2.1 DOM in Romanian

DOM in Romanian is sensitive to a variety of features: animacy, specificity, discourse prominence, referential stability. Since sensitivity to animacy can be overridden by specificity, discourse prominence, and referential stability in various contexts, one could argue the Romanian DOM system is not primarily triggered by animacy but rather by factors related to discourse-pragmatics (Ticio & Avram 2015). Animate and specific entities are generally DOM marked with clitic doubling and PE-marking (Farkas 1978; Dobrovie-Sorin 1994; Cornilescu 2000; Farkas & von Heusinger 2003; Mardale 2008; 2010; von Heusinger & Onea 2008; Ciovârnache & Avram 2013; Tigău 2010; 2011; Irimia 2020a, 2020b, 2023; Hill & Mardale 2021; Irimia & Guardiano 2024), homophonous with a preposition, as in (1):

Nouns occurring after PE-marking take a bare form unless they belong to a special class of nouns usually referring to a specific role (mama ‘mom’, tata ‘dad’, şef ‘boss’, a. o., Hill 2013) (2a). It is important to note that the interpretation of the bare noun is nevertheless definite. In cases where the nouns are followed by an adjective or determiners including demonstratives and possessives, among others (Dobrovie-Sorin 1994; Mardale 2008; von Heusinger & Chiriacescu 2009) (2b), the nouns present an enclitic definite article (2b).

Animacy also plays a role in DOM in Romanian. Animate objects are often PE-marked. In contrast, inanimate objects are usually not PE-marked. Nevertheless, inanimate objects can be PE-marked if they are interpreted as specific (Cornilescu 2000; Mardale 2008; Țigău 2010; 2011; Irimia 2020a, b, 2023), e.g., if they are dislocated (4a) or if they are demonstratives under ellipsis (4b).

Specific interpretations can result from DOM of inanimates, making the object prominent in the discourse. In (5a), where the object occurs in its default postverbal position, the rose, which is PE-marked and clitic doubled, is perceived as semantically upgraded to some animate entity. This effect can be further enhanced if, for instance, in (5b), the object pe trandafir ‘PE rose’ is dislocated from its default postverbal position in (5a) to the left of the sentence:

In addition to animacy, the distribution of DOM in Romanian is sensitive to referential stability (Farkas & von Heusinger 2003), defined as how the specificity of the referent varies with discourse properties. Farkas & von Heusinger (2003) propose a Referentiality stability scale, according to which the more referentially stable an object is, the more likely it will be marked (6).¹

Animacy and referential stability combine, giving rise to multiple constraints or preferences. Consequently, DOM is obligatory with proper names and definite pronouns, optional with animate definite DPs, partitives and indefinite DPs, and generally not allowed with inanimate definite DPs, partitives and indefinite DPs. Thus, for animate DPs, partitives and indefinite DPs, speakers either leave the direct object unmarked (7a) or use both PE-marking and clitic doubling (7b):

In the case of inanimate DPs, DOM and clitic doubling are usually not permitted, except for some cases where specificity takes precedence (see 4) or where the object is a personal pronoun or a relative pronoun, i.e., items which are prominent in Farkas and von Heusinger’s (2003) referential stability scale. Additionally, DOM can also mark demonstratives in an ellipsis context,² implying the presence of the previously mentioned noun that is resumed under ellipsis, as in (8). In (8), the referent of the elliptical demonstrative can be either animate (8a) or inanimate (8b):

While proper names and personal pronouns require clitic doubling, there seems to be some variation regarding clitic doubling in combination with PE-marking in animate definite DPs, with higher registers showing more clitic doubling. Avram and Zafiu (2017) and Avram (2019) provide experimental evidence from acceptability judgment tasks that there are currently two parallel DOM systems in contemporary Romanian, manifested in participants of the same age (19–21 years old): in optional contexts, conservative speakers allow both single pe and pe + accusative clitic as DOM, whereas innovative speakers only accept pe by itself as DOM (see Table 1).

2.2 DOM in Spanish

DOM in Spanish, on the other hand, is mainly triggered by animacy (Bossong 1991; Torrego 1998; Aissen 2003; Leonetti 2004; 2008; a. o.), unlike DOM in Romanian, which is triggered by discourse-pragmatic factors such as specificity, discourse prominence, and referential stability, as we have seen above. Animate objects are usually A-marked in Spanish (9a), as opposed to inanimate objects (9b). Moreover, definiteness seems to play a less important role than animacy, since indefinite human objects can be marked too, under certain conditions (9c) (Leonetti 2004; 2008). Several authors argue that specificity plays a role as well (see Brugè & Brugger 1996; Torrego 1998; 1999; Leonetti 2004). For instance, the object una secretaria ‘a secretary’ in (9c) is interpreted as specific when marked but as non-specific when left unmarked.

The contrast between Romanian and Spanish DOM regarding which feature is their main trigger leads to several mismatches in their distribution. For instance, while in Romanian personal and demonstrative pronouns are more prominent on the referential scale and DOM-marked regardless of animacy, in Spanish, animate objects are usually marked. This difference results in demonstrative pronouns being marked in Romanian in ellipsis contexts (10a) yet their marking in Spanish remains sensitive to animacy features (10b):

In addition to their distribution, DOM in Spanish is syntactically simpler than in Romanian: while DOM is often accompanied by clitic doubling in Romanian, at least for conservative speakers, in Spanish this is not the case unless the object is a personal pronoun, with some exceptions stemming from regional variations (e.g., Leonetti 2007; Sánchez & Zdrojewski 2013; Rinke et al. 2023). The participants in the present study have acquired Spanish in monolingual regions of Spain, where, to our knowledge, clitic doubling does not co-occur with DOM unless the object is a personal pronoun. In summary, Romanian and Spanish display different DOM systems: despite both of them being sensitive to animacy, Romanian DOM is mostly driven by discourse-pragmatics factors such as referential stability or discourse prominence (Farkas & von Heusinger 2003; Tigău 2010; Ticio & Avram 2015), while Spanish DOM is mainly triggered by animacy. Additionally, Romanian DOM usually presents clitic doubling while Spanish DOM does not. In our study, we explore the acquisition and maintenance of DOM in both Romanian and Spanish among Romanian-Spanish bilinguals by considering alternative strategies to express transitive structures.

2.3 DOM in Romanian in contact with other languages

3.1 Motivation and Research Questions

The present study explores whether there exist signs of L1 attrition and bidirectional cross-linguistic influence among first-generation Romanian immigrants in Spain, who engage with both Romanian and Spanish in their daily lives. This language pair is of particular interest due to both languages exhibiting DOM, unlike commonly studied language pairs such as Spanish and Romanian in contact with English. By examining narratives (in line with Montrul and Bateman 2020b and Avram et al. 2023) in both languages, we aim to extend previous findings from controlled tasks like elicited production and acceptability judgments (López Otero 2020; 2022), furthering our understanding of bilingual lexical access and language use patterns. The narratives are intended to provide a more naturalistic insight into the participants’ productive knowledge of DOM in their two languages.

Additionally, we collect participants’ information on lexical access and patterns of language use in order to explore within-group differences among the bilinguals. Lexical access is assessed using the Multilingual Naming Test (MiNT) (Gollan et al., 2012). Lexical access reflects the ability of bilingual individuals to retrieve lexical items efficiently, despite competition from simultaneous language activation. Studies have shown this metric to correlate strongly with both subjective and objective measures of language proficiency (Bedore et al., 2012; Gollan et al., 2012). Patterns of language use and extralinguistic factors, known to influence bilingual language maintenance, are evaluated using the Language Experience and Proficiency Questionnaire (LEAP-Q) (Marian et al., 2007). This tool collects data on language experiences and contextual usage, enabling us to measure language activation as a proxy for bilingual variability (Paradis, 1993; 1996; Putnam & Sánchez, 2013). Additionally, factors such as age of arrival and length of residence are examined to understand their impact on L1 attrition and L2 acquisition. In light of this, we posit the following research questions, each subdivided into two subquestions.

RQ1: DOM in L2 Spanish:

1. What transitive strategies do Romanian-speaking L2 speakers of Spanish present in their oral production in Spanish?

2. Are their transitive strategies in Spanish, particularly their use of DOM, shaped by extralinguistic variables?

Hypothesis 1a: We hypothesize that the Romanian-speaking L2 speakers of Spanish show use of transitive strategies in Spanish similar to the native Spanish speakers’, which would be congruent with a target-like acquisition of DOM in Spanish. Romanian-speaking L2 speakers of Spanish have acquired the animacy-driven nature of DOM in Spanish and their production is similar to that of monolingual Spanish speakers, in line with previous studies on the facilitative role of speaking an L1 featuring DOM (Montrul 2019; Montrul & Gürel 2015).

Hypothesis 1b: Nevertheless, the lack of activation for comprehension or production of Spanish may lead to variability in the L2 speakers’ A-marking use (Paradis 1993; 1996; Putnam & Sánchez 2013). Specifically, we expect the lack of activation of Spanish to result in at least some Romanian-Spanish bilinguals marking inanimate objects, particularly if they are ranked high in the referential stability scale (López Otero 2020; 2022).

RQ2: DOM in L1 Romanian:

1. What transitive strategies do Romanian-speaking L2 speakers of Spanish present in their oral production in Romanian?

2. Are their transitive strategies in Romanian, particularly their use of DOM, modulated by extralinguistic variables?

Hypothesis 2a: We hypothesize that the Romanian-speaking L2 speakers of Spanish show use of transitive strategies in Romanian similar to those of monolingual Romanian speakers, which would signal a target-like maintenance of DOM in Romanian, their native language. Following previous studies on the maintenance of Romanian as a home language in non-Romanian-speaking societies (Montrul & Bateman 2020a, b), we expect DOM in Romanian to remain comparable to that in Romanian monolingual populations.

Hypothesis 2b: However, the lack of activation for comprehension or production of Romanian may lead to variability in their use of PE-marking (Putnam & Sánchez 2013; López Otero 2020; 2022). Specifically, we expect the lack of activation of Romanian to result in at least some Romanian-Spanish bilinguals simplifying PE-marking in Romanian given its syntactic complexity by abandoning clitic doubling and favoring an animacy-driven distribution of their DOM due to cross-linguistic influence from Spanish, in line with previous findings (López Otero 2020; 2022).

3.2 Participants

A total of 58 subjects distributed across three groups participated in the study: 18 L1 Romanian L2 Spanish bilinguals residing in Spain (hereinafter bilinguals), 17 Spanish native speakers residing in Spain (hereinafter Spanish monolinguals, SM) and 23 Romanian monolingual native speakers living in Romania (henceforth Romanian monolinguals, RM). By means of a background questionnaire, the Language Experience and Proficiency Questionnaire (LEAP-Q) (Marian et al. 2007), we gathered information on the participants’ language background, their L2 acquisition process and their language patterns of use and exposure. The bilinguals (age range = 21–49; M = 30.89, SD = 8.04) (n = 18; female = 16) started learning Spanish after puberty (age of arrival and of onset of acquisition range = 15–37, M = 20.72, SD = 6.75) in a naturalistic setting in Spain (years of residence range = 6–17, M = 10.16, SD = 3.01). No cut-off point was initially established for participation in this study; nevertheless, previous studies on L1 attrition have established 6 years of residence (Tsimpli et al. 2004; Cairncross et al. 2023) or even 10 years (Schmid 2007) in the L2-speaking society. Additionally, none of the participants reported having studied Spanish in a classroom setting. Instead, some reported having learned other languages at school: English, French, and German. All participants reported that Spanish was their L2, and that their native and dominant language was Romanian. Their self-assessment results indicate that their overall Spanish skills are highly proficient (M = 9.01/10, SD = 0.65). When looking at specific language skills, results remain similar (speaking: M = 8.19, SD = 0.86; listening: M = 9.39, SD = 0.78; reading: M = 9.44, SD = 0.78). Their MiNT results suggest they are balanced bilinguals (range = 46–64/67; M = 57.28; SD = 3.97 for Spanish) (range = 52–65/67; M = 59.39; SD = 3.69 for Romanian). Their DELE results did not present much variability (range = 41–48/50; M = 45.83; SD = 2.01). A linear regression failed to find a correlation between the DELE and the MiNT (p = 0.9). This suggests that their DELE results do not present enough variability to capture nuances among this group of end-state L2 speakers, whereas the MiNT, which presents wider ranges in its results, might. Section 3.3. provides details on the DELE and the MiNT.

Their pattern of language use is consistent: most bilinguals (13/18) reported using Spanish more frequently than Romanian in their daily life. Specifically, some bilinguals (5/18) used Spanish more than 80% of the time. Three bilinguals, on the other hand, reported using both Spanish and Romanian equally, while only two bilinguals used more Romanian than Spanish, which they reported using only 40% of the time. More Romanian was used at home, with all bilinguals reporting using Romanian with their families while only one participant reported using both Romanian and Spanish at home. In addition, four participants reported using some Romanian and Spanish code-mixing.

Their sociolinguistic profile is also consistent: most bilinguals (15/18) were originally from Bucharest and Southeastern Romania while three participants came from different regions: two from Transylvania and one from Oltenia. All bilinguals resided in a monolingual Spanish-speaking region in Southern Spain. All participants had completed more than obligatory education: 11 had a university degree while 7 had completed technical/professional education.

The monolingual groups also featured consistent profiles: 17 SMs (age range = 19–50; M = 27.65; SD = 8.72) (female = 8) participated in the study. They were born and lived in the same monolingual Southern Spanish region as where the bilinguals were living. They had never lived abroad, and their knowledge of foreign languages was limited. Their MiNT results were slightly more accurate than the bilinguals’ (M = 62.94/67; SD = 2.16 for the SMs vs. M = 57.28/67; SD = 3.97 for the bilinguals). As opposed to the bilinguals, the majority of the SMs did not hold university degrees: only 6 SMs had graduated university while 11 had completed secondary education.

A group of 23 RMs (age range = 19–31; M = 21.67; SD = 3.11) (female = 22) participated in the study. They were college students living in Bucharest, the same region that most of the bilinguals were originally from. Like the SMs, they had never lived abroad and, although they all reported having studied English at school, Romanian was their dominant language. The RMs scored slightly higher than the bilinguals in the MiNT (M = 63.43/67; SD = 1.59 for the RMs vs. M = 59.39/67; SD = 3.69 for the bilinguals).

3.3 Tasks and procedures

The current study is part of a larger project including several tasks: two screening tasks (LEAP-Q, discussed above, and the Diploma de Español como Lengua Extranjera) and three experimental tasks (a Narrative Task, an Elicited Production Task and an Acceptability Judgment Task, presented to participants in the mentioned order). Among these experimental tasks, the current study relies solely on the narrative task. While the monolingual groups completed the narrative task only in their languages, bilinguals completed them in Spanish first and then in Romanian to avoid priming effects from their reportedly dominant language, Romanian, into Spanish, their non-dominant L2.

Diploma de Español como Lengua Extranjera (DELE). As in other versions of this test (e.g., Duffield & White 1999; Montrul & Slabakova 2003), its goal is to determine overall Spanish proficiency by assessing morphosyntactic and lexical knowledge via 50 multiple-choice questions (Cuza et al. 2013). This instrument is an adaptation of a vocabulary task from the MLA Foreign Language Test and a cloze test from the Diploma de Español como Lengua Extranjera (DELE) focusing on morphosyntax and lexicon. It is commonly referred to as the DELE by researchers in the field, yet it differs significantly from the DELE administered by the Instituto Cervantes, which includes sections on speaking, listening, reading and writing in addition to grammar. In our study, we use the term DELE to refer specifically to the adapted version described. Only the bilingual group completed the DELE, and they all scored 40 or more.

Multilingual Naming Test (MiNT). The goal of this test is to measure the participants’ lexical knowledge. The MiNT consists of 68 images of increasing difficulty presented individually. Participants are asked to identify and name the objects depicted in the images. The slide depicting the items “mortar” and “pestle” was considered one single item, resulting in the maximum MiNT score in this study being 67. The bilinguals completed the MiNT in Spanish and then in Romanian, while the monolingual groups completed it in their native languages.

Narrative Task. This task aimed to elicit semi-spontaneous use of DOM. Participants were asked to narrate the story depicted in the wordless children’s book One Frog Too Many (A Boy, a Dog, and a Frog) (Mayer & Mayer 2003). Gathering narrative data from “frog story” narratives, which feature a high number of animal characters, is consistent with previous studies on the acquisition of DOM in Romanian bilinguals (Avram et al. 2023). Participants completed the task first in Spanish, their non-dominant language, followed by their dominant language, Romanian.

3.4 Data coding and statistical analysis

For the present study, we examined the participants’ transitive structures and coded them after classifying them into several categories. In Romanian, these categories are (a) PE-marking with clitic doubling, (b) PE-marking by itself, (c) unmarked determiner phrase, (d) accusative clitic, and (e) clitic dislocation (right/left dislocation). See (11) and (12) for examples of these strategies as used by Romanian speakers. Based on a reviewer’s advice, we excluded any structures involving clitic dislocation from the category “PE-marking with clitic doubling” to more accurately identify genuine instances of DOM and clitic doubling. Clitic dislocation was instead treated as a separate, distinct category. While clitic doubling and clitic dislocation may seem similar superficially, they are syntactic phenomena of a different nature, as argued by a rich body of studies (e.g. Cinque 1990; Iatridou 1991; Anagnostopoulou 1994; 1997; 2005; Hill & Mardale 2021; a. o.). Clitic doubling keeps the canonical word order intact, using the clitic to emphasize or clarify the direct or indirect object, while clitic dislocation involves the co-occurrence of a clitic with a phrase that appears in a non-standard position (to the left or to the right of the clause), for purposes of topicalization or emphasis. Importantly, their distributional restrictions are different: doubled objects have the intonation and distribution of arguments, whereas dislocated objects have the intonation and distribution of peripheral elements. Notably, in Romanian, clitic doubling is subject to Kayne (1994)’s Generalization (i.e. it takes place only with marked DPs, i.e. DPs marked by DOM), while clitic resumption under dislocation takes place with any type of DP object.³ Prosody-wise, the dislocated phrase is set off from the rest of the sentence with an intonational break, while the clitic doubled object is not preceded by such a break. Additionally, other strategies used by the bilinguals included missing objects and clitic doubling without PE-marking, as shown in (13a) and (13b), respectively. We also coded the participants’ uses of transitive structures by attending to the animacy features of their referent: human, non-human animate, and inanimate.⁴

Note that the structure in (13b), i.e. clitic doubling of unmarked objects, is not a strategy found in Spanish. While its presence might seem to contradict the high proficiency of bilinguals in Spanish, it should be noted that this type of structure appears in only one single speaker, and very rarely (just twice). Overall, bilinguals prefer other strategies.

On the other hand, in Spanish, the participants’ transitive strategies were categorized into the same major classes Romanian structures were categorized into: (a) A-marking with clitic doubling, (b) A-marking by itself,⁵ (c) unmarked determiner phrase, (d) accusative clitic, and (e) clitic dislocation. We found no strategies in Spanish corresponding to the two additional classes observed only rarely among Romanian bilinguals (missing object and clitic doubling without PE-marking). As with the Romanian data, we also coded the participants’ uses of transitive structures by attending to the animacy features of their referent: human, non-human animate, and inanimate. See (14) and (15) for examples of these strategies as used by Spanish speakers.

The participants’ transitive strategies were the dependent variable in the statistical analyses of the present study whereas independent variables included the categorical variables group (bilinguals, SMs, RMs) and language (Romanian, Spanish) as well as the continuous variables age of onset of acquisition of Spanish, productive vocabulary knowledge, age in years, and years spent in Spain. Before running any statistical analyses, the categorical variables were sum coded, and the continuous variables were standardized.

In addition to descriptive statistics, a series of mixed-effects multinomial logistic regressions (MEMLR) and one generalized linear mixed model (GLMM) were performed to analyze our data. Specifically, sixteen MEMLRs were performed using the mblogit function in the mclogit package (Elff 2021) in R (R Core Team 2021) with transitive strategy type (PE-marking + CD, PE-marking, Unmarked DP, Clitic) as the dependent variable. Their independent variables, on the other hand, were different in every MEMLR: MEMLR1 included group (bilingual vs. RM) as dependent variable and focused on Romanian language data, MEMLR2 included group (bilingual vs. SM) as dependent variable and focused on Spanish language data, and MEMLR3 included language (Romanian vs. Spanish) and focused on bilingual data only in order to explore the effects of language on the transitive strategy choices among the bilinguals. The remaining MEMLRs focused on either the Spanish or the Romanian data from the bilinguals and explored whether their transitive strategies were shaped by extralinguistic predictors. The Spanish data were analyzed by examining predictors including age of onset of acquisition of Spanish (MEMLR_SP_4), age in years (MEMLR_SP_5), years spent in Spain (MEMLR_SP_6), amount of Spanish spoken daily (MEMLR_SP_7), productive vocabulary knowledge in Spanish or Spanish MiNT score (MEMLR_SP_8), and exposure to Spanish through reading (MEMLR_SP_9). The Romanian data were explored by investigating the role of similar predictors: age of onset of acquisition of Spanish (MEMLR_RO_4), age in years (MEMLR_RO_5), years spent in Spain (MEMLR_RO_6), amount of Romanian spoken daily (MEMLR_RO_7), productive vocabulary knowledge in Romanian or Romanian MiNT score (MEMLR_RO_8), and exposure to Spanish through reading (MEMLR_RO_9). An additional regression (MEMLR_RO_10) explored whether the strategies observed in the Romanian language dataset, particularly the use of PE-marking, PE-marking with CD and unmarked objects, were predicted by animacy (animate vs. inanimate), definiteness (definite vs. indefinite), specificity (specific vs. non-specific) or group (bilingual vs. RM). All these MEMLRs included participant as a random factor. Finally, a GLMM (GLMM1) was performed using the glmer function in the lme4 package (Bates, Kliegl, Vasishth & Baayen 2015) in order to explore variability in the use of clitic doubling in combination with PE-marking. This GLMM included response (PE-marking + CD or PE-marking) as dependent variable, and context (obligatory vs. optional) and group (bilingual vs. RM) as independent variables. The dependent variable was coded as ‘0’ versus ‘1’ while the independent variables were sum coded (‘–0.5’ vs. ‘0.5’). The dataset for this GLMM focused on the instances in which bilinguals and RMs produced the strategies under examination (i.e., PE-marking + CD or PE-marking). Two additional GLMMs explored the effects of animacy, definiteness, specificity and group in the use of DOM (1 = marking; 0 = non-marking). In these GLMMs, participant was also included as a random factor, interactions were explored, and nested model comparisons were assessed in order to find the best fit model. Appendices A and B include the R code used to analyze the data as well as their output.

4.1 Romanian results

The Romanian data included 409 instances of transitive structures produced by the bilinguals (n = 165) and the RMs (n = 242). Twenty of those were excluded because they featured clitic dislocations, clitic doubling yet no PE-marking or they presented a null object (see 12e, 13a and 13b above). Additionally, the RMs produced two passive constructions, which were not included in the analysis. Among animate objects, only thirteen objects referred to humans (precisely the boy in the story) while all the others referred to animals. The resulting 387 instances were organized into categories according to transitive structures, lexico-semantic nature of the object to which they referred, and whether they were produced by bilinguals or RMs. Tables 2, 3 and 4 below include the counts and proportions of the use of transitive strategies in Romanian among these two groups.

Table 2 indicates that the number of human objects is negligeable. Table 3 suggests that the transitive strategies employed by the two groups differ: the proportions indicate that RMs produce more PE-marking combined with clitic doubling as well as more unmarked nouns/DPs, whereas bilinguals used more clitics. Finally, Table 4 shows that both groups use the same strategies to refer to inanimate objects. MEMLR1 explored whether the two groups of Romanian speakers differ in their distribution of transitive strategies when referring to non-human animate objects. ‘Clitic’ was the intercept of this regression as it is the only strategy unrelated to the distribution of PE-marking. Table 5 below shows the results yielded by MEMLR1:

These results indicate that the two groups differ in the transitive strategies that they produce. Specifically, bilinguals are less likely to produce PE-marking combined with clitic doubling and unmarked nouns/DPs than their Romanian-speaking monolingual counterparts. These results confirm the tendencies observed in the descriptive statistics in Table 2.

GLMM2 and MEMLR_RO_10 examined the effects of animacy, definiteness, specificity and group on the use of PE-marking. GLMM2 included PE-marking versus unmarked noun/DP as dependent variable whereas MEMLR_RO_10 included PE-marking by itself, PE-marking with CD and unmarked noun/DP as dependent variable. MEMLR_RO_10 found that bilinguals are more likely to use DOM without CD than the RMs, in line with the results yielded by MEMLR1. GLMM2 failed to find significant effects.

With the goal of examining the alternance between noun objects, whether marked or not, and clitics, we analyzed the use of clitics regarding distance from their antecedents. We coded if clitics referred to an antecedent in the same sentence (coded as “0”), in the previous sentence (coded as “1”) and so forth. Only animate objects were examined as our focus was to explore alternative strategies to the use of DOM. Table 6 below shows the distribution of clitics across different degrees of distances from their antecedent in both groups.

Table 6 above shows that the bilinguals produce clitics presenting a longer distance from their antecedents than the RMs. It could be argued that RMs prefer referring to the antecedent by using either a marked or an unmarked full DP object. The bilinguals, on the other hand, feature clitics farther away from their antecedent, arguably to avoid the syntactic complexity of DOM.

4.2 Spanish results

The Spanish language data included 396 instances of transitive structures produced by the bilinguals (n = 200) and the RMs (n = 196). Seventeen of those were excluded because their interpretations were relational (e.g., Él ya tenía una rana ‘he already had a frog’; see Tippets 2010): seven of those were produced by the bilinguals and ten of them were produced by the SMs. None of them included DOM. Additionally, one bilingual speaker produced a passive construction, and a monolingual speaker produced a dislocated clitic in combination with a marked non-human animate object. Both structures were excluded from further analysis. Tables 7, 8 and 9 show the resulting 377 instances of transitive strategies included in the analyses, which were organized into the same four categories used to analyze the Romanian dataset.

MEMLR2 explored whether the two groups differ in their distribution of transitive strategies when referring to non-human animate objects. ‘Clitic’ was the intercept of this regression as it is the only strategy unrelated to the distribution of A-marking. Also, due to the low number in the ‘Clitic doubling and DOM’ category, this category was combined with the ‘DOM’ category. Table 10 below shows the results yielded by MEMLR2:

These results indicate that the two groups use these strategies at similar rates. The lack of significant group effects confirms the tendencies observed in the descriptive statistics.

GLMM3 examined the effects of animacy, definiteness, specificity and group on the use of PE-marking. GLMM3 included A-marking versus unmarked noun/DP as dependent variable. GLMM3 found that definiteness plays a role in the use of DOM in both groups of Spanish speakers (β = 2.59; SE = 1.16; z = 2.23; p = 0.03); in particular, definite objects receive more DOM than indefinite objects.

As with the Romanian data, we analyzed the use of clitics regarding distance from their antecedents following the same coding. Table 11 below shows the distribution of clitics across different degrees of distances from their antecedent in both groups.

As with the Romanian data, Table 11 shows that the bilinguals produce clitics featuring a longer distance from their antecedents than the SMs. One could argue that SMs prefer strategies other than clitics when the distance to the antecedent increases whereas the bilinguals might rely on clitic production as an avoidance strategy regarding the syntactic complexity of DOM.

4.3 Variability within the bilingual group

This section focuses on exploring predictors that may shape the variability found within the bilingual group, specifically within their Romanian production as no differences were found between their Spanish and the production of the SMs. MEMLR3, nevertheless, explored the effects of language on the bilinguals by examining the strategies that they used in both Spanish and Romanian to refer to non-human animate objects. For this analysis, DOM uses, with or without clitic doubling, were grouped into a single category. Counts and proportions of these strategies can be seen in Table 12 below.

In both languages, bilinguals favored the use of clitics, followed by DOM and finally by unmarked nouns or DPs. This similarity across languages was confirmed by the results yielded by MEMLR3, which did not find significant effects of language.

The remaining MEMLRs focus on either the Spanish or the Romanian data from the bilinguals only and explore whether their transitive strategies were shaped by extralinguistic predictors, containing information gathered with the LEAP-Q or the MiNT. For the Spanish data, MEMLR_SP_4 examined the role of age of onset of acquisition of Spanish (range = 15–37; M = 20.72; SD = 2.57) yet failed to find significant age of acquisition of Spanish effects on the bilinguals’ production of transitive strategies in Spanish. MEMLR_SP_5 explored the effects of age in years (range = 21–49; M = 30.89; SD = 8.04) on the bilinguals’ transitive strategies in Spanish, but no effects were found either. MEMLR_SP_6 focused on years spent living in Spain (range = 7–14; M = 9.61; SD = 2.57); however, no significant effects were detected. MEMLR_SP_7, which focused on how often bilinguals speak Spanish daily (range = 40–90/100; M = 63.61; SD = 15.89), did not find significant effects. The only predictor containing data from a source different than the LEAP-Q was productive vocabulary knowledge in Spanish or Spanish MiNT score. MEMLR_SP_8 explored the role of productive vocabulary knowledge in Spanish (or Spanish MiNT score: M = 57.28/67; SD = 3.97) on the bilinguals’ transitive strategies in Spanish and failed to find significant effects. Finally, MEMLR_SP_9 examined the role of exposure to Spanish through reading (range = 3–10/10; M = 8.47; SD = 2.21) and said strategies and also failed to find effects.

For the Romanian dataset, MEMLR_RO_4 investigated age of onset of acquisition of Spanish effects (range = 15–37; M = 20.72; SD = 2.57) on their transitive strategies in Romanian, but no significant effects of age of Spanish acquisition were found. MEMLR_RO_5 explored the influence of participants’ ages in years (range = 21–49; M = 30.89; SD = 8.04) on their transitive strategies in Romanian, yet no significant effects were observed. MEMLR_RO_6 examined the role of years of residence in Spain (range = 7–14; M = 9.61; SD = 2.57), but no significant effects were detected. MEMLR_RO_7, focusing on the frequency of Romanian spoken daily (range = 10–60/100; M = 33.33; SD = 16.00), did not reveal significant effects either. MEMLR_RO_8 investigated the impact of productive vocabulary knowledge in Romanian (or Romanian MiNT score: M = 59.39/67; SD = 3.69) on the bilinguals’ use of transitive strategies in Romanian, but likewise found no significant effects. Finally, MEMLR_RO_9 examined the role of exposure to Spanish through reading (range = 3–10/10; M = 8.47; SD = 2.21) and said strategies and also failed to find any effects.

Finally, a GLMM was performed to explore variability in the use of clitic doubling in combination with PE-marking in the bilinguals’ Romanian in comparison with the RMs’. This GLMM included response (PE-marking + CD or PE-marking) as dependent variable, and context (obligatory vs. optional) and group (bilingual vs. RM) as independent variables. For the purpose of consistency, this analysis focused only on non-human animate objects. Table 13 below shows the counts and proportions of the uses of PE-marking with and without CD across contexts and groups.

The raw data shows that the bilinguals use PE-marking without clitic doubling more frequently than the RMs (14.29% and 14.29% versus 1.35% and 2.70% for obligatory and optional contexts, respectively). Nevertheless, the GLMM did not find effects of context, group or an interaction between these predictors.

It is noteworthy to mention that not all the 32 instances of DOM that the bilinguals produced were grammatical: for example, divergences were found regarding gender agreement between the clitic doubling and the marked direct object (see 16), regarding the form of pe (see 17) or regarding the use of the definite article (see 18). Appendix C includes the 32 instances of PE-marking (with and without DOM) produced by the bilinguals along with the rest of the transitive strategies examined in this study.

4.4 Summary

Results show that the bilinguals’ transitive strategies in Spanish are consistent with those produced by the SMs; however, the picture is different in their native language: while RMs produce transitive strategies in a more distributed fashion, bilinguals produce fewer PE-marking and unmarked nouns and DPs and favor clitics. Additional observations indicated that the bilinguals use more clitics than the comparison groups when the clitics present a longer distance from the antecedent to which they refer. When comparing the transitive strategies produced by the bilinguals in their two languages, no significant differences were found. Additional analyses explored the role of extralinguistic factors as well as of productive vocabulary knowledge in both Spanish and Romanian; nevertheless, no effects were found. Finally, an analysis focusing on clitic doubling across contexts and groups was not able to pinpoint significant predictors. All R codes and outputs for all models are included in Appendices A and B.