by Rod Ellis
Phonetic context Stage
Learners Initial Vowel Liquid Nasal Vd stop VI stop
31 0 0 0 0 0 0 1
18 0 0 0 0 0 0 1
29 0 0 0 0 0 0 1
10 01 0 0 0 0 0 2
7 01 01 0 0 0 0 2
38 01 01 01 01 01 01 2
28 1 01 01 01 01 01 3
25 1 1 1 1 01 1* 3
17 1 1 1 1 1 1 4
9 1 1 1 1 1 1 4
Key
0 = /ð/ not supplied; 01 = /ð/ supplied variably; 1 = /ð/ supplied;
* = cell not conforming to implicational pattern
Figure 5.2 Implication scaling of /ð/ according to linguistic context (simplified from Trofimovich et al. 2007)
Trofimovich et al. also investigated two factors that determined the likelihood of learners supplying /ð/ in the different contexts—input frequency and cross-language similarity. They were more likely to supply it in a context where /ð/ occurred frequently in the input and where /ð/ was perceived as dissimilar to an L1 consonant. They were less likely to do so in a context where /ð/ occurred less frequently in the input and where /ð/ was perceived as similar to an L1 consonant. This study, then, indicates that input frequency and the learners’ perceptions of L1–L2 differences determined the likelihood of a phonological feature being supplied in different contexts.
Form-function studies
Before we examine examples of longitudinal studies that have investigated form-function mapping, a word on the meaning of ‘function’ is needed. Tarone (1988) noted that this term covers ‘pragmatic function’ (for example, requests and denials), ‘discourse function’ (for example, topic, and cohesion), ‘semantic function’ (for example, specific and non-specific reference), and ‘grammatical function’ (for example, subject and object). Tarone argued that studying form-function mappings in learners’ interlanguage ‘can reveal the linguistic system hidden in a learner’s apparently unsystematic use’ (1988: 54).
A good example of a form-function study is Huebner’s (1979, 1983) longitudinal study of Ge, a Hmong refugee acquiring English naturalistically in the United States. Huebner was able to show that although Ge’s form-function mappings were not the same as those in the target language (English), they were systematic and they changed over time. Huebner focused on Ge’s use of articles (da and zero). He based his analysis on Bickerton’s (1981) Semantic Wheel. This distinguishes two binary categories of the semantic function of noun phrases:
+/- information assumed to be known by the hearer (HK)
+/- specific referent (SR)
When combined, these yield four types of noun phrases, as shown in Table 5.1.
Type Standard English forms Examples
1 - SR/+ HK ‘the’, ‘a’, or zero ‘Lions are beautiful.’
2 + SR/+ HK ‘the’ ‘Ask the man over there.’
3 + SR/- HK ‘a’ or zero ‘She gave me a present.’
4 - SR/- HK ‘a’ or zero ‘He’s a nice man.’
Table 5.1 Noun phrase types
Initially, Ge used da mainly for (2) (i.e. to mark nouns as specific and known to the hearer). This resulted in apparent target-like accuracy although not completely; when the noun phrase in question functioned as a topic of the sentence Ge used the zero article. A month and a half later he used da for all four types of noun phrase. This constituted a stage where Ge overgeneralized da. Huebner described this stage as ‘flooding’. Five months into the study, he began to eliminate the use of da first from type (4) contexts and at 7 months from type (3) contexts. By the end of the study, Ge was using da at an 80 to 90 per cent level of accuracy. Huebner’s main point is that what would be seen as random use of articles in an obligatory occasion analysis turns out to be largely systematic in a form-function analysis.
However, Huebner’s study also suggests that learners pass through a stage of development where the linguistic forms at their disposal are used in free variation (i.e. randomly). In the ‘flooding’ stage, Ge’s use of da was not categorical. That is, he did not invariably use it with nouns. During this stage of the learner’s development, da appeared to be used randomly and randomly omitted. Huebner also acknowledged that even at the end of his one-year study, there was an unexplained residue of between ten and 20 per cent of non-target use. It is possible of course that there were other factors that constrained the use of da in a systematic way at this time but—to some extent at least—da was used randomly. I will return to the whole issue of whether variability in learner language is systematic or—in part—random later in this chapter.
Schachter (1986) also used a form-function analysis to re-examine the data for one of Cancino et al.’s (1978) learners (Jorge). She investigated whether his use of negative forms was as random as the original authors suggested. She argued that the variability was not explicable in terms of different situational requirements, as the data collection took place on a regular basis with the same situational constraints throughout. However, she found evidence of ‘a rich system, complex from the very beginning, which became even more so as time progressed’ (1986: 123–4). Schachter identified seven functions performed by Jorge’s productive negative utterances and found surprising regularity in his pairing of forms and functions. For example, the formula ‘I don’t know’ was always used to perform the same function of ‘no information’ (i.e. to indicate that the speaker is not in a position to confirm or deny whether something is the case), while ‘no’ + verb carried the ‘denial’ function (i.e. to assert that an actual, supposed, or proposed state of affairs does not hold for the speaker).
Some conclusions
In some respects, the Dynamic Paradigm provides a more insightful account of interlanguage variability than the Labovian paradigm as it demonstrates more clearly how horizontal variability relates to L2 development. The following are the main conclusions that can be drawn from the research:
When a linguistic form emerges in a learner’s interlanguage, it appears first in one linguistic context and then spreads systematically to other contexts.
The diffusion of a new form is also governed by both input frequency and the learner’s first language.
Learners construct form-function systems in the process of learning and using a second language. These systems are likely to differ from the form-function systems found in the target language.
The learner’s form-function systems evolve over time. Thus, at any stage of development, different form-function systems are likely to be observed.
Learners manifest free variation in the use of the linguistic forms at their disposal at some points in their development.
Point 5, however, is controversial as it runs contrary to the underlying assumption of the Labovian paradigm (i.e. that all variability is systematic). However, it is supported by the final approach we will now examine.
Dynamic Systems Theory
Dynamic Systems Theory (DST), like the other approaches to investigating variability, views variation as inherent in language systems but—unlike the other approaches—it does not seek to identify the specific social and linguistic factors that predict variation. Rather, it examines how the interaction of a whole range of external and internal forces contributes to L2 development in individual learners. As de Bot and Larsen-Freeman (2011) put it ‘instead of investigating single variables, we study patterns that emerge from interactions’ (p. 21). DST claims that ‘any account that focuses on one aspect only cannot but provide a gross oversimplification of reality’ (de Bot, Lowie, and Verspoor 2007: 18)NOTE 3.
DST views the L2 learner as a complex system involving complete interconnectedness: that is, ‘all variables are interrelated and therefore changes in one variable will have an impact on all other variables that are part of the system’ (de Bot, Lowie, and Verspoor 2007: 8). Two points follow: (1) systems are inherently dynamic (i.e. constantly changing); (2) it is not possible to predict the outcome of development over time. Another key tenet of the theory is that L2 development is ‘embodied’ (i.e. cognition
is not bounded within the individual but is socially constructed) and ‘situated’ (i.e. cognitive functions can only be understood in terms of the particular setting in which they were carried out). In these respects, DST is very similar to Complexity Theory (see Chapter 1).
I will focus on two assumptions of DST that are of special relevance to interlanguage variation: (1) variation is chaotic; (2) although change is constant, systems may temporarily settle into ‘attractor’ states that are relatively stable. These two characteristics are helpful for understanding two key aspects of interlanguage variation—free variation and systematic variation. Chaotic variation (i.e. variation that is impossible to predict) ‘may be caused by a range of factors, such as physical fatigue, memory overload, or temporary disfunctioning of a part of the brain’ (de Bot and Larsen-Freeman 2011: 15). It is an inevitable product of the system’s flexibility and adaptability to the environment. Chaotic variation, therefore, should not be seen as ‘noise’ (as it is in variational sociolinguistics), but as evidence of ongoing change.
However, at times, a system evolves into a more settled form. An attractor state is defined as ‘the state the system prefers to be in over other states at a particular point in time’ (de Bot and Larsen-Freeman 2011: 14). Such states are not permanent, but—depending on the strength of the attraction—may resist change for a period of time. In terms of L2 development, attractor states might be understood as stages where variability in some aspect of the system becomes temporarily more systematic. In an attractor state, specific linguistic forms that have already emerged in the learner’s system are now organized in such a way that they distinguish the different functions that are communicatively important to a learner. They are supplied more regularly in accordance with constraints that are both inherent in the system and social—as was seen in Ge’s progression from a ‘flooding stage’ to more systematic use of da. Thus, from the perspective of DST, L2 development is a continuous process that involves both free and systematic variation, both of which are found in different parts of the overall system at different times.
Dijk, Verspoor, and Lowie (2011) point out the essential difference between how variability is viewed in traditional sociolinguistic approaches and in DST. The former ‘have been mainly interested in discovering external causes of variability’ whereas ‘a DST approach is interested in variability to discover when and how changes take place in the process of development, how different subsystems develop and interact, and how different learners may have different developmental patterns’ (p. 60–1). They point out that learners do not learn all the subsystems of language at one time and that chaotic variation is not therefore a characteristic of the complete system, but only of some parts of it. For example, at the very beginning stage of L2 acquisition, variation of any kind is largely absent, but once the learner has learned how to construct short, simple sentences, they may switch attention to verb forms and begin to manifest a high level of free variation in the use of these. After some time, attractor states will emerge in the learner’s verb system, creating conditions for the development of some other system—such as complex sentences—which initially will display free variation. The aim of DST research, then, is to investigate how chaotic variation in a system can evolve into a stable ‘attractor’ state.
To illustrate this approach, Dijk et al. reanalysed the data from Cancino et al.NOTE 4. This analysis revealed the following:
The developmental curves for the individual learners investigated are quite different from the averaged curve for the whole group: in other words, inter-learner differences are pronounced.
Two of the learners investigated in detail displayed equal levels of variability, but—whereas in one of the learners the variability was essentially random—in the other, a significant ‘peak’ was evident at a given stage of development, suggesting that this learner had settled into an attractor state.
Dijk et al. concluded that it is misleading to average data for a group of learners and then to claim that there is a common sequence of acquisition. Their re-analysis of the data challenges Cancino et al.’s claim that there was a distinct sequence in the acquisition of English negation. It also suggests that different variability patterns reveal different kinds of development. Free variation occurs when no or very little development is taking place, whereas the presence of a significant peak in a learner’s use of a specific grammatical feature signals that the learner is ready to accommodate new constructions. They also noted that the patterns of variability of the six learners differed according to their age, with the children taking longer to produce peaks in the use of non-target forms than the teenagers, and the adults failing to show peaks to the same extent as the teenagers.
Research based on DST is still in its infancy. It faces several challenges. As de Bot and Larsen-Freeman (2011) noted, it is not clear whether it is valid to study any one linguistic feature (for example, L2 negation) separately from other linguistic features. The theory claims that the development of one subsystem of language interacts with the development of other sub-systems. Thus, how individual learners develop English negation is inextricably linked to the development of their general fluency, accuracy, and complexity in the use of English as a whole. It is difficult to see, then, how Dynamic Systems Theory can ever provide a full account of variability. That said, the theory affords some important insights; in particular, it challenges the assumption of variationist sociolinguistics that free variation is just uninteresting ‘noise’.
Conclusion
In this chapter, we have examined a number of studies that show how social and linguistic factors influence learners’ choice of L2 forms. We have seen that ‘change is only the temporal consequence of current variation’ (Widdowson 1979: 195). This is evident in a number of ways:
Linguistic forms may appear initially in one style (for example, the careful style) and then subsequently spread to another style (for example, the vernacular style).
Linguistic forms may only be supplied in one linguistic context initially, but then spread to other linguistic contexts over time.
Patterns of variability change as learners construct and then reconstruct their form-function systems as new forms enter their repertoire.
Chaotic variation can give way to more systematic variation as learners enter an attractor state, which may in turn constitute a transitional point allowing for further development.
These generalizations, however, are not supported by all the theories of variation we have examined in this chapter. In particular, there is disagreement about both the existence and role of free variation in L2 development and of whether there are universal trends.
The significance of free variation
Variational sociolinguists are generally dismissive of free variation. Preston (1996), for example commented ‘I am suspicious that language variation which is influenced by nothing at all is a chimera’ (p. 25). For Preston, free variation is only apparent because of researchers’ failure to identify the sources of systematicity. He was critical of my own early claims about the role of free variation in L2 development (see Ellis 1985). In contrast, both the Dynamic Paradigm and Dynamic Systems Theory (especially the latter) view free variation as an inherent feature of interlanguage systems and of significance for understanding how linguistic systems are developed.
In Ellis (1999), I drew on Towell, Hawkins, and Bazergui (1996) to propose a general model of how learners acquire specific linguistic forms (see Table 5.2). Free variation occurs in stage three and functions as the precursor of systematic variation in stage four. The model is not intended as a model of interlanguage development as—clearly—learners’ entire interlanguage systems do not progress neatly from one stage to another. Rather, stage-like progress is evident only in specific features. Thus, an individual learner may have reached a different stage for different linguistic features. The model inevitably constitutes an idealization: development is much more uneven and messy than the model suggests, given that sub-linguistic sy
stems interact and affect each other, as proposed by DST. Nevertheless, I would claim that the model captures the way in which linguistic forms enter and then develop in the subsystems that comprise interlanguage.
Stage Description
1 Non-linguistic
Learners operate in accordance with an ‘accept what I am offered’ strategy. During this stage they have no awareness of form-meaning mapping.
2 Acquisition
Learners operate an overgeneralization strategy, selecting one form and using it in contexts that in target-language use would require two forms.
3 Replacement
Learners allow an alternate form into their interlanguage but are unable to determine the functional differences between the two forms. This phase is characterized by free variation.
4 Interlanguage
Learners now begin to use the two forms systematically, but in accordance with interlanguage rather than target language norms. This stage may be characterized by categorical use or systematic variation.
5 Completion
Learners use the two forms in accordance with target language norms, which may be categorical or variable.
Table 5.2 Stages in the development of a grammatical sub-subsystem (based on Towell et al. 1996)