Highlights
-
• Despite receiving less HL input, S.A. children outperformed N.A. children in Greek
-
• S.A. showed a stronger cognate facilitation compared to N.A.
-
• S.A. produced a higher rate of felicitous referential forms compared to N.A.
-
• S.A. produced a higher rate of felicitous VS compared to N.A.
-
• Above and beyond HL, CLI may boost children’s HL acquisition.
1. Introduction
A considerable number of studies examining heritage language (HL) development focus on contexts where English is the majority language (ML). These studies show that child heritage speakers (HSs) tend to overproduce strong referential forms (pronominal/lexical) and preverbal subjects in contexts where reduced forms and postverbal subjects would be preferred or required in the home country variety (Montrul, Reference Montrul2015, Reference Montrul2023). Various non-mutually exclusive factors have been argued to contribute to these patterns, including cross-linguistic influence (CLI) and limited HL use (Polinsky, Reference Polinsky2018). Under the CLI explanation, child HSs transfer referential forms and preverbal subjects from their more dominant ML, a non-null subject/non-object clitic language with a rather rigid SVO word order. Under the limited HL use explanation, child HSs may lack sufficient input to consolidate HL structures. As a result, they may be resorting to strategies that are easier from a processing perspective, such as the use of SVO word orders and strong referential forms. One way to disentangle the effects of CLI from the effects of HL use would be to study the same HL under the influence of typologically different MLs (Chondrogianni, Reference Chondrogianni, Elgort, Siyanova-Chanturia and Brysbaert2023; Scontras et al., Reference Scontras, Fuchs and Polinsky2015). Existing cross-national/cross-linguistic studies, though, are scarce (e.g., Torregrossa et al., Reference Torregrossa, Andreou, Bongartz and Tsimpli2021; Zuban et al., Reference Zuban, Martynova, Zerbian, Szucsich and Gagarina2021), and conclusions about the role of CLI are difficult to draw.
The present study adds to this line of research by comparing child HSs with the same HL who speak MLs that differ across three language domains: (non-)cognate vocabulary items, third-person reference and word order. Specifically, we collected production data from a group of school-aged Greek-Spanish bilingual children who acquired Greek as an HL in South America (S.A.) and a group of age-matched Greek-English bilingual children who acquired Greek as an HL in North America (N.A.). As Greek is closer to Spanish than to English in all three domains of interest, the “Greek-English” and “Greek-Spanish” dyads are ideal for the study of CLI, HL use and their role in HL acquisition.
2. HL acquisition of reference and word order
Studies on HL acquisition of reference and word order show that child HSs in the country of residence may show different preferences in their HL from their monolingual peers in the country of origin (Montrul, Reference Montrul2015, Reference Montrul2023; Polinsky, Reference Polinsky2018). For instance, HSs of null subject/object clitic languages tend to overuse/overaccept strong referential forms (strong pronouns/lexical DPs) in contexts in which a null subject/object clitic would be pragmatically more appropriate (Argyri & Sorace, Reference Argyri and Sorace2007; Daskalaki et al., Reference Daskalaki, Chondrogianni and Blom2022; Montrul, Reference Montrul2018; Paradis & Navarro, Reference Paradis and Navarro2003; Shin et al., Reference Shin, Cuza and Sánchez2023; for an overview, see Serratrice & Hervé, Reference Serratrice, Hervé, Serratrice and Allen2015). Similarly, HSs of languages that allow multiple word orders tend to overuse/overaccept SVO in contexts in which a different word order would be preferred or required in the home variety (Argyri & Sorace, Reference Argyri and Sorace2007; Austin et al., Reference Austin, Blume and Sánchez2013; Chondrogianni & Schwartz, Reference Chondrogianni and Schwartz2020; Cuza, Reference Cuza2016; Daskalaki et al., Reference Daskalaki, Chondrogianni and Blom2022; Polinsky, Reference Polinsky2006).
While these patterns are well-reported, their origin is under debate. It is possible that they are due to CLI from the ML, which is often English, a language that, in the structures of interest, licenses solely strong referential forms and preverbal subjects (SV). It is also possible that they are due to (or further amplified by) diminished HL use. For example, Shin et al. (Reference Shin, Cuza and Sánchez2023) found that diminished use of HL Spanish was associated with a higher rate of non-felicitous lexical DPs in the HL (Spanish) of Spanish-English bilingual children. Similarly, Daskalaki et al. (Reference Daskalaki, Chondrogianni and Blom2022) found that diminished use of HL Greek was associated with a higher rate of ungrammatical preverbal subjects in the HL (Greek) of Greek-English bilingual children. One way to determine whether, above and beyond HL use, CLI also contributes to these patterns would be to compare different language combinations, where the MLs are either structurally different or closer to the HL (for related discussion, see Chondrogianni, Reference Chondrogianni, Elgort, Siyanova-Chanturia and Brysbaert2023; Daskalaki et al., Reference Daskalaki, Chondrogianni and Blom2022; Scontras et al., Reference Scontras, Fuchs and Polinsky2015; van Dijk et al., Reference van Dijk, van Wonderen, Koutamanis, Kootstra, Dijkstra and Unsworth2022). To date, very few studies have adopted this approach.
Fridman et al. (Reference Fridman, Polinsky and Meir2024), for instance, examined how CLI and HL input affect HL morphology, by comparing adult HSs of Russian in Israel and the United States across three phenomena: numerical phrases (encoded in Russian; absent in Hebrew and English), adjective-noun agreement (encoded similarly in Russian and Hebrew; absent in English) and morphological case marking (encoded differently in Russian and Hebrew; absent in English). They found CLI/ML effects solely in the domains in which the Russian and the equivalent Hebrew configurations displayed surface morphological similarity (adjectival agreement). By contrast, in the structures in which there was no surface morphological similarity between the Russian and the Hebrew configuration (case marking), the Russian-Hebrew group did not outperform the Russian-English group. As to the role of HL input, it only emerged as significant in the structure that lacked an equivalent configuration in the ML (numerical phrases). The authors concluded that the role of HL input might be more pronounced for structures that lack an ML equivalent configuration.
In the domain of reference, which is of particular relevance for the present study, the results are rather mixed with some cross-national/cross-linguistic studies finding evidence for CLI/ML effects (Serratrice et al., Reference Serratrice, Sorace, Filiaci and Baldo2012; Sorace et al., Reference Sorace, Serratrice, Filiaci and Baldo2009 for the younger group), others reporting null effects (Rinke & Flores, Reference Rinke and Flores2018; Torregrossa et al., Reference Torregrossa, Flores and Rinke2023) and others indicating CLI/ML effects for some syntactic contexts (Torregrossa et al., Reference Torregrossa, Andreou, Bongartz and Tsimpli2021). For instance, Torregrossa et al. (Reference Torregrossa, Andreou, Bongartz and Tsimpli2021) examined how HL experience and the typological properties of the ML affect bilingual children’s production of subject and object forms by comparing narrative data from Greek-English bilingual children in the United Kingdom/United States, Greek-German bilingual children in Germany and Greek-Albanian bilingual children in Albania. Results revealed HL experience effects for both subject and object forms, and ML effects solely for subjects. Specifically, the Greek-Albanian group (the only group whose other language was also a null subject/object clitic language) produced a higher rate of felicitous null subjects in Greek compared to the Greek-English and the Greek-German groups. In contrast, the three groups did not differ in the rate of felicitous object clitics. The authors concluded that the presence of clitics in Albanian does not seem to have the same facilitating role, possibly due to the clitic system in Albanian being morphologically less complex than the Greek one (it does not inflect for gender).
Turning to the domain of word order, Zuban et al. (Reference Zuban, Martynova, Zerbian, Szucsich and Gagarina2021) collected semi-spontaneous spoken and written narratives from Russian-speaking adolescents residing in Russia and of HSs of Russian residing in the United States and Germany. The two MLs differ in the degree of word order permutations that they allow, with German allowing V-final word order. Results revealed that in embedded sentences, Russian-English bilinguals used a higher rate of SVOs compared to Russian-German bilinguals and monolingual speakers of Russian; no differences between the groups emerged in main clauses. The authors concluded that the clause type/syntactic complexity and the properties of the ML may be important factors for producing different word order patterns in heritage Russian.
To sum up, existing cross-national studies on the acquisition of HL reference and HL word order are very few and give rise to conflicting results, possibly due to task differences (narratives vs. elicitation), modality (production vs. comprehension) and target domain/structures. Furthermore, not all studies take into consideration HL use, which, as demonstrated by the studies reviewed in our introduction, is a potentially confounding factor.
3. HL acquisition of vocabulary
With respect to vocabulary, studies show that child HSs tend to obtain lower accuracy scores and/or show slower reaction times in vocabulary tasks compared to their monolingual peers (e.g., Kubota & Rothman, Reference Kubota and Rothman2024). However, accuracy may increase on cognate words, that is on translation equivalents that are phonologically similar between the HL and the ML (e.g., Greek /e.li.ˈko.pte.ɾo/ – English /hɛl.ə.kɑːp.tɚ/ “helicopter”) compared to non-cognate words, that is compared to translation equivalents that are phonologically unrelated (e.g., Greek /fe.ˈga.ɾi/ – English /ˈmu:n/ “moon”). Cognate facilitation, widely reported across a variety of language pairs, has been taken to suggest that child HSs may use phonological overlap with ML words to understand and produce cognate items in their HL (Bosma et al., Reference Bosma, Blom, Hoekstra and Versloot2019; Chondrogianni & Daskalaki, Reference Chondrogianni and Daskalaki2023; Koutamanis et al., Reference Koutamanis, Kootstra, Dijkstra and Unsworth2024; Lindgren & Bohnacker, Reference Lindgren and Bohnacker2020; Sheng et al., Reference Sheng, Lam, Cruz and Fulton2016). Bosma et al. (Reference Bosma, Blom, Hoekstra and Versloot2019) found that the strength of cognate facilitation can be affected by the degree of phonological similarity of the cognate pair. Specifically, Frisian-Dutch bilingual children performed better with closely related cognates (Frisian /pus/ – Dutch /pus/ “cat”) compared to less closely related cognates (Frisian /wast/ – Dutch /vorst/ “sausage”). Koutamanis et al. (Reference Koutamanis, Kootstra, Dijkstra and Unsworth2024) showed that cognate effects may emerge even among less phonologically related languages, such as Greek and Dutch. Whether the degree of phonological similarity between cognate pairs of different language combinations may also affect the strength of the cognate effect remains to be investigated.
4. Open issues
The results of the studies reviewed so far are consistent with CLI from the ML to the HL. However, to convincingly argue for CLI, one would need to examine HSs of the same HL in two ML contexts: one where the ML is structurally and phonologically/lexically closer to the HL, and another where the HL and the ML are more distant in this regard. To this end, the present study is novel in three main respects: First, it compares child HSs of Greek who acquired Greek under the influence of two typologically distinct MLs: English (a non-null subject/non-object clitic language with a rigid SVO word order) in N.A. and Spanish (a null subject/object clitic language with a more flexible word order) in S.A. While a substantial number of studies have examined HSs of Greek in English-speaking countries, fewer studies have examined HSs of Greek in Spanish-speaking countries (e.g., Giannakou, Reference Giannakou2023; Giannakou & Georgiou, Reference Giannakou and Georgiou2024; Giannakou & Sitaridou, Reference Giannakou and Sitaridou2022; Zombolou, Reference Zombolou and Ihemere2011), and no study, to our knowledge, has compared these two bilingual populations. Taking a comparative, cross-national perspective enables us to determine the extent to which CLI may affect children’s HL acquisition. Second, it compares the two child HS groups across three domains: vocabulary (cognate and non-cognate items), third-person reference (subject/object forms in contexts favoring null subjects/object clitics) and word order (subject placement in contexts favoring/requiring VS). Taking an “across-domain” perspective, while keeping the modality and task type constant, will enable us to determine whether some domains are more susceptible to CLI than others. Third, in view of studies showing that HL use is a contributing factor (e.g., Shin et al., Reference Shin, Cuza and Sánchez2023), it takes into consideration individual differences in the frequency of HL use among our participants.
5. Cognates and referential form and placement in Greek, Spanish and English
In this section, we briefly review the properties of Greek (GR), non-Caribbean Spanish (SP) and English (ENG) in the three domains of interest: phonology (relevant for vocabulary acquisition), form of referential expressions (+/-null subjects; +/-object clitics) and placement of referential expressions (SV/VS). As will become evident, Greek phonology, referential expressions and subject placement patterns are closer to Spanish than to English.
5.1. Phonological properties of words
Greek and Spanish share the same set of five vowels (/i e a u o/) and are both syllable-timed languages. English, on the other hand, has at least 14 vowels, and it is a stress-time language (Hualde, Reference Hualde2014; Topintzi, Reference Topintzi2023). With regard to consonants, Greek, Spanish and English share several sounds, such as plosives at labial, dental/alveolar and velar positions (/p b t d k g/). At the same time, there are notable differences between the three systems. For instance, whereas in Greek the main realization of the rhotic is that of a tap /ɾ/, in Spanish the tap /ɾ/ and the trill /r/ are separate phonemes and in English the rhotic is generally realized as an alveolar approximant /ɹ/. As an example of how the differences/similarities between the three phonological systems may affect the perception of cognates, we may compare guitar /gɪ.ˈta:ɹ/ with its Greek (/ki.ˈθa.ɾa/) and Spanish (/gi.ˈta.ra/) translation equivalents. The Greek word differs from both the Spanish and the English one in terms of consonants. However, it is closer to the Spanish in terms of vowels and syllable structure: Both the Greek and the Spanish words share the same vowels (/i a/) as well as the same number of syllables and syllable structure (CV).
5.2. Reference
In referential forms, Greek is a null subject/object-clitic language, two properties that pattern with Spanish and differ from English (Mavrogiorgos, Reference Mavrogiorgos2010; Zagona, Reference Zagona2002). Focusing on third-person reference, in both Greek and Spanish, null subjects are typically used to maintain reference with a previous topic (1–2), whereas overt subjects are typically used to indicate topic shift (3–4), though this division of labor is not categorical (Alonso-Ovalle et al., Reference Alonso-Ovalle, Clifton, Frazier and Fernandez-Solera2002; Carvalho et al., Reference Carvalho, Orozco and Shin2015; Papadopoulou et al., Reference Papadopoulou, Peristeri, Plemenou, Marinis and Tsimpli2015) and is less pronounced in Spanish than in Greek (Giannakou & Sitaridou, Reference Giannakou and Sitaridou2020). Similarly, both Greek and Spanish allow for strong object pronouns (GR: aftos-afti-afto; SP: él-ella-ello/eso) and object clitics (GR: ton-tin-to; SP: lo-la), whereas English only allows for strong pronouns in subject (he-she-it) and object (him-her-it) positions. Object clitics are more felicitous with prominent discourse antecedents (5–6), whereas strong pronominal objects are typically used to convey focus or co-reference with a less prominent antecedent (7–8).
In addition to their different discourse distribution, Greek and Spanish strong and clitic pronouns have different morphosyntactic properties. Strong pronouns can occupy a postverbal position (on a par with regular lexical DPs), whereas object clitics are phonologically reduced pronominal forms that attach to the left of finite verbal forms in both languages. Both pronominal forms inflect for gender (GR: masculine, feminine, neuter; SP: masculine and feminine), number (GR/SP: singular and plural) and case (GR: accusative and genitive; SP: accusative and dative), whereas in Spanish, strong pronouns are also marked for Differential Object Marking (Mavrogiorgos, Reference Mavrogiorgos2010; Zagona, Reference Zagona2002) (Table 1).
Table 1. Pronominal subject and object alternations in Greek and Spanish and subject placement
5.3. Word order
Greek word order patterns with Spanish, and differs from English, in allowing for the alternation between preverbal (SV) and postverbal subjects (VS). However, there is a degree of micro-variation concerning both the distribution and the rate of VS acceptability. For instance, in Greek, “all new information” questions are compatible with both VSO and SVO responses, as in (9) (Spyropoulos & Revithiadou, Reference Spyropoulos, Revithiadou, Halpert, Hartman and Hill2009), whereas “wide focus” questions show a strong preference for Object clitic-Verb-Subject (ObjclVS) responses, as in (11) (Alexopoulou, Reference Alexopoulou1999). In Spanish, “all new information” questions trigger preferably SVO responses (10) (Hoot & Leal, Reference Hoot and Leal2023; Perpiñán, Reference Perpiñán2011), whereas “wide focus questions” are compatible with both ObjClVS and SObjClV responses (12), at least in Argentinian/Chilean Spanish (Giannakou et al., Reference Giannakou, Haska, Daskalaki and Chondrogianni2024; Giannakou & Georgiou, Reference Giannakou and Georgiou2024).
Finally, in embedded wh-dependencies, such as object-questions and object-relatives, both Greek (13–15) and Spanish (14–16) require VS (Cuza, Reference Cuza2013; Kotzoglou, Reference Kotzoglou2006), though this appears to be less strict in the case of object relatives (Guasti et al., Reference Guasti, Stavrakaki and Arosio2012; Perpiñán, Reference Perpiñán2011) (Table 1).
5.4. Previous elicited production studies targeting Greek cognates, referential forms and placement
Currently, there are four studies that used elicited production tasks similar to the current study (Vogindroukas et al., Reference Vogindroukas, Protopapas and Sideridis2009 for vocabulary, and Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019, for reference and word order) with different groups of Greek speakers to target vocabulary (cognates/non-cognates), subject/object form (in reference maintenance contexts) and subject placement (in wide focus declaratives, embedded wh-interrogatives and object relatives). We discuss them in the following sections.
5.4.1. Monolingual adult data
To investigate whether there is microvariation between Greek and Spanish in referential forms and subject-verb word order, Giannakou et al. (Reference Giannakou, Haska, Daskalaki and Chondrogianni2024) collected data from Greek-speaking monolingual adults from Greece and Spanish-speaking monolingual adults from South America (Chile, Argentina and Uruguay). Results revealed that even though both groups favored the use of reduced forms (in reference maintenance contexts) and VS (in wide focus and wh-dependencies), the Spanish group showed a more variable performance: In the conditions targeting reference, the Greek group produced exclusively reduced forms, whereas the Spanish group produced some lexical DPs (4% in Subject Contexts; 6% in Object Contexts). In the conditions targeting word order, the Greek group produced exclusively VS across structures, whereas the Spanish group produced some SVs (44% in Wide Focus; 8% in Embedded Interrogatives; 27% in Object Relatives).
5.4.2. Monolingual L1 acquisition of Greek
Daskalaki et al. (Reference Daskalaki, Chondrogianni and Blom2022) found that younger (3;6–5;9) and older monolingual Greek-speaking children (6;5–14;9), born and raised in Athens, Greece, used exclusively felicitous null subjects and object clitics as referential forms, largely in line with results reported in previous elicited production studies with Greek monolingual children (Chondrogianni, Reference Chondrogianni, Haznedar and Gavruseva2008; Varlokosta et al., Reference Varlokosta, Belletti, Costa, Friedmann, Gavarró, Grohmann, Guasti, Tuller, Lobo, Anđelković, Argemí, Avram, Berends, Brunetto, Delage, Ezeizabarrena, Fattal, Haman, van Hout and Yatsushiro2016). In the subject placement conditions (wh-embedded interrogatives and object relatives), the older monolingual group performed at ceiling in producing almost exclusively grammatical VS; however, the younger monolingual group occasionally omitted the subject, an error pattern associated with younger children (see also Cuza, Reference Cuza2016 for young Spanish monolingual children). Crucially, though, none of the monolingual groups produced any tokens of SV word order, a pattern favored by the Greek-English bilingual populations. The authors concluded that the overuse of SVs is most likely due to the special circumstances associated with the children’s bilingual experience.
5.4.3. HL acquisition of Greek
Recent studies with Greek-English bilingual children in Europe (Argyri & Sorace, Reference Argyri and Sorace2007) and North America (Chondrogianni & Daskalaki, Reference Chondrogianni and Daskalaki2023; Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019, Reference Daskalaki, Blom, Chondrogianni and Paradis2020, Reference Daskalaki, Chondrogianni and Blom2022) have replicated the differential patterns reported with child HSs of other null-subject/object clitic languages (see Cuza, Reference Cuza2016, and Shin et al., Reference Shin, Cuza and Sánchez2023, for heritage Spanish).
For instance, Daskalaki et al. (Reference Daskalaki, Blom, Chondrogianni and Paradis2020) focused on subject placement in wide focus declaratives and embedded wh-interrogatives. Based on data collected from child HSs of Greek born and raised in western Canada (6;03–18;04), they reported that children produced ungrammatical SV in both structures in different degrees: They produced a higher rate of SV in the wide focus condition (63%), where VS is a discourse preference, than in the embedded interrogatives (27%), where VS is a requirement. Daskalaki et al. (Reference Daskalaki, Chondrogianni and Blom2022) extended their investigation to a larger participant sample and a wider range of structures. Specifically, they collected data from child HSs of Greek-born and raised in western Canada and New York (6;5–14;09) to test them in two language domains: subject/object form in reference maintenance contexts (where null subjects and object clitics are preferred) and subject placement in embedded wh-dependencies (where postverbal subjects are required). The bilingual children showed variable performance in both domains: In subject/object form, they produced some non-felicitous lexical DPs (9.7% in subject contexts and 26.6% in object contexts), whereas in subject placement, they produced some ungrammatical SVs (34.5% in embedded wh-interrogatives and 40.3% in object relatives). The authors concluded that, whereas these results seem to be consistent with a CLI explanation, studying Greek in contact with a typologically similar language would elucidate the possible contribution of other factors, such as HL use. Finally, Chondrogianni and Daskalaki (Reference Chondrogianni and Daskalaki2023) tested the same group of children as Daskalaki et al. (Reference Daskalaki, Chondrogianni and Blom2022) but extended their investigation to the acquisition of cognate effects in vocabulary scores and subject placement in both embedded wh-interrogatives and wide focus structures. For vocabulary, they found that bilingual children obtained higher accuracy scores with cognates than with non-cognates, replicating the cognate facilitation effect widely shown in other language combinations (see e.g., Dijkstra et al., Reference Dijkstra, Grainger and van Heuven1999). In subject placement, bilingual children produced some SVs in wide focus structures, as well as in embedded wh-interrogatives, though to a lesser degree (in line with Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019).
Significantly, all three studies report HL use effects across domains. The higher the HL use, the higher the accuracy in the vocabulary task, and the higher the rate of felicitous reduced forms and postverbal subjects. Thus, their results are consistent with both a CLI and an HL use interpretation. To understand the relative contribution of these two factors, we investigate in the present study Greek under the influence of a typologically similar (Spanish) and a typologically dissimilar (English) language, while accounting for HL use.
6. Present study
The overall aim of the present study is to determine how CLI and HL use, differentially and in tandem, affect HL acquisition. To this end, we extended previous studies on the acquisition of Greek as an HL in North America (Chondrogianni & Daskalaki, Reference Chondrogianni and Daskalaki2023; Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019, Reference Daskalaki, Blom, Chondrogianni and Paradis2020, Reference Daskalaki, Chondrogianni and Blom2022) in two main respects. First, instead of comparing the Greek-English North American (N.A.) group with a monolingual control group, we compared it with a Greek-Spanish bilingual group in South America (S.A.). Second, differently from the previous studies that either focused on cognate effects and word order (Chondrogianni & Daskalaki, Reference Chondrogianni and Daskalaki2023) or on reference and word order (Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019, Reference Daskalaki, Blom, Chondrogianni and Paradis2020, Reference Daskalaki, Chondrogianni and Blom2022), we compared the two bilingual groups across all three subdomains: (non)-cognate vocabulary items, reference (form of referential expressions) and word order (placement of referential expressions). Specifically, we asked:
-
1. How do N.A. and S.A. children compare in:
-
a. cognate effects across the two languages?
-
b. the form and placement of Greek referential expressions?
-
2. Does the quantity of Greek language use at home affect child HSs’ performance in:
-
a. cognate effects across the two languages?
-
b. the form and placement of Greek referential expressions?
Importantly for our study, the S.A. group received less HL input compared to the N.A. group but was exposed to an ML that is typologically closer to Greek across all three domains of interest. This configuration allows us to make the following predictions: For research question 1, we predict that if there is a facilitation due to CLI, then the S.A. children may outperform the N.A. children, despite being disadvantaged in terms of HL input. We perceive facilitation as positive CLI that may occur due to exposure to similar phonological and structural patterns in the ML. Our rationale is the following:
-
(i) Greek-Spanish cognates are phonologically closer than Greek-English cognates (see Supplementary Appendix 2 and Method). If this degree of phonological proximity boosts cognate acquisition, then the S.A. group may show a stronger cognate facilitation effect compared to the N.A. group (Bosma et al., Reference Bosma, Blom, Hoekstra and Versloot2019).
-
(ii) In reference, Greek, like Spanish (and unlike English), favors the use of reduced forms (null-subjects and object clitics) in reference maintenance contexts. If the use of reduced forms in the ML facilitates the acquisition of reduced forms in the equivalent HL structures, then the S.A. group will produce a higher rate of null subjects and object clitics than the N.A. group (Torregrossa et al., Reference Torregrossa, Andreou, Bongartz and Tsimpli2021).
-
(iii) In word order, Greek, like Spanish (and unlike English), favors postverbal subjects in finite declarative sentences and wh-dependencies (though, as shown in Giannakou et al., Reference Giannakou, Haska, Daskalaki and Chondrogianni2024, in different degrees). If VS use in the ML facilitates VS acquisition in the equivalent HL structures, then the S.A. group will produce a higher rate of grammatical/felicitous VS than the N.A. group.
For research question 2, we predicted that if the amount of Greek use influences HL outcomes, children’s performance will improve as a function of Greek use regardless of ML. In particular, a higher frequency of HL use at home will be associated with higher accuracy on Greek vocabulary (both cognates and non-cognate items), a higher rate of felicitous reduced forms (null subjects/object clitics) and a higher rate of grammatical/felicitous postverbal subjects. Lastly, we entertained the possibility that the HL use effect may be less pronounced in the S.A. group, revealing facilitative CLI due to higher similarity between the HL-ML target structures. This is based on Fridman et al. (Reference Fridman, Polinsky and Meir2024), who found that HL use effects were less robust for the phenomena that were encoded similarly in the two languages of bilingual children.
7. Method
7.1. Participants
We compared data from two groups of child HSs of Greek: a group of 27 Greek-English bilingual children residing in North America (N.A.) (15 from western Canada and 12 from New York) and a group of 29 Greek-Spanish bilingual children residing in South America (S.A.) (18 from Argentina, 10 from Chile and 1 from Uruguay). In both sites, recruitment was facilitated by Greek HL schools and Greek cultural community centers.
All children in both groups were either born and raised in the host country or moved from another country before the age of three, as per our selection criteria. Furthermore, no children in our sample used a third language at home (besides Greek and the ML) or had a language impairment diagnosis. Descriptive statistics for both groups are presented in Table 1.
The N.A. children were on average 120.1 months old (SD: 28.15, R: 78–187) and belonged to different generations of immigration: Seven children were second generation (both parents first generation), 11 children were 2.5 generation (one parent first/other parent second) and nine children were third generation (both parents second). All children were enrolled in a mainstream English-speaking school and attended Greek language classes in community schools operating at weekends (for approximately four hours per week).
The S.A. children were on average 113.3 months old (SD: 33.23, R: 60–190). Most of these children (N=22) belonged to mixed families (one parent of Greek origin/other parent of Argentinian/Chilean/Uruguayan origin). Of the remaining five children, two were 2.5 generation and three were third generation. In terms of schooling, 10 children attended mainstream Spanish-speaking speaking schools, whereas 18 children attended Spanish-English bilingual schools, and one child attended a Spanish-French bilingual school. In addition, all children attended Greek language classes for a few hours per week, either in HL schools or through private tutoring.
The two groups were tested in different time periods: The S.A. children were tested through Zoom toward the end of the pandemic (in 2022 and 2023). The N.A. children were tested in person between 2016 and 2018 and were selected from a larger sample of 62 children (Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019), so as to match the S.A. children on age. To match the two samples, we closely age-matched the N.A. participants from the larger sample in Daskalaki et al. (Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019) to each S.A. participant (with an age difference of no more than 12 months). This resulted in the subsample of 27 N.A. participants that we use in the present study. Two of the youngest S.A. participants could not be age-matched to any of the N.A. participants. However, they were included in the final analysis, as the two groups did not differ in age overall.
7.2. Tasks
7.2.1. Parental questionnaire
To obtain background information about the family language history and HL/ML use, we employed a parental questionnaire (ALEQ_Heritage; Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019; adapted from Paradis, Reference Paradis2011). This includes demographic questions (age, SES, generation of immigration), and questions indexing Greek language exposure, such as the children’s age of systematic exposure to the ML through schooling (AoA), the amount of Greek language use at home, both at the time of testing and in early childhood (Current/Early Amount of HL Use), the frequency of Greek language activities (GR richness) and the length of visits from and to Greece over the last four years (See Descriptives in Table 2).Footnote 1
Table 2. Participants’ characteristics (mean, standard deviation, range)
Notes: Age = chronological age; AoA = age of systematic exposure to majority language schooling; Current Amount of HL Use = the proportion of HL input and output that the child received from and directed to other family members at home, at the time of testing; calculated on a scale from 0 to 4 (0 = HL almost never/ML almost always; 4 = HL almost always/ML almost never); Early Amount of HL Use = the proportion of HL input that the child received from other family members at home, before the age of three; calculated on a scale from 0 to 4 (0 = HL almost never/ML almost always; 4 = HL almost always/ML almost never); HL richness = the proportion of extra-curricular activities in Greek; calculated on a scale from 0 to 2 (0 = almost never; 1 = at least once per week; 2 = everyday); SES = years of maternal education; trips to GR = the total number of weeks that the child spent in Greece in the last four years; trips from GR = the total number of weeks friends/relatives from Greece spent in the host country in the last four years.
Independent samples t-tests revealed that the two groups were matched on age (t(53.54) = 0.82, p = .41). However, they differed in SES (t(51.52) = −2.15, p = .03), as S.A. mothers had, on average, more years of education compared to N.A. mothers. The two groups also differed on most measures indexing Greek exposure, with the S.A. group receiving overall less Greek than the N.A. group. In particular, the S.A. group had a lower AoA of the ML (suggesting a shorter period of exposure to HL Greek and less time to consolidate the HL), used Greek less often with family members both in early childhood and at the time of testing and had shorter visits from and to Greece in the last four years. The overall lower Greek language exposure among S.A. children could be due to a variety of reasons. Notably, the majority of S.A. children had a mixed family background (only one parent was Greek/of Greek descent). This may have put them at a relative disadvantage in terms of HL acquisition compared to N.A. children who, in principle, could use Greek with both parents. Furthermore, the S.A. children were tested toward the end of the pandemic, which means that in the target period, travel restrictions were in place. Therefore, they were more constrained in their visits from and to Greece compared to the N.A. children.
7.2.2. Picture naming task
To assess children’s expressive vocabulary in Greek, we used a picture naming task by standardized with monolingual Greek-speaking children (Vogindroukas et al., Reference Vogindroukas, Protopapas and Sideridis2009). In this task, children were presented with a total of 50 black-and-white flashcards and were asked to name the object depicted on the flashcard. Thirteen items in the task consisted of cognate words between Greek and English. For Argentinian-Uruguayan Spanish (River Plate Spanish), there were 13 cognates, while for Chilean Spanish, 12. Of these cognate items, 11 were shared across the three languages (e.g., GR /bluza/, ENG /blaus/, SP /blusa/). There were also two items that were cognates only in Greek-English, and another two that were cognates solely in Greek-Spanish (see Supplementary Appendix 2). To classify words as cognates between Greek and English or Spanish, we relied on the phonological resemblance between the Greek and the English/Spanish word, following the IPA Levenstein’s distance that measures the number of phoneme substitutions and deletions between languages, as well as the syllable structure similarity (Dijkstra et al., Reference Dijkstra, Grainger and van Heuven1999). According to this measure, the mean Levenstein distance for cognates between Spanish and Greek was .76 (range = .50–1, SD = .17) (the closer the value is to 1, the higher the similarity between cognates), and for the English-Greek cognates was .47 (range = .29–80, SD = .15), confirming that cognates in Spanish and Greek had higher phonological similarity than the English-Greek cognates (see Supplementary Appendix 2).
To score the children’s responses, we followed the vocabulary scoring conventions developed by Haman et al. (Reference Haman, Łuniewska, Pomiechowska, Armon-Lotem, de Jong and Meir2015): correct responses were given a value of “1” and included target forms, as well as regional variants (e.g., /xo.xlios/ instead of /sa.li.gari/ “snail”) and synonyms (e.g., /xr.isa.fi.ka/ instead of /ko.smi.ma.ta/ “jewelry”). Erroneous responses were given a value of “0” and consisted primarily of no responses. Mispronounced words and words with wrong inflections were coded as correct. Children were administered all items, and no discontinuity rule was applied.
7.2.3. Elicited production task
To examine children’s knowledge of the form and placement of referential expressions in Greek, we used Daskalaki et al.’s (Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019) elicited production task. The task consists of four conditions with eight items per condition: one condition targeting subject form (the Topic Continuity condition), two conditions targeting subject placement (the Embedded Interrogative condition and the Object Relative condition) and one condition analyzed for both object form and subject placement (the Wide Focus condition).
For each condition, participants were shown a picture of animated characters, and were subsequently asked a question that was meant to prompt the production of the target structure. In the Topic Continuity (TC) condition, participants were shown the picture of a character involved in a certain activity. Subsequently, they were asked to give a reason for the character’s activity (e.g., giati pije sto periptero o Jianis? “Why did Janis go to the kiosk?”), and were instructed to begin their response with the word epidhi “because.” The felicitous response in Greek involved the use of a null subject pronoun to maintain reference with a salient subject antecedent.
In the Wide Focus (WF) condition, participants were presented with a picture depicting one animate character acting on an object (e.g., a dog and a ball) and a second animate character (e.g., a girl) watching the activity. They were, then, asked a WF question of the type Ti ejine i bala tis kopelas? “What happened to the girl’s ball?” The felicitous answer in Greek involved an object clitic to refer to the previously mentioned salient antecedent (see the preferred responses in the three languages under consideration in Table 3).
Table 3. Target responses in Greek, Spanish and English
Turning to subject placement, in the Embedded Interrogative (EI) condition, participants were presented with a picture depicting a grandparent who complained about not remembering his/her grandchild’s activities. Participants were then prompted to complete a sentence of the sort i jaja den thimate ti … “Grandmother doesn’t remember what …”. The target structure involved an embedded interrogative, which in Greek requires a VS.
In the Object Relative (OR) condition, participants were presented with a picture depicting one animate character acting on an object and a second animate character pointing to the object (e.g., a girl eating an ice-cream and a boy pointing to the ice-cream). Subsequently, participants were asked to explain what the boy was pointing to and were instructed to start their reply with the sentence To aghori mas dhixni to [paghoto] pu ‥. “The boy is pointing to the [ice-cream] that …” The target structure was an object related to the VS word order. Finally, the Wide Focus (WF) condition, discussed above, was also analyzed for subject placement, offering a context in which Greek VS word order is preferred (the preferred/required responses in the three languages under consideration are summarized in Table 3).
In terms of coding, all responses with a null subject (TC) and with an object clitic in (WF) were coded as correct and given a value of “1,” and all responses with a strong referential form (pronominal/lexical) were coded as incorrect and were given a value of “0.” In the conditions targeting subject placement (EI, OR, WF), all responses with VS were coded as correct and were given a value of “1,” whereas responses with preverbal or null/omitted subjects were coded as incorrect and were given a value of “0.” Incomprehensible responses, responses that did not include the target structure, and responses with missing subjects, missing verbs or English/Spanish verbs were coded as “n/a,” and were excluded from the analysis. This corresponded to 7.48% of the responses (67 of 896 datapoints) in the conditions targeting subject/object form, and to 8.63% of the responses (116 of 1,344 datapoints) in the conditions targeting subject placement.
7.3. Procedures
7.3.1. Testing
The N.A. children participated in one hourly session that consisted of a battery of tasks, including the vocabulary task (Vogindroukas et al., Reference Vogindroukas, Protopapas and Sideridis2009) and the elicited production task (Daskalaki et al., Reference Daskalaki, Chondrogianni, Blom, Argyri and Paradis2019). The parental questionnaire (ALEQ_Heritage) was administered over the phone or through face-to-face interviews, either in Greek or in English. The S.A. children were tested for approximately 60 minutes over Zoom. At the end of the online session, parents completed the ALEQ_Heritage either in Greek or in Spanish. In both the N.A. and S.A. studies, all the tasks, apart from the questionnaire, were video/audio-recorded.
8. Analyses and results
8.1. Analyses
Data analysis was carried out using the R statistical software (version 4.3.2; R Core Team, 2021), first descriptively and then inferentially using mixed-effects regression models using the lme4 package (Bates et al., Reference Bates, Maechler, Bolker and Walker2015). Visualizations were performed using the yarr package in R (Phillips, Reference Phillips2017). To answer RQ1 and 2, we ran three generalized mixed-effects logistic regression models with a binomial distribution, each model corresponding to one of the three target domains (vocabulary, form of referential expressions and placement of referential expressions) with accuracy (1=accurate, 0=non-accurate) as the dependent variable. In the model targeting vocabulary, fixed effects included group (N.A., S.A.), GR use (operationalized as a continuous variable), cognate status (cognate, non-cognate) and the interaction between the three. In the models targeting the form and placement of referential expressions, fixed effects included group (N.A., S.A.), GR use, condition (TC, WF for reference; EI, OR, WF for subject placement). To determine whether the effect of condition and HL use was modulated by the target group, we also looked for interactions between condition and group and HL use and group. All three models included random intercepts for Participant and for Items. Because the two groups differed in SES, it was included as a covariate in the analysis. Post hoc comparisons were carried out using the emmeans package (Lenth, Reference Lenth2023).
To increase the conceptual simplicity of our models and avoid overfitting, we followed backwards selection on our fixed effects, removing those interactions or terms that were not contributing to the model. At each step, the reduced model was compared with the previous model using a log likelihood ratio test, and the reduced model was retained when it did not entail a significant loss of model fit. All continuous fixed effects/covariates were scaled and centered. The data and the scripts are available on OSF.
8.2. Results
8.2.1. Cognate effects
On the vocabulary task, the N.A. children had an overall accuracy of .41 (range = 0–1, SD=.49) and the S.A. children of .39 (range =0–1, SD=.49); the two groups did not differ from each other (p>.1). To investigate the effect of cognate status and group (N.A., S.A.) on children’s performance, we calculated children’s accuracy on cognates and non-cognates. The N.A. children had a mean accuracy of .64 (range= 0–1, SD= .48) on cognates and .34 (range = 0–1, SD=.47) on non-cognates, whereas S.A. children’s accuracy was .79 (range=0–1, SD=.40) and .25 (range=0–1, SD=.44) on cognates and non-cognates, respectively (Figure 1).
Figure 1. Mean proportion correct on cognates and non-cognates in the N.A. and S.A. group.
The mixed effects regression model revealed a main effect of cognate status (E=1.65, SE=.33, z=3.51, p=.0004), an effect of HL use (E=.79, SE=34, z=2.31, p=.02) and a significant interaction between cognate status and group (E=1.11, SE=.32, z=3.54, p=.0004). There was no overall difference between the two groups (E=−0.79, SE=.45, z=−1.78, p=.08) and no effect of SES (E=.15, SE=.22, z=.70, p=.483). The cognate by group interaction was due to the cognate effect being much stronger in the S.A. (E=3.46, SE=.46, z=7.47, p<.00001) compared to the N.A. group (E=2.5, SE=.73, z=3.43, p<.001). A subsequent analysis for each group separately revealed that in the S.A. group, there was an overall effect of HL use, such that higher HL was associated with higher accuracy (E=.73, SE=.24, z=3.07, p= .002). However, the effect was less pronounced for cognates than for non-cognates (E=−.67, SE=.18, z=−3.75, p< .001). In the N.A. group, on the other hand, we only observed an overall facilitation of HL use irrespective of cognate status (E=2.49, SE=.73, z=3.39, p<.001) (Figure 2).
Figure 2. Interaction between HL use and cognate status for the N.A. and the S.A. children.
8.2.2. Form and placement of referential expressions
The descriptive analysis (Figure 3) showed that the N.A. group had an overall accuracy of .89 with null subjects in the TC condition (R=.33–1; SD=.19), and .68 with object clitics in the WF condition (R=0–1; SD=.39). Similarly, the S.A. group had an average accuracy of .88 with null subjects (R=.37–1, SD=.18) and .73 with object clitics (R=0–1, SD=.34) in the TC and the WF conditions, respectively.
Figure 3. Proportion of accurate responses in the conditions targeting the form of referential expressions.
In contrast, in the conditions targeting the placement of referential expressions, the N.A. group produced fewer VS structures compared to the S.A. group (Figure 4) with an average accuracy of .3 in the WF (R=0–1, SD=.38), .66 in the EI (R=0–1, SD=.35), and .55 in the OR condition (R=0–1). The S.A. group, on the other hand, had an average accuracy of .58 in the WF (R=0–1, SD=.31), .84 in the EI (R=0–1, SD=.31) and .86 in the OR (R=0–1, SD=.25).
Figure 4. Proportion of accurate responses in the conditions targeting the placement of referential expressions.
Subsequently, we ran two mixed effects logistic regressions, one for the conditions targeting form (TC, WF) and another one for the conditions targeting placement (WF, EI, OR). For the form of referential expressions, results revealed main effects of Condition (accuracy was lower in the WF condition targeting object clitics), a marginal effect of HL use (children who used Greek at home more often were more likely to produce accurate responses), and a marginally significant interaction between Group and Condition, suggesting that the difference between null subjects (TC) and object clitics (WF) was less pronounced in the S.A. group. A likelihood ratio test revealed that the more complex model (with the interaction) was significantly better than the simple model (without the interaction) (p = .043). Therefore, the complex model was retained. The estimates for the optimal model are presented in Table 4.
Table 4. Optimal model for the form of referential expressions
Note: *p<.05; ***p<.001
Post-hoc comparisons revealed that the two groups differed neither in the TC condition (p=.73) nor in the WF condition (p= .36).
Table 5 presents the estimates for the optimal model for the placement of referential expressions. There were main effects of Group (the S.A. group was more accurate than the N.A. group), Condition (accuracy was lower in the WF condition than in the other two conditions) and HL use (children who used Greek at home more often were more likely to produce accurate responses). There was also a significant interaction between Group and Condition, suggesting that the difference between the WF and the EI condition was less pronounced in the S.A. group (p = .004).
Table 5. Optimal model for the placement of referential expressions
Note: *p<.05; **p<.01; ***p<.001
Post-hoc comparisons revealed that the two groups differed significantly in all three conditions. In particular, the S.A. group was significantly more likely than the N.A. group to produce an accurate postverbal subject in the WF condition (p = <.001), the EI condition (p = .036) and the OR condition (p = .001).
To unpack the interaction between group and condition, we ran two separate models, one for the N.A. and one for the S.A. group. It was found that in the N.A. group, accuracy both in the EI condition (E=4.01, SE=.48, z=8.28, p<.001) and the OR condition (E=3.08, SE=.46, z=6.66, p<.001) was significantly higher than accuracy in the WF condition. Changing the reference level further showed that accuracy in the OR was lower than accuracy in the EI (E=−.92, SE=.37, z=−2.4, p=.01). Similarly, in the S.A. group, accuracy both in the EI (E=2.94, SE=.3, z=9.94, p<.001) and the OR condition (E=2.59, SE=.29, z=8.79, p<.001) was higher than accuracy in the WF. However, changing the reference level revealed no significant differences between the OR and the EI condition (E=−.35, SE=.28, z=−1.23, p=.21).
9. Discussion
This study examined the effect of CLI and HL use by comparing Greek child HSs speaking different MLs (English and Spanish) across a range of language domains (vocabulary, reference, word order). Specifically, we analyzed production data collected from a group of Greek-Spanish bilingual children in South America (S.A.) and a group of Greek-English bilingual children in North America (N.A.). Crucially, the S.A. group had lower HL use than the N.A. group; however, the S.A. children were exposed to an ML that is typologically closer to Greek across all three domains of interest (Giannakou et al., Reference Giannakou, Haska, Daskalaki and Chondrogianni2024). We hypothesized that if, above and beyond HL use, CLI facilitates HL acquisition, then the S.A. children may outperform the N.A. children, despite receiving overall less HL input. In particular:
-
(i) The S.A. children may show a stronger cognate effect compared to the N.A. children
-
(ii) The S.A. children will produce a higher rate of reduced forms compared to the N.A. children
-
(iii) The S.A. children will produce a higher rate of post-verbal subjects compared to the N.A. children.
Turning to the role of HL use, we predicted that HL use would have a positive effect across groups and domains, though it may turn out to be less crucial for the structures in the S.A. group, where there is higher lexical/phonological and structural overlap between Greek and Spanish. Our results were partially in line with our hypotheses.
9.1. Cognate effects
In the vocabulary task, the S.A. group displayed a stronger cognate effect compared to the N.A. group, despite the finding that the two groups did not differ in their overall vocabulary accuracy. This supports our hypothesis that the stronger phonological similarity between Greek-Spanish cognates compared to Greek-English cognates may facilitate the acquisition of cognate vocabulary items. Recall from our introduction that existing studies have shown that cognate effects depend on the degree of phonological similarity between different cognate pairs of the same language combination (e.g., Bosma et al., Reference Bosma, Blom, Hoekstra and Versloot2019). Our results complement these studies by showing that the degree of phonological similarity between the same cognate pairs across different language combinations may also be relevant.
What is more, child HSs’ vocabulary, independently from group, benefited from HL use at home (in line with Chondrogianni & Daskalaki, Reference Chondrogianni and Daskalaki2023; Dosi et al., Reference Dosi, Papadopoulou and Tsimpli2024). Further analyses for each group separately revealed that, whereas HL use affected equally cognates and non-cognates in the N.A. group, in the S.A. group, cognates benefited less from exposure than non-cognates. The finding that for the Greek-Spanish cognates, exposure mattered even less suggests that the closer phonological overlap between Spanish and Greek cognates, and the potential positive phonological/lexical CLI from Spanish to Greek, may counteract fluctuations in HL use.
9.2. Reference
In the conditions targeting third-person reference, there were no significant differences between the two groups. These results are in line with other cross-national/cross-linguistic studies that did not find CLI/ML effects with clitics (Rinke & Flores, Reference Rinke and Flores2018; Torregrossa et al., Reference Torregrossa, Flores and Rinke2023) and partially in line with Torregrossa et al. (Reference Torregrossa, Andreou, Bongartz and Tsimpli2021), who found CLI/ML effects with null subjects, though not with object clitics. It is important to acknowledge that even though these null results do not allow us to make a strong claim in favor of CLI, they may still be consistent with the hypothesis that typological proximity may facilitate HL acquisition of reference. This is because with less HL use, S.A. children managed to perform as well as the N.A. children. We will return to this point in our concluding remarks.
Turning to the role of HL use, in both language groups, there was a marginally significant effect of HL use. That is, independently of the properties of the ML, there was a trend for the rate of felicitous reduced forms (null subjects and object clitics) to increase as a function of HL use.
We also found that both groups were more likely to produce a null subject (in the Topic Continuity condition) than an object clitic (in the Wide Focus condition) and that they both tended to replace reduced forms with lexical DPs rather than with overt pronouns (in line with Daskalaki et al., Reference Daskalaki, Chondrogianni and Blom2022; Torregrossa et al., Reference Torregrossa, Andreou, Bongartz and Tsimpli2021). Note that a similar subject-object discrepancy has been reported in Daskalaki et al.’s (Reference Daskalaki, Chondrogianni and Blom2022) elicited production study, as well as in a number of studies using narratives (Andreou et al., Reference Andreou, Knopp, Bongartz, Tsimpli, Roberts, McManus, Vanek and Trenkic2015; Serratrice, Reference Serratrice2007). This discrepancy could be related to the inherent complexity of object clitics. Clitics are complex in the sense that they require the coordination of pragmatic knowledge (related to their discourse distribution), morphological knowledge (related to their inflection) and syntactic knowledge (related to their placement) (Serratrice, Reference Serratrice2007). It could also be related to the fact that in our task, subject antecedents (in the TC condition) were animate, whereas object antecedents (in the WF condition) were inanimate. Given that animate antecedents are more salient than inanimate antecedents, they may favor the use of reduced forms. Indeed, Shin et al. (Reference Shin, Cuza and Sánchez2023) in a recent study with child HSs of Spanish found that animate object antecedents were more likely to be retrieved by clitics compared to inanimate antecedents.
9.3. Word order
While in the conditions targeting reference, there were no significant differences between the two groups, in the conditions targeting word order, the S.A. group outperformed the N.A. group. That is, the S.A. children were more likely than the N.A. children to produce grammatical VS in the targeted structures, a result supporting our hypothesis that the availability of VS in the ML may facilitate the use of VS in the HL equivalent structures.
In addition to the ML effects, we found that, in both language groups, the rate of felicitous VS increased as a function of HL use. Thus, word order benefited from HL use independently of the typological proximity of the ML.
Lastly, as revealed by the interactions between group and condition, the two language groups differed not only quantitatively but also qualitatively. While the S.A. group showed a two-way distinction between the narrow syntactic conditions (EI and OR) (higher accuracy scores) and the discourse condition (WF) (lower accuracy scores), the N.A. group showed a three-way distinction: They obtained higher accuracy scores in the EI condition, followed by the OR condition, and, finally, the WF condition. In this regard, the S.A. variety of Greek seems to be developing a system that stands between the monolingual variety of Greek (no distinction across conditions) and the monolingual variety of Spanish (three-way distinction across conditions) (see Language Properties).
10. Conclusions
Taken together, the results of this study support three main conclusions. Our first conclusion is that CLI is at play in HL acquisition. Despite having less HL use and less time overall to consolidate the HL, as the different profiles of the two groups revealed (Table 1), the S.A. children outperformed the N.A. children in the conditions targeting word order and performed comparably with the N.A. children in the conditions targeting reference. The S.A. children were also facilitated by the phonological overlap in cognates in the domain of vocabulary. These results suggest that, above and beyond HL use, the use of VS, reduced forms and the lexical overlap in the equivalent ML structures (and more generally the availability of reduced forms, VS and similar phonology in the ML) may facilitate children’s HL acquisition.
Second, the language domain also matters. The absence of facilitation in the domain of reference could be taken to suggest that, above and beyond similarity at an abstract level (licensing of null subjects through verbal agreement/availability of object clitics), similarity at a surface/phonological level may also be relevant (Fridman et al., Reference Fridman, Polinsky and Meir2024). For instance, even though both Greek and Spanish license preverbal clitics, Greek clitics sound different (GR: ton-tin-to; SP: le-la) and realize more gender distinctions than Spanish clitics. Independent support for the role of surface similarity comes from the domain of morphology. As discussed in our introduction, Fridman et al.’s (Reference Fridman, Polinsky and Meir2024) study on heritage Russian found CLI effects solely in the domains in which the HL structure and the equivalent ML structure displayed both abstract and surface similarity (adjectival agreement in HL Russian and ML Hebrew). By contrast, in the structures in which there was no surface similarity (case marking in HL Russian and ML Hebrew), the Russian-Hebrew group did not outperform the Russian-English group.
Our third conclusion concerns the role of HL use. In line with the studies reviewed in Chondrogianni (Reference Chondrogianni, Elgort, Siyanova-Chanturia and Brysbaert2023) and van Dijk et al. (Reference van Dijk, van Wonderen, Koutamanis, Kootstra, Dijkstra and Unsworth2022), our results suggest that CLI and HL use are not mutually exclusive. Whereas CLI explains why the S.A. children performed on average better than the N.A. children, the amount of HL use explains the observed within-group variability.
10.1. Implications
Turning to the broader implications of our study, our results inform theories of CLI and the conditions under which it may take place. We argued that the S.A. children outperformed the N.A. children due to ML facilitation. Facilitation can be perceived either as positive CLI or as diminished negative CLI. Under the former interpretation (which is what we have been assuming in the present study), the S.A. children outperformed the N.A. children because they benefited from the presence of the same pattern (VS) in the ML. Under the latter interpretation, the S.A. children outperformed the N.A. children because the rate of competing structures (SVs) was lower in Spanish than in English. Under either interpretation, our results suggest that CLI may take place between two systems that allow the same range of options (SV and VS) but in different contexts and degrees (from Spanish to Greek).
Our second point is methodological in nature. We showed how adopting a cross-national approach allows us to disentangle the effects of CLI from other aspects of the child’s bilingual experience. However, cross-national studies are not always easy to coordinate and, as we discuss below, they have their own challenges. Other ways to isolate the effects of CLI would be to target HL structures that differ in the degree of dissimilarity with the ML (see, for instance, Cuza, Reference Cuza2016) or to target ML structures that are not easier from a processing perspective and could not be treated as the default option to which children resort due to reduced HL use. For instance, an increase in VS in the Spanish of Spanish-Greek bilingual children would be most likely due to CLI from Greek rather than to reduced HL use.
Lastly, cross-national approaches could be complemented with approaches that include HL data from the parents (e.g., Daskalaki et al., Reference Daskalaki, Blom, Chondrogianni and Paradis2020). The inclusion of parental data would enable us to disentangle the effects of CLI from the effects of parental input quality, since some of the properties found in the children’s HL (SVs and lexical DPs, in our case) may already be present, to a smaller degree, in the HL variety spoken by the parents.
10.2. Limitations and further research
There are several limitations to this study. First, differently from the N.A. children who were tested in person, the S.A. children were tested online. This may have put the S.A. children at a disadvantage due to a lack of immediate face-to-face interaction. Another limitation concerns our tasks. The vocabulary task was not designed to examine cognate effects or the role of the degree of phonological similarity between Greek-English and Greek-Spanish cognate pairs. As a result, there was no even distribution between cognate and non-cognate items, or between cognate pairs showing different degrees of phonological similarity across language combinations. Similarly, the task that we used to test object clitics was originally designed to test subject placement in WF declaratives. Thus, it might have been cognitively demanding for bilingual children; in addition to producing an object clitic to refer to a previously mentioned antecedent, they had to decide on the placement of the subject.
A final issue concerns the cross-national comparison. Heritage communities may differ from each other considerably for a variety of reasons, including differences in their density, HL schooling opportunities, the political landscape, the immigration depth of their speakers and the social network structure of the community (see Lease et al., Reference Lease, Shin and Bird-Brown2022; Shin & Van Buren, Reference Shin and Van Buren2016). Further research is needed to examine how these factors interact with language-level factors related to the typological properties of the ML.
Supplementary material
The supplementary material for this article can be found at http://doi.org/10.1017/S1366728925100722.
Data availability statement
Additional data that support the findings of this study are openly available in OSF at https://osf.io/mhy6s/?view_only=c2c8c12191714437bac6e950ed159bc1.
Acknowledgements
We thank the Killam Research Operating Fund (University of Alberta) and the Advanced Research Collaborative Fellowship (City University of New York) for funding data collection in Canada and the United States. We would also like to thank the Greek families in South America (Chile, Argentina and Uruguay) and North America (Canada and the United States) for their participation and enthusiasm. Finally, we are grateful to Eliana Kavgadoulis and Katerina Katehis for assisting with the data collection.
Competing interests
The authors declare none.
Open access
