9.3.1 DOM omission in other Latin American heritage speakers

In order to find out whether the patterns of erosion in the first-generation immigrants were related to the specific Spanish variety tested (heritage Mexican Spanish in the Chicago area) is evidence of more widespread ongoing language change in the Spanish spoken in the United States, we collected data from a small sample of Spanish heritage speakers (n = 15) and first-generation immigrants (n = 13) from other Latin American countries (Colombia, Ecuador, El Salvador, Guatemala, Nicaragua, Peru, Puerto Rico, and Venezuela) tested in Chicago and in New York. These data have not been published elsewhere and were collected to follow-up on the findings on Mexican Spanish presented in Chapter 6. The participants completed the same tests and questionnaires as the Mexican groups discussed in Chapter 6. Three of the heritage speakers came to the United States at ages 6, 7, and 9; the rest were born and raised in the United States. Because the group was small, it was not divided into simultaneous and sequential bilinguals. Table 9.4 presents basic information about these participants: mean age at testing, mean age of acquisition (AoA) of Spanish and English, length of residence in the United States (LOR US), mean level of education completed, and mean SES, which combined level of education and occupation of parents. We compared these two groups with the three groups reported on in Chapter 6 (Tables 6.2–6.9). The groups had similar level of education completed (F(4,100) = 0.777, p = 0.543) and SES (F(4,100) = 1.224, p = 0.305). We call these two groups the Latin American participants for short, which is to be understood as participants from Latin America except for Mexico.

All participants completed the same language background questionnaire discussed in Chapters 6, 7, and 8. Table 9.5 presents the participants’ mean self-ratings in Spanish and English, globally and by skill and their scores on the standardized written proficiency test.

The three heritage speakers groups (simultaneous bilingual Mexican heritage speakers, sequential bilingual Mexican heritage speakers, and Latin American heritage speakers) were significantly more fluent in English than the two first-generation immigrant groups (Mexican and Latin American immigrants) (F(4,100 = 17.99, p < 0.0001). The three heritage speaker groups did not differ from each other (each p > 0.5) and neither did the two first-generation immigrant groups (p = 0.112). The opposite pattern obtained for Spanish self-ratings, which also differed by group (F(4,100) = 4.737, p < 0.0001). The first-generation immigrants were not different from each other (p = 0.584) and neither were the heritage speaker groups (all ps > 0.05). The five groups did not differ in their written proficiency scores (F(4,100) = 1.97, p = 0.317).

As in the Mexican groups, more first-generation immigrants (58.85%) than heritage speakers (13.33%) from Latin America indicated that Spanish was their preferred language. Except for three participants, all other heritage speakers from Latin American Spanish varieties were simultaneous bilinguals. Therefore, like the Mexican-origin simultaneous bilinguals, the Latin American heritage speakers felt that Spanish was more like a second than a native language. That is, 92.31% of the Latin American first-generation immigrants considered Spanish as their native language compared to 46.67% of the Latin American heritage speakers.

As with the Mexican heritage speakers, the main source of Spanish input for the Latin American heritage speakers were the parents (caregivers): 73.33% were spoken to in Spanish, 26.27% in both languages. Confirming typical patterns, the percentage of heritage speakers that used the heritage language with parents (60%) and grandparents (93.3%) was higher than with siblings and friends. With siblings and friends, the heritage speakers had used both English and Spanish, but they had used more Spanish than English with siblings than with friends. Most of the Latin American heritage speakers (93.33%) had watched TV in Spanish, 86.67% had been read to in Spanish, 80% had been encouraged to speak Spanish, and 80% of them had been corrected by their caregivers when they did so.

As with the Mexican heritage speakers, the percentage of heritage speakers from other Latin America countries using Spanish declined over time during the school-age period, as they began to use more English and use of Spanish became less frequent. Most participants reported using Spanish often during the school age period, very few used Spanish seldom or stopped using it altogether. All the Latin American heritage speakers reported using Spanish with their parents throughout their lifetime, although the percentage of speakers who reported communicating with siblings and others in Spanish decreased over time, as with the Mexican heritage speaker groups. More Latin American heritage speakers (between 70% and 80%) than Mexican heritage speakers (between 29% and 30%) reported using English only with their Spanish-speaking friends during elementary school, middle school, and high school.

Figure 9.6 Mexico and Latin America groups: Mean percentage accuracy on animate specific direct objects in the oral narrative task

I now turn to the results of the linguistic tasks, which compare the Mexican and Latin American groups.

9.3.2 Results of the linguistic tasks

In the oral narrative task, there were significant differences between the five groups (F(4,100) = 8.713, p < 0.001) in their accuracy producing animate specific objects with DOM: the Latin American first-generation immigrants were more accurate than all the heritage speaker groups (each p > 0.05) but were no different from the Mexican first-generation immigrants (p = 0.815). Although they scored lower and omitted required DOM, the Latin American heritage speakers did not differ from the two groups of Mexican heritage speakers.

Figure 9.7 shows the individual scores by group. Only three first-generation Latin American immigrants achieved 100% accuracy with DOM on this task, whereas the Latin American heritage speakers, like the Mexican heritage speakers, ranged in accuracy between 0% and 100%, omitting DOM in required contexts similar to the other heritage speaker groups.

Figure 9.7 Mexican and Latin America groups: Individual results on DOM marking with animate specific direct objects in the oral narrative task

Figure 9.8 displays accuracy percentages on animate, specific direct objects in the elicited oral production task. The first-generation Latin American immigrants did not differ from the other groups (F(4,100) = 1.354, p = 0.255). Figure 9.9 displays the individual results, which show that the Latin American immigrants scored in the range of 53–100%, just like the first-generation Mexican immigrants.

Figure 9.8 Mexico and Latin America groups: Mean accuracy percentages on animate specific direct objects in the Spanish elicited oral production task

Figure 9.9 Mexico and Latin America groups: Individual results on DOM marking with animate specific direct objects in Spanish elicited production task

Lastly, Figure 9.10 illustrates mean acceptability ratings for grammatical and ungrammatical sentences with animate specific direct objects. All participants gave very high scores to grammatical sentences with DOM-marked animate objects and unmarked inanimate objects (close to 4 = perfectly acceptable). However, acceptability ratings for ungrammatical sentences with unmarked animate objects did not differ by groups (F(4,100) = 0.997, p = 0.413).

Figure 9.10 Mexico and Latin America groups: Mean acceptability ratings of animate and inanimate objects with and without DOM in the Spanish bimodal acceptability judgment task

On the comprehension task, the five groups performed alike on the auditory (F(4,99) = 1.854, p = 0.125) and written versions (F(4,99) = 1.854, p = 0.125). The five groups also performed alike on DOM marking on the written production task (F(4,100) = 0.689, p = 0.601).

The collective data from all tasks show that DOM exhibits variability in usage and is vulnerable to omission in Spanish heritage speakers in general, confirming results of previous studies of heritage speakers in the United States (Montrul, Reference Montrul2004; Montrul and Bowles, Reference Montrul and Bowles2009). Therefore, the patterns of language change with Spanish DOM attested in the Mexican heritage speakers reported in Chapter 6 and in previous work appear to generalize to other Spanish heritage speakers in the United States. Moreover, DOM is omitted by adult first-generation immigrant as well, suggesting generalized L1 attrition.

Because DOM omission with animate specific direct objects co-exists in the grammars of Spanish-speakers in the United States with the grammars of correctly marked objects, there is variation and fluctuation (optionality) with the use of the a marker. Although this optionality might have been due to incomplete acquisition or L1 attrition of DOM in childhood, unmarked animate specific direct objects may be, in fact, a stabilized feature in a new variety of Spanish, similar perhaps to what Sharma (Reference Sharma2005a) showed for the distribution of definite and indefinite articles in Indian English and Otheguy and Zentella (Reference Otheguy and Zentella2011) for the rate of overt subject pronouns in the Spanish of New York. This is because, at least in Spanish (but not in Hindi and Romanian), there is no relationship between DOM omission and general proficiency in Spanish, or age of acquisition of Spanish, or Spanish language variety. Therefore, unmarked animate direct objects appear to be on their way to become a stable feature of US Spanish.

To summarize, DOM is vulnerable to change in situations of language contact, as we reviewed in Chapter 4 and confirmed in Chapters 6–9. We have discussed how it evolved diachronically along the animacy and prominence scales in Spanish and in Romanian, and we confirmed that the expansion of DOM to inanimate objects in some varieties of Spanish is taking place. Table 9.6 summarizes the main findings.

We next explore the linguistic and situational factors that contribute to the differences and similarities in DOM variability in the three languages.

9.4.1 Grammatical complexity

Grammatical complexity is a very controversial and elusive construct in linguistics but has recently been invoked to describe structural changes in heritage grammars (Polinsky, Reference Polinsky2018). To give an example of how complexity has been operationalized and measured, McWhorter (Reference McWhorter2007) considers three dimensions: overspecification (degree of grammaticalized semantic distinctions), structural elaboration (number of rules or foundational elements required to generate surface forms), and irregularity (degree of morphological regularity and suppletion). Extending this measurement to animate specific direct objects, languages that mark DOM are more overspecified in their expression of direct objects and present more grammatical elaboration than grammars that do not mark DOM.

As we have seen in Chapter 3, DOM encodes semantic notions such as animacy, definiteness, and specificity, among others, and is checked in an additional functional projection above the VP, higher than the projection for unmarked objects (López, Reference López2012). Compared to a non-marked object, the DOM object must move outside of the VP to be overtly case marked by the DOM marker in the specifier of the vP. Instead of staying in situ close to the vP, it moves out due to its referential properties which likens it to a nominative and a dative subject (see also López, Reference López2012). Unmarked direct objects and dative indirect objects, by contrast, receive only structural case in the VP. Alternatively, under more current theories, the object may move because it is affected and marks specificity – as in languages that have object shift – and not necessarily for case reasons. (Argumental accusative and dative clitics in Spanish and in Romanian are morphological manifestations of accusative and dative structural case). Carrying the semantic features of animacy and specificity, DOM is both semantically and syntactically more complex than unmarked objects and indirect objects because it involves more structure in addition to movement. Unmarked inanimate objects and non-specific objects, receive case structurally lower in the VP, like all objects in English. Therefore, syntactically, DOM objects are structurally more complex than unmarked objects.

O’Grady et al. (Reference O’Grady, Kwak, Lee and Lee2011) provide a cognitive explanation for the complexity of DOM, capitalizing on the nature of form-meaning mappings. The DOM markers in the three languages studied have other functions in addition to being DOM markers. So, we have one form (a, ko, pe) with multiple meanings and functions (direct object, indirect object, dative subject, locative), or what O’Grady et al. call an opaque form-meaning mapping. At the same time, while all indirect objects and dative experiencers are obligatory marked in these languages, only a subset of direct objects (animate, specific) is obligatorily marked. Such opaque mappings require very high frequency of occurrence in the input to be reliably acquired and mastered, and it is precisely an adequate amount of input during the school-age period, if not earlier in life, that is missing in heritage speakers. Limited input and use of the language would explain why DOM, which lies at the morphology–syntax–semantics interface, is so vulnerable in heritage languages under the pressure from English (a language with no DOM). Thus, both syntactic and cognitive approaches to DOM can capture the complexity of this construction.

9.4.2 Majority language transfer

Transfer from the dominant contact language cannot be denied, and a move away from linguistic complexity for animate, specific objects is further reinforced by English. Although English does not have DOM, it has both prepositional datives (Juan gave a book to Maria), analogous to the Spanish dative construction Juan (le) dio un libro a María (except for the lack of a dative clitic), and double object datives (Juan gave Maria a book). According to (Torrego, Reference Torrego1988), the English double object construction and Spanish DOM are roughly equivalent in their semantic and syntactic characteristics, except that Spanish marks the affected object overtly while English does not (see Torrego for specific arguments). Furthermore, although English has topicalizations, it does not have clitic left dislocations, like Spanish and Romanian, or scrambling like Hindi, which are structural options that signal specificity and reinforce DOM in these languages. Finally, English has nominative-marked experiencers with stative psych verbs (Mary likes John), while Spanish, Hindi, and Romanian have dative experiencers with these verbs. These differences between English and Spanish, Hindi, and Romanian may also contribute to the weakening of DOM-marked objects in these languages when they are the weaker language.

At the individual, psycholinguistic level, under communicative pressure, the cognitive structure of English appears to impose itself on the structure of Spanish, Hindi, and Romanian in heritage speakers in the United States. Significantly more input and use of English than of Spanish, Hindi, and Romanian throughout the lifespan of the heritage speakers has reinforced both the structures that English and the other three languages share and that are linguistically less complex (nominative experiencers, unmarked objects). Thus, the structural changes going on in the individual grammars of heritage speakers are also affected by English, since the change happens to be in the retreat rather than in the advancement of overtly marked case (i.e., DOM). In Chapter 4, we discussed more evidence that DOM is omitted when the target or heritage language is in contact with languages other than English that also do not exhibit DOM: for example, Spanish in contact with French (Grosjean and Py, Reference Grosjean and Py1991) and with Dutch (Irizarri van Suchtelen, Reference Irizarri van Suchtelen2016), and Romanian in contact with varieties of Italian that do not have DOM (Cohal, Reference Cohal2014). Similar loss of case resulting from language contact and incomplete acquisition in bilinguals is reported by Lightfoot (Reference Lightfoot1991, Reference Lightfoot1999), who discusses the loss of the inherent dative case in Middle English, and by Polinsky (Reference Polinsky2006), with the loss of the genitive of negation and other semantically based cases, such as the instrumental case in Russian heritage speakers.

Admittedly, more direct support for the dominant language transfer hypothesis would have to be sought by examining heritage speakers of Spanish in contact with a language that exhibits DOM: if in this case heritage speakers do not move in the direction of losing DOM, then the likelihood of dominant language influence is greater. There are several studies of L2 and L3 acquisition with learners of different L1s, mentioned in Chapter 4, that support the transfer hypothesis for L2 acquisition. For example, Turkish-speaking and Romanian-speaking learners of Spanish as L2 attain mastery of DOM earlier and more reliably in L2/L3 Spanish than English-speaking L2 learners of Spanish (Montrul, Reference Montrul2019; Montrul and Gürel, Reference Montrul, Gürel, Perpiñán and Judy2015). D’Alessandro (Reference D’Alessandro2021) and Frasson et al. (Reference Frasson, D’Alessandro and van Osch2021) found that heritage varieties of Italian that have DOM preserve it in situations of contact with Spanish in Latin America.

9.4.3 Language processing and lexical access

Another similarity between the three studies is that the heritage speakers performed better on comprehension than on production and grammatical judgments, which points to the difference in cognitive resources required for production and comprehension. There are two possible reasons for the production–comprehension discrepancy found in our study: the tasks and the grammar. In the production tasks, participants were given the words and they had to verbally construct sentences adding the necessary morphology. In the comprehension task, they were given three pictures and they had to match a very short simple sentence with one of the pictures. It takes more processing resources to access the lexicon, activate the relevant features, bundle them, and map them to morphophonological forms to produce sentences with the required morphology under time pressure in an experimental task (especially if it is oral) than to comprehend sentences when all the words are given. Language production involves bottom–up processing, while language comprehension is top–down, relies on heuristics, and requires fewer cognitive resources (Paradis, Reference Paradis2004). The fact that heritage speakers in the three languages displayed more variability in production than in the comprehension/written tasks supports the observation that heritage speakers have the relevant linguistic competence but exhibit online processing limitations when using their weaker language (Pérez-Cortés et al., Reference Pérez-Cortés, Putnam and Sánchez2019; Polinsky, Reference Polinsky2018; Putnam and Sánchez, Reference Putnam and Sánchez2013), even when they may have relatively advanced proficiency in the heritage language, like the Romanian speakers in our study.

Neurocognitive and psycholinguistic research on bilingualism has shown that elements of both grammars can be simultaneously active to a greater or lesser degree in the minds of individual speakers (Grosjean, Reference Grosjean2008). There is connection and interaction between the lexicons of the two languages, and the strength of lexical connections between the two languages can change over time and in response to input (Linck et al., Reference Linck, Kroll and Sunderman2009). Less frequent use of one of the grammars and continued competition from the more active dominant grammar over processing resources may lead to the restructuring of the weaker language. With these assumptions, Putnam and Sánchez (Reference Putnam and Sánchez2013) proposed that functional features must be activated during language use for comprehension and production, and as in second language acquisition, there is transfer and/or reassembly of features from the dominant language to the heritage language. Pérez-Cortés et al. (Reference Pérez-Cortés, Putnam and Sánchez2019) maintain that variability and structural differences in heritage language grammars arise from asymmetries in lexical access, which include the functional features of syntactic projections, and the building of syntactic representations formed in early childhood. Infrequent use of the heritage language during childhood leads to lower proficiency in adulthood, which in turn affects the fast and efficient activation and inhibition of the formal features, functional projections, and morphosyntactic representations in the heritage language. Variable outcomes in heritage language acquisition at high and intermediate levels of proficiency stem from reduced or inhibited access to linguistic representations during language production. These representations are not always easily accessible in language comprehension, which requires less activation and fewer impulses to the neural substrate than production (Paradis, Reference Paradis2004).

Building on Pérez-Cortés et al. (Reference Pérez-Cortés, Putnam and Sánchez2019), Montrul (Reference Montrul2021) advanced the Differential Access Model as another viable theoretical explanation of how differences from baseline grammars could arise. This model may account for the categorical and variable behavior of heritage speakers and some adult immigrants in the realization of DOM in Spanish, Hindi, and Romanian, in the following way: in the diachronic syntax and in the language acquisition literature, optionality of rules or co-existence of two forms have been captured under the notion of multiple grammars at the individual and societal levels (Amaral and Roeper, Reference Amaral and Roeper2014; Kroch and Taylor, Reference Kroch, Taylor, van Kemenade and Vincent1997; Roeper, Reference Roeper1999; Yang, Reference Yang2000) (see Chapter 4), and this idea can be extended to explain the syntactic representations of heritage speakers with respect to DOM. As in situations of diachronic change in general, multiple grammars co-exist for an extended period until one of them is not generated anymore (Lightfoot, Reference Lightfoot2013). This process is gradual and takes many years, such that multiple I-grammars for the same linguistic phenomenon (the old ones and the new ones) co-exist for a long period of time until the demise of the old grammar is unsupported by diminishing frequency of samples in the E-language (Amaral and Roeper, Reference Amaral and Roeper2014; Kroch, Reference Kroch1989). In individual speakers, co-existing grammars of DOM occur at the representational level (syntax) and at the processing level, i.e., parsing for comprehension and production (Sharwood Smith et al., Reference Sharwood Smith, Truscott, Hawkins, Herschensohn and Young-Scholten2013) (I-language).

Competing grammars are also at play in language acquisition (Yang, Reference Yang2000). Universal Grammar provides a population of grammars that are accessed by the language learner. Acquisition can be viewed as a variational process in which the distribution of grammars changes as an adaptive response to the linguistic evidence in the environment and, depending on each grammar’s weight in the input, one grammar will eventually rise to dominance. The variation in this case is between categorical rules. The same process applies in bilingual grammars, where the grammars for each language compete in the input.

Competing structural representation of objects in Spanish/Hindi/Romanian, and in English would work as follows. Heritage speakers have two active grammars that compete for selection during language use. In English, there is no object split: all direct objects are generated in the VP and stay there, as in Figure 9.11, regardless of whether the object is human and specific (Mary) or inanimate and specific (the movie). (EA stands for external argument).

Figure 9.11 Syntactic representation for animate and inanimate direct objects in English

Spanish/Hindi/Romanian, by contrast, split objects into two types. Prominent objects, which are animate and specific as in the Spanish example in Figure 9.12a, are, following (López, Reference López2012), base-generated in the VP but move to a higher position for DOM marked objects in ⍺P, above the VP, where the [+animate] feature is checked. Assuming the basic tenets of Distributed Morphology (Halle and Marantz, Reference Halle, Marantz, Hale and Keyser1993), there is a functional category ⍺P with an array of features. This functional category receives a phonetic matrix (zero or a in Spanish) depending on the animacy of the object. Inanimate objects stay in the object position in the VP and do not move to ⍺P, as for example in Figure 9.12b illustrates. (Other accounts of Romanian propose that the features of DOM are in the DP or ApplP, as discussed in Chapter 3.)

Figures 9.12 Syntactic representation for animate and inanimate direct objects in Spanish

Heritage speakers have knowledge of two separate but interacting syntactic systems (in our case, one for Spanish/Hindi/Romanian, one for English). The two grammars for objects in Figure 9.11 and Figures 9.12a and b are generated by Universal Grammar and compete in the input of heritage speakers of Spanish/Hindi/Romanian in the United States.

The results of the oral tasks in the three languages show that there are three types of speakers: those who produced DOM with 100% accuracy, those who did not produce animate objects with DOM at all (0% accuracy), and those who sometimes produced DOM and sometimes omitted it with animate specific objects (20–80% accuracy, i.e., variability). We will assume that those who produced DOM with 100% accuracy have the mental representation in Figure 9.11 for English objects and in Figures 9.12a and b for Spanish/Hindi/Romanian animate and inanimate objects. These heritage speakers, who are consistent in their production of DOM, have activated grammar in Figure 9.11 for English and grammar in Figures 9.12a and b for Spanish/Hindi/Romanian because they were exposed to sufficient input to select the syntactic representations in Figures 9.12a and b for Spanish/Hindi/Romanian.

Those who produced 0% DOM could be assumed to have had very little exposure to, and less use of, Spanish/Hindi/Romanian growing up and, as a result, they select the grammar of English to use for the heritage language as well. Perhaps they have eradicated DOM from their grammar of Spanish/Hindi/Romanian and they represent animate and inanimate objects in Spanish/Hindi/Romanian in the VP, as they do in English, using the structure in Figure 9.11 for English and for Spanish/Hindi/Romanian, as in Figure 9.13. This would be a representational difference at the level of features and functional projections.

Figure 9.13 Syntactic representation of animate objects in Spanish heritage speakers

But those heritage speakers and immigrants who show inconsistent behavior, ranging from 20% to 80% accuracy with obligatory DOM, might still be accessing and activating both grammars, and the grammar for English in Figure 9.11 and the grammar for the heritage language in Figures 9.12a and b are in constant competition. Another possibility, yet to be explored, is that they have developed their own DOM system, assigning DOM on different semantic or pragmatic criteria than baseline speakers (see Sharma, Reference Sharma2005b, on the article system in non-native varieties of English). Therefore, while the syntax of objects in English does not instantiate an intermediate category ⍺P between vp and VP, for these speakers, the syntax of Spanish/Hindi/Romanian objects does. However, due to weaker lexical links related to reduced language use, the formal features of ⍺P, in the case of Spanish [+ animate] may be weakly activated, deactivated, or dormant, so heritage speakers do not assign these features any morphophonological content (zero), thus omitting a, as in Figure 9.14b.

Figures 9.14 Syntactic representations of animate objects with activated and dormant animate features

Recall that in Romanian, DOM co-occurs with CD. For the cases of CD omission in CD+DOM contexts, one possibility is that heritage speakers who made these errors access the ApplP analysis (Hill and Mardale’s (Reference Hill and Mardale2019, Reference Hill and Mardale2021) proposal described in Chapter 3) efficiently or consistently, but due to cognitive pressure and slow lexical retrieval, they fail to spell out [F_mark] overtly through the clitic in ApplP, a surface morphological problem (Perez-Cortes et al., Reference Hill and Mardale2019). Pe is retained in these cases because it has semantic features, whereas the clitic has discourse features. The other possibility is that because both the ApplP and the DP analyses are available for DOM, some heritage speakers, under communicative and cognitive pressure, activate the DP instead of the ApplP to check DOM for names (and modified DPs), and bundle both [particularized] and [F_mark] under pe (as was done in Old Romanian). The pressure to establish minimal domains (Scontras et al., Reference Scontras, Polinsky and Fuchs2018) explains the bundling of the features [personalized] and [F_mark] under one head (D) by the Romanian heritage speakers with lower proficiency in the present study. When both pe and CD are omitted, the heritage speakers may still access the ApplP or the DP, but fail to realize the marking overtly, following Putnam et al. (Reference Pérez-Cortés, Putnam and Sánchez2019). The possibility that the little variability attested is more likely to arise from lexical and representational access, rather than by a simpler structural or featural representation, is supported by the fact that the comprehension of CD and DOM-marked names and DPs was almost at ceiling for most speakers, with a very few exceptions.

Exposure to the heritage language and the amount and frequency of heritage language use will determine whether the heritage language-specific grammar is eventually eliminated and replaced by the other grammar, or whether both grammars continue to co-exist in the mental representations of these heritage speakers, as long as they use both languages. The amount of input will guide which grammar of direct objects is activated and selected for comprehension and production. Frequency of language use, as in the variational account of L1 acquisition (Yang, Reference Yang2000), will determine which linguistic representation wins and is adopted. With heritage speakers, fluctuations in input could lead to permanent optionality rather than complete eradication or consistent use of the Spanish grammar.

Dealing with multiple grammars at the level of lexical representation and access already implies significant linguistic complexity in handling two systems for bilinguals whose native language has weakened. The apparent simplification of forms at the level of morphological expression, does not necessarily mean that the underlying linguistic representation has restructured and simplified as well through loss of features or elimination or neutralization of functional projections. At best, this explanation cannot be applied to all heritage speakers. In sum, there is individual variability in speed of lexical access and in the variability of morphological expression in oral production. There is also indeterminacy demonstrated in acceptability judgments. Whether differences in DOM expression from baseline grammars are due to structural simplification or to lexical access and competition between grammars may depend on where on the proficiency continuum heritage speakers may fall. The Romanian heritage speakers were statistically more proficient than the Spanish heritage speakers and the Hindi heritage speakers in this study.

The first-generation Mexican and Latin American immigrants who also omitted DOM with animate, specific direct objects, exhibited psycholinguistic behavior in line with Hicks and Domínguez’s (2020) Attrition via Acquisition model, according to which the change undergone by the L1 grammar in late sequential bilinguals does not necessarily involve the “loss” of existing options, since the L1 grammar itself remains active in processing for both production and comprehension. The changes observed, in our case production and acceptance of unmarked animate, specific objects, are likely to supplement – rather than replace – the existing grammar. None of the attrited first generation immigrants had 0% accuracy in DOM production (unlike some of the heritage speakers), suggesting that both their grammars were still active.

9.4.4 Acoustic salience of the DOM markers

One of the driving hypotheses of this study was that omission of Spanish DOM could be related to the low acoustic salience of the marker a, and this hypothesis motivated the choice of Romanian and Hindi as comparison groups, given that the DOM markers in these languages, the preposition pe and the post-position -ko, are more acoustically salient than the Spanish DOM-marker a. The issue of perceptual salience in the acquisition and retention of morphology in heritage grammars has been discussed at length by Kim et al. (Reference Kim, O’Grady and Schwartz2018), Montrul et al. (Reference Montrul, Bhatt and Girju2015), and Polinsky (Reference Polinsky2018). The idea is that morphemes that are more perceptually salient are acquired and retained better than morphemes that are acoustically less salient. Low perceptual salience may account for why heritage speakers tend to omit obligatory morphology in required contexts and misunderstand them in comprehension. Polinsky (Reference Polinsky2018) has claimed that this property explains a lot about morphological errors, including patterns with DOM. Hindi and Romanian were chosen for this study because their DOM markers, being CV syllables, are more perceptually salient than the Spanish marker a: pe (Romanian), and -ko (Hindi). In Spanish, many verbs in the 3rd person singular present tense of the -ar conjugation (ayudar “help”) end in the vowel /a/, and the DOM marker is also /a/ (Ayuda a Juan “He/she helps Juan”) and the coalesce, making it difficult to perceive the DOM marker. As a CV syllable, Romanian pe and Hindi -ko seem to be more acoustically perceptible than a and therefore more noticeable. If acoustic salience is an issue (Polinsky, Reference Polinsky2018: pp. 165–169), it is not surprising to see that Romanian and Hindi DOM are preserved more than Spanish DOM in heritage speakers.

We did find that DOM was omitted more in Spanish than in Romanian and Hindi, which is consistent with the hypothesis that phonological salience plays a role in degree of DOM erosion (see also Montrul et al., Reference Montrul, Bhatt and Girju2015). However, the study also included other syntactic and semantic contexts where the preposition a is also used and is equally perceptible, such as with indirect objects and dative experiencer subjects (with gustar-type verbs). In the tests that included these other constructions, we found that a, pe, and -ko were omitted more often when they marked animate, specific direct objects (DOM) than when they marked indirect objects or locatives in the case of Romanian, suggesting that perceptual salience cannot be the only reason why DOM morphology is omitted, or even a reason at all: the syntax, the semantics, and the polyfunctionality (one-to-one, vs. one-to many-correspondences) of the morpheme are also relevant. In this respect, it seems that linguistic complexity, and the fact that DOM is a true interface phenomenon regulated by syntax, semantics, and pragmatics, is a better linguistic explanation for the patterns observed with erosion of DOM in Spanish, Hindi, and Romanian than the acoustic salience one. In fact, Longobardi’s (Reference Longobardi2001) Inertial Theory of diachronic syntax proposes that syntactic changes are forced by changes in the phonology, the semantics, and finally in the interfaces. Rinke (Reference Rinke, Fischer and Gabriel2016) emphasizes the role of interfaces in the modeling of diachronic change.

9.4.5 Other language-particular properties

Perhaps other structural properties of the languages contribute to DOM omission or retention. For example, Spanish and Romanian have definite articles, and definite articles mark definiteness and specificity in these languages. By contrast, Hindi does not have definite articles. The indefinite determiner in Hindi is the numeral ek “one.” Specific and non-specific definite NPs, including generics, have no articles. Definiteness and specificity can be marked by case marking (-ko) or word order, for example by scrambling, in Hindi. This means that, structurally speaking, -ko is needed more in Hindi than perhaps a and pe are needed in Spanish and Romanian to mark definite and specific direct objects.

Another likely cause for the better preservation of DOM in Romanian compared to Spanish could be the co-occurrence of CD with DOM, with animate, specific direct objects, and the fact that, according to Hill and Mardale’s (Reference Hill and Mardale2019, Reference Hill and Mardale2021) analysis assumed in this study, Romanian DOM is structurally and typologically different from Spanish DOM. Hill and Mardale claim that Romanian DOM takes place in the DP within the VP (nominal spine), whereas Spanish DOM takes place in a functional projection above vP (in the verbal spine). Furthermore, the syntactic status of the marker is different in the two languages (a preposition in Spanish but not in Romanian). Although Kayne’s generalization may apply to Spanish, it does not for Romanian, even though there is a close relationship between pe-marking and CD. CD reinforces the use of pe in Romanian maybe because the features of pe must be checked and valued by the clitic. Although some definite DPs do not take pe, if they are clitic-doubled they must have pe, because pe brings a more specific reading. Diachronically, pe lost the ability to value discourse features in K, the feature bundle [particularized] and [F_mark] split, and the clitic in ApplP acquired this valuing function for the discourse features in [F_mark], as we also saw in our data. The Romanian heritage speaker groups produced several instances of pe-marked animate specific direct objects without CD, suggesting that they may be bundling [particularized] and [F_mark] and valuing them with pe, as in Old Romanian. Therefore, it is likely that the other structural properties of Romanian, like the co-occurrence of DOM with accusative CD, may contribute to the higher preservation of DOM in Romanian heritage speakers compared to the Mexican and Latin American Spanish heritage speakers.

In addition to the structure of the language and dominant language transfer, there are several other potential external factors related to language vitality, attitudes, efforts for language maintenance, and exposure to the language in the families that may explain the overall the degree of erosion of DOM in the three languages.

9.5.1 Size and vitality of the language

In terms of the sociolinguistic characteristics of the three immigrant groups, with almost 60 million Spanish speakers, the Spanish speakers have a wider speech community in the United States than the Hindi speakers and the Romanian speakers in the United States (Carreira, Reference Arechabaleta Regulez and Montrul2021; Gambhir and Gambhir, Reference Gambhir and Gambhir2013). Although not uniformly, of all heritage languages in the United States, the Spanish-speaking community has more access to Spanish in media and advertising, and in instruction in elementary school through bilingual programs and in middle and high school as foreign language (Carreira, Reference Carreira, Beaudrie and Fairclough2012). For example, according to their answers to the extensive language background questionnaire administered to supplement the results of the experimental tasks, all the heritage speakers tested had access to Spanish as a second language in middle school and high school whereas the Hindi and the Romanian heritage speakers did not, unless instruction of some sort was provided by their parents or religious organizations/groups. The Romanian heritage speakers learned to read and write Romanian at home, so literacy activities were performed by the family. However, there seems to be an inverse relationship between the size of the speech community and the degree of vitality and the degree of DOM erosion in the data presented in this book.

9.5.2 Patterns of language use

All speakers reported using English and the heritage language, and the use of English increased with age. Yet all the heritage speakers reported using the heritage language the most with parents and grandparents, followed by siblings, and finally other friends. In terms of language use throughout the lifespan, the Hindi speakers used Hindi less frequently, preferring mostly English. The Spanish and the Romanian speakers reported higher use of the heritage language in the family than the Hindi speakers, including more interactions with friends who spoke the heritage language. The Romanian families appeared to provide more literacy opportunities to their children at home than the Spanish-speaking families, while the Hindi speakers were the most disadvantaged in terms of literacy. They were the least literate of the three groups largely because their language uses a different script, whereas the Spanish and Romanian scripts are easier to read for English speakers. Although all the heritage speakers seemed to have used both languages to some extent, the Hindi and the Romanian speakers reported the highest use of English with the exception of the Spanish heritage speakers. Thus, in terms of overall heritage language use and availability, Spanish has an advantage over Hindi and Romanian, because it is spoken by more speakers and used in a wider variety of contexts. Yet it shows the highest degree of erosion of DOM.

We also found an age of acquisition effect in the Romanian and Hindi heritage speakers but not in the Spanish heritage speakers. This pattern was evident from the proficiency measures (self-ratings and written cloze test) as well: in Spanish, there were no overall proficiency differences between simultaneous and sequential bilingual heritage speakers while in Romanian there were. One reason for this difference between Spanish, on the one hand, and Romanian and Hindi, on the other, may be related to the patterns of immigration of the three communities. The sequential bilingual heritage speakers from Romania immigrated to the United States later in childhood (5–14 years, mean 9) and had a longer length of residence in Romania (average 10.63 years) than the sequential bilinguals who immigrated as children from Mexico between the ages of 5–12 (mean 6.7) and lived in Mexico for an average of 6.8 years. The sequential bilinguals from Romania had longer exposure to their native language (3.8 years more) than the sequential bilinguals from Mexico, explaining, in part, the difference in their proficiency scores relative to the other groups and DOM omission between the Spanish-speaking sequential bilinguals and the Romanian-speaking sequential bilinguals.

We also found that the sequential bilingual Romanian heritage speakers spoke more Romanian with their parents and friends than the sequential bilingual Spanish heritage speakers spoke Spanish to their parents and friends during middle school. They reported having used Romanian often or seldom, and it was mostly when speaking with their parents and siblings, although they had frequently used it with English. However, when asked what language they preferred to speak, none said Romanian. The fact that the adult Romanian and Hindi groups did not seem to show signs of attrition whereas the first-generation adult immigrants from Mexico and the Latin American immigrants did may have also influenced these results. When the sequential bilingual Romanian heritage speakers spoke Romanian to their parents, the input from their parents was target-like, but when the sequential bilingual Spanish heritage speakers spoke to their elders, who also produced unmarked animate, specific direct objects, the input reinforced the grammar of unmarked objects activated by many Spanish heritage speakers.

9.5.3 Education and SES

Another important difference between the three groups is education and SES. The Hindi heritage speakers and the Hindi-speaking adult immigrants had the highest level of education when compared with the Mexican and Romanian immigrants. They came largely from professional backgrounds (doctors, engineers, scientists, lawyers, etc.), with profiles like the ones described by Gambhir and Gambhir (Reference Gambhir and Gambhir2013). By contrast the Spanish and the Romanian heritage speakers came from predominantly working-class backgrounds, as judged by the reported occupation of their parents in the United States, although the Romanian immigrants seemed to have higher level of education than the Spanish immigrants. Whether and how working class and economic experiences affect language development is a matter of controversy. Several studies have shown that working class children have smaller vocabularies and different profiles of syntactic development than children of professional families (Hart and Risley, Reference Hart and Risley1995; Huttenlocher et al., Reference Huttenlocher, Vasilyeva, Cymerman and Levine2002). Studies on adults suggest that educational and SES background have deterministic effects on language proficiency and linguistic competence in monolingual and bilingual speakers (Dąbrowska, Reference Dąbrowska2012) and commentaries therein (Pakulak and Neville, Reference Pakulak and Neville2010). Yet, the studies documenting differences in morphology and syntax are about later, rather than earlier language development. DOM is a feature of early language development because it seems to be acquired by children before age 3 (Chapter 4). At the same time, we see a marked difference between the Spanish and the Romanian speakers, even though many of the Romanians also came from working-class families. It is consequently not certain that the working-class status of the parents can account for our results. Certainly, SES is an issue that remains to be verified in Spanish by testing Spanish heritage speakers who come from professional families. It appears that maternal SES is relevant for the acquisition of the majority language but not for the development of the heritage language in school-age children (Gathercole, Reference Gathercole, Eilers and Oller2002).