Multilingual acquisition property by property: A view from a wider perspective

The keynote article by Marit Westergaard (Westergaard, 2021) is an interesting contribution to the field of language acquisition and multilingualism, especially because it attempts to provide a wider perspective by embracing first, second and third language acquisition. The main argument put forward by Westergaard (2021) is that all the above types of acquisition ‘are fundamentally the same process, based on learning by parsing’ and that language development takes place in small steps, i.e. property by property. The article outlines two models proposed earlier by Westergaard, i.e. the Micro-cue Model of first language (L1) acquisition (Westergaard, 2009, 2014) and its extension to the multilingual context, i.e. the Linguistic Proximity Model (LPM) of L3 acquisition (Westergaard et al., 2017). Additionally, the author argues for the Full Transfer Potential (FTP) and presents both trheoretical and empirical evidence for this claim throughout the article. In this commentary I would like to address major claims put forward by Westergaard, to support them with my own research data from the domain of phonology as well as to raise some questions as to potential limitations of the approach, and to point to avenues for future theoretical and methodological refinement.

The most important aspect of Westergaard (2021), in my view, is the proposal that language acquisition takes place property by property based on structural linguistic similarity. Westergaard (2021: 399) argues that ‘the genetic endowment for language learning makes humans sensitive to fine linguistic distinctions in monolingual as well as all types of multilingual language development, which means that the acquisition process does not involve setting or re-setting (macro-)parameters or copying whole grammars.’ With this idea being central to the article, and, in fact, the proposed models, we may wish to elucidate what is actually meant by the concept of the property-by-property transfer. Westergaard (2021: 398) offers an explanation, according to which it ‘simply means that it should be possible to find effects of both languages at every stage of L3 acquisition, not that it always obtains, regardless of the nature and number of properties investigated’. What comes across as highlighted in the provided definition is the temporal aspect, i.e. ‘at every stage of acquisition’ as well as its scope, which I interpret as encompassing various properties or linguistic features across linguistic domains that may behave differently, i.e. transfer independently from different sources. What seems less clear is the proviso ‘not that it always obtains’, which calls for further elaboration. One may also advise caution with the reference to ‘both languages’ as it seems too restrictive; I would suggest referring rather to ‘previously acquired languages’, without limiting their number. Such an approach would be in line with the acknowledged understanding of L3 acquisition as learning of the third and/or additional languages (e.g. de Angelis, 2007). This seems to be Westergaard’s (2021) understanding as well since in her comments contributing to the debate on bilingual advantage earlier in the text she states ‘what property-by-property transfer captures then is the insight that the more languages the learner knows, the easier it will be to learn further languages, since the existing repertoire will be larger’ (p. 397), therefore the use of the term ‘both languages’ seems slightly inconsistent with this stance. It is somewhat unclear to me, though, how the link between the multilingual repertoire size and its facilitative impact on subsequent language acquisition is assumed to be directly related to or stem from the property-by-property transfer. Thus, this thought may require further elucidation.

Another fundamental merit of Westergaard’s (2021) approach, in my opinion, is that LPM does not make a principled distinction between transfer and crosslinguistic influence. It seems a (welcome) detour from previous claims made by some scholars in the generative paradigm of L3 acquisition, according to whom these are two different processes; the former being representational transfer (i.e. copying or reduplication of a representation) and the latter consisting in temporary influence of one language onto another (see, for example, the Typological Primacy Model; Rothman, 2015). I second Westergaard’s (2021) view that there is no need to assume such a fundamental distinction which, albeit much debated, has a rather weak theoretical foundation and fails to stand empirical testing (see current psycholinguistic research).

Westergaard (2021) further presents a convincing line of argumentation disfavouring full (wholesale) transfer, based both on theoretical reasoning and limited empirical evidence. I am in complete agreement with this stance, as it seems to be also corroborated by research in the domain of third language acquisition of phonology, as I discuss. Referring to cognitive economy as a somewhat elusive concept that has been used differently in linguistic argumentation, Westergaard seems right in maintaining that ‘it is not at all clear that transferring a lot once is cognitively more economical than transferring a little many times’ (2019: 393) to address the wholesale vs. property-by-property transfer dispute. Although cognitive burden is obviously considerable in multilingual acquisition and processing, it seems unrealistic that for the wholesale transfer to occur from one language, inhibition processes would be necessary to suppress completely the activation of the other system or systems (see Wrembel, 2015).

However, what I am not fully convinced by is the notion of the Full Transfer Potential (FTP) put forward in the article. I am somewhat uncertain about how to reconcile this concept, which potentially relates to the idea of full transfer, with an earlier criticism wedged against it. How exactly does the Full Transfer Potential complement the notion of the property-by-property transfer? The rationale offered by Westergaard reads as rather tentative since she argues that FTP means that ‘anything’, rather than everything, ‘may transfer’. A more plausible explanation is the following: ‘all previously acquired languages remain active and may influence the L2/Ln [second language or subsequent languages], which I refer to as Full Transfer Potential’ (Westergaard, 2021: 380). In my view, this aspect of the proposal may be worthwhile further development and discussion.

Coming back to Westergaard’s idea that language acquisition takes place ‘incrementally in a step-wise fashion’, I fully endorse this view as it resonates well with my own work on the role of native and non-native languages in shaping the phonological acquisition patterns in L3 (e.g. Wrembel, 2015). In a series of parallel studies conducted on four groups of multilingual learners with different language repertoires, I examined three selected measures of phonetic performance: (1) accentedness ratings of perceived global accent in the L3, (2) acoustic measurements of voiced onset time (VOT) as a correlate of foreign accentedness and (3) oral protocols of metaphonological awareness. The results demonstrated that it is difficult to provide a unified account for the sources of crosslinguistic influence (CLI) in L3 as the process of phonological acquisition proved to be dynamic and complex. It appeared that the source of CLI largely depended on the phonological system and phonetic realization of the segments or processes in question, allowing for CLI from either of the prior languages. Consequently, the idea of a full transfer from only one language system was not corroborated by the data, and the cross-linguistic influence attested for L3 speech production (e.g. Wrembel, 2015; Wrembel et al., 2019b) as well as perception (e.g. Balas et al., 2019; Wrembel et al., 2019a), was found to be gradual and structure dependent, which is in line with the view that transfer is selective and works property by property, as proposed by Westergaard (2021) as well as Slabakova (2017) in her Scalpel Model.

In other related studies into L3 phonological acquisition, scholars have attempted to validate the assumptions of the proposed L3 theoretical models, even though they mostly stemmed from the morphosyntactic domain and were developed within the generative paradigm, i.e. the Cumulative-Enhancement Model (CEM): Flynn et al., 2004; the L2 Status Factor (L2SF): Bardel and Falk, 2007; the Typological Primacy Model (TPM): Rothman, 2015; the Linguistic Proximity Model (LPM): Westergaard et al., 2017; and the Scalpel Model (SM): Slabakova, 2017. No conclusive evidence has been shown in the domain of L3 phonology for any of these models in particular; Hammarberg and Hammarberg’s (2005) case study lent support to the L2SF, Wrembel’s (2012) and Lloyd-Smith et al.’s (2017) foreign accentedness studies offered partial confirmation of the TPM, while Kopečková’s (2016) study on early L3 production of vowels was in line with the predictions of the LPM.

Further support for Westergaard’s LPM approach was evidenced in a recent large-scale project (Kopečková et al., under review), in which we aimed to test the predictions of the existing L3 acquisition models by extending them to the context of initial L3 phonological acquisition. To this end, we investigated three phonological features (i.e. rhotics, final obstruent devoicing and the /v/ vs. /w/ contrast) with a relatively different standing in two parallel groups of young sequential multilinguals learning their foreign languages in a school setting: first language (L1) Polish, second language (L2) English, third language (L3) German vs. L1 German, L2 English, L3 Polish. The results demonstrated that the strongest predictor of CLI sources employed in the learners’ speech production was the phonological feature under investigation. For instance, the L1 German learners resorted to their L1 to produce the rhotic and final devoicing in their L3 Polish, while they relied on their L2 English to produce /w/ in L3 Polish. On the other hand, CLI from the L2 was more frequent in the L1 Polish group, especially in the rhotics production rather than final devoicing in L3 German. These findings can be best interpreted as corroborating (at least partially) the assumptions of LPM in that CLI is clearly structure-dependent and it occurs from both previously learned languages, based on perceived structural similarity of the specific features.

The results of our phonologically oriented study (Kopečková et al., under review) did not fully corroborate the conclusions reached by Puig-Mayenco et al. (2020) in their comprehensive overview of research into L3 morphosyntactic acquisition. In the phonological domain, at the early stages of acquisition of L3 speech the main source of CLI proved to be L1-driven. Further, a substantial number of hybrid forms as well as high inter- and intra-individual variation was attested; while some learners resorted to the L1 as the only source of CLI, others used their L2 and yet others exhibited multiple sources of CLI. To be able to explain better the sources of variability in the generated data, we would need to take into account various transfer conditioning factors stemming from, on the one hand, differential learnability of specific linguistic properties, misleading/insufficient input or construction frequency, as proposed by Slabakova (2017), and, on the other hand, universal preferences, context of use, saliency of input as well as forces implicit in human vocalization and perception such the ease of articulation and the clarity of perception (see, for example, Dziubalska-Kołaczyk and Wrembel, 2017, under review).

Another main insight in Westergaard (2021) that I find highly relevant is the attempt to assume a broader picture on the theoretical modeling of language acquisition. The models proposed by Westergaard seem generally capable of extending to various contexts of bi/multilingualism, as she seeks to holistically explain language acquisition in all its forms. What is also worth noting is the dynamic potential and a developmental perspective embedded in her proposals; ‘the LPM ‘maintain[s] that as the exposure to an L3 grows, the role of overall typological proximity should decrease, while the role of more abstract structural similarities should increase’ (Westergaard et al., 2017: 677). Generally, this approach resonates well with my own take on modeling multilingual acquisition. In search of an optimal explanatory framework, which could account for the complexity of the acquisition process in a multilingual mind, we recently proposed the Natural Growth Theory of Acquisition (NGTA) (Dziubalska-Kołaczyk and Wrembel, 2017, under review). NGTA is informed by natural phonology and enhanced by complexity theory with inductive support from our data. The theory assumes a gradual dynamic emergence of Ln phonology, shaped by the input from L1 and other languages (L2, Ln), and influenced by universal preferences understood as preferability generalizations as well as typology and context of use. NGTA is conceived as a general theory of language acquisition, which allows to model the acquisition of phonology as well as morphology and other language domains. One of the main tenets of NGTA is that the relative impact of linguistic variables (i.e. L1, Ln, preferability generalizations) is moderated by the configuration of extralinguistic factors in a given acquisition situation. Further, the complexity of the acquisition process is assumed to be dynamic and grow as the function of time and language learning experience. Motivated by a distinction between System 1 and 2 proposed by Kahneman (2011), we also claim that acquisition proceeds at two levels: the automatic and meta-level. The former is involuntary and instinctive, as manifested by articulatory routines and phonetic perceptual constraints; the latter is conscious and cognitively-based, as manifested by any aspect of meta-awareness. On the whole, I can see several parallels between Westergaard’s (2021) and our approaches despite the fact that they stem from different theoretical frameworks.

However, what appears to be more debatable is the claim that constitutes the backbone of the keynote paper, namely that ‘there is no fundamental difference between L1 acquisition on the one hand and L2/L3/Ln acquisition on the other.’ Even though I would be inclined to agree with the rationale behind this statement, namely that ‘both child and adult learners are sensitive to fine linguistic distinctions, and language development takes place in small steps’ Westergaard (2021: 379), I would be reluctant to fully endorse it, even with the proviso ‘that the L3 learner has more resources available at the outset of learning’ (Westergaard, 2021: 395). We should bear in mind that the settings of L1 vs. L2/L3/Ln acquisition may be fundamentally different (i.e. naturalistic vs. formal) and the course of learning may be differentiated by social-psychological factors as well. Several psycholinguistic experiments have demonstrated that implicit and explicit linguistic knowledge are neurolinguistically distinct and have different memory sources (Paradis, 2004), the former being based on procedural knowledge, the latter on declarative knowledge. Therefore, implicitly learned L1 is stored in procedural memory, whereas the L2/L3/Ln, if learned in an instructed setting, are based on explicit knowledge and are stored in declarative memory. Following from this, it seems difficult not to assume any principled distinction between the processes of first vs. second and subsequent language acquisition.

As for methodological considerations, unlike some scholars who consider a ‘mirror-image design’ as optimal research methodology for L3 studies (see Puig-Mayenco et al., 2020), Westergaard (2021) favours a design in which we can make a comparison between L3 learners and two groups of L2 learners with L1s that are the same as the previously acquired languages of the L3 group. According to her, only this methodology guarantees that we could pinpoint the exact contribution of additional language/s involved in L3 acquisition. However much I support this design, it may not always be feasible to implement it, as English tends to prevail as the L2 in educational systems across Europe and beyond.

When dwelling upon the significance of extralinguistic variables for language acquisition, such as age, recency of use, instruction, and language dominance, I get the impression that their role, and especially that of the order of acquisition, is somewhat deemphasized in Westergaard’s approach. However, as it transpires from research on L2/L3 phonology (for an overview, see, for example, Cabrelli Amaro and Wrembel, 2016), the order of acquisition and the so-called language status tend to be important predictors in this process. I therefore consider that potential variability resulting from different acquisition situations (as stipulated in NGTA) and an array of extralinguistic factors at play need to be further addressed. Moreover, individual differences should be accommodated in the theoretical modeling of language acquisition, ultimately leading to a greater emphasis on predicting individual paths of acquisition and explaining variable trajectories of language development.

To my mind, it is of paramount importance to expand the explanatory potential of a proposed model to various acquisition situations and different linguistic domains. Future studies into L3 acquisition should attempt to foster more across-domain investigations and feed into a wider interdisciplinary framework. It seems that Westergaard’s (2021) contribution constitutes an important step in the right direction.