Bilingual acquisition as the locus of syntactic change

1 Introduction

Languages change over time, and grammars change too. This is not news, yet the mechanisms of change are less well understood than we would hope. An adequate theory of change must account not only for the properties of a given variety but also for their immutability or vulnerability to variation and change. Syntax is acknowledged as a domain where change does not happen easily; cf. Longobardi’s inertial theory (2001). Still, some properties are more resilient than others.^[1]

The question I address here is: When syntactic properties are particularly resistant, what causes them nonetheless to change? My focus is on grammatical phenomena I consider to be part of the syntactic core, primarily OV/VO order and V2 (verb-second placement). This terminological choice calls for a justification since the core-periphery distinction, proposed by Chomsky (1986), is an object of controversy (Culicover 2013). The notion of syntactic core that I adopt covers a set of properties that are constitutive of grammatical systems and that play a crucial role in emergent grammars. They are essential for the definition of grammatical parameters (Chomsky 1981, 1986) and provide criteria for a typology of languages. Perhaps more importantly, they are the first syntactic properties to emerge in first language (L1) acquisition. Indeed, Tsimpli (2014) argues that they are acquired early because they are syntactic core properties. It should not come as a surprise that phenomena with these characteristics are particularly resistant to reanalysis. Parameter Theory distinguishes between macro- and microparameters, and core properties reflect macroparametric options.^[2] These are claimed to “have highly pervasive effects on the grammatical system” and to be “unlikely to be subject to reanalysis by language acquirers … ” (Biberauer and Roberts 2017: 149).

In what follows, I briefly explain the role attributed to language acquisition in diachronic change. I then summarize results of empirical acquisition studies that can corroborate or undermine these scenarios. I finally propose grammatical and acquisitional prerequisites for a plausible scenario of diachronic change affecting syntactic core properties.

2 The language learner as the locus of syntactic change

There exists broad consensus among researchers in historical linguistics, particularly among those adopting a generative framework, that the language-learning child is the main agent of change. This hypothesis is based on the fact that major alterations of grammars do not happen across the lifespan of adults. To the extent that mature grammars are at all affected by change, this concerns peripheral rather than core properties; cf. Sankoff (2005, 2019. Designating language learners as the locus of change is a venerable idea in historical linguistics. Paul (1975 [1880]) argued that first language acquisition plays a crucial role in language change, and Andersen (1973) implemented this idea in a model of diachronic change – an inevitable conclusion in cognitively oriented linguistic theorizing where the object of study is the mental grammar of the individual rather than the collective knowledge of the speech community. Consequently, the individual must be the locus of change, and if alterations of fundamental grammatical properties never occur in mature grammars, these changes must happen in the process of acquisition when learners reconstruct the grammars underlying the primary linguistic data (PLD) to which they are exposed.

Shifting the task of explaining grammatical change from historical linguistics to the study of acquisition is not without consequences for the latter, for the idea of transmission failure imposes a task on acquisition theory that stands in conflict with its genuine research interest, namely, to explain what enables children to reconstruct the mental grammars of their parents’ generation. Transmission failure implies that learners diverge from the grammar of the previous generation in their interpretations of constructions encountered in their linguistic environment. This assumption contradicts the hypothesis that adequate exposure from birth is all it takes for children to become fully competent speakers of a language and that uniformity of development and ultimate success are defining characteristics of L1 acquisition. This hypothesis is strongly supported by empirical findings, and reports on acquisition failure are extremely rare in the L1 literature; cf. Meisel (2011b). Language use of learners does, of course, temporarily differ from adult speech, but the systems ultimately converge, and this happens very early and fast for core syntax.

One way out of this paradoxical situation is to search for elements in the input data that might trigger an analysis diverging from that of the parents’ generation. Lightfoot (2006, 2017 suggested that variation in the PLD could have this effect. A frequently defended hypothesis assumes, on the other hand, that only language or dialect contact can produce such an effect. Note that in these cases, learners perform correct analyses, relying, however, on “wrong” cues.

Alternatively, it has been proposed that transmission failure occurs if the information that is necessary for learners to reconstruct the previous generation’s mental grammars is difficult to detect or partly inaccessible in the PLD. This is an approach taken in historical linguistics over past decades: various learner-external as well as -internal factors have been suggested that allegedly trigger reanalyses of grammatical properties in intergenerational transmission, most importantly structural ambiguity. The idea is that the PLD contain ambiguous information, obliging children to decide between grammatical analyses because a particular construction can be assigned more than one grammatical interpretation. This choice between competing analyses is argued to be decided by the construction’s frequency of occurrence or by its structural complexity.

These are potentially crucial factors for explanations of core syntactic changes. Unfortunately, these scenarios refer only cursorily or not at all to insights obtained by L1 research. They rely on grammatical arguments alone in their attempts to identify conditions favoring novel analyses. Yet in order to make a convincing case for transmission failure, morphosyntactic and acquisitional facts need to be taken into account; see Meisel (2011a) or Rinke and Meisel (2009). In what follows, I will point out shortcomings of scenarios that rely exclusively on syntactic arguments, using the change from V2 to non-V2 as an illustration.

The ambiguity argument refers to the fact that V2 as well as non-V2 grammars generate SVO surface order in main clauses although subject and finite verb occupy distinct structural positions in the two grammar types, CP (V2) or TP (non-V2). Only if a constituent other than the subject appears initially, does this underlying difference become apparent in surface word order: in V2 languages, the subject then follows the finite verb (XVSO), yet it precedes the verb in non-V2 languages (XSVO). A first observation casting doubts on the ambiguity argument is that there exists no empirical evidence from L1 research indicating that surface ambiguity of SVO sequences trigger reanalysis. To the contrary, it has been demonstrated again and again that children learning a V2 language like German acquire the V2 option during the earliest phase of syntax development, fast and virtually without errors, with no sign of confusion; cf. Clahsen (1982) and more recently Westergaard (2008) or Tsimpli (2014).

This is where the frequency argument kicks in. Adams (1988), for example, who analyses OF (Old French) as a V2 language, attempts to explain the transition to non-V2 in Modern French due to an increase in the number of SVO patterns in children’s input during the Middle French period. Thus, the high frequency of SVO surface strings putatively reduces the number of telltale signs for the underlying V2 grammar. Yet again, insights from L1 research do not support the claim that increasing numbers of ambiguous patterns trigger reanalysis, as long as the number of unambiguous ones does not drop below a critical threshold; cf. Kiparsky (1997: 464). In fact, Westergaard (2008) has shown that even children acquiring a mixed V2/non-V2 system acquire both orders early, irrespective of very different input frequencies.

An alternative approach combines structural ambiguity and complexity, suggesting that the less complex among competing analyses wins out over the more complex one in the course of acquisition. Roberts (1993), for example, claims that reanalysis of French as a non-V2 language is due to the fact that verb movement to T° is more economical than movement to C° and is, therefore, preferred, for children adopt a “Least Effort Strategy” (LES). LES is considered a principle of acquisition, but it depends on a notion of economy defined in strictly grammatical terms (Economy of Derivation). According to a more recent version (Roberts 2007: 233), learners prefer more economical representations, defining economy in terms of the number of formal features of functional heads (Feature Economy, FE).

Nevertheless, from an acquisition perspective, this is not a satisfactory solution either, for difficulty of learning must be explained in terms of principles operative in learning processes. Whether featural or derivational economy translates into learning or processing complexity is, of course, an empirical question. It is not an implausible hypothesis, but one that still lacks psycholinguistic support. Similar qualms arise concerning the LES, for it is not obvious that this is indeed an acquisition strategy applied in other domains, rather than an ad hoc solution for this particular problem. Again, the necessary psycholinguistic evidence is lacking. Moreover, empirical and theoretical considerations raise doubts about this approach. Firstly, it is a mystery what could ever entice language learners to opt for a less economical option if a more economical one is available. Yet children acquiring Germanic V2 languages succeed in doing so without apparent effort, producing XVS sequences as soon as they use multi-word utterances. This is even more remarkable if we consider that speech directed to very young children seems to consist primarily of main clauses, exposing learners of V2 languages mostly to SVO strings. Secondly, as pointed out by Hale (1998), in order to assign the simplest parse to an input string, learners need to posit a numeration, determine that this numeration can converge at LF, and posit the appropriate features on the functional heads, allowing convergence at PF. This, however, is not a licit procedure since “Economy of Derivation is relevant only to the evaluation of derivations involving the same numeration. It cannot, therefore, be invoked to choose between these two competing hypotheses since they involve different numerations.” (Hale 1998: 14). In fact, such a procedure is not only excluded on theoretical grounds (Lightfoot 2017: 517, among others), known language processing mechanisms may not enable children to compare and evaluate derivations either; cf. Jacobson (1998).

To avoid misunderstandings, my conclusion is not that syntactic core properties in general or the V2 constraint in particular never change. If this were the case, we should not find variation across languages in this respect, as we actually do. The claim is that change of syntactic core properties happens rarely and not in intergenerational L1 transmission triggered by structural ambiguity under the above-mentioned conditions of decreasing frequency of telltale constructions and/or economy of the reanalyzed option. One question that remains to be addressed is whether L1 learners develop grammatical knowledge different from that of their parents’ generation if at least part of the PLD contain structures that might trigger alternative analyses, as in dialect or language contact settings. Language contact has indeed been claimed to be a likely condition for this to happen, and I will address this issue in Section 3.

3 Searching for structural reanalysis in bilingual acquisition

The principal idea pursued here is that the plausibility of any scenario of grammatical change must be assessed based on what is known about principles and mechanisms of language acquisition. Considering how widely held the view is that grammatical change happens during L1 acquisition, it is surprising that arguments in its favor are based almost exclusively on grammatical considerations or presumed acquisition mechanisms. In fact, the question whether empirical studies of child language support the idea of acquisition failure in cases of structural ambiguity, low input frequency, etc., does not meet with much interest in historical linguistics. But the cognitive mechanisms operative in acquisition are the same today as thousands of years ago. Consequently, what happens in acquisition today might have happened in past centuries, and what is impossible today was impossible then. I will, therefore, examine some results from acquisition research that shed light on the role of language learners as agents of grammatical change, focusing on findings obtained by studies of various types of bilingual acquisition.

4 Successive bilingualism: Second language speakers as agents of change

The question raised above was whether structural reanalysis happens if the prerequisites for the development of native competences are not met, early age of onset of acquisition and sufficient exposure to PLD. The available evidence suggests that drastically reduced quantity of exposure to the PLD results in high error numbers, but not in changes of early acquired morphosyntactic knowledge. Turning to the other factor, early onset of acquisition, we are looking at a different picture. Summarizing succinctly the vast amount of research on this topic, one can say that second language (L2) acquisition differs substantively from L1 development. What is controversial is whether these differences indicate that L2 learners have only partial access to Universal Grammar (UG) and must, therefore, resort to inductive learning in cases where L1 learners are guided by UG. I will not revisit this lengthy and inconclusive debate. It should suffice to mention the empirically well-documented fact that morphosyntactic phenomena that emerge early in L1 development represent particularly difficult L2 learning tasks; see Meisel (2011b: Ch. 4.4).

Children acquiring L1 German, for example, initially use both OV and VO. Yet already during this phase of variable word order, typically before age 2;0, they clearly prefer OV. Shortly afterward, still at around age 2;0, when most of their utterances consist of only two words (MLU 2.0), they distinguish between finite and nonfinite verbs, placing finite ones before objects or adverbs and moving them into V2 position where required; cf. Clahsen (1982). Importantly, although they occasionally fail to move finite elements, they do not raise nonfinite ones. Finally, they consistently place verbs in clause-final position as soon as they use subordinates (2;6–3;0).

L2 learners, on the other hand, treat German as an SVO language. Note that this adherence to VO is not particular to learners whose L1 is a VO language; it is also attested in learners who speak an OV language as L1; cf. Meisel (2011b: 105–107). Since superficial SVO order exists in German main clauses containing a single verbal element, the VO grammar generates target-conforming simple main clauses. Yet when a constituent other than the subject appears in clause-initial position, the result is ungrammatical (*XSVO), as are constructions where auxiliaries or modals are combined with nonfinite verbs (*SvVO). L2 learners then proceed through an acquisition sequence that is identical across individuals yet different from the L1 sequence, cf. Meisel et al. (1981). First, they place nonfinite verbs in final position (SvOV); successful learners take another step and invert subject and verb, mimicking V2 (XvSOV). Interestingly, they do not carry over the knowledge about OV order to subordinate clauses but use almost exclusively SVO order again. In naturalistic L2 acquisition, only the most successful learners reach the SV-inversion stage, and they continue to also use *V3. Equally importantly, emergence of verb movement does not correlate with acquisition of finiteness markers on verbs. Rather, even successful learners place nonfinite verb forms in finite positions. In other words, the tight connection between finiteness and verb raising in developing native grammars is not found in the linguistic knowledge of L2 learners of German.

In sum, at least some grammatical phenomena that are acquired early and without apparent effort by monolingual and bilingual L1 children count among the most vulnerable ones in L2 acquisition. Non-native grammars can, therefore, differ temporarily or permanently from native ones, even in core properties, and L2 learners are indeed possible agents of changes affecting the syntactic core, provided they play a significant role in language transmission. McWhorter (2007) refers to such cases as “language interrupted” and claims that L2 acquisition is the source of reductions of structural complexity not witnessed when transmission is “uninterrupted” by non-native acquisition. Although he does not rely on results from empirical L2 research, he is clearly in agreement with the hypothesis first suggested by Weerman (1993), according to which L2 speakers are the source of structural reanalysis; see also Kroch and Taylor (1997) and Meisel (2011a). Note that under this scenario, change is not a result of transmission failure; it is triggered by speakers who are themselves learners with an imperfect command of the target language. Thus, the triggering data are not structurally ambiguous; they rather contain divergent evidence, structures that are not generated by the target grammar; see also Lightfoot (1997).

Having identified L2 speakers as possible agents of diachronic change, the question arises as of which age of onset of acquisition successive bilinguals acquire an L2 kind of knowledge. Recent research on childhood bilingualism shows that AO affects the course of acquisition and ultimate attainment of another language at an earlier age than previously assumed, namely no later than between 3;0 and 4;0; see Meisel (2009). Only some aspects of phonology and morphosyntax are subject to maturational effects and not all of them simultaneously. Consequently, the knowledge attained by early successive bilinguals is largely identical with that of L1 grammars, but it also shares crucial properties with adult L2. Later, AO increasingly leads to similarities with L2.

Determining a turning point as of which successive bilingualism can be classified as child L2 acquisition (cL2) depends on which grammatical features are taken into account, and it can only be approximative. However, Sopata and Długosz (2022), analyzing elicited production data by Polish-German children, conclude that the turning point for this group can be fixed at 3;0, confirming the age range of 3;0–4,0. The grammatical phenomena affected by age-related changes include case and gender marking and inflectional morphology as well as verb placement in main and subordinate clauses. The fact that as of approximately AO 3;6 successive bilinguals resemble adult L2 learners in their acquisition of finiteness and inflectional morphology generally and was documented in previous publications; see Meisel (2009). The same age range was established for verb placement. Sopata (2011), for example, found that Polish–German children (AO 3;8–4;7) resemble L2 learners in their use of German OV and V2 order. We can, therefore, conclude that successive bilinguals exposed to the other language at age three or later qualify as potential agents of change, not only adult L2 learners.

5 Toward a plausible scenario of diachronic change

Since early versus late onset of acquisition is a criterion that differentiates vulnerable properties from less vulnerable ones, it is clear that a plausible scenario of diachronic change must take developmental schedules into account. The claim is not that early acquired features are typically lost late, but that developmental sequences reflect different acquisition timelines of various structural properties. Inductive learning, involving identification of language-particular forms and extraction of structural information contained in the PLD, requires extensive exposure to preferably salient and unambiguous input data. It is, therefore, a slower process prone to individual variation, resulting in protracted developmental patterns. Yet language learning children can also rely on knowledge available to them before any linguistic experience. The activation of this genetically transmitted knowledge requires minimal exposure to the PLD; it, therefore, happens fast and early; cf. (Carroll 1989).

Let me illustrate how the latter kind of “learning” interacts with grammatical properties of the phenomena to be acquired in shaping the emergence and the resilience of morphosyntactic properties. German L1 development can serve as an example again; cf. Clahsen (1982) and Meisel (1994). Multi-word utterances emerge at around age 1;10 (MLU 1.75), initially combining two elements, nominal, adverbial or verbal ones. This statement already credits children with a considerable amount of grammatical knowledge, most importantly that lexical items extracted from the PLD are syntactically categorized and organized. This seems to happen instantaneously and error-free: Verbal elements never carry nominal inflection or vice versa, indicating that they are indeed analyzed as syntactic categories and that category attribution is not achieved by trial and error. This was to be expected under the assumption that syntactic categorization, structure dependency, finiteness, etc. are part of the genetically transmitted knowledge.

In terms of Parameter Theory (PT), the claim is that parameters and their values are available from the start, but choosing target-conforming values requires structural information that learners must detect in the PLD. The directionality parameter is set as soon as children begin to produce multi-word utterances, still before age 2;0. Finite verb movement emerges immediately afterward, at around age 2;0 (MLU 2.0). As soon as three-word or longer utterances are produced, the V2 effect is recognizable, i.e., verbs are raised to the head of CP in V2 languages. Importantly, only finite verbs are moved, indicating that [±finiteness] is part of early grammars. Crucially, erroneous V2 placement has never been reported to occur in the speech of children acquiring a non-V2 language, not even in bilinguals acquiring a V2 and a non-V2 language simultaneously; cf. Meisel (2011b).

Note that the setting of the directionality parameter and the linguistic behavior of language acquirers informs us about the structural knowledge that they bring to the acquisition task. As mentioned in Section 4, German children initially use OV as well as VO, but OV is the preferred option from early on. This is an indication that VO is not more basic than OV, as argued by Kayne (1994), nor vice versa. If this was the case, we would expect to find, contrary to fact, that, universally, the basic order is exclusively used at the initial stage.

What we do find rather supports the explanation proposed by Haider (2013). According to his basic branching constraint (BBC), the structural build-up of phrases is universally right-branching. This appears to favor basic OV order. However, “the branching restriction is independent of the linearization restriction of head and complement, that is, head-final or head-initial order. The linearization follows from the directionality parameter for identification by a head” (Haider 2013: 4). The choice between the values of this parameter is of interest as a potential source of variation and change. Haider (2013) characterizes it as a case of competition between the structural simplicity of the configuration generated by the BBC and the processing complexity of the linearization imposed by the parameter. OV is structurally less complex because it is a direct instantiation of the BBC and because canonical directionality is in harmony with the branching direction of head-final phrases. VO is structurally more complex because it induces a shell structure and implies a mismatch between canonical directionality and right-branching. Yet VO has a processing advantage, for it presents the head early, thus facilitating parsing. Haider (2013: 9) concludes that OV and VO are “equally costly or equally cheap outcomes of cognitive selection for parser friendliness, otherwise one of the two types would have vastly outnumbered the other in the course of grammar change over the past millennia.”

This equilibrium also seems to exist in L1 acquisition where one might have expected parsing to work in favor of VO since parsing simplicity is crucial for learnability, and SVO is apparently easier to parse than SOV. Such a bias actually exists in L2 acquisition where learners commonly misanalyse OV languages as VO, but never VO languages as OV. Interestingly, diachrony also tends to work in one direction, i.e., change from VO to OV is rare. Haider (2013: 121) observes that “we lack any evidence that an Indo-European language has ever changed from a strict SVO language into a non-SVO language,” confirming a similar observation by Kiparsky (1996). If correct, this corroborates the claim that (2)L1 children are not the agents of change in directionality, for they set the parameter from early on to the target value, even bilinguals acquiring an OV and a VO language, as in Basque–Spanish or German–French bilingualism.

Considerations like the ones alluded to suggest strongly that a plausible scenario of grammatical change must take developmental schedules of structural properties into account. The ones that require minimal experience with the PLD are implemented early in developing grammars. Examples mentioned include head directionality, verb movement, finiteness, and the target-conforming movement, all implemented in emerging grammars shortly before or after age 2;0. This was to be expected, assuming that syntactic core phenomena are instantiations of macroparameters that are set before microparameters; see Tsimpli (2014). More importantly, in the present context, acquisition as well as attrition studies have demonstrated that early acquisitions are firmly entrenched in mental grammars; see Section 3.2. In fact, variable use has only been observed during initial phases, as mentioned above, concerning the use OV/VO. Interestingly, initial variability in verb raising results from an occasional failure to move finite elements, whereas nonfinite verbs are never raised erroneously; see Meisel and Müller (1992).

In sum, the resilience of these phenomena is a well-established and empirically confirmed fact. However, if we want to understand why this is so, alluding to their nature as narrowly syntactic phenomena will not suffice. A plausible scenario of diachronic change must explain how grammatical properties interact with acquisition principles and processing mechanisms, resulting in the resilience of some phenomena and in the vulnerability of others.

6 Conclusions

The goal of this paper has been to review and assess the relevance of empirical studies of different acquisition types for a theory of grammatical change. Based on the observation that certain fundamental structural properties are resistant to diachronic change and do not change across the lifespan of adult native speakers, an increasingly popular hypothesis in historical linguistics regards language acquisition as the source of alterations in these grammatical areas, identifying L1 learners as agents of diachronic change. Arguments in support of this idea consist almost exclusively of plausibility assumptions derived from principles offered by the adopted theoretical framework. This raises the question whether empirical findings of acquisition research can corroborate this hypothesis and ensuing claims.

The conclusion to be drawn from the review of research on first language acquisition is that the quest for change of syntactic core properties in L1 development has not been successful. As pointed out by Roberts (2017), change must be well-motivated. Yet neither structural ambiguity nor opacity constitute the required triggering factor, not even in combination with increasing frequency of ambiguous constructions or with decreasing frequency of unambiguous ones. Nor does low structural complexity of an alternative analysis suffice as motivation, not even if it could be shown to imply reduced processing complexity. I am, therefore, led to the conclusion that for reanalysis to happen in the syntactic core, the triggering information must be contained in the PLD, as suggested by cue-based approaches (Lightfoot 2006, 2017). This is why bilingualism has been claimed to be a possible setting for syntactic change to happen, assuming cross-linguistic interaction within individuals when the alternative trigger is offered by the other language. Yet this effect is not found in simultaneous or early successive bilinguals who normally do not differ from L1 children in course and ultimate attainment of acquisition. Language contact is thus not a sufficient cause either.

Weaker languages, on the other hand, do exhibit grammatical properties that deviate from L1 norms. Here, however, alterations are not triggered by the other language. Rather, they result from insufficient exposure to the target PLD, as is also argued to be the case with heritage languages. They differ from the baseline in morphological and syntactic properties, and such differences are possible sources of diachronic change; cf. Polinsky (2018) or Montrul (2023). Yet although HL speakers are potential agents of morphosyntactic change – these alterations do not involve reanalysis of syntactic core properties.

In sum, the quest for evidence of change of fundamental properties of grammar has led to the conclusion that age of onset of acquisition is a crucial factor and that only second language acquisition exhibits such effects. Non-native grammars can indeed differ from native ones in core properties, and L2 learners are, therefore, possible agents of such changes, provided they play a significant role in language transmission. This need not imply that they are a socially dominant group within a speech community. The perhaps most likely setting for this to happen is when child or adult L2 speakers provide a significant part of L1 children’s input. Thus, L2 speakers are not only themselves potential agents of diachronic change, they can also cause (2)L1 children to become agents of change affecting core properties.