2.1 The Cumulative-Enhancement Model

The Cumulative-Enhancement Model (CEM) (Berkes & Flynn, Reference Berkes, Flynn, Cabrelli Amaro, Flynn and Rothman2012; Flynn et al., Reference Flynn, Foley and Vinnitskaya2004) is based on the idea that multilingual acquisition is inherently non-redundant, and predicts that previously acquired linguistic knowledge will only transfer when it is facilitative to the L3 and subsequent language (Ln) acquisition process. Transfer may occur from any previously acquired linguistic system, but, crucially, only if it results in target-like production and comprehension in the L3/Ln. According to the CEM, overarching typological/structural similarities between the L3 and the L1 or L2 grammars do not play an inhibitory role in the transfer process.

Evidence for the CEM comes from Flynn et al.'s (Reference Flynn, Foley and Vinnitskaya2004) study of head directionality in relative clauses by L1 Kazakh/L2 Russian learners of L3 English. In the study, L3 learners transfer head-directionality from L2 Russian at the initial stages of L3 acquisition, thus converging on the proper L3 structure. Transfer from Kazakh would be non-facilitative, since Kazakh is left branching and English is right branching. These results, while consistent with the CEM's predictions, are also consistent with the predictions of the L2 Status Factor (discussed below). On the basis of these data alone, we cannot differentiate between the predictions of these two L3 transfer models.

Although the CEM makes testable predictions, it raises several questions about the nature of transfer as it is conceptualized here. First, to ensure that only facilitative transfer takes place, the CEM would have to predict that linguistic transfer happens on a structure-by-structure basis, perhaps uneconomically (in the sense of Rothman, Reference Rothman, Baauw, Dirjkoningen and Pinto2013). This conceptualization of (L3) transfer is fundamentally different from both competing L3 models and the predominant models of SLA (e.g., Schwartz & Sprouse, Reference Schwartz and Sprouse1996; Vainnika & Young-Scholten, Reference Vainnika and Young-Scholten1996), all of which predict transfer of a complete linguistic system. Furthermore, what appears to be “facilitative transfer” into the L3 may simply be acquisition: the work done by the parser in determining whether a given linguistic property is appropriate for transfer into the L3 seems to presuppose that a representation for that property already exists in the L3 grammar (to ensure compatibility), and transfer from the L1 or L2 would then be superfluous.

2.2 L2 Status Factor

The L2 Status Factor (L2SF) (Bardel & Falk, Reference Bardel and Falk2007; Falk & Bardel, Reference Falk and Bardel2011) predicts that the L2 serves as the source of transfer at the initial stages of multilingual acquisition. This transfer may be facilitative or non-facilitative, depending on whether the linguistic structures of interest are shared (i.e. representationally similar) between the L2 and the L3. The authors, citing recent neurolinguistic claims (Paradis, Reference Paradis2004; Ullman, Reference Ullman2001, Reference Ullman and Sanz2005), propose that the L2 has a special status as the source of default transfer due to differences in storage between the L1, which is stored in procedural memory, and the L2, which is stored in declarative memory. If this is true, it follows that the L2 would be more accessible than the L1 as a source of transfer during the initial stages of L3/Ln acquisition.

Evidence for the L2SF comes from Bardel and Falk (Reference Bardel and Falk2007) and Falk and Bardel (Reference Falk and Bardel2011). In the 2007 study, the authors tested verbal negation in two groups of bilingual learners learning L3 Dutch and L3 Swedish, both V2 languages (languages in which the finite verb obligatorily occupies the second position of the matrix clause). One group consisted of learners whose L2 was a V2 language, while the other group consisted of learners whose L1 was a V2 language. The group with a V2 language as an L2 performed significantly better with post-verbal negation, suggesting that the syntactic structure of negation was transferred from the L2 and not the L1.

In their 2011 study, the authors tested the acquisition of object pronouns in L3 German with mirror-image groups of adult French–English bilinguals. Results showed transfer primarily from the L2 in both bilingual groups, resulting in facilitative transfer for one group and non-facilitative transfer for the other. This supports the authors’ claim that the L2 is the primary source of transfer, regardless of whether such transfer is facilitative. While these results support the L2SF, it is unclear how this model accounts for L1 transfer in L3 acquisition, which has been observed (e.g., Hermas, Reference Hermas2010; Lozano, Reference Lozano, Mardsen, Pourcel and Whong-Barr2002).

2.3 Typological Primacy Model

The Typological Primacy Model (Rothman Reference Rothman2010, Reference Rothman2011, Reference Rothman, Baauw, Dirjkoningen and Pinto2013) predicts that transfer is determined by the structural similarity between the L3 and the L1 or L2. It is argued that transfer is not partial or selective (in the CEM sense), but rather, similar to L2 acquisition (Schwartz & Sprouse, Reference Schwartz and Sprouse1996),Footnote ¹ is holistic. Specifically, the TPM proposes that the linguistic parser, after sufficient input early in L3 acquisition, selects a linguistic system to transfer based on four dimensions of the input, in order of decreasing influence: (i) the lexicon, (ii) phonological/phonotactic cues, (iii) functional morphology (and its semantic import) and (iv) syntactic structure. In the case of Spanish–English bilinguals learning Brazilian Portuguise (BP) as an L3, the many similarities between the Spanish and BP lexicons trigger the transfer of Spanish, and the other cues do not need to be considered (Rothman, Reference Rothman, Baauw, Dirjkoningen and Pinto2013; published November 13, 2013 – this issue).

There are several studies showing evidence for the TPM, many of which address the acquisition of L3 Brazilian Portuguese (BP) (e.g., Montrul, Dias & Santos, Reference Montrul, Dias and Santos2011; Rothman Reference Rothman2010, Reference Rothman2011; Rothman & Cabrelli Amaro, Reference Rothman and Cabrelli Amaro2010). Rothman (Reference Rothman2010), for example, examined word order in L3 BP in mirror-image groups of Spanish–English bilinguals. The results show that both groups transfer word-order preferences from Spanish, despite the fact that Spanish results in non-facilitative transfer and English would have resulted in facilitative transfer. Rothman argues that Spanish transfer occurs in both groups due to its structural and typological similarity to L3 BP.

2.4 Summary of L3 transfer models

Each model that we have discussed here makes unique predictions for transfer at the initial stages of L3 acquisition. The CEM predicts that transfer comes from either the L1 or the L2, but will always be facilitative. The L2SF predicts that transfer comes exclusively from the L2, but, unlike the CEM, such transfer can be facilitative or non-facilitative. Finally, the TPM predicts that transfer is based on psychotypological/structural similarity, and thus comes from either the L1 or the L2 and can be facilitative or non-facilitative.

While each model makes explicit predictions for the initial stages of the L3, testing learners at the initial state has practical challenges. Completion of a test in the L3 requires, at the minimum, some knowledge of the lexical items used in the experimental task. However, if we take seriously the notion that cross-linguistic differences are specified in the lexicon, any lexical learning could result in linguistic development. These models do not preclude linguistic development, and make implicit predictions for what artifacts might be seen during the acquisition process. These predictions are similar to the explicit ones made for the initial stages: for mirror-image groups of the same language pairings learning an L3 property instantiated in either the L1 or L2, the CEM predicts that no development will be necessary (as transfer is only facilitative), the L2SF predicts developmental asymmetries between groups, and the TPM predicts no asymmetries, but leaves room for the possibility of development stemming from non-facilitative transfer. These differences in the developmental predictions of the three L3 models will prove critical in the interpretation of our results.

3.1 Montrul, Dias and Santos (Reference Montrul, Dias and Santos2011)

Montrul et al. (Reference Montrul, Dias and Santos2011) tested two groups of successive Spanish–English bilinguals at the initial stages of L3 acquisition of BP to examine the three transfer models above. Participants completed both an oral production task and an acceptability judgment task. We discuss only the first task here, as it is the results of this task that provide insight related to the current study.

In the oral production task, participants listened to the story of Little Red Riding Hood in BP and then retold the story using picture cues. The goal of the task was to compare their object expression to a BP control group. Object expression was selected as a point of comparison for the three transfer models because these languages differ in this domain, particularly with respect to Differential Object Marking (DOM), null object licensing, clitic doubling, and clitic placement. These differences make it possible for non-facilitative transfer from Spanish to BP to take place and contrastively test the predictions made by the TPM and the CEM. Participants in both bilingual groups produced DOM and clitic doubling, suggesting transfer from Spanish. Only the TPM, which predicts that the most structurally/typologically similar language will serve as the source of transfer into the L3, accounts for this non-facilitative transfer.

3.2 Iverson (Reference Iverson, Pires and Rothman2009)

Iverson (Reference Iverson, Pires and Rothman2009) aimed to use L3 acquisition as a testing ground for evaluating the predictions of Full Access (e.g., Schwartz & Sprouse, Reference Schwartz and Sprouse1996) and Representational Deficit (e.g., Hawkins & Hattori, Reference Hawkins and Hattori2006; Tsimpli & Dimitrakopoulou Reference Tsimpli and Dimitrakopoulou2007) approaches to adult SLA, claiming that the availability of a property/feature for transfer into an L3 is contingent upon its acquisition in the L1 or L2. Showing transfer of a property into an L3 which could only be transferred from an L2 learned in adulthood can be taken as evidence of the possibility of acquisition. This claim can be corroborated by comparable behavior in the L3 of simultaneous and successive bilinguals of the same language pairing.

Iverson tested two groups of Spanish–English bilinguals acquiring L3 BP: (i) heritage bilinguals, who began acquiring Spanish before puberty and (ii) successive bilinguals, who did not begin acquiring Spanish until after puberty. Both groups were studying abroad in Brazil at the time of the study. Each group completed two experimental tasks which gauged their knowledge of gender features and noun-drop. Spanish, like BP, has noun-drop – as a reflex of having grammatical gender – while English does not. If participants in the successive bilingual group were to show knowledge of noun-drop in L3 BP, that knowledge could only be transferred from their Spanish grammar.

On both tasks, the two experimental groups performed like the BP natives, providing evidence that the acquisition of formal features, in this case, gender features, is not subject to a maturational critical period.

4.1 What is Differential Object Marking?

DOM is the name given to the use of an overt morphological (case) marking which distinguishes certain objects from others. Importantly, DOM exists in Spanish, but not in English or BP. In Spanish, certain direct objects are preceded by the preposition a, whereas in English and BP there is no overt accusative case marker. The use of DOM, therefore, is illustrative when considering transfer at a Spanish/English/Portuguese interface.

4.2 Differential Object Marking in Spanish

The distribution of DOM in Spanish has been discussed frequently across linguistic paradigms, with many factors proposed as deterministic in its use.Footnote ² As Rodriguez-Mondoñedo (Reference Rodríguez-Mondoñedo2007), writes, however, “there are at least two dimensions that everybody has to contemplate in one way or another: specificity and animacy” (p. 89), and these two are the most commonly discussed (see e.g., Aissen, Reference Aissen2003; Leonetti, Reference Leonetti2004; Torrego, Reference Torrego1998; Zagona, Reference Zagona2002). Of these, animacy has been called the “dominant trigger for DOM” (Leonetti, Reference Leonetti2004, p. 75).

In the present study, the use of DOM is tested across four object types, based on the most commonly referenced paradigm for DOM in Spanish, in which [+animate, +specific] objects are obligatorily marked and all other object types ([+animate, –specific], [–animate, +specific], [–animate, –specific] are not obligatorily marked (see Table 1).

Table 1.

Differential Object Marking in Spanish (Rodríguez Mondoñedo, Reference Rodríguez-Mondoñedo2007).

Objects which are [+animate, +specific] are obligatorily marked with an a preceding the direct object, as in (1)–(2):

1. (1)

   

2. (2)

   

As demonstrated, specificity can be indicated in different ways. In (1), María, a proper name, is [+animate, +specific]. A specific referent is typically also implied when the object is used with a definite article, such as la niña “the girl”.

However, in (2), we see that in certain cases the object can be considered [+specific] even with an indefinite article (una secretaria “a secretary”, rather than la secretaria “the secretary”). In these cases, DOM can be used to differentiate between two distinct meanings, which, as Leonetti writes, “indicates that the presence of ‘a’ tends to be associated with specific readings” (p. 75). He also notes that this distinction is somewhat asymmetrical, in that the specific reading is only available in the presence of DOM, while the non-specific reading is available either in the presence or absence of DOM. In (2a) then, when DOM is used, una secretaria “a secretary” refers to a specific or non-specific secretary, while in (2b), lack of DOM indicates a non-specific secretary. This is not uncontroversial, as others (e.g., Guijarro-Fuentes & Marinis, Reference Guijarro-Fuentes and Marinis2007) claim that the use of DOM entails specificity.

Inanimate objects are unmarked, regardless of specificity, as in (3)–(4).Footnote ³

1. (3)

   

2. (4)

   

Like Rodriguez-Mondoñedo (Reference Rodríguez-Mondoñedo2007), we acknowledge certain limitations of the paradigm represented above (Table 1). This analysis is not uncommon in the literature (see e.g., Rodriguez-Mondoñedo, Reference Rodríguez-Mondoñedo2007; von Heusinger & Kaiser, Reference von Heusinger, Kaiser and von Heusinger2003; Zagona, Reference Zagona2002) and serves to clearly distinguish the [+animate, +specific] category from other object types, which is useful given the purpose of the present study. However, we would like to highlight the fact that DOM is possible, although not obligatory, in the case of [+animate, –specific] objects (see e.g., Aissen, Reference Aissen2003; Leonetti, Reference Leonetti2004; Torrego, Reference Torrego1998).

Leonetti (Reference Leonetti2004) suggests that there is no systematic correlation between the object marker and specificity of the object, despite the fact that the presence of DOM has a tendency to be associated with specific readings. Examples like (5) and (6) show DOM in cases of non-specific indefinites. In (5), DOM obligatorily precedes indefinite, non-specific pronouns alguien “someone” and nadie “anyone”. In (6), DOM is used with the non-specific bare plurals trabajadores “workers” and candidatos “candidates”. In the case of bare plurals, though, DOM is optional. As the examples show, the sentences are grammatical with or without DOM preceding the object.

1. (5)

   

2. (6)

   

Leonetti argues that the use of DOM in these cases is not due to any [+specific] feature, but rather is based on a [+human] feature which ‘overrides all other factors in the DOM system in Spanish’ (Leonetti, Reference Leonetti2004, p. 85). Among [+animate] objects, he suggests that optionality of DOM occurs only with non-pronominal DPs (as in (6)) and with a small group of verbs.

Following Torrego (Reference Torrego1998) and Zagona (Reference Zagona2002), Guijarro-Fuentes and Marinis (Reference Guijarro-Fuentes and Marinis2007) suggest that DOM is restricted to verbs with an agentive or causal subject. This is illustrated in (7), where in (a), the object is obligatorily marked because the subject el paciente “the patient” is agentive, whereas in (b) the object is obligatorily unmarked because the subject la situación “situation” is not agentive.

1. (7)

   

Guijarro-Fuentes and Marinis (Reference Guijarro-Fuentes and Marinis2007) also suggest that DOM is obligatory when the object is understood as affected (physically, psychologically, or in terms of location), as in (8). In (8a), the object un delincuente “a criminal” is considered unaffected and thus DOM is optional. In (8b), however, the object el delincuente “the criminal” is considered to be affected physically, and thus DOM is used obligatorily.

1. (8)

   

Finally, Guijarro-Fuentes and Marinis (Reference Guijarro-Fuentes and Marinis2007) claim that the use of DOM is restricted by the aspectual class of the predicate, as shown in (9) and (10). They claim that telic verbs require obligatory use of DOM, as in (9), whereas non-telic verbs require obligatory use of DOM only in the case of a [+animate] subject, as in (10).

1. (9)

   

2. (10)

   

As detailed above, DOM in Spanish involves many factors. Nonetheless, there is consistency in certain sub-contexts: DOM is obligatorily used with [+animate, +specific] objects, and is overwhelmingly prohibited with [–animate] objects. The methodology for the current study is based on the distributional paradigm presented in Table 1, in which objects are divided into four types based on animacy and specificity only. While these are the main variables examined in this study, we do take into consideration the other factors presented here. We elaborate on this in Section 5.3 below.

4.3 Prior studies of acquisition of DOM in Spanish

The acquisition of DOM has been shown to be problematic for both L2 learners (Bowles & Montrul, Reference Bowles, Montrul, Bruhn de Garavito and Valenzuela2008; Guijarro-Fuentes, Reference Guijarro-Fuentes2012; Guijarro-Fuentes & Marinis, Reference Guijarro-Fuentes and Marinis2007, Reference Guijarro-Fuentes, Marinis and Collentine2009) and heritage bilinguals (Montrul, Reference Montrul2004; Montrul & Bowles, Reference Montrul and Bowles2009).Footnote ⁴ Guijarro-Fuentes and Marinis (Reference Guijarro-Fuentes and Marinis2007) tested L2 learners’ ability to distinguish between grammatical and ungrammatical usage of DOM across six conditions manipulated for both the structure and the semantics of the predicate. Relevant for the present study are the first three conditions, testing the following object types: [+animate, +specific], [–animate, +specific], [–animate, –specific] and [+animate, –specific].

The L2 learners performed at chance across all conditions in a contextualized acceptability judgment task. While advanced L2 learners performed at chance with [+animate, +specific] objects and [+animate, –specific] objects, they did show sensitivity to animacy, performing well above chance on sentences with [–animate] objects. The authors attribute this result to the relative simplicity of this condition or perhaps explicit instruction (see Bowles & Montrul, Reference Bowles, Montrul, Bruhn de Garavito and Valenzuela2008), as the second language learners only need to learn the animacy restriction on DOM to perform accurately. In general, the results of the study show that DOM proves difficult for native English speakers to acquire in L2 Spanish.

Like L2 learners of Spanish, heritage bilinguals also display divergence from native Spanish speakers with respect to DOM. Rodríguez-Mondoñedo (Reference Rodríguez-Mondoñedo2008) has shown that monolingual Spanish-speaking children acquire DOM with 98% accuracy as early as 2;9 (i.e. at two years and nine months of age). It is possible that heritage bilinguals acquire DOM in early childhood and later suffer attrition which results in variable usage of this property. Alternatively, heritage bilinguals may not converge on monolingual norms of DOM because they are exposed to qualitatively different input.Footnote ⁵

Montrul and Bowles (Reference Montrul and Bowles2009) show that even advanced heritage bilinguals diverge from monolingual native speakers with respect to their usage and acceptance of DOM in Spanish. In an acceptability judgment task, 16 of 30 advanced proficiency heritage bilinguals and 20 of 23 intermediate proficiency heritage bilinguals accepted sentences with a [+animate, +specific] direct object without DOM, diverging significantly from the monolingual controls. In a second acceptability judgment task given to different participants, the authors found that 24 of 30 advanced heritage bilinguals and 27 of 28 intermediate heritage bilinguals accepted ungrammatical sentences with DOM. Interestingly for the present study, all groups of heritage bilinguals in the second experiment accepted DOM with [–animate] objects, which the authors attribute to differences in input, language change (Company, Reference Company, Bernabé, Berenguer, Cantarero and de Torres2002) or perhaps the complexity of DOM itself.

5.1 Research questions and hypotheses

The present study aims to examine what morphosyntactic transfer effects obtain in the L3 acquisition of BP by successive and heritage bilinguals of Spanish and English. We adopt the three research questions (RQs) and hypotheses as set out below.

• RQ1. Which of the three proposed models best predicts and explains the results of all three groups tested?

We predict evidence of transfer from Spanish in all three bilingual groups, based on the prior work of multiple authors, particularly those working with BP (Montrul et al., Reference Montrul, Dias and Santos2011; Rothman Reference Rothman2010, Reference Rothman2011), which would be consistent with the predictions of the TPM and inconsistent with predictions of both the CEM and the L2SF.

• RQ2. Following from Iverson (Reference Iverson, Pires and Rothman2009), is there evidence based on these data to support complete acquisition of a new linguistic property (DOM) in the L2 in adulthood?

Our prediction of transfer from Spanish (see RQ1), irrespective of its L1/L2 status, has implications not only for L3 acquisition but also, indirectly, for L2 acquisition. As Iverson (Reference Iverson, Pires and Rothman2009) explains, in order to transfer any linguistic property (such as DOM), the learner must have a mental representation of this property, meaning it has been acquired. We predict evidence of transfer from Spanish in all groups, including the non-native Spanish bilinguals. Such transfer implies that the L2 Spanish group has acquired DOM in their L2 and supports the Full Transfer/Full Access Model for SLA (Schwartz & Sprouse, Reference Schwartz and Sprouse1996).

• RQ3. What do the results of the heritage speakers, specifically in terms of Spanish transfer, tell us about the syntactic representation of Differential Object Marking in their Spanish grammar?

As we outline in our predictions for RQ2, we believe evidence of transfer from Spanish in this experiment is an indication of an underlying mental representation of the Spanish syntax in the speakers tested. We predict that the transfer observed in the heritage bilinguals may look different than the transfer of the other two groups due to evidence that DOM is represented differently in the Spanish of heritage speakers (Montrul & Bowles, Reference Montrul and Bowles2009). Divergent results from the heritage bilinguals, while indicating transfer from Spanish, may indicate transfer of a system in which DOM is applied in a different distributional pattern.Footnote ⁶

5.2 Participants

There were 43 participants in the study, all Spanish–English bilinguals in their first semester of studying BP as an L3.Footnote ⁷ Participants were recruited from Accelerated BP courses at a number of large public and private universities across the United States.

The first two groups of L3 participants tested were successive bilinguals of the same language pairing.Footnote ⁸ The L1 English group (n = 14) consisted of native English speakers (mean age 21.5 years, range 19–24) who began acquiring Spanish as an L2 at age 12 or later. The L1 Spanish group (n = 9) consisted of native Spanish speakers (mean age 24.33 years, range 19–34) who began acquiring English as an L2 at age 9 or later.Footnote ⁹ In addition to the “mirror-image” groups, a third bilingual group was also tested. The heritage Spanish group (n = 20) consisted of US residents (mean age 20.9 years, range 19–24) exposed to Spanish from birth and English before the age of 7 years. There were also control groups for each of the three languages: Spanish (n = 26), English (n = 24) and Brazilian Portuguese (n = 15).

5.3 Grammaticality judgment task

Participants completed a grammaticality judgment task (GJT) via surveygizmo.com, rating 120 BP sentences on a scale of 1 (ungrammatical) to 4 (grammatical). Sentences were presented in audio and written formFootnote ¹⁰ simultaneously, one sentence at a time. Participants were unable to revise previously rated sentences.

There were 64 target sentences, 16 DOM-related fillers and 40 unrelated fillers. Each of the 64 target sentences consisted of a DP subject followed by a transitive verbFootnote ¹¹ and a direct object of one of four types: (i) [+animate, +specific] (ii) [–animate, +specific] (iii) [–animate, –specific] and (iv) [+animate, –specific]. For each object type, half of the sentences did not include the differential object marker a – which is ungrammatical in Spanish for [+animate, +specific] objects, but grammatical for other object types in Spanish and across the board in BP – and half included a – which is ungrammatical in BP and only obligatory for [+animate, +specific] objects in Spanish. Since BP allows for the optional use of the definite article before proper nouns, which could possibly interpreted as DOM by L3 learners, it was important to include related filler sentences (n = 16) that contained a definite article used with [+animate, + specific].Footnote ¹² Half of these exhibited DOM, while half did not. Table 2 contains examples of sentences in GJT in all three languages.

Table 2.

Sample sentences from grammaticality judgment task by object type.

The goal of this study was to investigate two major variables – animacy and specificity. Additional factors that play a role in the use of DOM in Spanish were controlled for in the task. In the [+animate, +specific] condition, all objects were singular, male, proper nouns (i.e. personal names). In the [+animate, –specific] condition, all of the objects were singular (eliminating the possibility of bare plurals), masculine, human nouns, preceded by the indefinite article.

Immediately following the Portuguese task, participants completed Spanish and English versions of the task. Participants who scored outside the native control range for either the Spanish or the English task were excluded from the study. Participants were also excluded from the study if they had knowledge of another language other than English or Spanish, or more than one semester of BP exposure (recall footnote 9)