L1, L2 and L3: Same or different?

This text comments on the Keynote article ‘Microvariation in multilingual situations: The importance of property-by-property acquisition’ by Marit Westergaard, who argues for Full Transfer Potential within the Linguistic Proximity Model in third language (L3) acquisition. The commentary points at some theoretical and methodological issues related to the Linguistic Proximity Model, e.g. the age factor in language learning, the role of metalinguistic knowledge and proficiency in L3 learning, and the lack of predictive power of the model.


I Introduction
The study of the L3 (third language, or additional languages), with its roots back in the 1950s has experienced a boom in the last decades. That boom has echoed in the generative branch of language acquisition studies. In recent years, a number of L3 models have been suggested, one of the latest being the Linguistic Proximity Model (LPM; Westergaard et al., 2017). The main claims of Westergaard's keynote (Westergaard, 2021) are that (1) the acquisition of all languages (L1, L2, L3) occurs by fundamentally the same UG-driven process; (2) they are acquired property by property; (3) in case of L3 acquisition, both the first language (L1) and the second language (L2) potentially influence the target language, and (4) transfer from both the L1 and the L2 can be either facilitative or nonfacilitative for L3 acquisition. Furthermore, it is claimed that variation in the L3 is comparable to variation in dialects.
The keynote starts with an overview of the study of L1 and L2 syntax departing from Westergaard's own Micro-cue Model of L1 acquisition and the claims that L2 and L3 learners are as sensitive to fine linguistic distinctions as L1 learners and that all language development takes place step by step. In L2 and L3 learning, Westergaard argues for what she calls a Full Transfer Potential. The prospect that all previously acquired or learned languages 1 potentially transfer into the L3 is intuitively attractive and also realistic considering what is known from a quantity of studies on language acquisition and learning, which to varying degree have shown transfer from the L1 into the L2 and, in case of additional learning, from both the L1 and the L2 into the L3. In fact, the idea of potential transfer is basically what most L3 research is about, and researchers have been struggling during the last decades with the question of which factors force either the L1 or the L2 to take on the stronger role in L3 learning. Strangely enough, Westergaard does not consider this question until the very end of the article. When it finally comes up, the attempts to answer it made by many of the authors already quoted in the article are not acknowledged.
What is then an L3? One often quoted definition is the one from Hammarberg (2001: 22): 'In order to obtain a basis for discussing the situation of a polyglot, we will here use the term L3 for the language that is currently being acquired, and L2 for any other language that the person has acquired after the L1' (emphasis added). From this definition it follows that L3 acquisition takes place when the learner has already acquired (at least) one L2, and that the L2 has been acquired later in life than the L1. In De Angelis (2007) this is further problematized when defining what characterizes third language acquisition. She states that 'it may seem obvious to many that the prior knowledge of a nonnative language is a variable that needs to be properly controlled' (2007: 6; emphasis added) and further discusses what effect the experience of already having learned a foreign language has on the L3 acquisition process. In summary, both Hammarberg and De Angelis (as well as many others) point out that it is the previous experience of learning a non-native language that distinguishes L3 learning from L2 learning. In the model suggested by Westergaard, this distinction is not made, which in itself does not have to be a problem, but we argue that her definition of a L3 needs to be clearly spelled out. We will come back to why below. In the following, we will discuss some further critical points of the keynote. As Westergaard acknowledges (2021: 396), previous criticism of the LPM has been put forward by Rothman et al. (2019: 143) concerning its lack of predictive power. We agree with this critique. Assuming not only that all background languages (L1 and L2) can exert influence on the L3 learning process but also 'both facilitative and non-facilitative influence from one or both previously acquired languages' (p. 16) makes it impossible to falsify the model, since whatever influence the learner exhibits, or does not exhibit, will be in line with the predictions of the model. In empirical support of her model, Westergaard briefly presents two studies (2019: 19-20), the first being her study from 2017  of bilingual children (2L1 Norwegian-Russian) learning English as what is argued to be an L3. First, as stated above, we miss a definition of what Westergaard calls an L3. We are not convinced that the target language in this study should be defined as L3 only because there are three languages involved; on theoretical grounds that are developed below, it is more reasonable to call the current target language an L2 since the bilinguals in question possess two L1s acquired simultaneously in an informal way and setting . Hence, this study can hardly serve as support for a general model of L3 learning. Second, there is no information in Westergaard (2021) about the number of participants or how the items were presented, i.e. what kind of test was used. The reference to the other study brought up to corroborate the model, an unpublished poster by Mitrofanova and Westergaard (2018), suffers from similar shortcomings: the participants are called bilinguals, but nothing is said about at what age they acquired their L2, and no information is given about the number of participants nor about method. We find it hard to acknowledge how these two studies, of which the second is impossible to replicate, can serve as empirical evidence for the LPM, since there is no clear indication of what exactly is found in the data that corroborates a model for L3 learning.

II Methodological issues
Westergaard (2021) also acknowledges a critical point made by Schwartz and Sprouse (2019). They have argued that piecemeal transfer 'in L3A [third language acquisition] is fundamentally incompatible with generative grammar, because elements of natural language grammars . . . necessarily interact' (Schwartz andSprouse, 2019: 1, in Westergaard, 2021: 396). According to Westergaard, L3 learning (as L1/L2 learning) is learning by parsing. In Westergaard's Figure 5 (Westergaard, 2021: 395), this is illustrated with micro cues from both the L1 and the L2, which together with UG build up the L3 grammar, step by step. According to Schwartz and Sprouse, this is incompatible with the nature of UG. Our understanding of Westergaard is that as there is microvariation in native speakers' UG-driven language, this holds also for L3 processing. However, we find it hard to understand what is meant by 'this view reflects the variability that is typical of early L3 grammars' (Westergaard, 2021: 396): how are these two phenomena comparable, and how could this be measured in order to confirm the predictions made by LPM? To equate the well-known variation of interlanguage with (micro)variation in L1 or in certain dialects of native speakers is farfetched.

III About the age factor in additional learning
Does it make sense to stretch Westergaard's Micro-cue Model from 2009 on child L1 learning to L2 and L3 learning? We do not think so, because what is usually referred to as L3 learning is conditioned by factors that do not apply in L1 acquisition. What Westergaard et al. (2017) show in their empirical study can probably tell something about simultaneous bilinguals learning their first 'foreign language', i.e. English in school, comparing them to monolingual peers, but is not relevant to the study of how postpuberty learning of an L2 affects an L3.
Is there then, really no difference between L1, L2 and L3? If all languages are learned before the critical period for language learning (CP), this assumption is easy even for us to accept. If L2 and L3 are, instead, learned at a different age (i.e. after the CP, with all that this implies in terms of metalinguistic knowledge and awareness, cognitive development, encyclopedic knowledge, etc.), we find the LPM hard to endorse. Stepping into the debate on what characterizes an L3, basing one's account on a study on young learners who are bilinguals from early age is comparing apples and oranges.
Regarding age, Westergaard (2021: 394) claims that 'relevant studies [to the LPM?] could include simultaneous bilinguals who do not have a clear L1 and L2'; order of acquisition does not matter. This is indeed in line with the assumption that there is no fundamental difference between acquisition or learning of L1/L2/L3. However, using data from simultaneous bilingual acquisition before the age of 3 years (i.e. young learners with two L1s learning an L2), assuming that two first languages in these children are the same as a L1 and a L2, in order to back up a theoretical model for L3, and defending the methodology with a debatable theoretical assumption, begs the question.

IV About proficiency levels in the L2 and the L3
Another factor that has been vividly discussed in L3 research is that of proficiency in both the L2 and the L3. Studies have investigated what low versus high proficiency levels in an L2 imply for the L3 learning process, as well as what role the L3 proficiency level per se has for transfer into the L3. It was already claimed in Williams and Hammarberg (1998) that a background language in which the learner has a high proficiency level is more likely to influence the L3. Sánchez (2014Sánchez ( , 2020, among others, have investigated how various levels of proficiency in the L3, as well as in the L2, affect syntactic transfer in the L3. Sánchez (2014) found a significant difference between three groups of L3 learners that was clearly attributed to their proficiency level. Also, as suggested in Falk and Bardel (2011: 77), proficiency level in the L2 must be of importance for transfer, as the learner must reach 'a sufficiently high level in their L2' in order to be able to transfer a structure. Westergaard (2021) leaves no space in the LPM for the impact of proficiency on the L3 learning process. The L3 learner gradually builds up a new grammar: 'There is no second step, the process simply continues' (Westergaard, 2021: 395) and 'parsing will lead to stable linguistic representations' (p. 18).

V About structural similarity and transfer
In the theoretical construction behind the LPM, structural similarity rather than genetic relationship between the languages involved is a condition for transfer, structural similarity meaning that structures are similar in a particular, influencing, background language and in the target language. It is a commonly agreed position in transfer studies that such objective similarity is involved in facilitative transfer. In the LPM, both types of transfer are allowed by (or in) incremental property-by-property Ln acquisition (as seen above). How then, will the LPM account for non-facilitative transfer? In an amendment of the 2017 version of the model, it is acknowledged in the keynote that, somehow, the learner can get fooled by 'surface typological (lexical) similarity at early stages' (Westergaard, 2021: 394), exactly as modeled by Rothman (2015) with his Typological Proximity Model (TPM) as a step in the parser's way through L3 development. This seems to be the only way to explain non-facilitative transfer in the LPM framework and is a big step closer to the TPM. The one significant difference between the two models that remains is the Full Transfer in the TPM vs. the Full Transfer Potential in the LPM.
Finally, Westergaard discusses other factors that could influence the learning of an L3 and mentions Bohnacker's (2006) study, which reported non-facilitative transfer from an L2. Westergaard (2021: 399) concludes: 'the influence here is arguably from the L2, and as far as I can tell, neither the LPM nor the TPM can account for these findings. ' Westergaard (2021: 399) proposes that future research should 'investigate which factors decide which slice of the linguistic resources to use at which point in the L3 acquisition process -one or both of the previously acquired languages of the general linguistic endowment', a somewhat anachronistic suggestion, since this is exactly what studies on, for example, the L2 status factor have been aiming at over the last decade, whereas the LPM does not comprise, and cannot confirm, any factors (see, for instance, Falk, 2007, 2012;Falk et al., 2015).

VI Concluding remarks
There are differences between acquiring an L1 and learning a non-native language (L2, L3, etc.). These differences are related to age of acquisition and context of learning, metalinguistic knowledge and awareness, formal learning situations, among others. L3 learning cannot be explained with data from simultaneously bilingual children who learn their first foreign language. An important factor in L3 learning is metalinguistic knowledge about languages. This develops with age and is especially activated in learning around and after the critical period, when learning languages in formal contexts.
In the title of the keynote, 'Microvariation in multilingual situations: The importance of property-by-property acquisition', Westergaard (2021) refers to multilingual situations; prototypically these make reference to situations in which individuals have access to more than one language, for instance bilinguals who, on a regularly basis, make simultaneous or alternated use of two or more languages during conversation. These situations do not necessarily refer to the learning of yet another language. The LPM may be useful and important when explaining what conditions bilinguals who live in multilingual situations and learn their first non-native language, and how the two L1s constantly offer the learners a Full Transfer Potential.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.