Choosing between theories: Can L3 acquisition really decide?

In her thought-provoking article, Westergaard (2021) presents the micro-cue model of first language (L1) acquisition. She argues that the same model (with modifications relating to language transfer, in particular Full Transfer Potential or FTP) applies to subsequent acquisition situations, including second language (L2) and third language (L3). According to the micro-cue model, all language acquisition involves grammar building, via parsing, using micro-cues from UG or previously acquired languages to develop micro-cues specific to the language being acquired. Here I will focus on Westergaard’s argument that L3 data can be used to choose between L2 transfer theories, in particular the Full Transfer Full Access hypothesis (FTFA; wholesale transfer) versus FTP (property-by-property transfer). I will suggest that use of L3 data to distinguish between theories is not as straightforward as Westergaard assumes, hence that it is not clear that the predictions of FTP are in fact testable.

There is considerable overlap between FTFA and FTP, as can be seen in (16) in Westergaard’s article. In particular, both theories assume that the L1, together with UG, comprises the initial state of L2 acquisition, both assume acquisition by parsing (although this is implemented rather differently), both assume access to UG when the L1 grammar fails to provide a suitable analysis/parse, and both assume gradual development of the L2 grammar. Major differences between the two theories include the following: (1) FTFA argues for initial transfer of the L1 grammar in its entirety whereas according to FTP all L1 micro-cues are fully available but aren’t necessarily used; (2) related to the first point, FTP assumes, unlike FTFA, that transfer takes place property by property, and not just in the initial state.

Westergaard points out, however, that one cannot choose between the two theories on the basis of L2 acquisition data. Where there is evidence for full transfer, one cannot tell whether transfer took place in a wholesale or piecemeal fashion. Furthermore, although not discussed by Westergaard, where there is evidence at some stage for transfer of only some properties, this could be because the interlanguage grammar has been partially restructured. In other words, only some aspects of the L1 are maintained beyond the initial state, consistent with the claims of FTFA. Alternatively, transfer was property-by-property, with some properties not transferring, consistent with FTP.

Westergaard argues that L3 acquisition can provide the test case to determine which transfer theory (wholesale or property-by-property) is correct. Assuming that the claims of FTFA extend to L3 acquisition, the L1 grammar will transfer in its entirety; there will be subsequent restructuring on the basis of L3 input. According to FTP, on the other hand, the learner can start off by deploying micro-cues from either of the previously acquired languages, L1 or L2, and develop micro-cues specific to the L3. Thus, there would seem to be a relatively straightforward difference between the two theories: full L1 transfer initially (FTFA) versus evidence of transfer from both prior languages, or from either (FTP). However, matters are not so straightforward.

Let us first consider the initial state. In order to test between the two transfer theories, it is necessary to examine the initial state, contrary to Westergaard’s assumption that there is no ‘special status of the initial stages’ (Westergaard, 2021: 394). It is in fact only in the initial state that one would expect to see a difference between the predictions of these theories. As an example of the problem of distinguishing between FTFA and FTP, consider a study by Grüter and Conradie (2006). Much L2 research is, inadvertently, L3 research. That is, L2ers of language X are recruited, and the fact that they have previously acquired language Y as an L2 is ignored. Grüter and Conradie’s article is ostensibly a study of L2 German acquisition by speakers whose L1s are Afrikaans or English. But it turns out that the Afrikaans speakers all knew English, so German was in fact the L3 for them.

The study focuses on the initial state of the German grammar, using a methodology devised by Grüter (2006) to explore knowledge of word order alternations pertaining to wh-questions. These can be ambiguous between SVO or OVS order, with differences in interpretation depending on the assumed word order: Was beisst die Katze ‘What is biting the cat’ (SVO, subject interpretation) / ‘What is the cat biting’ (OVS, object interpretation). The same ambiguity holds for Afrikaans but not for English. All participants were beginners in German, with around 30 hours of instruction at the time of testing. Results of an interpretation task showed that the L1 Afrikaans group performed similarly to a German native speaker (NS) control group, either recognizing potential ambiguity or choosing only one of the interpretations (predominantly object), whereas the English speakers mostly did not recognize the ambiguity and chose subject interpretations for sentences that would require this interpretation in English.

These results appear to provide quite direct and convincing evidence in favour of FTFA, as argued for by Grüter and Conradie (though they do not treat the Afrikaans speakers as an L3 group): (1) there was transfer from L1 English to L2 German, since the test items were treated like their English counterparts; (2) there was transfer from L1 Afrikaans to L3 German, since ambiguity was either recognized or, when not recognized, learners preferred the same interpretation as the NS group. There is nothing that the L3ers did that suggests an effect from the L2 English.

While the data clearly support FTFA (and, incidentally, argue against the L2 Status Factor of Bardel and Falk, 2007), the absence of evidence of L2 cues cannot, in fact, be used to disconfirm FTP. This is because of a modification introduced by Westergaard for the L3 situation, namely the Linguistic Proximity Model (LPM), which takes account of linguistic proximity, defined somewhat differently from the Typological Primacy Model (TPM) of Rothman and colleagues (for an overview of TPM and other L3 theories, see Puig-Mayenco et al., 2020).

The claim of the LPM is that the potential for transfer (either positive or negative) is enhanced when the learner perceives an abstract similarity between the L3 input and either of the previously acquired languages, such that transfer from that language will predominate. Indeed the name, Full Transfer Potential, shows that full transfer is not unexpected on this theory. In other words, evidence for full L1 transfer in the L3 initial state would not be inconsistent with LPM. In effect, linguistic proximity can result in micro-cues from only one of the prior languages being adopted, at least initially. When L1 and L3 are similar, as is the case for Afrikaans and German in the example above, the LPM allows for L1 transfer to predominate, rendering no difference between FTFA and FTP. This raises the question of whether the FTP is actually testable once LPM is taken into consideration.

In order to distinguish the two theories more precisely, we need situations where the L1 is not the linguistically closest, i.e. where the L2 and L3 are closer. In the case of the above example, L1-English–L2-Afrikaans–L3-German would be required. This, together with L1-Afrikaans–L2-English–L3-German would constitute what Puig-Mayenco et al. (2020) refer to as mirror imaging, which, for some reason, Westergaard rejects as the preferred means for testing LPM (Westergaard, 2021: 397). But mirror-imaging would seem to be essential for testing FTFA versus FTP. In this situation, FTFA predicts transfer from English, not Afrikaans, while FTP would presumably allow both or either, most likely Afrikaans (because of the LPM). I am not aware of any data that address the L1-English–L2-Afrikaans–L3-German combination with respect to German wh-questions, so it remains an open question as to whether or not FTFA would be supported (for examples of other relevant mirror image research, see Puig-Mayenco et al., 2020).

Let us now turn to stages beyond the initial state. Here we have more or less the same problem that Westergaard identifies for L2, namely that learner behaviour can be rendered consistent with either theory. FTFA does not expect wholesale transfer to be replaced, in one fell swoop, by the L2 or L3 grammar. While transfer is across the board in the initial state, restructuring is not, since development away from the L1 grammar depends on how the input interacts with the current grammar; different grammatical properties may be affected differentially. This means that acquisition stages beyond the initial state will involve a mixture of L1 properties and non-L1 properties, which might superficially resemble property-by-property transfer. At any particular point in time, one cannot tell.

What about presence of L2 properties in the L3 interlanguage grammar? This might seem to be in conflict with the claims of FTFA. However, for FTFA, after the initial state, it is possible for the learner to arrive at a grammar (often temporarily) that is neither the L1 grammar nor the grammar of the language being acquired, as in Schwartz and Sprouse (1994), where Cevdet (L1-Turkish–L2-German) at a certain stage adopts an analysis of case-checking of post-verbal subjects, which is consistent with neither of these languages but, rather, with French. ¹ The implication, then, is that, once beyond the initial state, FTFA does not preclude the possibility of the adoption of analyses consistent with a grammar that is neither that of the L1 nor the L3, so possibly the L2.

Consider again the hypothetical example of L1-English–L2-Afrikaans–L3-German. The best evidence in favour of FTFA would be evidence of the English grammar and only the English grammar in the initial state, parallel to what Grüter and Conradie (2006) and Grüter (2006) found for L1 English L2 German. Would evidence of apparent L2 influence from Afrikaans constitute counter-evidence to FTFA? In fact, if the L1 English group were to show appropriate recognition of ambiguity of German wh-questions, one could not exclude the possibility that L3 acquisition was taking place, rather than L2 transfer.

Westergaard tries to get round such problems by looking at different properties in the same grammar at a particular stage (for example, V2 and subject–auxiliary inversion). But, as already mentioned, since nothing in FTFA dictates that transfer will be overcome at the same time for different properties, even such cases cannot decide between the two theories.

In conclusion, while conceptually one can understand the differences between FTFA and FTP, empirically it is much harder to demonstrate how these differences work out, and where the theories yield testable predictions that allow one to distinguish between them. In particular, it is not clear what would constitute counter evidence to FTP. Westergaard has already pointed out that, using only L2 data, it is not possible to decide between the two theories. I have suggested here that L3 data are equally problematic. A potentially plausible theory may prove to be untestable by means of either L2 or L3 data.