that matter – have advantages because they have a larger repertoire of linguistic structures to draw from. This latter scenario would predict no differences between early and late bilinguals. Again, testing typologically distinct languages might provide some clarification: if early bilingualism comes with general language learning advantages, early bilinguals should have advantages over late bilinguals when learning new properties (e.g., sounds or structures that are inexistent in their earlier acquired languages).

14.5.3 Methodological Challenges

One aspect that should be pointed out in relation to working with HS L3ers is the importance of testing the phenomenon of investigation in the background languages, and not only in the L3. This point was made early on for studies with late sequential L3ers (Reference Rothman and Cabrelli AmaroRothman & Cabrelli Amaro, 2010), but it is equally important in the case of HS L3ers. This is because CLI from the ML into the HL is quite common, and certain properties may not be monolingual-like for numerous reasons. Obviously, this can affect the extent to which properties are available for transfer into the L3.

To provide an illustration, Reference Giancaspro, Halloran and IversonGiancaspro and colleagues (2015) tested HSs of Spanish and L2 learners of Spanish in the US in their acceptance of DOM in L3 Brazilian Portuguese, and found more Spanish-like behavior in the Spanish L2ers than in the HSs. However, when examining the HS Spanish data, it was clear that the HSs were not sensitive to certain cues, relying only on animacy as a cue for DOM, but not on specificity, as would usually be the case in monolingual Spanish (Reference Giancaspro, Halloran and IversonGiancaspro et al., 2015: 201). Without the HL data, it would have been impossible to know that the HSs were in fact transferring from their Spanish grammatical system. Similarly, in a study of VOT in German–Italian bilinguals speaking L3 English, Reference Geiss, Gumbsheimer, Svenja, Lloyd-Smith and KupischGeiss and colleagues (2021) measured VOTs in all three languages. They found language-specific patterns for Italian (high rates of prevoicing and short lag VOT) and for German (comparatively lower rates of prevoicing and long lag VOT). Had these speakers merged VOT systems in their early acquired languages, it would have been impossible to distinguish between German and Italian as transfer sources.

A further caveat in some of the relevant studies laid out herein is that the typologically closer language was also the dominant/more proficient language, meaning that the potential effects of typological proximity and dominance cannot be separated. One way of isolating potential effects of language dominance, while excluding typological distance as a factor, to be explored in future studies, is to choose scenarios in which all three languages are typologically similar (e.g., Swedish, German, English; for a study on conceptual lexical transfer in sequential L3ers, see Reference SuhonenSuhonen, 2020) or typologically very distinct (e.g., Sinhala, Tamil, English). Of course even in these scenarios an individual (or her parser) might perceive certain language pairs to be typologically closer than others because the parser does not know anything about genealogical relatedness. Nonetheless, such cases have hardly been explored and might bring fresh evidence.Footnote ⁷

Finally, a case that we mentioned at the beginning of this paper but have not yet discussed in detail is that proposed by Reference PolinskyPolinsky (2015), namely HSs relearning their HL, which has the potential to be viewed as a subcategory of HS L3 acquisition. In her article, Polinsky examines various pre-existing (and some new) data on HSs in the US context relearning their HL as adults. Polinsky’s line of argumentation is that, when HSs come to re-learn their HL in a formal setting, the variety they encounter is often different from that which they heard at home. This could be because the classroom variety is a standard variety, while exposure in the home was to a dialectal variety. Structured classroom input can also constitute the first exposure to literacy. Despite the typological vicinity between the HL acquired in childhood and the formal variant being acquired in the classroom, morphosyntactic transfer from the HL can still occur.

Thus, Polinsky argues that the relearning of a HL in this context bears much similarity to L3 acquisition. One difficulty with this approach is that it is not immediately clear at which point HSs become “re-learners.” While it seems fairly clear that this definition applies to populations such as those studied by Reference Oh, Jun, Knightly and AuOh and colleagues (2003), who were HSs of Korean with only limited exposure to Korean during childhood, it is less clear whether it applies to any adult-aged HS in a formal learning setting. Early bilinguals who acquire a HL with a lot of dialectal variation are potentially informative in this regard. Italian HSs are a case in point: dialects in Italy are not simply variants of Standard Italian, and some dialects are mutually incomprehensible (e.g., Sicilian and Piemontese). In this setting, one can easily imagine that a HS “re-learns” the Standard variety when she decides to study Italian in a formal setting. However, the details of how exactly this can be studied still need to be worked out in detail. One problem is that even a HS who has only spoken a dialect at home has had some contact with the standard language, for example because their parents might have adopted a more “neutral” way of speaking, or through engagement with Italian language through other sources such as the Internet and media.

15.3.1 Scenario 1: Return from an L3 Environment (in Childhood) and L3 Attrition

The first scenario where the return to a bilingual speaker’s homeland may affect an L3 emerges in the case of bilingual speakers who lived in an L3 environment and, at a certain moment of their life, moved (back) to the L1 or the L2 environment. This is the case of bilingual couples who move to or meet in a country where neither of their family languages is spoken. Their children grow up with both their parents’ family languages at home and start to acquire the dominant environmental language as an L3, often in the school context. They lose daily contact with this third language if the family moves back to (one of) the home country(ies).Footnote ¹ For instance, this situation is frequently experienced by families with academic jobs who get a position at an international research center or university.

If studies on returnees from L2 environments are rare, studying returnees moving back from L3 contexts is even more scarce, or almost nonexistent. We report the results from two exploratory case studies that analyze language development in three children after moving back from an L3 context to one of their native language environments. We will then discuss the main research questions related to L3 development that arise from this type of study.

One of the first studies looking at the development of an L3 in returnee children is Reference CohenCohen’s (1989) investigation of the performance of his own two English–Hebrew bilingual children after a one-year stay in Brazil, where the children acquired Portuguese as their L3 and were able to converse fluently in that language. They learned Portuguese mainly through social interactions outside their private Israeli school, for instance through a twenty-one-day excursion with monolingual Brazilians almost six months after living in São Paulo. The children moved (back) to Israel at the ages of almost ten and fourteen and lost contact with Portuguese. Data collection started one month after leaving Brazil and was repeated three and nine months later. Cohen investigated attrition effects in the productive L3 lexicon of the returnees based on a storytelling task (using the well-known Frog, Where Are You? booklet). The results show a decrease in the total amount of lexical items, but also its diversity to retell the story during the nine-month testing period in which the participants lacked contact with Portuguese. Furthermore, greater attrition effects were observed in the younger than in the older sibling, but attrition appeared to be restricted to productive skills since both children recognized most of the words they did not use, even eleven months after the onset of noncontact. The author argues that a prolonged period of lack of contact with the target L3 leads to forgetting processes, even though not to a complete loss of lexicon. In addition, L3 attrition is modulated by age of loss of input, which in turn is linked to degree of literacy.

The second study, by Reference Yildiz and KoyuncuogluYildiz and Koyuncuoglu (2017), investigates attrition effects in the L3 Turkish of an L1 English–L2 French bilingual child from Canada who lived for eleven months in Turkey when she was about six years old. She acquired a high level of fluency in oral Turkish by attending a Turkish-speaking nursery school. After moving back to Canada, she completely lost contact with her L3. Data collection occurred via Skype once a month during eight months through free conversation. Data collection started one month after the child returned to Canada. The study aimed at identifying attrition in Turkish morphology and vocabulary by analyzing code-switching utterances. According to the authors, the “findings have revealed the subtle yet significant changes” in the child’s Turkish (Reference Yildiz and KoyuncuogluYildiz & Koyuncuoglu, 2017: 298), particularly with respect to structurally assigned morphology. For instance, the child started to use analytical structures instead of bound morphemes in the later stages of data collection. The authors acknowledge that a longer period of observation would have been useful to better understand the attrition patterns in Turkish.

One crucial question that arises from these kinds of studies, which investigate the development of an L3 in bilingual returnees who lived for a certain period in the L3 environment, aims at understanding if processes of L3 attrition in bilingual child returnees follow the same trajectories as in L2 attrition. To shed light on this research inquiry we need to compare the findings from L3 studies with the studies on L2 attrition in returnees, even though both types of research are scarce. Thus, much more studies are needed to gain a more coherent picture of this type of bilingual language development.

Most L2 attrition studies involving child returnees are case studies with a reduced number of participants (see Reference Flores, Schmid and KöpkeFlores, 2019 for an overview; an exception are Kubota’s studies on Japanese returnees). Similarly to the two L3 attrition studies, several L2 attrition studies on child returnees present longitudinal analyses of the attrition trajectory in the L2 from the moment of return until some months (and no longer than two or three years) later (Reference KuhbergKuhberg, 1992; Reference Reetz-Kurashige and HansenReetz-Kurashige, 1999; Reference Tomiyama and HansenTomiyama, 1999, Reference Tomiyama2000; Reference Yoshitomi and HansenYoshitomi, 1999; Reference Snape, Matthews, Hirakawa, Hirakawa and HosoiSnape et al., 2014; Reference FloresFlores, 2015). They focus on lexical knowledge, morphosyntax, semantics or, more recently, processing costs and cognitive function (Reference KubotaKubota, 2019; Reference Kubota, Chevalier and SoraceKubota et al., 2020a, Reference Kubota, Chevalier and Sorace2020b, Reference Kubota, Heycock, Sorace and Rothman2020c). Essentially, the degree of L2 attrition observed in these studies varies significantly, which is due to differences in the age of return, the degree of contact with the L2 after moving back, the linguistic domain under analysis, or even the method used. Still, most studies report a decline of proficiency that starts some months after the change of dominant environment, which is more significant in younger returnees.

Overall, the attrition trajectories reported for the L2 seem to be similar to the processes observed in L3 attrition. After some months lacking regular contact with the target language, attrition effects start in the productive lexicon in the form of more frequent instances of code-switching, and move selectively to other linguistic domains. For instance, Isabella, the participant observed by Reference Yildiz and KoyuncuogluYildiz and Koyuncuoglu (2017), started to produce deviant system morphemes in her L3 Turkish within seven months of incubation (i.e., lacking context with the L3). The same is observed in the case of Ana, the participant analyzed by Reference FloresFlores (2015), who started to show attrition effects in her L2 German five months after her return to Portugal. The degree of attrition will depend in both cases, L2 or L3 returnees, on the same factors: the level of proficiency attained in the target language before the moment of return and the amount of exposure to this language after return, in addition to individual factors which may further impact on the attrition process (e.g., motivation to maintain contact with the target language after moving away from its environment). An interesting question, open to investigation, is whether a potential period of stabilization of linguistic knowledge, in the sense of Reference FloresFlores (2010), would apply in the same fashion to an L3. In other words, does an early acquired L3 need the same amount of input and time as an L2 to be acquired and consolidated in the speaker’s mind? Or does the fact that it is an L3 play a role in early language consolidation, precisely because it is in competition with two other languages? From what we already know about multilingual language acquisition in childhood, we would predict that this stabilization period would be similar in L2 and L3 acquisition as long as the language input conditions are similar.

Despite the logical assumption that L2 and L3 stabilization periods should pattern similarly, the fact that bilingual speakers who return from an L3 environment have two childhood languages instead of only one is indeed a crucial variable that distinguishes L2 from L3 returnees. The central question that arises from this particular language scenario is whether the presence of two native languages will accelerate the attrition process due to the competition of two languages in use against the one non-used language. Several studies on 2L1/ L2 returnees have shown strong effects of crosslinguistic influence in various language domains after the returnee moved back to the country of origin. The HL, which now becomes the dominant language, starts to influence the attrited L2 (or 2L1) in several domains. Reference FloresFlores (2012), for instance, revealed such effects of crosslinguistic influence from Portuguese into the returnees’ eroded German in the domain of object expression. After losing regular exposure to German, the group of Portuguese–German returnees that she investigated started to omit object pronouns in German in contexts that do not allow for object omission. This process is interpreted as crosslinguistic influence from Portuguese, which is a null object language. A relevant question that arises from these observations is whether in L3 returnees the processes of crosslinguistic influence would be even more pronounced, since the speaker has two childhood languages competing with the attrited L3. We will explore this question from a theoretical perspective in Section 15.3.3.

15.3.2 Scenario 2: Reacquiring a Lost Language – A Case of Reactivation or L3 Acquisition?

The second scenario involves situations of return or of double return with a focus on the language that starts to be the dominant language after the change of environment. On the one hand, this is the case of the native language that is acquired as the HL while the bilingual individual lives in the migration context. As soon as s/he moves (back) to the country of origin, this HL becomes the dominant environmental language (for a deeper discussion of these cases, see Reference Flores, Schmid and KöpkeFlores, 2019). On the other hand, in the current global mobile society, many families move back to the country of migration after (failed) attempts to return to the homeland. In these cases of double return, the returnee child experiences attrition of the L2 after moving to the L1 context, followed by re-immersion in the L2 context after going back to the context of migration (a situation discussed by Reference FloresFlores, 2020).

What is the relationship of these scenarios with L3 development? The crucial fact in these scenarios is that a language, acquired through naturalistic exposure in childhood, has gone through a period of reduced use or complete deprivation at a certain moment of the bilingual speaker’s life. When the speaker is (re-)immersed in the context where this language is the majority language, exposure is regained and re-acquisition sets in. This raises a central question, namely: is the reacquisition of a previously acquired and lost L1 or L2 an instance of reactivation of a latent language, or is it relearning? In the latter case, we may argue that the language is relearned in the same fashion as it would be as an L3.

This hypothesis has been discussed by Reference PolinskyPolinsky (2015) with a slightly different population, so called “heritage-speakers-turned-learners” (163), heritage speakers who lose daily contact with their HL due to extensive exposure to the environmental language and decide to formally re-learn the HL as adults in a classroom setting. Polinsky asks whether heritage speakers make better L3 learners of their latent HL than speakers who learn this same language as a late L2. If they show considerable advantages, then we may conclude that a latent, non-used native language is never lost and, in cases of renewed contact, this language is reactivated, not newly learned as an L3. If, on the other hand, this population evidences performance similar to that of “traditional” late L2 learners, we may conclude that a language that has not fully stabilized in childhood must be relearned as an L3. Polinsky concludes that “heritage-speakers-turned-learners” may not reactivate their whole HL in cases of renewed contact; instead, they have to relearn certain areas of language structure, particularly morphosyntax. These speakers may therefore show only “selective advantages in mastering the phonetics and phonology of their L1/L3” (165). Polinsky recognizes, however, that her conclusions are based on limited data, since “the work on re-learning of L1 as L3 is still in its infancy” (165).

A further population that has been a test case for this question of “reactivation-versus-relearning” are internationally adopted children. These are children who are adopted by families that live outside the child’s country of birth and move to their new homeland, where they are completely deprived from their L1 from this moment on. As a result, these speakers are posited to effectively lose their L1 and replace it with their L2 (Reference Pallier, Dehaene and PolinePallier et al., 2003; Reference Ventureyra, Pallier and YooVentureyra et al., 2004). Often, they enroll in language courses to relearn their native language when they are adults. Currently, there is no consensus on the question of whether the L1 is effectively erased from the speakers’ brain (as suggested by, for instance, Reference Pallier, Dehaene and PolinePallier and colleagues [2003]) or if they maintain L1 remnants that are activated when the speakers are re-exposed to this language. Research with international adoptees show that there are indeed remnants at least at the phonetic level, which may boost the reactivation of the attrited L1 (Reference Hyltenstam, Bylund, Abrahansson and ParkHyltenstam et al., 2009; Reference ParkPark, 2015; Reference Choi, Broersma and CutlerChoi et al., 2017). The prediction arising from this observation is that the re-exposure to an apparently lost native language will lead to a learning process that differs from traditional non-native (L2/L3) acquisition.

Moving this discussion to the population of (double) returnee bilinguals, again, the research addressing questions of relearning and reactivation is scarce. We selected two studies on Turkish returnees from Germany in an attempt to discuss this topic, even though indirectly. The first study is by Reference Treffers-Daller, Daller, Furman and RothmanTreffers-Daller and colleagues (2016), who analyzed a property that is challenging in the acquisition of Turkish by non-native speakers, namely collocations and fixed expressions (involving yap- and et- “to do”), and asked if bilingual returnees have difficulties in fully mastering this property after returning to Turkey. The results show that the returnees did not use target-like collocations in their HL Turkish before or right after their return but developed native-like knowledge within seven years of re-immersion in Turkey. Even though they do not compare the target group with L2 learners of Turkish, they conclude that the returnees activated their native language’s grammatical system with increasing input, which is distinct from that of a non-native language. Overall, the finding in this study contradicts the idea that some properties of a native language that are latent or not fully stabilized may be relearned as a non-native language/L3 in a case of increased input. However, there is also evidence that not all linguistic properties of a HL converge toward the monolingual/homeland norm, even several years after the return to the country of origin. This is shown by Reference Kaya-Soykan, Antonova-Unlu and Sagin-SimsekKaya-Soykan and colleagues (2020) in a study involving fifteen Turkish returnees. They analyzed the participants’ production and perception of evidentiality markers in their heritage Turkish eleven years after their return from Germany. The authors concluded that even after “many years of residing in Turkey the language behaviour of the returnee participants still differed from monolingual Turkish speakers” (16). Of course, this conclusion does not imply that the HL is like an L3 in this domain of evidentiality markers, but it hints that there are properties of a native language that have to be fully stabilized during the optimal phase in order for it to be acquired.

In addition to the development of the HL upon return, we may also look at situations of double return. Recall that, in this case, bilingual returnees move to the homeland, typically in childhood, and spend a period of time in the homeland environment, where they lose contact with the L2. After this time in the homeland, the speakers move back to the country of migration and are re-immersed in the dominant L2 environment. This situation was investigated by Reference Flores, Schmid and KöpkeFlores (2019) in a case study on a bilingual Portuguese–German speaker who moved to Portugal at the age of nine and went back to Germany after a four-year stay in Portugal. During her stay in Portugal, considerable effects of attrition were detected at various levels of morphosyntactic knowledge; eleven months after being re-exposed to German, the rate of attrition in case, gender, and plural marking decreased significantly. The author argues that re-stabilization of native linguistic knowledge occurred, which is in line with the idea that a latent language system is reactivated shortly after the speaker is immersed in the target language’s environment.

15.3.3 Scenario 3: L3 Development after Return and the Role of the Dominant Language Environment – A Contribution to the Theoretical Discussion of L3 Acquisition Models

The third and final scenario we would like to consider is how the other languages (L1 or L2 or 2L1) of a returnee may influence L3 development/attrition upon return to the homeland. In Section 15.3.1, we discussed the trajectories of L3 attrition (with the limited evidence available in the literature) and described potential factors that may affect this process in light of findings from L2 attrition literature. What we did not consider, however, is the potential crosslinguistic influence that may occur from the L1/2L1 or L2 to the L3. For example, let us illustrate a hypothetical case of Japanese–English bilingual children who moved to France and acquired French as an L3 in the community. When these children move back to Japan, some may receive the majority of their input in Japanese by entering a Japanese school system, while others may enroll their children in English as a medium (EMI) schools, thereby getting ample English exposure in school. Given that French is not widely spoken in Japan, it is likely that under this scenario, the children’s L3 French will undergo changes, perhaps as a result of influence from the L1 (Japanese) or the L2 (English). In such cases, what factors may drive the potential effects of multilingual crosslinguistic influence? Does dominance or linguistic structure play any role? We would like to review some models and studies from L3 acquisition literature for insight into this line of inquiry.

Several formal models have been proposed in the L3 acquisition literature to explain the source of CLI and/or transfer (for discussion on distinctions between CLI and transfer, see Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019; Reference Schwartz and SprouseSchwartz & Sprouse, 2021; for a detailed overview of these models, see Chapter 1 of this volume; but we will use the term “influence” to include all instances of effects from one language to the other). These models, however, mainly focus on the initial stages of sequential L3 acquisition. One of the earlier hypothesis established in the field, the L2 Status Factor Hypothesis (L2SF) (Reference Bardel, Falk, Cabrelli Amaro, Flynn and RothmanBardel & Falk, 2012), assumes that the individual properties of the L2 will influence the L3. Following Reference ParadisParadis (2004), the authors assume that this is because the same memory system, namely, declarative memory is used to sustain explicit knowledge of grammar in the L2 and L3, while speakers use procedural memory to sustain implicit or automatized knowledge of their native grammar. Due to the fact that L2SF does not explicitly state when influence can occur and also confers maturational constraints (i.e., focuses on adult L2/L3 learners), it is uncertain whether it can be applied to predict the nature of CLI in the child returnee population. However, if we were to follow this hypothesis, we would expect syntactic properties of their L2 English to influence their L3 French, regardless of whether they are more dominant (both in terms of proficiency and exposure) in their L2 English or L1 Japanese.

While the L2 Status Factor Model focuses on the sequence of acquisition, the Typological Primacy Model (TPM) (Reference RothmanRothman, 2011, Reference Rothman2015; Reference Rothman, González Alonso and Puig-MayencoRothman et al., 2019) predicts that the structural similarity determines the source of (full) transfer to the L3. The idea here is that, in the initial stages of L3 exposure, linguistic information such as lexicon, phonology, morphology, and surface syntactic structures in the L3 is assessed by the parser against both L1 and L2 grammars. The parser determines the grammar that is holistically closest (i.e., structurally similar) to the L3 grammar and thus is chosen as the best candidate to be transferred onto the L3. Although the TPM restricts its prediction to L3 initial stages, Reference González Alonso and RothmanGonzález Alonso and Rothman (2017) call for the need to examine whether models of L3 acquisition can be applied to developmental stages of L3 interlanguage grammar.

An interesting piece of evidence that aligns with the TPM in the developmental trajectories of an L3 comes from Reference HoppHopp (2019), who tested the applicability of L3 acquisition models to developmental stages in a child population by administering L3 English sentence repetition and oral production tasks in Turkish–German heritage speakers. These children were already learning English as a L3 (once a week) from grade 1 and were tested longitudinally over time, at the end of grade 3 and the end of grade 4 (one year in between). The results showed that, for both sentence repetition and oral production tasks, Turkish–German heritage speakers had greater difficulty (compared to the German “monolingual” controls who were also learning English) with linguistic properties that are dissimilar between German and English, while in terms of shared properties, they showed comparable performance to the controls. Crucially, properties that are different between Turkish and German/English showed no signs of influence in L3 English. Hopp’s finding suggests that child heritage speakers holistically transferred the structurally similar language (German) to their L3 (English), which is line with the TPM. One important aspect that Hopp points out, nonetheless, is that it is impossible to tease apart the effect of dominance from typological proximity since the heritage speakers in his study were more dominant in the majority language, German, rather than their HL, Turkish.

In fact, studies that tested the role of dominance in L3 influence are largely inconclusive (Reference Fallah, Jabbari and FazilatfarFallah et al., 2016; Reference Fallah and JabbariFallah & Jabbari, 2018; Reference Lloyd-Smith, Gyllstad, Kupisch and QuagliaLloyd-Smith et al., 2018; Reference Puig-Mayenco, Miller, Rothman, Cho, Iverson, Judy, Leal and ShimanskayaPuig-Mayenco et al., 2018, Reference Puig-Mayenco, Rothman and Tubau2020). For instance, Reference Fallah and JabbariFallah and Jabbari (2018) compared L3 English performance of three groups of bilingual young adolescents in their initial stages of L3 acquisition: (1) Mazandarani–Persian bilinguals who are dominant in their L1 Mazandarani, (2) Mazandarani–Persian bilinguals who are dominant in their L2 Persian, and (3) Persian–Mazandarani bilinguals who are dominant in their L1 Persian. The results from a grammaticality judgment task and elicited production task on attributive adjectives showed that language dominance (measured by quantity of exposure) was the only factor that sufficiently explained the observed differences in performance among the three groups. That is, group (1), who is more dominant in Mazandarani, accepted and produced more attributive adjectives in prenominal position, while groups (2) and (3), who are more dominant in Persian, allowed more attributive adjectives in postnominal position. However, the authors do point out that it is difficult to determine which language of the two (Mazandarani–Persian) are more similar to English, rendering it difficult to make strong predictions about the role of structural proximity on L3 transfer, posited by the TPM (and other models such as the Cumulative Enhancement Model [Reference Flynn, Foley and VinnitskayaFlynn et al., 2004; Reference Berkes and FlynnBerkes & Flynn, 2012]).

As shown in the studies highlighted, it appears to be extremely difficult to tease apart the influence of external or participant-related factors such as dominance or proficiency (for literature on effects of proficiency, see Reference Lloyd-Smith, Gyllstad, Kupisch and QuagliaLloyd-Smith, 2018) from internal structures of the language on L3 acquisition/development. Moreover, to understand the full picture of how CLI is manifested at the initial stages of exposure to L3 and how this may or may not carry over to later stages of learning or acquisition, there is a need “to start data collection at the initial stages and track these same L3 learners over developmentally long periods of time” (Reference González Alonso and RothmanGonzález Alonso & Rothman, 2017: 689). Although limited, there is some work that suggests that factors such as language dominance affect the rate of L3 development past the initial stages of L3 influence (Reference Puig-Mayenco, Rothman and TubauPuig-Mayenco et al., 2020).

We propose that returnees will be an excellent sample to solve these aforementioned issues. First, attrition studies on returnees form one of the few areas of research in which a longitudinal approach is widely adopted, mostly due to the fact that it is crucial to establish a baseline at the onset of their return to pinpoint the changes at the individual level that occur upon linguistic transition. Second, by tracking the development of their L3 from the moment of arrival to the host country to several years after their return to the home country, we will be able to precisely measure how the change in language dominance affects manifestations of L3 CLI. Going back to the Japanese–English–French returnee example, if typological proximity is the driving force of L3 influence at both initial and developmental stages, then it is predicted that English will always influence French, regardless of the changes in language environment. In contrast, if dominance is the sole or overriding factor, then we would expect to see influence from English to French while the returnees are living in the host country (since Japanese would hypothetically be only spoken in the family unit) and, crucially, there should be a gradual shift in the source of influence from L2 English to L1 Japanese (on L3 French) once the returnees move back to Japan. Alternatively, it could be the combination of both of these factors – for instance, Reference Puig-Mayenco, Rothman and TubauPuig-Mayenco and colleagues (2020) found initial L3 influence from a structurally similar language, followed by an effect of dominance in the developmental stages. The major challenge with such an experimental design, however, is to test children immediately after their arrival to the host country and track them longitudinally until they are back for some time in the home country. Such studies would span over a period of at least three to four years, which is largely impractical and costly. However, we believe that longitudinal studies, coupled with such an interesting and unique population, will provide further insights into how an L3 is acquired and how it may interact with L1 and L2. As acknowledged by Reference Bardovi-Harlig and StringerBardovi-Harlig and Stringer (2010), there are several methodological constraints on conducting proper research on L2 attrition in the context of remigration. As has been made clear in this chapter, these constraints are even bigger in research on L3 attrition.