Attrition,addition,and age

It has long been proposed that language attrition and language acquisition must be related at some level (Jakobson, 1941). With empirical evidence from lexical access and processing, phonetics, and the syntax-discourse interface, Schmid and Köpke (2017) have recently contextualized attrition within bilingualism, emphasizing its empirical and theoretical relationship to first and second language acquisition. In their keynote article, Hicks and Domínguez (2020) (henceforth H&D) also attempt to relate acquisition and attrition by advancing a formal model of grammatical attrition that integrates formal principles of generative grammar (including specific mechanisms of feature reassembly) with a more fleshed out psycholinguistic model of first language acquisition that invokes an inference module, at the level of intake. Their ‘attrition by acquisition’ model, H&D claim, can successfully resolve the fundamental paradox of language attrition: that grammatical change, including loss, is in principle possible and supported by our inborn grammatical machinery but extremely rare and generally confined to very specific grammatical domains.

Innovative in H&D’s attrition by acquisition model is their emphasis on the role of intake, a construct that has long been discussed in second language acquisition theory (Carroll, 2001; Krashen, 1982; VanPatten, 1996), but has only now been invoked in L1 acquisition (Lidz and Gagliardi, 2015). Whereas perceptual input is acoustic and visual signals, intake is what the mind extracts and makes sense from these signals, and the subset of the input that ultimately feeds the grammatical machinery. Adapting the L1 acquisition model proposed by Lidz and Gagliardi (2015), H&D expand on the function of the inference procedure that mediates between perceptual intake and acquisitional intake. It is at this level where input is processed and integrated by the learner/attriter. When grammatical representations are adultlike or nativelike the inference engine is no longer engaged, as there is nothing to learn or restructure; when representations are partial or incomplete, perceptual representations compatible with the input are generated by UG. In H&D’s model, both children and attriters are assumed to engage the inference engine to effect grammatical change. In future work, the specific functioning of the inference module in grammatical language change could be extended to explain aspects of second language acquisition that are also paradoxical but less rare than attrition in adults, such as fossilization. Fossilization, when second language grammars do not restructure any further and stay in a non-native stage, could be seen as the opposite of attrition because despite optimal input, the inference module seems not to be engaged or fails to become engaged to change grammatical representations. Elucidating more precisely the role and operation of the inference engine in first language (L1) and second language (L2) acquisition and attrition contexts and the factors that contribute to its malfunction would be a fruitful venue to pursue in future work to build a more general theory of acquisition and attrition.

An important question in language attrition research is what specific form attrition takes, and transfer effects of the L2 onto the L1, more than any other available UG option, are often the dominant pattern. That is why I concur with H&D’s characterization of attrition in a bilingual context as ‘addition’ or in some cases ‘disuse’ of available grammatical options, or fluctuations between two options. In their example, bilinguals who overuse overt pronouns in contexts where null pronouns are more felicitous, have not lost the grammatical features of their L1 altogether, but they may be entertaining the features of the L2 grammar concurrently. I have found these same patterns in my own work on maintenance, incomplete acquisition, and attrition of differential object marking (DOM) in Hindi, Romanian and Spanish as heritage languages (Montrul, 2014; Montrul et al., 2015; Montrul and Sánchez-Walker, 2013). In languages that have DOM, DOM is the overt morphological marking (with a case marker, a preposition or a postposition) of direct objects that must be distinguished from subjects on semantic and pragmatic prominence scales (Aissen, 2003), as in Spanish Juan vio a Maria (John saw DOM Maria), where the preposition ‘a’ is the object marker. In Romanian, the DOM marker is the preposition pe (Ileana a văzut pe Beatrice, Ileana saw DOM Beatrice), and in Hindi the postposition -ko (Mira-ne Ramesh-ko dekhaa, Mira-erg Rakesh-DOM saw, ‘Mira saw Rakesh’). English is not a DOM language and all objects are unmarked. The syntactic analyses that have captured the complexity and prominence of DOM in the clause structure posit that DOM objects move to a higher functional category than unmarked objects, attracted by the animacy or specificity features of DOM (Torrego, 1998; López, 2012). So, all DOM languages have some objects with semantic features that occupy a dedicated functional projection and objects without features that stay and receive case in situ. Non-DOM languages have only one projection for all objects.

Montrul et al. (2015) found that the heritage speakers of Spanish, Hindi and Romanian and adult Spanish-speaking immigrants all showed incorrect acceptance of unmarked animate, specific objects. At the same time, they also correctly accepted marked animate, specific objects. These patterns also obtained in language production, where some speakers omitted DOM to various degrees, producing both correctly marked objects and incorrectly unmarked objects (Montrul and Sánchez Walker, 2013; Montrul et al., 2019; Montrul and Bateman, in press).

We can explain their behavior by assuming that the heritage speakers and immigrants who consistently judged and produced DOM-marked objects have a linguistic representation of these objects in a higher structural projection above the VP, as proposed by Torrego and López. For these speakers, unmarked objects stay within the vP. Now, when these speakers use English, they do not mark animate specific direct objects and project a structure where animate, specific direct objects, like all objects, receive structural case within the vP. We can imagine that these bilingual speakers have a grammar of marked objects for Hindi, Spanish and Romanian and a different grammar for unmarked animate, specific direct objects in English, which does not involve movement with overt morphological marking. In other words, these bilinguals are not attrited and generate two grammars for animate and specific objects, one for the DOM heritage language and one for the non-DOM language, English, and they use one grammar for each language, keeping both grammars separate, or inhibiting one grammar when they use the other. In contrast, the heritage speakers and adult immigrants who incorrectly accept and produce unmarked animate, specific direct objects in Spanish, Hindi and Romanian (attrited) also have two grammars of animate specific, direct object, except that because English is more dominant than the heritage language in these speakers, the higher cognitive and psycholinguistic activation level of English with respect to the level of activation of the heritage language (Putnam and Sánchez, 2013), leads them to recruit the grammar of animate, specific direct objects of English – a representation with no marking and no movement – for the heritage language as well, instead of using the grammar of Spanish. Those who produce and accept DOM probabilistically are entertaining the two grammars for the heritage language, which would explain the optionality or fluctuation. In essence, all these bilingual speakers generate multiple UG-sanctioned grammars in the sense of Kroch and Taylor (1997), Amaral and Roeper (2014) and Yang (2002), and the coexisting grammars of DOM occur at the representational and processing levels (Sharwood Smith and Truscott, 2014). As in situations of diachronic change in general, different grammars coexist for an extended period of time until one of them is not generated or entertained anymore (Lightfoot, 1991). The bilingual speakers who are consistent in their acceptability judgments and their production, either because they rejected all or most sentences with DOM or because they accepted all or most of them, have activated one or the other grammar of DOM (the one of the heritage language or the one of English). Those who consistently accept unmarked objects in the heritage language and no longer produce DOM objects may have already eradicated the DOM-grammar, while those bilinguals who show more inconsistent acceptability ratings and optionality in production are still entertaining and activating both grammatical options for animate specific direct objects in the heritage language (the one for English in a the lower vP and the one for the heritage language above the VP). In essence, taking into account the interaction of the two languages of the bilinguals, attrition looks like addition and less than loss, and drives us away from the deficit view of attrition which results when we only look at one language, the shrinking language, in bilinguals and compare it to that of speakers living in a predominantly monolingual environment. I concur with H&D that incomplete acquisition and attrition in heritage speakers and the first generation of immigrants lead to structural changes and different end state grammars that can be described and explained using the available tools of theoretical models developed to account for native, monolingual grammars.

Where I think H&D’s model is not entirely successful, or at best not clear, is in their claim that their model by itself accounts for why attrition is rarely attested in adults and heavily constrained to very specific grammatical domains. Missing from this model is the maturational or age related component of attrition, which has been shown to strongly determine the extent and frequency of attrition observed in the literature. Age of acquisition is the strongest prediction of extent of L2 acquisition and language loss (Bylund, 2009; Flores, 2010; Montrul, 2008), especially when it comes to morphosyntax, the focus of H&D’s model. Actual language loss is rare in adults but it is not in children exposed to the same environmental conditions, and this important fact is not acknowledged in this model. L1 attrition in a bilingual context starts off as minor, but perceptible changes in language use (in phonology, vocabulary, morphosyntax, interfaces, processing) can lead to eventual drastic loss of linguistic competence and performance of any sort. This dramatic outcome, however, has only been attested when input and use of the first language is abruptly interrupted and discontinued, as in very specific cases of very young children adopted internationally (Pierce et al., 2016). When input is still available, complete loss is unlikely, but it is positively correlated with age (the younger the child, the higher the loss), as in heritage language speakers. Because these bilingual children in non-adoption contexts continue to have access to L1 input, even if exposure is suboptimal, their degree of attrition is less severe than in the adopted contexts but significantly more pervasive in frequency and extent than in adults (Montrul, 2016). Postpuberty, language attrition becomes rarer, and actual examples of grammatical or morphosyntactic attrition, other than the overt/null subject distribution, have been hard to come by (but see Montrul, 2017). A comprehensive model of attrition and acquisition must account for age effects along the lifespan, language entrenchment and susceptibility to input. H&D’s model is a good first step in the right direction.