1. Bilingual acquisition and grammatical change

Language development happens in many different contexts and takes on a considerable number of different forms, and the language sciences have branched out into a corresponding number of disciplines, each dedicated to one type of development, e.g. monolingual and bilingual first language acquisition, second language acquisition, impaired language acquisition (SLI), the genesis of pidgins and creoles, and diachronic change. In fact, it may not be unreasonable to also include instances of regression rather than expansion, e.g. language attrition or loss. Each of these phenomena is commonly studied by different subfields of linguistics. This diversification is justified by the fact that in each of these cases different factors are operative in determining the course of development. But there can be no doubt that important commonalities also exist, shared in part or totally by the various types of development. Consequently, the division of labour among research disciplines may represent an obstacle to achieving an adequate understanding of the driving forces and the mechanisms determining linguistic development. In fact, it is very likely that deeper insights into the mechanisms of development can only be attained by taking into consideration not merely the particularities of one type but also the commonalities across several or all of them.

In this paper, I intend to discuss issues which are potentially of common interest for studies of grammatical development in language acquisition, especially in bilingual settings, on the one hand, and for diachronic language change, on the other. In both cases, an adequate treatment of the phenomena under investigation depends on whether we can identify structural areas particularly susceptible to change, the actual processes of reanalysis, and, crucially but most difficult, language internal as well as external factors causing such modifications of the system. If this happens at a remote point in time, a direct observation and analysis of such processes is obviously impossible. But since it can be argued that bilingualism is possibly a necessary though not sufficient cause and certainly a necessary consequence of grammatical change, as will become apparent in what follows, my claim is that insights gained by the study of contemporary bilingualism – whether acquired simultaneously or successively – can contribute in essential ways to an explanation of diachronic change. In other words, by examining the present, we may be able to explain what happened in the past (see Labov, Reference Labov and Heilmann1975), and, reversing the perspective, chronologically distant developments can put problems on current research agendas which might otherwise be neglected. In the one case we have relatively easy access to the results of developments but not to the processes which have brought them about, and in the other, we are able to study ongoing processes but we do not necessarily know their ultimate results.

The issue I want to focus on is the problem of the learnability of morphosyntactic systems. The very notion of grammatical development implies that speakers of a language, at a given point in the course of development, rely on a system which differs in at least one aspect from those at earlier or later points. Consequently, different generations of speakers of this language must acquire systems which differ from the ones used by the previous generations, and each state of the developing system is subject to learnability conditions. Formulated in this way, this statement may appear trivial, but its interest should immediately become apparent if one looks more closely at the conditions for the implied process of change to become possible. Adopting the theoretical perspective suggested by the theory of Universal Grammar (UG), an obvious requirement is that each of the grammatical systems in question conforms to the principles of UG which, by definition, spell out the properties of the human language faculty. Structural considerations alone, however, will not suffice. Rather, it is necessary to demonstrate that the structural changes in question are actually implemented in the grammatical knowledge system of the speakers of the language concerned. In other words, it needs to be shown that the mental representations of such knowledge system can be reorganized accordingly.

Note that by adopting this perspective focusing on the reorganization of mental grammars, this study is designed to investigate change at the level of grammatical knowledge or of I-language, an internal entity of the individual, as opposed to E-language, “E” suggesting “external”, i.e. the overt products in language use; see Chomsky (Reference Chomsky1986). Changes reflecting different types of use, e.g. an increase or decrease in the frequency of occurrence of specific forms or constructions are thus not the object of our discussion. Nor will cases be dealt with where surface devices encoding a specific grammatical notion change but the underlying notion remains unaffected, e.g. if overt case markers are dropped or replaced by other kinds of devices. Some instances of grammaticalization also correspond to this second case. If, for example, referential lexical items turn into markers encoding an already available grammatical function, this does not involve a change of the type to be discussed here. To avoid possible misunderstandings, these are not only legitimate but indeed crucial issues for diachronic studies of language – yet they are not directly relevant to the question pursued in this paper, which is concerned with the emergence of new grammars.

We can speak of a “new” grammar if it differs from the previous one in at least one of the structural options which it generates. The type of variation across grammars is constrained by the nature of the human language faculty, and any adequate theory of grammar needs to specify the properties shared by all human grammars as well as those in which grammars of various languages differ. The latter are the ones which concern us here since any feature occurring in human grammars can, in principle, emerge in the course of development of a specific language, and children may be expected to explore all the options offered by the Language Making Capacity (LMC). These options are the ones which also define the range of variation across grammars. The theory of UG attempts to explain this variation in terms of grammatical parameters.

Universal Grammar comprises invariant principles designed to capture formal properties common to all grammars, as well as parameterized principles. The idea is that some principles of UG do not account exhaustively for the grammatical properties to which they refer; instead, parameterized principles offer several options, i.e. parameters are left unspecified by UG and must be set to one of the possible values in each individual grammar. It is important to note that the principles as well as the parameterized options are given by UG. This can be illustrated by the probably most frequently discussed example, the Null-Subject Parameter. A principle of UG states that a sentence necessarily contains a structural subject. Languages differ, however, in whether they require that this structural position be lexically filled or not. This option, namely the possibility to grammatically license a lexically empty subject position, distinguishes null-subject from non-null-subject languages and can be thought of as the result of a syntactic parameterization.

An important feature of the notion of parameter as suggested by Chomsky (Reference Chomsky1981, p. 6) is that parameters and their possible values are defined at an abstract level of grammatical structure rather than in terms of surface properties of the target language. As a consequence, setting a parameter to a specific value typically causes a cluster of superficially unrelated grammatical properties to appear in the language. Subsequent developments of Parameter Theory (PT) have led to doubts about the theoretical well-foundedness of classical parameters like the Null-Subject Parameter, more recently referred to as “macroparameters”, favouring instead more narrowly and locally defined ones. The idea of “microparameters” goes back to the suggestion by Borer (Reference Borer1984) and Chomsky (Reference Chomsky1989), who proposed that parameters should be defined exclusively in terms of properties of functional elements. More specifically, it has been argued that parameters refer to uninterpretable features of functional heads, like T (Tense) or C (Complementizer). The placement of finite verbs can serve as an illustration of what this means. It has been suggested that the feature [+F] (finiteness) may be located in either T, as in Romance languages, or in C, as in German. A principle of UG requires that finite verbs be moved to a functional head above VP; the parametric choice states that [+F] may be instantiated in either of these functional elements. A comprehensive and widely accepted version of Parameter Theory is currently not available, but it seems that both micro- and macroparameters need to be made available by grammatical theory; see the discussion in Biberauer (Reference Biberauer2008). Importantly, both types of parameters can account for the clustering effect, i.e. the fact that a number of surface phenomena depend on the setting of a single parameter. What matters for the present purpose is that grammatical variation – synchronically across languages as well as diachronically in language acquisition and historical change – can be described and hopefully explained in terms of parametric variation.

Returning now to the issue of emerging grammars in language development, ontogenetically or phylogenetically, a crucial observation which can serve as the starting point of our discussion is that parameter settings do not change across the lifespan. It has been shown repeatedly that parameters once fixed on a specific value cannot be reset in the course of first language development; see, for example Clahsen (Reference Clahsen1991) and Meisel (Reference Meisel, Fletcher and MacWhinney1995) for a summary of this debate. More importantly, in the present context, there is no evidence that syntactic parameters set to the target value of a native language can be changed in the course of adult life. Patterns of language use, on the other hand, may change, e.g. in response to alterations of the speech norms of the community. As Sankoff (Reference Sankoff, Ammon, Dittmar, Mattheier and Trudgill2005) and Sankoff and Blondeau (Reference Sankoff and Blondeau2007) have demonstrated, individuals may, in fact, adapt their language use during adulthood to innovative patterns resulting from generational change. Such lifespan changes may have profound consequences, but they do not involve reanalyses of grammars, i.e. we do not find evidence suggesting that mental representations of parameterized grammatical knowledge are subject to modifications after childhood. Even attrition of syntactic knowledge only seems to affect a person's ability to use the knowledge developed early on in life; see Flores (Reference Flores2008, Reference Flores2010).

The conclusion which I draw from these observations is that grammatical change involving reanalyses of syntactic patterns and leading to the reorganization of mental grammars, is most likely to occur in the process of language transmission. This is in tune with the claim commonly made in generative studies of diachronic change, according to which first language (L1) learners are the locus of change. The well-attested robustness of the LMC, however, suggests that in order for this to take place, the ambient language(s) and thus the primary linguistic data accessible to the learners must contain information triggering the required analyses. Bringing together two claims defended in the literature, that children acquiring their L1 are the locus of change and that language contact is a crucial if not a necessary factor triggering change, I suggest turning to bilingual acquisition as a plausible setting favouring morphosyntactic change. Yet contrary to what has been suggested in the literature on historical linguistics, I will argue that the simultaneous acquisition of two or more languages, i.e. bilingual first language acquisition, typically leads to a kind of grammatical competence which does not differ qualitatively from that of the respective monolingual L1 learners. My claim is that cross-generation reanalysis of grammars is most likely to happen in successive acquisition of bilingualism or if second language speakers provide a substantial amount of input in monolingual or bilingual first language development.

The structure of this paper is as follows: First, I outline some hypotheses about diachronic change characterizing current work in historical linguistics, leading to the claim that the language learning child can be seen as the principal agent of grammatical reanalysis. Section 3 argues that identifying language acquisition as the locus of grammatical change does not necessarily lead to a more adequate understanding of the emergence of new grammars. Only if triggers of reanalyses can be discovered in the primary linguistic data, will we get closer to this goal. A brief summary of three frequently mentioned causes of grammatical reanalysis, changing frequencies in language use, exposure to primary data containing ambiguous or contradictory evidence, and language contact, concludes that neither of these can plausibly be argued to suffice as an explanation of reanalysis. Section 4 therefore explores the claim according to which diachronic reorganization of grammars results from acquisition failure. Three scenarios are examined where this is likely to happen. A review of studies dealing with cases where one of the languages of bilinguals appears to be “weaker” than the other, leads to the conclusion that they do not exhibit imperfect transmission nor do they show cross-linguistic influence between grammatical competences. If, on the other hand, the onset of acquisition is delayed, or if children are primarily or exclusively exposed to sustained input from second language (L2) learners, incomplete acquisition may indeed be the result. Finally, in cases of reduced access to the primary linguistic data, the available evidence is inconclusive. In consideration of these findings, I hypothesize in Section 5 that grammatical change of the sort discussed here is most likely to happen either if substantial parts of the speech community are themselves L2 learners, or if the data available contain evidence not compatible with any of the grammars being acquired, e.g. in case of input from L2 learners of the target language(s). This entails that both child and adult learners can be agents of morphosyntactic change. The important role attributed to L2 learners, however, leads to the prediction that parametric changes occur less frequently than is normally assumed. This prediction is corroborated by the fact that a number of alleged instances of diachronic changes of settings of syntactic parameters can be explained differently. The main claims and findings are briefly summarized in Section 6.

2. The developmental problem in diachronic change

The developmental problem, as it is understood today, refers to the notion of a grammatical system constituting the core of the linguistic competence of the speakers of a linguistic community. It is therefore only fairly recently that historical linguistics has begun to focus on the issues related to such concerns, when learnability became a primary concern of linguistics after the cognitive turn in the 1950s and 1960s. In this respect, current research possibly shares more common interests with studies carried out in the nineteenth and early twentieth century than with early structuralist work, especially since philological research, which was almost entirely dedicated to the study of historical change of languages, included approaches which, in modern terms, can be qualified as “psycholinguistic” ones.Footnote ¹

However, a hypothesis crucial for the current debate was put forth already in the nineteenth century by Paul (Reference Paul1880) who suggested that processes operating in first language acquisition play an essential role in an explanation of language change.

More recently, Andersen (Reference Andersen1973) proposed an influential model of diachronic change in which a similar claim is made. Since children do not have direct access to the mental grammars underlying the language use of their parents’ generation, they need to reconstruct this grammar, based on the output produced by the adults and guided by their innate language making capacity. This process results in grammatical change if the language learning children interpret the data differently, as compared to the adults’ grammar and develop a grammatical system which can be understood as involving a reanalysis of parts of the target grammar.

Importantly, reanalysis need not lead to differences in use. In other words, although this scenario implies the existence of two different grammars across generations, the language use based on these grammars need not differ, i.e. the output may be superficially identical in both cases, in at least some contexts. This can be illustrated by the use of SVO order in main clauses of verb-second (V2) languages, like German, and in non-V2 languages, like most modern Romance languages; see Kaiser (Reference Kaiser1998, Reference Kaiser2000, Reference Kaiser2002). In V2 languages, only a single major constituent can precede the finite verb. Traditionally, this fact is captured by stating that both the initial constituent and the finite verb appear in the pre-clausal field (“Vorfeld” in the terminology of the German Structuralist tradition). In more recent grammatical theorizing, this corresponds to the Complementizer Phrase (CP), whereas the clause proper is constituted by the Tense Phrase (TP). Here, the V2 effect is explained as resulting from movement of the finite verb to C the head of CP, whereas in non-V2 languages it is moved to the head of TP, i.e. it remains within the clause proper. The potentially problematic case for language acquisition emerges when the target-conforming structural position of the verb (in the pre-clausal field, i.e. in CP, rather than within the clause, i.e. in TP) is not unambiguously signalled by surface order. This situation arises when the initial position is occupied by the subject. In this case, the resulting surface order is SV, in V2 as well as in non-V2 languages, because in the latter both the subject and the verb appear in the pre-clausal field. Thus, in spite of substantially different structural properties, the surface order of subject, verb, and object may be identical in both cases. Only if a constituent other than the subject is initialized, e.g. an object, an adverb or an embedded clause, will the underlying differences become apparent in surface word order: in V2 languages the subject then follows the finite verb in second position (XVS) and it precedes the verb in non-V2 languages, resulting in a sequence with the verb in third position (XSV). However, even an XSV sequence may be ambiguous, as argued by Platzack (Reference Platzack, Battye and Roberts1995), because, subsequent to cliticization of pre-verbal subjects, these are claimed to have been incorporated into the finite verb. However, children exposed to utterances of this type may fail to analyze the subject clitic (SCL) as incorporated into the finite verb and interpret it, instead, as an independent constituent; see Kaiser (Reference Kaiser2002, p. 94). Thus, children may have analyzed a Middle French sentence as in (1) by assigning a clause-internal position to the verb, as opposed to adults in whose grammar the verb is raised to the pre-clausal field (C), assuming Old or Middle French were indeed V2 languages; see (2).

1. (1)

1. (2) Adult grammar

   [_CP (XP) [_C′[_COMP SCL+V_i]] [_TP t_i]]

   Child grammar

   [(XP) [_TP SCL V]]

Leaving further details aside, we may retain three points from the discussion in the literature on language change briefly alluded to here. The first consists of the hypothesis, broadly accepted at least among researchers following the generative approach, asserting that the language learning child is the locus of change. The second point refers to possible causes of grammatical changes. In the generative literature, a frequently entertained hypothesis is that structural ambiguity is a major reason for this to happen. The third point concerns the actual process by which the transition from one system to the other may occur. In this case, a commonly held view is that children analyze the primary data encountered in their linguistic environment differently from the grammars actually underlying the adult language use. All this demonstrates clearly that the task of explaining the developmental problem in diachronic language change has been transferred to language acquisition studies since grammatical reanalysis must be understood as a process happening in the course of acquisition. From this, in turn, it follows that the plausibility of the scenario suggested as an explanation of grammatical change needs to be tested against what is known about the mechanisms of language development.

Note that, from what has been said so far, it might appear as if the adequate point of reference for diachronic change was monolingual first language acquisition. In reality, however, the contexts which might possibly favour structural change are bilingual situations. At any rate, the scenario outlined above implies that different grammars are used across generations, although this may not be immediately apparent in language use.Footnote ³ Assuming this to be an adequate account of the facts of diachronic change, one must conclude that it leads, at least temporarily, to cross-generation bilingualism which is likely to persist for extended periods of time. Moreover, one may reasonably question the assumption that all children facing this kind of learning context will be induced simultaneously to reanalysis, all focusing on the same structural properties of the target grammar, and all of them performing the same changes. More plausibly, one should also expect different grammars to emerge even within the same generation. These considerations necessarily lead to the conclusion that explaining language change by reanalysis in the course of language acquisition results in bilingualism which, in turn, is a prerequisite for diachronic change to occur, triggered by system-internal factors.

But even if it can be argued that grammatical change necessarily happens in bilingual settings, the question remains as to whether language contact or language internal variation will suffice as a cause for grammatical reanalysis. Remember that this seems to imply that children fail to develop the grammatical knowledge underlying the speech of their parents’ generation, speech providing the input to their acquisition process. In what follows, I will argue that, at least as far as the setting of parameters is concerned, “transmission failure” (see Section 4, below) is unlikely to happen in simultaneous first language acquisition. Only in successive acquisition of bilingualism might L2 learners fail to reconstruct the target grammar, based on the information provided by the primary linguistic data. Alternatively, monolingual or bilingual children may develop a grammar distinct from that of the previous generation if they are exposed to an L2 variety of the target language. In this case, however, change is not the result of transmission failure but due to exposure to different triggers. The latter scenario corresponds to the suggestion by Weerman (Reference Weerman1993), Kroch and Taylor (Reference Kroch, Taylor, van Kemenade and Vincent1997), Lightfoot (Reference Lightfoot1997), or Kaiser (Reference Kaiser1998, Reference Kaiser2000, Reference Kaiser2002) who argued that reanalysis only happens when children are exposed to data containing conflicting evidence. Since this situation arises only if L2 learners are involved, Weerman (Reference Weerman1993) and later also Kroch and Taylor (Reference Kroch, Taylor, van Kemenade and Vincent1997) have suggested that what causes grammatical change is the exposure of children to speakers who are themselves “second language learners with an imperfect command” of the target language. In other words, the crucial data are not merely structurally ambiguous, they contain structural information favouring analyses which cannot be generated by the adult native grammar.

3. (Bilingual) first language acquisition as a testing ground for theories of diachronic change

Let me now turn to the question of whether the scenario of language change outlined in the preceding section can indeed be corroborated by findings from acquisition research. Recall that the crucial issue under review here is the developmental problem, i.e. the explanation of the transition from one system to another. By shifting the problem from diachrony to ontogeny, one faces an at least apparent paradox which has been termed the “logical problem of language change” (Niyogi & Berwick, Reference Niyogi and Berwick1995). As Brandner and Ferraresi (Reference Brandner, Ferraresi, Brandner and Ferraresi1996, p. 14) put it:

Note that by addressing this conceptual difficulty in the context of language acquisition studies, we need to reverse the habitual perspective aiming at an explanation of what enables children to acquire successfully the adult grammatical system, and we must attempt, instead, to explain in a non-ad-hoc fashion the possibility of unsuccessful or, rather, partially successful acquisition. What is intended by the expression “partially successful” is that the children acquire grammatical knowledge which is not fully equivalent to the grammars underlying the production of the utterances which served as input for the learning process.

Such a change of explanatory goals is by no means a trivial matter. It is generally assumed in acquisition research that first language development happens relatively fast and uniformly across individuals, and that it is ultimately successful in non-pathological cases. These characteristics – rate, uniformity and success – are commonly attributed to the guidance by the innate language faculty. In the theoretical framework adopted here, Universal Grammar plays an essential role in this, since all human grammars, including developing ones, must conform to the principles of UG. In this way, one can account not only for the uniformity of development but also for rate and success, since a substantial part of grammatical knowledge need not be learned inductively, by trial and error, but is assumed to be genetically transmitted. From this it follows that the content of principles of UG is not learned in the usual sense. But, as mentioned in the preceding section, some of these principles are underspecified by UG, i.e. they offer more than one option, and the implication for acquisition is that the child has to set these parameters to the value required by the target grammar. In other words, although the options offered by the parameterized principles are given by UG, the choice results from the child's linguistic experience. Parameter setting may therefore be understood as experience-driven triggering of implicit knowledge available prior to experience. For this reason, it is a conceptually and empirically important domain of acquisition research: it represents the intersection of innate and inductively acquired knowledge. Assuming this perspective, language acquisition is characterized, first, by universally invariant properties defined by non-parameterized principles, and secondly by language particular aspects which need to be learned by experience. Parameter setting thus represents a third type of acquisitional process exhibiting both characteristics. What matters in the present context is the question of how parameter setting is triggered. As Roberts (Reference Roberts1993, pp. 158f.) has pointed out, reanalysis, as illustrated by (2), above, can be thought of as

Reanalysis need not necessarily imply changes of parameter values, but, conversely, it is a necessary condition for changes in parameter settings to happen, and it is this type of change which characterizes radical or “catastrophic” or “bumpy” (Lightfoot, Reference Lightfoot1997, Reference Lightfoot2006) diachronic change, for only analyzed structures, or reanalyzed ones in this context, can serve as triggers for parameter setting. Raw data do not function as input for learners, they are merely acoustic events (in case of oral productions); only once they have been assigned a structural interpretation by learners, will they serve as input for learning processes; see Carroll (Reference Carroll2001). This observation applies to all types of possible input, including triggers for the setting of parameter values, i.e. they need to be defined in terms of their structural properties, as is also argued by Fodor (Reference Fodor1998).

An important question, with respect to the issues at stake here, is whether parameters can be set to a wrong value in normal cases of monolingual or bilingual first language development. In the literature on monolingual acquisition, it has been argued repeatedly that the notion of parameter requires that for every possible setting of parameters, there must exist a specific structural configuration triggering it, see Meisel (Reference Meisel, Fletcher and MacWhinney1995) for a summary of the arguments supporting this argument. In fact, Fodor (Reference Fodor1998) has presented strong support for the idea of “unambiguous triggers”. This, in turn, has occasionally been understood as excluding the possibility of erroneous settings. Note, however, that this need not be correct. If ambiguous data of the sort illustrated by example (1) above do occur in the child's linguistic environment, it is, in principle, possible that the child learner arrives at only one of the possible analyses. If this happens to be one which differs from that required by the adult grammar, i.e. in case of reanalysis, and if it contains the structural configuration which functions as the trigger for a specific parameter setting, the result is an instance of an erroneous setting of a parameter triggered by an unambiguous trigger. In other words, primary data which are ambiguous in that they allow for more than one structural interpretation may, in principle, cause erroneous parametric choices based on grammatical input functioning as an unambiguous trigger. In view of the importance of this issue for an understanding of language development, it is surprising that only very few studies explicitly address the question of whether such situations actually arise in first language acquisition; but see Fritzenschaft, Gawlitzek-Maiwald and Tracy (Reference Fritzenschaft, Gawlitzek-Maiwald and Tracy1990), Müller (Reference Müller and Meisel1994), or Tracy (Reference Tracy1991, Reference Tracy1995).

Let us now return to the most frequently suggested solutions to the developmental problem in language change, keeping in mind these brief considerations regarding the setting of parameter values and the necessary triggering experience. The main question is what causes changes to happen in (bilingual) language acquisition which can be characterized as instances of reanalysis; see Lightfoot (Reference Lightfoot1991). The prime candidates nominated as possible causes for this type of change are (i) changes in the frequency of use of particular constructions, (ii) structural ambiguity of constructions, and (iii) exposure to conflicting evidence in situations of language or dialect contact.

Frequency, although a much debated issue in historical linguistics, is by itself a poor candidate as a factor causing developmental changes, if the hypothesis is that reanalysis happens in a context where the children's input is the output of a “homogeneous” grammar. This is not to say, of course, that a minimum number of occurrences might not be a necessary prerequisite. But estimates of minimum frequency based on theoretical considerations are not available, and it is even difficult to make an educated guess. Lightfoot (Reference Lightfoot1997), nevertheless, attempts to do just that. Reviewing studies on word order patterns, he concludes that a frequency of 17% is not sufficient to trigger the setting of a parameter to a specific value, whereas 30% of occurrences do seem to have the expected effect. He concludes that “somewhere between 17% and 30% is a phase transition” (Lightfoot, Reference Lightfoot1997, p. 179). In my view, this line of argument is highly problematic. Most importantly, I see no justification for the implicit claim that the kind of data available for diachronic studies, especially medieval texts, might reveal any reliable information about frequency of use in colloquial language, let alone child-directed speech, given that it is difficult enough to discover solid facts about spoken language. Even more seriously, language acquisition research is not even able to specify a quantitative threshold for parameter settings in contemporary analyses, and analyses of grammatical development have demonstrated again and again that low frequency items are acquired without problems. Decreasing frequency alone is therefore very unlikely to be a sufficient cause of grammatical restructuring. Most importantly, even if we were able to quantify the decrease in frequency of a given construction, it is by no means clear what the implications for language acquisition might be. Unless the primary linguistic data contain some information about alternative options, the child has no choice but to continue analyzing the infrequent one.Footnote ⁴ Finally, the validity of this proposal is further weakened by methodological problems; see Kaiser (Reference Kaiser1998, Reference Kaiser2002), who mentions, among other things, the narrow data base on which the estimate by Lightfoot (Reference Lightfoot1997) is calculated.

My claim, thus, is that a change in frequency of particular phenomena in the primary linguistic data does not, by itself, suffice as a cause for reanalysis. Rather, the grammatical nature of the data needs to have changed as well, i.e. the learners must be offered alternative options in the input. Put differently, it seems improbable that bilingualism could arise in monolingual settings, even if one takes into account the fact that monolingual settings also exhibit a certain amount of variability of the primary linguistic data and that there is no truly “homogeneous” input. This variability has led Roeper (Reference Roeper1999) to postulate the idea of “universal bilingualism”. Similarly, Tracy (Reference Tracy1998), who, referring to register variation in languages, concludes that this implies “syntactic diglossia within individual authors” – in other words, dialectal ‘multilingualism’; see Kroch (Reference Kroch and Baltin2001, p. 722) for a similar argument. Note, however, that the decisive factor is the presence of conflicting evidence in the child's linguistic environment, more specifically structural triggers for different parameter settings. In conclusion, then, frequency changes may only be expected to cause reanalysis if reinforced by one of the other two factors mentioned, structural ambiguity or presence of more than one grammatical system in the linguistic environment of the learner.

Consider then the second possible factor causing change, namely structural ambiguity. It should not be difficult to see that this hypothesis too is problematic in a number of ways. Most importantly, in order for it to be considered as a plausible solution to the developmental problem, it must predict in a non-ad-hoc fashion which of the possible analyses learners will prefer. It has been argued that this can be achieved if one assumes that, faced with structurally ambiguous constructions, children opt for the less complex or simpler solution; see Roberts (Reference Roberts1993), among others. Surprisingly, however, this solution attempts to explain the preference for one of the options in terms of principles of UG (Economy of Derivation) and defines the Least Effort Strategy (LES) in syntactic terms, referring to length of syntactic chains, although it is recognized that “simplicity”, in this context, relates to principles of acquisition, notably the LES. This is clearly not satisfactory, for one would expect learning principles to be defined in terms of learnability or processing considerations and to be empirically corroborated by psycholinguistic and acquisition research. Given that this kind of evidence is lacking, an explanation of the learner's implicit choice in case of structural ambiguity is a mere stipulation when it refers to the principle of economy. Perhaps even more importantly from the perspective of language acquisition, it is not obvious how this approach would explain how the less economical option could ever be chosen by the language learning child, as is clearly the case in the acquisition of V2 languages which exhibit the kind of ambiguity illustrated by example (1). If one was to postulate a subsequent reanalysis driven by further experience with the primary linguistic data, this would run into further serious acquisition problems since “unlearning” of acquired grammatical knowledge is a notoriously difficult problem for which no satisfactory solution has been proposed as yet. In other words, psycholinguistic considerations do not support the proposal of economy-driven change as a convincing solution to the problem of explaining morphosyntactic change.

Yet even if one puts such considerations aside, one finds that from a theoretical perspective, this proposal does not fare much better. In fact, Hale (Reference Hale1998) as well as Lightfoot (Reference Lightfoot2000), referring to the choice between overt versus covert verb movement, argue that reanalysis cannot be explained in terms of the LES. Hale (Reference Hale1998, p. 13) observes that in order for learners to be able to assign the simplest parse to an input string, they need to posit a numeration, determine that this numeration can converge at L(ogical)F(orm), and posit the appropriate features on the functional heads, allowing convergence at P(honetic)F(orm). He concludes: “As Chomsky (Reference Chomsky1995: 227) has pointed out, 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, Reference Hale1998, p. 14).

Some objections against this approach might perhaps be overcome by combining the ambiguity with the frequency argument, claiming that high frequency leads to reanalysis if the type of construction triggering a different setting of a parameter, as compared to the value required by the grammar of the previous generation, occurs significantly more frequently in the child's linguistic environment. Leaving aside the problem of explaining what might cause such shifts in frequency (see Kaiser, Reference Kaiser2002, pp. 114ff.), this suggestion is still flawed by a number of serious weaknesses. An obvious one is that the argument only holds if the most frequently used pattern is not ambiguous, for, otherwise, both parameter values are favoured by the increasing number of uses. Most importantly, the primary data incontestably still contain unambiguous evidence favouring the setting of the parameter to the value required by the grammar of the previous generation – evidence which the child would have to ignore. The latter assumption, however, is in conflict with findings in acquisition research which has shown that high frequency is not a necessary requirement for children to be able to set parameters; see, for example, Carroll (Reference Carroll1989), Fodor (Reference Fodor1998) and Meisel (Reference Meisel, Fletcher and MacWhinney1995). In fact, this is precisely what is generally seen as an essential difference between inductive learning as opposed to triggering, and Fodor (Reference Fodor1998) argues convincingly that, in case of structural ambiguity, the acquisition mechanism refrains from fixing the parameter on one of the possible values until unambiguous evidence is encountered. In sum, if certain constructions allow for more than one grammatical analysis, our current understanding of processes of first language acquisition suggests that this should not entice the child to opt for the apparently simpler or less complex solution but for the one favoured by unambiguous evidence, even if this happens to be used significantly less frequently.

The third factor suggested as an explanation of the developmental problem in language diachrony which I want to consider here is language or dialect contact, i.e. the presence of more than one grammatical system in the child's linguistic environment. The basic idea is that, in contrast with structural ambiguity, conflicting evidence may constitute sufficient cause of reanalysis, provided that triggers for opposing parameter values are contained in the input because learners encounter data from two or more different systems in the acquisitional setting. In some respects, this situation is not entirely different from the one just discussed, and some of the objections raised there also apply here. First of all, if more than one analysis of the input is possible, no matter whether this is due to ambiguity or to a situation of language contact, there is no a priori reason for the child to opt for a specific one of them. At best, one should predict that the children of the new generation will choose one or the other analysis. In this case, one still needs to explain why the one differing from that of the previous generation eventually wins. Recall that neither frequency nor simplicity provide a solution to this problem. More importantly, the presence of conflicting information concerning the value a parameter should be set to, is, quite obviously, a defining characteristic of bilingual first language acquisition. We can therefore refer to a considerable body of research carried out over the past decades. For reasons of space, I must refrain from summarizing even the most pertinent results; see Meisel (Reference Meisel, Cenoz and Genesee2001) for a critical overview. The punch line, with respect to the present discussion, is that conflicting input typically does not lead to the children incorporating fragments (e.g. specific parameter values) of one grammar into the other, as is assumed by Lightfoot (Reference Lightfoot1999, p. 12), who imagines children to “converge(d) on a single grammar . . ., in a kind of creolization”. All the available evidence suggests that fusion of grammatical systems is not what normally results from the simultaneous acquisition of two or more languages. Rather, it indicates that children are capable of differentiating the grammatical systems from early on, and conflicting evidence on parameter values has been argued to be the major cause of the differentiation of systems and for not treating this type of variability as a property of a single system; see Meisel (Reference Meisel1993).

In sum, this short review of three possible causes for grammatical reanalysis – changes in the frequency of use of particular constructions, structural ambiguity of constructions, or exposure to conflicting evidence in contact situations – suggests that none of them, taken alone, suffices as an explanation of grammatical change as a change of parameter settings. The evidence scrutinized here is, of course, not sufficient to definitely rule out that morphosyntactic change can be caused by one of these factors, or to deny the possibility of internally motivated change. Rather, the purpose of this brief look at possible causal factors has been to ask whether they can help to explain reanalysis in the course of transmission of grammatical knowledge, if we assess their possible roles in the light of what is actually known about mechanisms of grammatical development, rather than entertaining stereotypical notions of language acquisition. The answer is that none of them qualifies by itself as a plausible explanation of change when examined from the acquisition perspective. Dialect or language contact, for example, may indeed be a necessary cause of grammatical change, see Thomason and Kaufman (Reference Thomason and Kaufman1988) for a detailed discussion of this issue, but in view of the results of extensive research on first language development in bilingual settings, one must conclude that it is certainly not a sufficient condition for reanalysis to happen. In the following section, I will therefore examine simultaneous and successive bilingual acquisition in more detail, in search of additional factors which, in combination perhaps with the ones mentioned above, might account for grammatical reanalysis resulting in morphosyntactic change.

4. The quest for transmission failure

The present discussion proceeded from some reflections on the developmental problem as it presents itself with respect to studies attempting to explain language change. It has become apparent that the notion of “reanalysis” plays a crucial role in a possible solution to the problem and that children acquiring first languages are suspected to be the principal agents in this process of reorganizing grammars. The surprising result, in my judgement, of the ensuing brief review of possible causes of the changes in question is that apparent solutions become increasingly opaque rather than transparent as one examines them from an acquisitional perspective. The relative ease and the ultimate success of grammatical development in first language acquisition as well as the incontestable ease with which children acquiring two or more languages simultaneously succeed in differentiating grammatical systems, precisely these positive results turn out to be major obstacles for explanations of language change which is “by definition a failure in the transmission across time of linguistic features” (Kroch, Reference Kroch and Baltin2001, p. 699). Phrased a bit more crudely, we apparently have to rely on the possibility of systematic failure in acquisition in order to be able to explain grammatical reanalysis in the course of L1 development, i.e. profound and radical changes of the underlying system rather than merely quantitative shifts in usage. In fact, the insight that grammatical systems are considerably less prone to radical changes than might appear is not entirely new in historical linguistics. From a generative perspective, this amounts to saying that parametric changes are not as common as had previously been claimed; see Kaiser (Reference Kaiser1998, Reference Kaiser2002). Longobardi (Reference Longobardi2001), in his Inertial Theory, even asserts that syntactic change should, ideally, not happen at all and must therefore be caused by external factors. Since we know that, in the reality of the history of languages, such changes do arise, we must continue our quest for the causal factors. Consistent with the claim that bilingualism is a necessary condition for change, the previous discussion incites us to investigate the implications of this hypothesis. As mentioned at the end of Section 2, I will argue that one possible scenario implies that monolingual or bilingual children encounter the crucial structural information in L2 varieties of the target languages in their linguistic environment. This, however, is not an instance of “transmission failure”; rather, the triggering experience for reanalysis is provided by successive bilinguals. But before turning to this case, I will discuss in more detail the possibility of failing transmission of grammatical knowledge, scrutinizing simultaneous as well as successive bilingualism in search for cases where bilingual acquisition is not entirely successful, thus putting a fairly neglected issue on the research agenda, as can also be deduced from the remarks in the following passage:

Kroch summarizes clearly the questions which need to be answered, but he limits his attention to monolingual acquisition. In multilingual acquisition, however, it is indeed possible to study cases where one of the languages acquired is used less proficiently, and it is only reasonable to inquire whether it has also been acquired less successfully. Much of the research on early bilingualism over the past thirty years investigated whether it is possible to develop a competence and a proficiency in use in each of the languages qualitatively equivalent to that in the respective monolinguals. These studies have demonstrated beyond any reasonable doubt that this is indeed possible and that bilingual children achieve these goals without going through a phase of confusion, fusion, or whatever threatening scenarios had previously been developed; see Meisel (Reference Meisel, Cenoz and Genesee2001, Reference Meisel, Bhatia and Ritchie2004). From this, however, it does not follow that less successful cases cannot exist. But it is important to know that under the normal conditions of first language development, simultaneous acquisition of two or more languages can easily be attained. In fact, only based on this knowledge are we able to adequately formulate questions as to what exactly happens in cases where children do not succeed in the expected fashion. In order to do so, it is necessary to carefully describe the structural properties in which the language use of these individuals appears to differ from that of other bilinguals and from monolinguals. To the extent that such differences can be qualified as differences in grammatical competence, one can proceed by investigating the causes for what might possibly be a lack of ultimate success.

Multilingual acquisition, however, not only allows us to investigate possible cases of incomplete acquisition in one of the “first” languages of individuals developing more than one competence simultaneously, it also enables us to assess the role of what is referred to in the above quitation from Kroch as “the critical age period”. By analyzing successive acquisition of bilingualism, it should become possible to decide whether delayed onset of acquisition can result in transmission failure. In Section 4.2, below, I will argue that this is indeed the case, whereas it is unlikely to happen in simultaneous acquisition of bilingualism, as I hope to show in Section 4.1.

5. On reanalysis and diffusion of change by first and second language learners

In the first sections of this paper, I have argued that the developmental problem as it presents itself in studies of diachronic change remains largely unsolved. The further discussion led to the hypothesis that bilingual language acquisition might be a possible source of grammatical reanalysis in cases of “catastrophic change”, involving changes of the values of grammatical parameters. But this resulted in a paradoxical situation since research on the simultaneous acquisition of two or more languages has demonstrated that bilinguals, just like monolinguals, normally acquire successfully the languages they are exposed to. In the preceding section, we have then seen several scenarios where the language use of bilingual children deviated from the target norms, but which, nevertheless, did not provide plausible evidence of grammatical transmission failure.

In fact, the perhaps strongest result of the discussion is that the human Language Making Capacity is an extraordinarily robust device. Transmission failure is therefore much less probable and occurs less frequently than is commonly assumed in diachronic linguistics. The Language Making Capacity can, for example, cope with a significant reduction of input. Quite obviously, there must be a minimal threshold, below which the quantity of input will not suffice to trigger parameter setting. Yet exposure to the primary linguistic data during the optimal periods defined by the maturational schedule is significantly more important than the quantity of exposure. Transmission failure is most likely to happen in case of delayed acquisition. Importantly, these windows of opportunity probably begin to close as early as at approximately 3 or 4 years of age, and some changes may indeed happen earlier, especially in phonology. Consequently, child as well as adult second language learners can be the agents of change.

Thus, reconsidering grammatical change in the light of insights gained by language acquisition research yielded results which should incite us to revise at least some of the working hypotheses adopted at the outset of this discussion. Let me first mention one assumption which does not have to be modified, namely that morphosyntactic change of the type captured by the notion of parametric change will only happen in the course of language transmission. As has been argued in Section 1, no currently available research finding suggests that such changes might occur over the lifespan of speakers, affecting their L1 grammatical knowledge. This, however, does not necessarily imply that child L1 learners are indeed the principal agents of grammatical change. In view of the above-mentioned robustness of the LMC, transmission failure is unlikely to happen, especially in monolingual L1 development, although it cannot be ruled out for principled reasons. After all, as has been mentioned in Section 3, it is not impossible that parameters might be set to target-deviant values in L1 development, even if the L1 literature reports on less than a handful of cases where this is claimed to have occurred. Moreover, as discussed in Section 3, changing frequencies in the use of specific constructions or the occurrence of ambiguous structural information cannot be excluded categorically as possible causes for syntactic change. The latter case, however, would not qualify as an instance of transmission failure since the cues – in the sense of Lightfoot (Reference Lightfoot1991 and subsequently, e.g. 2006) – which trigger the change are in fact present in the primary linguistic data. These considerations amount to saying that we cannot preclude from the outset the possibility of internally motivated changes in grammar. But in view of the fact that studies investigating L1 acquisition offer virtually no support for the claim that parameters are set differently in developing grammars as opposed to the mature grammars of the caretakers, the assumption that child L1 learners are the principal agents of grammatical change is not warranted.

This is why I suggested examining bilingual acquisition as a possible cause of syntactic change, thus combining the idea of transmission failure with the widely accepted hypothesis of an influential role of language contact in diachronic change. Yet as summarized above, simultaneous acquisition of bilingualism is no more likely to lead to grammatical reanalysis than monolingual L1 development, not even in settings which are far from being optimal for language development. Thus, from the point of view of acquisition research, the only scenario according to which diachronic change as a result of transmission failure can plausibly be expected to occur is the one which I have described under the heading of “successive acquisition of languages”, in Section 4.2, i.e. when children acquire one of the ambient languages as a second language. This includes at least some of the speakers of heritage languages mentioned in Section 4.3. In these cases, child second language acquisition is the locus of change, but the same considerations apply, of course, to adult second language learners. Consequently, the answer which I propose to the question of who are the agents of change is: second language learners, children or adults, at least as far as change via transmission failure is concerned.Footnote ⁶

On the other hand, second language acquisition can also be the trigger of grammatical change when L2 learners provide the input for subsequent generations of first language learners. This refers, of course, to the proposal by Weerman (Reference Weerman1993), adopted by Kroch and Taylor (Reference Kroch, Taylor, van Kemenade and Vincent1997), mentioned in Section 2, according to which L2 input can cause the kind of changes we have been looking for. Under this scenario, children are supposed to be exposed to “wrong” evidence in the speech of second language learners. The claim thus is that the speech of second language learners which serves as input for first language learners contains the triggers necessary for reanalysis, possibly because L2 learners transfer parameter settings from the grammar of their first language into their L2 interlanguage. Note that this scenario presupposes that these L2 speakers represent the only or the strongly predominant source of primary linguistic data for the children acquiring the language and that they are engaged in interactions with the children of a type which qualifies them as the preferred linguistic role models for these learners. For if native data of the target language were simultaneously accessible, the children would still have access to unambiguous triggers, setting parameters to the same values as in the grammar of their parents’ generation. In sum, if L2 acquisition is the only locus of transmission failure, grammatical reanalysis can also happen in mono- or bilingual L1 development if the triggering structural information (cue for change) is contained in the samples of speech to which learners are exposed – a situation most likely to occur if L2 learners of the target language are the linguistic role models.

However, if second language acquisition seems to make a significant contribution to the solution of the transmission problem, it apparently also leads to new problems concerning the diffusion of the changes; see Labov (Reference Labov2007). The question is how L2 learners can possibly influence the entire speech community. In order for the scenarios sketched out here to gain in plausibility, language-external factors causing linguistic change need to be taken into account, and it must be shown that the social and demographic factors characterizing imaginable settings favourable for change can indeed be expected to prevail. These include a broad and sustained influence of the L2 speakers on the new generation, already during the transmission period, in the case of change triggered by L2 speakers. Similar conditions must be met if child L2 learners are the agents of change. Both scenarios require that the appropriate settings are sufficiently common and long lasting in order for the possible changes to be able to spread.

Studies investigating the history of specific languages will have to demonstrate that these conditions can indeed be met and prevail long enough in order to have an effect on a generation of learners. But I think one can, in fact, cite examples which suggest that the scenarios developed here are not as unusual or improbable as might appear at first sight. To mention a contemporary example, the majority of today's Basque speakers in the Basque Country are arguably L2 learners of the language, and this is probably also true for the larger part of the caretakers of Basque children. And although studies of ongoing grammatical changes in Basque are scarce, at least some research of this type exists, and we do not have to rely entirely on anecdotal evidence. Almgren and Barreña (Reference Almgren and Barreña2005), for example, report that bilingual as well as monolingual children increasingly use present tense forms in order to refer to future events, a use which is not acceptable in present-day adult Basque and which seems to be an innovation due to the influence by “new” speakers of Basque.

The Basque situation is certainly not unique; it may, in fact, be more common than one would suspect, as has been pointed out by one reviewer, since similar processes can be observed in various parts of the world as a result of efforts to revitalize endangered languages. In the case of Basque, political changes brought about the specific linguistic situation. But in small societies with only several hundred people per language and in which exogamy is institutionalized, so that every marriage brings a speaker of some other language into the community, such customs can result in a situation where L2 learners can indeed influence the linguistic community as a whole.Footnote ⁷ My hypothesis is that in larger and more complex societies, situations in which L2 learners exert a major influence on a language are most likely to emerge in periods of substantial demographic changes. Contemporary labour migration may appear to be such a case, but it is primarily the minority language which is affected here. Majority languages or more generally autochthonous languages are likely to be affected only in constellations of profound upheavals of these societies (and linguistic communities), e.g. as a result of the loss of large parts of the population. The migration of the peoples would certainly count as an example, and one can speculate that it had an impact on the formation of the various Romance languages. To mention another example, it is perhaps not a coincidence that we find that several European languages, e.g. French and English, underwent substantial changes during the 14th century and that grammatical innovations abound in texts written during the third quarter of the century. This is precisely the period when the Plague led to massive migrations, primarily within countries, but the “other language” may of course also be another dialect. The disease reached Sicily in 1347 and subsequently spread towards Western Europe. Its various attacks between 1350 and 1550 resulted in repeated decimation and subsequent growth of populations in many parts of Europe, as is observed by Johansson (Reference Johansson1997), who discusses the example of eastern Normandy (Johansson, Reference Johansson1997, pp. 133ff.), where the population level fell around the year 1385 to 45% as compared to 1314 and rapidly grew to 65% around 1410, although it did not exceed 80% of the 1314 level around 1550.Footnote ⁸ These dramatic demographic changes may well have led to situations like the ones described above. I will refrain from further speculations, hoping that the ones presented here will suffice to show that it is plausible that situations may arise where L2 speakers of a language will exert an important influence on this language.