Experimental and other empirical research on language is faced with the fact that language performance exhibits a high degree of variability at all linguistic levels. Variability is found across languages, across speech communities within one language, across individuals within one speech community and even within the same individual. Bilingual language use adds a further source of variability to this already complicated picture. On the other hand, there are aspects of language and language use that are constrained, stable, or robust and that are less (or not at all) subject to variability, for example, possible options that are not chosen in any language or kinds of error that are never produced. Several familiar ways of dealing with the variability of language use and its limits have turned out to be unsatisfactory. One approach has been to simply abstract away from variability with constructs such as the ‘ideal speaker–hearer’ (who – to our knowledge – nobody has met so far). Another strategy is to average across individuals, which sometimes results in arbitrary mean scores or mean activation patterns that are hard to replicate for individuals, even for those who took part in a given study. A third solution when confronted with variability in language use is to take it at face value, positing that every language, every speech community, and even every individual is different, an approach that essentially gives up on discovering any kind of generalizations. While none of these strategies appears to us to be particularly fruitful, the problem of how to deal with variability in language performance and its limits remains.

A large body of research into bilingualism has revealed that language processing is fundamentally non-selective; there is simultaneous, graded co-activation of mental representations from both of the speakers’ languages. An equally deep tradition of research into code switching/mixing has revealed the important role that grammatical principles play in determining the nature of bilingual speech. We propose to integrate these two traditions within the formalism of Gradient Symbolic Computation. This allows us to formalize the integration of grammatical principles with gradient mental representations. We apply this framework to code mixing constructions where an element of an intended utterance appears in both languages within a single utterance and discuss the directions it suggests for future research.

Goldrick, Putnam and Schwarz (Goldrick, Putnam & Schwarz) have offered an account that must delight all linguists who have spent an inordinate amount of their professional time working on understanding the grammar of bilingual language use – just how, and why, do bilinguals produce the utterances that they do. The computational approach Goldrick et al. propose blends the grammatical principles (of the optimality kind) with general processing constraints to yield patterns of code-mixing that may be sparse – the emergence of doubled elements in bilinguals’ utterances – but certainly needing an explanation. Their Gradient Symbolic Computation model is, in fact, the first robust account of the presence of doubled elements, i.e., an element of the utterance is doubled, appearing in both languages within a single utterance (see 1 below). Their account, however, does force them to make an important, but vulnerable, theoretical assumption: that grammatical principles can refer to language membership. It recalls, albeit obliquely, Belazi, Rubin and Toribio's much discussed ‘Language’ feature needed to mobilize their Functional Head Constraint (see Bhatt, 1997, for an empirical and theoretical critique of the Functional Head Constraint).

Research on bilingualism has documented profound brain plasticity by which the bilingual experience reconfigures the cognitive system. These effects include temporary as well as more enduring ones, and parallel activation of a bilingual's two languages may well be a key factor at the root of these observed changes. Recent recommendations (Green, 2011) have emphasized that research on code-switching in particular could provide a fruitful avenue for investigating the nature of how a bilingual speaker selects words and ultimately produces an utterance. Findings to date illustrate that if anything, the reach of co-activation is more extensive than previously thought, extending to the phonology and syntax of languages. While the degree of permeability may compound the perceived difficulty of the selection process, it is also a testament to the documented mental agility of bilinguals.

Goldrick, Putnam and Schwarz (henceforth Goldrick et al.) base their computational account of code mixing on two premises, (1) that code mixing is subject to grammatical principles and (2) that “mental representations in bilingual speakers incorporate blends of structures from each language”. While there is copious evidence for the first premise, the second is far less well empirically supported. The authors imply that the second premise, the assumption of blend representations, is supported by Kroll & Gollan's (2014) review of speech planning in two languages, but this is not the case. Kroll & Gollan (2014:1) do indeed discuss evidence for the “parallel activation of the bilingual's two languages” but in terms of competition between alternatives rather than blended representations. The notion of blend is of course familiar to code-switching researchers from work on bimodal bilingual code mixing, as Goldrick et al. acknowledge, but here the term refers to “simultaneously produced English words and ASL signs” (Emmorey, Borinstein, Thompson & Gollan, 2008: 43).

Bilingual speakers can use one of their languages in a given interactional context or switch between them when addressing different speakers during the same conversation. Depending on community usage bilingual speakers may insert single lexical forms from one language into the morphosyntactic frame of another or alternate between languages at clause boundaries. They may also engage in dense code switching with rapid changes of language within a clause during a conversational turn (Green & Li, 2014). These varieties of language use configure the same speech production mechanism and so a theory of code-switching must be part of a theory that accounts for the range of bilingual speech. Does the proposal described in Goldrick, Putnam and Schwartz (2016) meet these criteria?

Code-mixing (CM) is a striking example of how two languages are active simultaneously in bilingual production. Gradient Symbolic Computation (GSC) proposes a formalism to account for the systematicity of CM patterns by integrating psycholinguistic notions of bilingual co-activation with generativist accounts of grammar. We applaud the attempt to bridge research traditions and all efforts to capture the systematicity of variation, and the interaction between processing and grammatical constraints in bilingual production. However, the descriptive and predictive scope of the current proposal remains somewhat unclear, as does its connection to existing accounts.

Goldrick, Putnam and Schwarz (2016) propose a very explicit account of code mixing, which nicely draws implications from the well-established findings of coactivation during bilingual language production, and that code mixing is constrained by grammatical principles. This precise characterization will undoubtedly serve as a useful basis for further research on code mixing, second language sentence production, and syntactic learning. However, there are three issues with the account that require further elaboration.

The keynote article (Goldrick, Putnam & Schwartz, 2016) discusses doubling phenomena occasionally found in code-switching corpora. Their analysis focuses on an English–Tamil sentence in which an SVO sequence in English is followed by a verb in Tamil, resulting in an apparent VOV structure:

1. (1)

The paper to which this commentary responds (Goldrick, Putnam & Schwarz, 2016) represents a big step forward in a field which was showing signs of stasis – the study of the grammatical properties of intra-sentential code-switching or code-mixing – and this for two reasons. First of all, it explicitly links the insights from the grammatical study of code-mixing to the rich array of results from psycholinguistic research in the domain of bilingual language processing, making use of the Gradient Symbolic Computation framework (GSC; Smolensky, Goldrick & Mathis, 2014). Second, it provides a set of tools to handle problems in the domain of code-mixing having to do with simultaneous representations in the domain of bilingual complexes.

New theories are a constant of the now vast literature on code-mixing (CM). The Gradient Symbolic Computation model proposed by Goldrick, Putnam and Schwartz (Goldrick, Putnam & Schwartz) will appeal to many, especially those who already espouse constraint-based approaches to grammar. As variationist sociolinguists, we particularly welcome the model's incorporation of “relative probabilities of certain structures”, a feature we believe can enhance our chances of capturing actual CM behavior. We also applaud Goldrick et al.’s efforts to integrate experimental findings on co-activation with grammatical principles. Our questions concern the utility of “doubling constructions” to showcase the model, and by extension, the degree to which it can account for bilinguals’ spontaneous production of CM. A historical perspective on the field shows that none of the myriad theories of CM, often inspired by competing sets of grammatical principles, has yet achieved broad acceptance. In the absence of any widely endorsed evaluation metric – still sadly lacking -– how are we to decide amongst them?

Goldrick, Putnam and Schwarz (Goldrick, Putnam & Schwarz) argue that code-mixing in bilingual production involves not only combining forms from both languages but also – crucially – integrating grammatical principles with gradient mental representations. They further propose an analysis of a particular case of intrasentential code mixing – doubling constructions – framed within the formalism of Gradient Symbolic Computation. This formalism, in their view, is better suited to accounting for code mixing than other generative language models because it allows the weighting of constraints both in the choice of particular structures within a single language and in blends of structures in code-mixed productions.

In their keynote article, Goldrick, Putnam and Schwarz (2016) present a computational account of code-mixing. Although they review literature on the co-activation of lexical representations and cognate facilitation effects in bilingual language processing, their model remains silent on how it interfaces with lexical factors, and how lexical factors impact code-switching. One such lexical factor is cognate status, which has been found to affect code-switching, as demonstrated in corpus analyses (e.g., Broersma & De Bot, 2006) and psycholinguistic experiments (Kootstra, Van Hell & Dijkstra, 2012). For example, using the structural priming technique to examine the role of lexical factors in code-switching, Kootstra et al. asked Dutch–English bilinguals to repeat a code-switched prime sentence (starting in Dutch and ending in English) and then describe a target picture by means of a code-switched sentence (also from Dutch into English). They observed that bilinguals' tendency to switch at the same position as in the prime sentence was increased when the prime sentence and target picture contained cognates.

The Gradient Symbolic Computation (GSC) model presented in the keynote article (Goldrick, Putnam & Schwarz) constitutes a significant theoretical development, not only as a model of bilingual code-mixing, but also as a general framework that brings together symbolic grammars and graded representations. The authors are to be commended for successfully integrating a theory of grammatical knowledge with the voluminous research on lexical co-activation in bilinguals. It is, however, unfortunate that a certain conception of bilingualism was inherited from this latter research tradition, one in which the contrast between native and non-native language takes a back seat.

Our keynote article “Coactivation in bilingual grammars: A computational account of code mixing” (Goldrick, Putnam & Schwarz) aimed to provide a framework that would begin to unify psycholinguistic and formal grammatical approaches to code mixing. We situated our account within a large body of psycholinguistic and phonetic evidence suggesting that, under many conditions, multiple representational elements simultaneously occupy (to varying degrees) a single position within a linguistic structure. The presence of such blends in multilingual cognition is not compatible with many formal grammatical approaches that assume mental representations are necessarily discrete.

Seminal research which indicates that bilingualism leads to enhanced executive control (Bialystok, Craik, Green & Gollan, 2009; Bialystok, Craik & Luk, 2012) has recently been challenged (de Bruin, Treccani & Della Sala, 2014; Hilchey & Klein, 2011; Paap & Greenberg, 2013). We suggest that this discrepancy in findings is attributable to differences between bilinguals. Although the present results showed no significant differences in executive control between the monolinguals and bilinguals, those bilinguals who used their L2 more frequently in their daily lives were significantly more likely to evidence enhanced executive control over those who rarely used their L2.

There is ample evidence that native and non-native listeners use lexical knowledge to retune their native phonetic categories following ambiguous pronunciations. The present study investigates whether a non-native ambiguous sound can retune non-native phonetic categories. After a brief exposure to an ambiguous British English [l/ɹ] sound, Dutch listeners demonstrated retuning. This retuning was, however, asymmetrical: the non-native listeners seemed to show (more) retuning of the /ɹ/ category than of the /l/ category, suggesting that non-native listeners can retune non-native phonetic categories. This asymmetry is argued to be related to the large phonetic variability of /r/ in both Dutch and English.

Cued language switching is used to examine language-control processes by comparing performance in language-switch trials with performance in repetition trials. In 1:1 cue-to-language mappings, language repetitions involve cue repetitions and language switches involve cue switches. Hence, the observed switch costs might reflect cue-switch costs rather than language-related control processes. By introducing a 2:1 cue-to-language mapping, we dissociated language switches (cue and language switched vs. cue switched, but language repeated) and cue switches (repeated language, with vs. without switched cue). We found cue-switch costs, but language-related switch costs were substantial, too, presumably reflecting language-control processes in cued language switching.

This study examines the strength and direction of lexical-grammatical associations within and between first and second languages (L1 and L2) in a longitudinal sample of sequential bilinguals. Thirty-three children who spoke Vietnamese (L1) and English (L2) completed picture-naming and story-telling tasks in each language at four yearly intervals. Hierarchical linear modeling across Years 1–4 revealed bidirectional within-language associations and a unidirectional cross-language association from the L1 to L2. Results suggest a conditional relationship between languages in which the L1 supports L2 growth, but not vice versa. Findings contribute to defining pathways for L1 and L2 learning across domains and languages.

Word-initial /s/-consonant clusters do not occur in Spanish. Confronted with such sequences (e.g., in loanwords), Spanish speakers tend to perceive an illusory initial /e/, ‘repairing’ the illicit sequence. In two experiments, both conducted in Spanish with Spanish-sounding nonwords, we ask whether knowledge of English, which has no restriction against this sound sequence, weakens this pattern of perceptual repair in fluent Spanish–English bilinguals, and whether the effects of English depend on language dominance. In both identification and discrimination tasks, bilinguals exhibited weaker perceptual repair effects relative to Spanish monolinguals. This was true even for bilinguals dominant in Spanish, though the weakening was more pronounced for English-dominant bilinguals. These results show that conflicting phonotactic systems can jointly influence bilinguals’ perceptual repair of the acoustic signal in the more restrictive language, even when it is the bilingual's dominant language, suggesting a degree of integration and mutual influence of knowledge between both their languages.