Introduction: Convergence and divergence in functional and formal approaches to SLA

I Introduction

It is generally recognized that linguistic approaches to second language acquisition (SLA) draw on various theoretical frameworks. Often, these frameworks are seen as incompatible. In this special issue, we seek to explore the ways in which these approaches can be brought together in providing explanations for acquisition that rely both on input and on existing linguistic knowledge, whether it be formal or functional.

In the early days of SLA, typological linguistics and Labovian variationist sociolinguistics provided important theoretical frameworks and analytical tools for interlanguage research. Eckman’s (1977) Markedness Differential Hypothesis, which predicted ease and difficulty of different types of phonological and grammatical structures, and relative clause acquisition research in relation to the Noun Phrase Accessibility Hierarchy (e.g. Gass, 1979) were informed by typological linguistics. Work on interlanguage variation (e.g. Tarone, 1979) drew on Labov’s work on sociolinguistic variation (e.g. careful vs. vernacular style) to explain different linguistic forms produced by second language (L2) learners depending on context. In the mid-1980s Chomsky’s (1981) parameter-setting approach was extended to SLA (Liceras, 1986; White, 1985) and, since then, the dominant linguistic framework in which learner language was investigated has been mostly formal, generative linguistics (for summary of such work, see White, 2003). ¹

In contrast, first language (L1) acquisition research had two different approaches, which Ingram (1989) called Language Acquisition and Child Language, the former often practiced by linguists, and the latter by psychologists. The former was more theory-driven, often within the generative framework (e.g. Hyams, 1986; Radford, 1990), the latter more data-driven, and often informed by typological linguistics, e.g. the cross-linguistic L1 acquisition research edited by Dan Slobin (e.g. Slobin, 1985). Both these research traditions had different research methods and interests as if there were two fields of child language research (Bennet-Kastor, 1986), pursuing their respective research agendas.

This situation started to change since the mid-1980s in that these two different camps started to compete in some domains of inquiry. First, connectionists presented a highly empiricist alternative to the nativist generative approach in the domain of regular vs. irregular morphology (in particular, past tense morphology in English, also discussed in Yang and Montrul’s article in this issue; see Pinker and Mehler, 1988; Rumelhart and McClelland, 1986). The debate between connectionists and nativists extended to other areas, such as the validity of the poverty-of-stimulus argument (e.g. Elman, 1990).

The field of L1 acquisition research has seen a major change since 2000, mainly due to the research by Michael Tomasello and his associates. Based on a comprehensive review of previous research in syntactic development, Tomasello (2000) showed that, contrary to the assumptions of the continuity hypothesis, children do not have an adult grammar right from the beginning when they start using a particular construction (past tense, passive, transitive); rather, they restrict that construction to particular verbs, as captured in Tomasello’s the Verb Island Hypothesis.

Tomasello (2003) went one step further. It has been pointed out that formal approaches to language acquisition emphasize a theory of language (property theory) and linguistic representations, while functional approaches are primarily concerned with a theory of development, or change (transition theory); for the property/transition theory distinction, see Gregg (1993). Tomasello (2003), however, proposed to change this state of affairs by assuming construction grammar (Fillmore et al., 1988; Goldberg, 2003) as the theory about the target that children need to acquire. Construction grammar is a cognitive linguistic theory that considers language to be a pairing of form and meaning/function. All kinds of form–meaning associations are considered to be constructions: words (book), morphemes (-ed), periphrastic constructions (be going to), idioms (put up with) syntactic constructions (ditransitive), with syntactic constructions themselves carrying a (non-compositional) meaning (e.g. open me a beer, even though the verb open does not have a meaning of transfer). Tomasello (2003) argued that language acquisition consists of learning constructions bottom up, first with small number of items (item-based representation), then semi-productive patterns and, finally, fully productive constructions without the help of domain-specific innate knowledge of language through social-intentional cues and social interaction. Hence, abstract grammatical knowledge emerges from language use, the interaction of input and innate aspects of general cognition (not specific to language). Also, Tomasello (2004) has questioned the falsifiability of the Universal Grammar (UG) hypothesis, for which various versions of UG are proposed by researchers without specific criteria for evaluating them.

However, formal linguists counter that some limit on the hypothesis space – based on limited crosslinguistic variation, knowledge of structure dependence and constraints on form and interpretation – must underlie the human language capacity (for a review, see Crain and Thornton, 2006). The research program is to provide an ever more precise characterization of that hypothesis space and how that plays out in child language development.

Although some generativists – for example, Steven Pinker, who is arguably one of the most influential generative psycholinguists – formerly argued that words and rules are two completely different ingredients of language – words being memorized while rules being productive and possibly innate (Pinker, 1999), Pinker appears to have shifted his position. Pinker and Jackendoff (2005), and later Jackendoff and Pinker (2005), appear to now acknowledge that construction grammar is a better account of knowledge of language than Chomsky’s (1995) minimalist program (Jackendoff and Pinker, 2005: 220–22). ²

Thus, the functional perspective and the formal perspective differ fundamentally in their assumptions and in some instances in the phenomena they seek to explain. The functional perspective relies more on data that derive from production and corpora and are especially strong at lexical and some morphological acquisition patterns. This approach is reluctant to cede cognitive status to ‘unseen’ elements and abstract hierarchical structure due to these empiricist roots. For example, Tomasello (2003: 7) denies any status for such elements as ‘empty categories’ and gaps, which are very important in formal theories. Moreover, functional approaches have not addressed hierarchical structure in phonology to the extent that formal theories have.

The formal framework has taken a different approach, first arguing from a rationalist perspective and subsequently valuing experimental data more than production data because such data may inform researchers about what is not possible in grammars. This is something production data alone cannot do and has been accepted by SLA researchers for over 30 years (Cook, 1986). One of the major contributions of such approaches is the fine-grained nature of the predictions that can be made for real-time processing of elements that are not adjacent to one another: the so-called ‘long distance’ (but covertly local) dependencies in wh-questions, relative clauses in some languages, and hierarchical structure in phonology. Indeed, the evidence in favor of empty categories in filler-gap processing and their relevance in SLA is now very strong (for data and reviews, see Featherston, 2001; Felser et al., 2012; Juffs and Rodríguez, 2014).

Thus, to some extent, arguments over theory have derived from the kinds of data that researchers have sought to explain, for example frequency effects in use vs. knowledge of abstract categories in adult SLA. While not underestimating the challenges of forces drawn up on each side (see, for example, Evans and Levinson, 2009, and commentaries), the task researchers now need to address is how perspectives can tease apart what is learned from input and what is contributed by the internal, possibly universal, cognitive resources. Some of these resources may involve abstract hierarchies that formal theories are excellent at providing an account of; other resources will involve frequency tracking. These suggestions are certainly not new, as researchers such as Lila Gleitman in her comments in Gene Searchinger’s (1993) Human Language Series or more recently Yang’s (2008) review article have made clear, but it is worth re-emphasizing them here.

In the field of SLA, where the generative approach has been the dominant linguistics framework in the last three decades, a shift has also taken place. This is arguably reflective of the change in the field of linguistics and L1 acquisition discussed above, which has traditionally influenced SLA research. In addition, Nick Ellis’s work has been quite influential in promoting functional approaches. Since the mid-1990s Ellis has been a major proponent of usage-based approach in SLA. In a series of influential articles (see, among others, Ellis, 1998, 2002, 2003) he advocated a connectionist, emergentist, and usage-based approach with construction grammar as linguistic framework (along the line of Tomasello’s approach in L1 acquisition discussed above).

Thus, currently two approaches – formal, generative, nativist approach and functional, usage-based, connectionist approaches – offer competing accounts of language acquisition, both first and second. ³ Needless to say, neither approach has yet uncovered the mechanisms of language acquisition and, given the ultimate goal of understanding its mechanism, both approaches should be able to make unique contributions to the field instead of two separate ‘fields’ (Bennet-Kastor, 1986) of language acquisition pursuing their respective interests. The articles in this special issue explore how each approach can contribute to our understanding of the mechanism, and try to understand the area of convergence – where the two approaches have converged, and of divergence – where they differ still. By identifying the areas of convergence and divergence, we will be in a better position to recognize the areas of cooperation as well as healthy competition (more on this below).

To summarize, the main perspectives described above – formal/generative approach and functional/cognitive approaches – emphasize different components of the language learning process: innate linguistic knowledge vs. general learning processes and input. Because of their different theoretical assumptions both approaches have also tended to focus on different empirical evidence. The purpose of this special issue is to foster dialogue between these views and to consider how the both approaches explain similar empirical phenomena in second language acquisition. Thus, this special issue is unique in trying to emphasize how the two approaches can actually complement each other.

II The articles

This special issue includes two articles taking a formal approach (Yang and Montrul, 2017, this issue; Bardovi-Harlig and Stringer, 2017, this issue), two articles taking a functional approach (Zyzik, 2017, this issue; Paquot, 2017, this issue), and one article testing both approaches (Shantz, 2017, this issue).

Yang and Montrul revisit the issue of how the dative alternation can be acquired in English, proposing a theoretical solution that marries top-down nativist assumptions with bottom-up statistical learning. They first consider proposals by Yang (2016); namely, the Tolerance Principle, a formal model of detecting productivity from distributional properties of input data, and the Sufficiency Principle, which lets child learners retreat from overgeneralizations. They then discuss L2 studies on dative alternation since Mazurkewich (1984) and examine whether these principles can account for the L2 acquisition of the dative alternation. They further consider how the same principles would operate in L1 attrition (when input conditions change dramatically), and advance a possible research agenda to examine the role of input in L2 acquisition.

Bardovi-Harlig and Stringer report on a large-scale study on the relationship between formulaic expressions and creative rule-based knowledge. Formulaic expressions have been a major research area in usage-based approaches, especially to show that syntax emerges from unanalysed items. Based on data from an oral conversation-simulation task and an aural-written elicited imitation task, Bardovi-Harlig and Stringer argue that formulaic language does not drive the acquisition of syntax; rather, the acquisition of syntax as an independent process drives changes in the production of conventional expressions, which goes against the usage-based acquisition model such as Ellis (2003) and Tomasello (2003).

Shantz investigated the effect of frequency and grammaticality in L2 processing of multi-word sequences by 56 L1 Mandarin Chinese learners learning English in the USA using a self-paced reading experiment (with 56 native English speakers as control) in order to test whether the rule-based or the usage-based approach explains the data better, by manipulating grammaticality and frequency contrasting sentences including phrases (e.g. I thought he was/*were vs. I believe he was/*were, the latter being less frequent). He found that although the native speakers in the control group showed a categorical grammaticality effect, the L2 learners showed proficiency effects and U-shaped development of grammaticality. Shantz suggested that the data are not totally compatible with either approach, and that both approaches need to be considered to understand SLA process better.

Zyzik reviews research on subjectless expressions in L2 Spanish conducted within the generative approach and the functionalist approach and identifies areas of convergence and divergence. Noting that there is much more extensive research from the generative approach compared to the functional approach, Zyzik finds that both approaches have come to emphasize the role of ‘discourse’ (an interface phenomenon under the generative approach), in their explanations of why L2 learners show optionality in the grammatical domain.

From a usage-based perspective, Paquot investigated use of three-word lexical bundles by French and Spanish learners of English as a foreign language (EFL) that have discourse or stance-oriented function (e.g. that is why, will certainly be, for this reason) to explore the role of L1 frequency effect in L2 acquisition. Word combinations were extracted from argumentative essays in learner corpora, which were compared with the frequency of their translation equivalents in French and Spanish native speaker corpora. Contra previous studies that did not find strong influence of L1 frequency, Paquot found strong and positive correlations between the frequency of a lexical bundle in the EFL learners’ writing and L1 frequency of their equivalents, with different patterns of use across the two L1 learner populations. She argues that more attention should be paid to the role of L1 frequency in usage-based L2 research, which tends to focus more on frequency in the input in the target language (L2).

III Convergence and divergence

The articles included in this special issue highlight important areas of convergence and divergence. There is convergence in the domains of inquiry. As was the case of past-tense debate noted above, where formal and connectionist approaches competed since 1980s, recently formal and functional approaches ‘compete’ to investigate which approach is more tenable. This ‘competition’ generates intensive research in a particular domain and is likely to create new insights that may not be available otherwise. The dative alternation (Yang and Montrul, 2017, this issue) is an important issue not only in language acquisition with respect to recovery from overgeneralization (Bowerman, 1987; Pinker, 1989), but also for the theory of grammar (Goldberg, 1995). Yang and Montrul present an important proposal in this domain, to which Goldberg offered an alternative account for L1 acquisition, which relies on statistical preemption: learners avoid using a construction if an alternative formulation has been systematically witnessed instead (Goldberg, 2016) and for L2 acquisition, in which only highly advanced learners can use statistical preemption (Robenalt and Goldberg, 2016).

Another linguistic area of where research is converging is what is called ‘semi-productive’ linguistic units, such as idioms, formulaic expressions, or lexical bundles. Three articles in the special issue investigate this domain (Bardovi-Harlig and Stringer, 2007, this issue; Shantz, 2017, this issue; Paquot, 2017, this issue). This area has been important because the formal (generative) approach mostly presumes a dichotomy between rules and lexical items, which are memory-based (e.g. Chomsky and Halle, 1968; Pinker, 1999). This is where construction grammar and generative grammar disagree, in that construction grammar assumes a continuum between productive rules and lexical items, with many semi-productive units in between. As noted above, both Tomasello (2003) and Ellis (2003) propose bottom-up learning in this area, from lexically-based learning to low-scope patterns to productive construction as a fundamental process in language acquisition. ⁴ Research on formulaic language has not been very popular in the generative approach but, given the current theoretical importance of this domain, more research should be conducted in future by formalists. ⁵

Conversely, research on subjectless expressions has been mostly dominated by the formal generative approach under the theory of Principles and Parameters (i.e. the Null Subject Parameter), as noted by Zyzik, and therefore more research from usage-based, functional perspectives are needed. In particular, how learners achieve native/adult-like competence in rejecting particular forms (zero vs. pronominal vs. lexical forms) needs to be investigated.

Other than research domains, an important area of convergence is the role of input frequency to learn linguistic rules and recognize exceptions (Yang and Montrul, 2017, this issue). Although generative grammar noted the importance of input as triggering primary data, it did not traditionally consider input frequency as an important factor in achieving adult competence in specific domains of morphology and syntax. Yang’s (2004) proposal that emphasizes both input frequency and innate universal grammar is an important development. This work shows that the role of UG as originally conceived may be more restricted than previously thought, but still needed to account for aspects of language acquisition that elude other potential explanations. An important issue that needs to be addressed is whether we really need UG (and in what form) to explain first and second language acquisition of the dative alternation, or statistical information without UG can explain them (as suggested by Goldberg).

Divergence of the two approaches is not particularly evident in the articles included in the special issue. Methodological preferences are sometimes noted, in that the formal approach prefers experimental data while the functional approach prefers naturally occurring data. However, this is just a tendency, and both approaches employ any type of data that help answer their research question (e.g. Tomasello and Ellis both use experimental data extensively, while many generative researchers use natural production data, e.g. Hilles, 1986; Hyams, 1986).

An important divergence is of course epistemology. The formal approach often takes innate endowment as given, while the functional approach does not. This results in very different approaches in interpretation of data, as well as different assumptions and terminology, which makes dialog between the two difficult. What both approaches need in order to advance our understanding of the acquisition of L1 and L2 is to avoid theory-internal, or discipline-specific terminology and concepts but to appeal to a wider audience regardless of their background. In doing so, both approaches should compete to see which approach can propose a better account of the acquisition data in the same linguistic domains. In particular the formal approach should propose why innate knowledge is necessary (and in what form) ⁶ while the functional approach should propose a realistic account of how acquisition is possible without innate knowledge. Such healthy competition will advance our knowledge greatly, which is already happening in the domain of dative alternation as noted above. Perhaps a researcher who is uncommitted to either approach (such as Shantz, 2017, this issue) should, or a research team consisting of researchers from both approaches could, attempt to investigate which hypothesis fits the data better, and how their hypotheses can be improved.

The field of language acquisition has seen many changes, and the time is ripe for a healthy competition and discussion across different approaches, which is already happening in the field of L1 acquisition (see Ambridge and Lieven, 2011, who compared and evaluated research from both approaches in various linguistic domains). Rather than engaging in two different ‘fields’ of language acquisition as we used to, competition, dialog, and collaboration would be a more reasonable approach. That is sorely needed to tackle the daunting task of understanding the mechanism of language acquisition: both first language and second language.