The Multiple Grammars Theory and the nature of L2 grammars

Attempts to provide representational accounts of second language (L2) optionality are always welcome because they provide us with tools to determine whether and to what extent non-native grammars are idiosyncratic. In their keynote article, Amaral and Roeper (this issue; henceforth A&R) have also devised a specific way of accounting for how an L2 (or an L3 or an Ln grammar) is added to the initial first language (L1) representation. However, we are left with the need to determine more specifically what is actually added; whether the end result is a grammar with the same features (or parametric properties) as the input L2 grammar, as well as what the predictive power of the model may be. In this commentary, I will refer to previous L2 research dealing with optionality in order to:

The degree and type of optionality that characterizes L2 grammars has been at the core of the L1/L2 divide since at least the late 1970s. Some early characterizations of L2 syntactic optionality inspired performance failure accounts (Adjémian, 1976; Epstein et al., 1996), but representational accounts, and specifically parametric variation were already subsumed in Liceras’ (1983, 1986) notion of ‘permeability’ and in Pérez-Leroux and Liceras’ (2002) implied notion of ‘hidden code-switching’ between two parametric options. Under these accounts, both options of a given parameter are fully present in the L2 grammar, although the various properties need not have the same status in terms of how they are implemented. However, in Liceras (1983, 1986) non-native idiosyncrasy was defined precisely as the permanent availability of the two options.

Further attempts to provide accounts of the differences between L1 and L2 syntactic variability can be found in Tsimpli and Rousseau (1991), Liceras et al. (1998, 1999), Prévost and White (1999, 2000) or Robertson (2000). Prévost and White (2000) and Robertson (2000) claim the divergence in L2 grammars is due to performance. For example, they argue that the missing inflection and the missing determiners that characterize L2 grammars are due to the problem that L2 learners encounter when mapping specific morphological forms to abstract categories. Therefore, under these accounts, at the abstract level, the L2 grammar has the same properties as the ‘corresponding’ native grammar. However, in the approach taken by Tsimpli and Rousseau (1991) and Liceras et al. (1998, 1999) the L2 grammar does not necessarily implement the parametric options of the L2 but keeps the L1 options and incorporates UG principles to accommodate new input.

For example, Tsimpli and Rousseau (1991) argue that the null referential subject in the L2 English grammar of Greek or Spanish speakers (Ø live in Chamberí) may be an instance of pro if the null subject option is transferred from the L1 (pro live in Chamberí) or an instance of PRO (PRO live in Chamberí) if the L2 does not have the Φ-features that would account for the identification of the null subject. They further argue that because pro is always licensed via the L1, the overt subjects that occur in the L2 English of these speakers are actual Φ-features (the equivalent of Greek or Spanish verbal agreement person markers); thus, a realization of agreement that identifies pro (we pro live en Chamberí). This analysis was proposed by Roberge (1986, 1990) and Authier (1992) for French subject clitics, and if the English subjects were clitics in the Greek of the Spanish L2 English grammars, they would have the properties of French subject clitics. This is a testable hypothesis though easily rejected, at least superficially, because these L2 English pronouns do not seem to have the properties of French clitics. I say superficially because even though, unlike French, those L2 English grammars may not have lexical items differentiating strong (moi, toi …) and clitic (je, tu …) pronouns, the same lexical item may display two different values. In fact, this is also relevant when interpreting both the production and the comprehension OPC (Overt Pronoun Constraint) data elicited by A&R since it is not obvious that the null and overt subjects in the L2 grammar are like those in L1 Spanish. This is an issue that the MGT, being a representational theory, should address. In fact – and even if it is not possible on the basis of production or even comprehension or interpretation data to determine the categorial status of a given functional category in an L2 grammar – we are bound by the fact that similar performance may not be a reflection of similar competence.

Within the classical view of parametric theory but taking into consideration the refinements introduced in relation to the nature of empty categories, Liceras et al. (1998, 1999) argue that the Spanish null subjects produced by L1 speakers of [–null subject] and [topic-drop] languages such as English or Japanese (respectively), are neither instances of a null constant as proposed for child language (Rizzi, 1994), nor Japanese-like pro (as in Speas, 1994) but rather instances of a default option (à la Lebeaux, 1988): a pro generated within the Spec-VP, along the lines proposed by Pierce (1992) for early L1 French. The advantage of this proposal is that learners can make use of a UG principle to license pro and use their L1 identification resources, something that was evidenced, among other things, by the distribution of overt pronouns in these L2s. The MGT contemplates adding grammars but does not seem to address the status of default options in relation to the grammars that may be added, including the L1 grammar. In fact, rather than resorting to default options, we might be able to come up with interesting hypotheses regarding the status of weak and overt pronouns in L2 grammars. For example, we can draw on proposals linked to the feature specification of weak and strong subject pronouns, such as Camacho’s (2008) adaptation of Cardinaletti and Starke’s (1999) tripartite typology of pronouns for Dominican Spanish and Brazilian Portuguese, which according to this author display two independent lexical entries for pronouns: an ella (‘she’) which has the features [3rd person, singular, weak, +ref] and an ella which has the features [3rd person, singular, weak, –ref ].

What stands out about these latter proposals is that while L2 grammars are instances of natural languages and always display the rules and mechanisms that are found in a given grammar, the implementation of L2 parametric options, functional categories, and features does not necessarily coincide with that of the ‘corresponding’ input or target grammar. The specific illustration of the MGT that A&R depict in their diagram 10 (‘Parameter setting with two values’) and diagram 11 (‘Subject form for Spanish and English bilinguals’) seems to contemplate only the two options of the parameter with the novelty that, for each of them, the other language has a restricted realization as in (1) and (2).

(1) SPANISH: [+null subject] productive; [–null subject] contextual

(2) ENGLISH: [–null subject] productive; [+null subject] lexical

However, A&R suggest that the sub-grammars may have other configurations. Does this imply that all combinations are possible? For instance, can a sub-grammar have both productive [+null subject] as in L1 Spanish and productive [–null subject] as in English but not the two ‘secondary’ dimensions? If this is the case, how will that grammar handle the contextual and lexical realizations of English and Spanish subjects? Do these authors consider the possibility that L2 grammars may be configured along the lines of the so-called partial null subject languages such as Finnish or Hebrew (Holmberg, 2009; Vainikka and Levy, 1999), or that L2 grammars may display a distribution of null and overt subjects like that of Dominican Spanish or Brazilian Portuguese?

Lardiere (2008, 2009) advanced a proposal that highlights the need to move away from parameters and puts the emphasis on features, depicting L2 grammars as being different from the ‘corresponding’ L1s in how features are assembled within a given functional category. She specifically argues that some unexplained and rather pervasive instances of optionality in Patty’s L2 English grammar – such as the omission versus suppliance of indefinite determiners – stem from the fact that in Chinese but not in English the feature [specificity] is bound up with the feature [Number]. In other words, Patty does not seem to have de-linked the features [Specificity] and [Number] in her L2 grammar. Thus, this proposal clearly complies with the ‘no complex rules requirement’ advocated for by A&R, since the only requirement would be to avoid linking specificity and plurality in the L2 English grammar. Along this same line of reasoning, the lexical and contextual options in A&R’s depiction of the null subject parameter in bilingual grammars could be defined in terms of features of pro and overt pronouns, respectively. In fact, the English null subjects that justify the [+null subject] lexical option are always instances of pleonastic pro, which happens to have a clearly specified [–referential] feature.

I would also like to address a representational account of L2 optionality, the adoption of the Competing Grammars Hypothesis (Kroch, 1994, 1997; Kroch and Taylor, 1997; Yang 2002), which also constitutes the basis for the MGT and questions the predictive power of this theory. In fact, in Zobl and Liceras (2005, 2006), Perales et al. (2009), Perales and Liceras (2010) and Liceras (2009) we hypothesized that L2 developing grammars mirror the steps of the diachronic changes that take place when a language has adopted a new parametric option. We first tested this hypotheses with respect to two different diachronic changes: V2 and V-movement, on the one hand, and the categorial status of clitics, on the other (Zobl and Liceras, 2005, 2006). In the transition from Old English to Modern English, some of the parametric shifts that led to the loss of V2 and V-movement in Old English are shown in (3).

(3) Old English Modern English

 a. OV →VO

 b. I-final →I-medial

 c. +V2 →–V2

 d. +verb raising →–verb-raising

We hypothesized that the acquisition of English by native speakers of German or Dutch (which very much resemble Old English with respect to these parametric properties) would follow the same steps as the loss of verb-second and verb-movement in English (Pintzuk, 1999; Pintzuk et al., 2000). Namely, VO would be acquired first and Verb raising last, and there would be periods where the two options for each property would show different degrees of variability, as in diachronic change. In the case of the status of clitics, Modern Spanish clitics underwent a categorial change with respect to Old Spanish clitics. According to Fontana (1994), clitics are phrasal categories (XP) in English and Old Spanish but they are heads (Xº) in Modern Spanish. This status implies that in the transition from Old Spanish to Modern Spanish, the properties associated with XP clitics were lost to those associated with Xº clitics, as shown in (4).

(4) Old Spanish Modern Spanish

a. +interpolation →–interpolation

b. clitic/V and V/clitic→clitic/V

c. –clitic doubling →+clitic doubling

We hypothesized that the acquisition of these three properties in Modern Spanish by speakers of languages whose pronouns are XPs (English or Chinese), as with Old Spanish clitics, would proceed along the lines illustrated in diachronic change. While the L2 data on the loss of V-movement and V2 showed evident parallels with the diachronic change which led to the loss of these phenomena in English, the data on the acquisition of clitics showed a very different picture, since two of the properties which characterized the Spanish grammar with pronominal XP’s (interpolation and placement ‘promiscuity’) were not found in the Spanish L2 grammar of speakers with pronominal XP’s. However, these L2 data were different from the English data in two important respects: (1) they were spontaneous data rather than grammaticality judgment data, and (2) the pronominals in the participants’ L1s were not exact replicas of Old Spanish pronouns because, while both English and Chinese have pronominal XP’s, they are not phonological clitics, as was the case in Old Spanish. In order to elicit comparable data and to be able to compare the diachronic changes to L2 development we elicited grammaticality judgments from L1 Czech speakers (Perales et al., 2009) and from L1 English speakers (Perales and Liceras, 2010) learning Spanish. We took the feature approach depicted in Table 1 to define the three properties and showed that the properties of clitics derived from the [+/–phon] feature (interpolation and ordering with respect to the verb) were acquired earlier and more successfully than properties derived by the [+/–XP] feature (clitic doubling).

OPEN IN VIEWER

Table 1.

From English and Czech to Modern Spanish object pronouns: a feature account.

Full pronouns	Simple clitics	2nd position clitics	Categorial clitics	Agreement markers
[–phon]		[+phon]		[+phon]
[+XP]		[+XP]		[–XP]
English	English	CzechSerbo-CroatianOld Spanish	ItalianPortugueseFrench	SpanishMacedonianBulgarian

Source. Perales and Liceras, 2010: 330.

The feature approach allows us to consider stages in the categorial nature of clitics, which are neither realized in Czech, Modern English nor in Modern Spanish but which could nonetheless be realized in the Spanish L2. The experimental data points to the fact that the L1 English speakers go through a stage where their grammar resembles the Czech (and the Old Spanish) grammar while the L1 Czech speakers go through a stage where there grammar resembles the French grammar. Clitic development in the two L1s differs in that the two properties related to phonology are acquired in the reverse order (interpolation first for L1 English and clitic placement first for L1 Czech), while it goes hand in hand with respect to the problematic status of clitic doubling (Liceras, 2009). Thus, while our Competing Grammars Hypothesis, and especially the comparison of diachronic change and language development, provides a useful tool for analysing L2 grammars because it deals with optionality as a natural phenomenon (one that can emerge in any language contact situation), there is a clear difference between the type of competition that takes place in diachronic change and in L2 acquisition: (1) the latter shows a more powerful sensitivity to the properties of the target language (there is obviously a clear target language only in L2 development), and (2) it does not show a clear pattern of substitution of one option for the other.

In conclusion, the MGH can be taken as work in progress that confronts the field of non-primary language acquisition with the need to revisit old and recent attempts to provide a representational account of L2 optionality.