by Rod Ellis
Sometimes style shifting in learner language is influenced by learners’ explicit knowledge of L2 rules. Regan (1996) investigated the style shifting in ne omission in the negative constructions of seven advanced learners of L2 French. She collected data before the learners spent time in France and after they returned. At time 1, the learners deleted ne more in their casual style than in their careful style. However, style made less of a difference at time 2 (one year later). This was because the learners now omitted it more regularly in their careful style. Regan suggests that this may have occurred because the learners believed that it was acceptable to delete ne and so extended this to their careful speech. By time 2—when their deletion rates were very close to those of native speakers—they appear to have abandoned this pedagogical rule. In other words, they had learned an explicit rule—’when you are being formal in French you retain ne but in casual speech you delete it’—but, as a result of exposure to French in France, rejected it.
Regan’s (1996) study investigated advanced L2 learners. But do less-proficient learners style-shift? Dewaele (2004) drew on a study by Dewaele and Regan (2002) of a mixed proficiency group of Dutch learners of L2 French to suggest that there is a U-shaped pattern of development in the omission of ne. Beginnerand intermediate-level learners manifest a simplification strategy; they omit ne categorically in favour of the more salient pas in their negative utterances. Intermediate-level learners produce ne + verb + pas constructions categorically. Only very advanced learners demonstrate style-shifting, omitting ne in the casual style. In other words, omission of ne occurs regularly with beginner level learners, rarely with low-advanced learners, and then at a higher rate with very advanced learners. Thus, it seems that style-shifting only becomes apparent when learners have achieved sufficient proficiency.
These studies suggest that style shifting in learner language is a complex phenomenon. Sometimes the same pattern of style shifting as in native-speaker speech occurs. However, learners also sometimes style shift where native speakers do not, influenced perhaps by the explicit instruction they have received. Style shifting does not always occur when it might be expected. Sometimes it results in greater use of the target variant in careful than in casual speech. Also, native-like style shifting may only be possible when learners have attained knowledge of the variant forms and internalized the sociolinguistic rules that govern their use.
The situational context covers a whole host of factors (for example, time, topic, purpose, and tone). SLA research has examined only a few of these situational factors. Thus, although the studies I have considered were informed by a sociolinguistic model designed to account for how situational factors affect learners’ language use, they have shed only limited light on this. The narrow focus on stylistic variation is also problematic for another reason. Style-shifting draws heavily on the notion of ‘attention’ which, arguably, is more of a psycholinguistic than a social phenomenon.
The effects of linguistic context
There is much clearer evidence of the effect that linguistic context has on learners’ use of variant forms. Preston (2002) observed ‘linguistic influences are nearly always probabilistically heavier than sociocultural influences’ (p. 146).
Dickerson’s (1975) study mentioned above also investigated the effects of linguistic context on Japanese learners’ production of English /z/. Dickerson examined four different phonological environments and found that in the dialogue reading task, the learners used the correct target-language form whenever /z/ was followed by a vowel, but were progressively less accurate in the other environments (for example, when /z/ was followed by consonants such as /m/ and /b/). Dickerson collected data at three different points in time and found that—although the learners improved in their ability to use target-language /z/ in the more difficult linguistic contexts over time—the environmental effects observed at time 1 were also evident at time 3. Dickerson’s study suggests that the linguistic environment has a prevailing effect on learners’ use of a target language phonological form.
The complex effect that the linguistic environment has on L2 phonological features is demonstrated in Bayley’s (1996) study of /t/ and /d/ deletion at the end of English words by Chinese learners of English. Bayley found that the learners, like native speakers, were less likely to delete /t/ or /d/ following a liquid (for example, /l/) than an obstruent (for example, /s/) or a nasal (for example, /n/). However, unlike native speakers, the preceding phonological environment had no effect. The learners also differed from native speakers in another way. /t/ and /d/ deletion was affected by the grammatical category of a word: it was more likely in uninflected words like ‘mist’ than in regular past tense verbs such as ‘missed’. Thus, both phonological and grammatical processes were involved interactively. Whereas semi-weak irregular verbs (for example, ‘left’) were subject to both processes, the past tense marking of strong verbs (‘talked’) was affected only by the grammatical process. Bayley concluded that ‘Chinese-English interlanguage diverges from native-speaker English when competing rules are involved’ (p. 116).
Hansen-Edwards (2011) reported similar results. She collected data from two conversational interviews with limited-proficiency Chinese learners of English conducted six months apart. She found that following linguistic constraints on /t/ and /d/ deletion were more powerful than preceding constraints. Overall, the learners displayed stable patterns of deletion with little change over. None of the learners had fully acquired target-like deletion patterns, but they did manifest native-like use of deletion with some patterns. Hansen-Edwards suggested that the reason for this may have been L1 transfer. She wondered whether L2 learners ever fully acquire native-like patterns, a point also raised in Chapter 2 when we looked at the role of age.
Other studies have shown that the linguistic context also affects grammatical variables. In Ellis (1988), I investigated the effects of linguistic context on copula -s in the English speech of three classroom learners (one Portuguese and two Punjabi-speaking) over a two-year period. The learners used three variants of copula -s (zero, full, and contracted). The target-language variants (full and contracted copula) were used more consistently when the preceding subject was a pronoun than when the preceding subject was a noun. Conversely, the interlanguage variant (zero copula) occurred more frequently when the subject was a noun. In other words, emergence of the target-language variants occurred initially in environments involving a closed class of items and then spread to those involving an open class. However, none of the learners achieved the same pattern of variability for copula use found in native-speaker speech even after two years. In other words, they did not acquire the target-language variable rule.
These studies indicate that learner language is influenced by the linguistic environment, sometimes in the same way as in native-speaker language use, but at other times in unique ways. They also show that, over time, learners approximate more closely to the pattern of variability evident in native speakers but that, even at very advanced levels, learners may not manifest exactly the same pattern of variability as native speakers. As Bayley noted in the conclusion to his study, ‘even highly proficient learners … are unlikely to perform like native speakers’ (1996: 116). The explanation for this may lie in the influence of the learners’ first language, the inherent difficulty of some target forms, and the learner’s starting age.
Multiple influences on L2 variation
Bayley (2005) is critical of much of the research that has investigated interlanguage variation because it has sought a single, over-arching explanation. This is true of the studies we have considered above—they investigated either single factors affecting stylistic variation or the effects of a limited set of linguistic factors. It is clear, however, that interlanguage variation is subject to multiple influences and that variationist research ideally needs to adopt the ‘principle of multiple causes’ (Young and Bayley 1996: 3).
One of the earliest studies to adopt a multi-factor approach is Young’s (1988) st
udy of one linguistic variable—plural -s—in the speech of twelve Chinese learners of English. The learners were interviewed in English twice, once by a native English speaker and the second time by a fellow Chinese speaker. Data relating to four general factors were obtained: (1) the context of the situation (in particular the extent to which each learner converged with the interviewers in terms of social factors such as ethnicity, sex, education, and occupation); (2) the subjects’ proficiency in English (whether ‘high’ or ‘low’); (3) the linguistic context (whether the plural nouns were definite or animate, whether the noun phrase was definite or indefinite, and the phonological environment), and (4) redundancy in plural marking (whether plural -s was omitted because plurality was indicated by some other linguistic device, such as a numeral). Young found that all four general factors accounted for the variability present in the data. However, some specific factors (for example, the definiteness of a noun phrase and the following phonological segment) had no effect on plural -s marking, irrespective of the learners’ proficiency. Also, there were differences between the high- and low-proficiency learners for some of the factors. For example, the preceding phonological segment influenced plural marking only for the low-proficiency learners whereas social convergence with an interlocutor was only a factor for the high-proficiency learners.
Dewaele (2004) reported a multi-factor study investigating L2 learners’ and native speakers’ deletion of ne in negative utterances. In this study, the learners took it in turns to interview each other and were also interviewed by a native speaker. Dewaele focused on exogeneous variables (i.e. age, gender, and frequency of speaking French) and an endogenous variable (i.e. whether the interlocutor was another learner or a native speaker). Out of a total of 991 negations, there were 331 cases of omission of ne. Neither the age nor the gender of the learners had an effect on omission. A personality variable—extraversion—had a marginal effect. The main factor was the frequency of speaking French. Those learners who reported using French more frequently outside of the classroom were much more likely to omit ne. Also, interestingly, omission rates were much lower in the interviews between learners than in the interviews with a native speakerNOTE 2.
Several studies have investigated the factors influencing tense marking on verbs in different languages. Low-proficiency learners commonly over-extend the present form of a verb in past-time contexts and—even when past-time forms begin to be used—they vary in non-target-like ways. Two factors that have been shown to be influential are the inherent lexical aspect of a verb and discourse grounding. The Aspect Hypothesis (Andersen 1991) predicts that past-tense marking depends on whether a verb refers to a state (for example, ‘be’ or ‘like’), an activity (for example, ‘play’ or ‘read’), an accomplishment (for example, ‘paint a fence’) or an achievement (for example, ‘die’ or ‘close’). Learners have been found to mark achievement and accomplishment verbs for past-time more consistently than state or activity verbs. The Discourse Hypothesis claims that past-tense marking in narratives is more likely to occur when the event is foregrounded than when it is backgrounded. Again, several studies have demonstrated that this is the case. Most of the studies, however, have investigated the predictions of the Aspect and Discourse Hypotheses separately. Howard (2004) investigated both hypotheses by examining temporal marking on verbs by Irish learners of L2 French. He first confirmed the predictions of the two hypotheses and then examined whether the two factors ‘joined forces’. The learners were more likely to employ perfective tense marking in contexts involving an achievement verb and a foregrounded event and least likely in contexts involving a state verb and a backgrounded event.
These studies testify to the importance of examining multiple factors affecting interlanguage variation. They demonstrate that interlanguage variation is multidimensional. It is the product of the complex interplay of a range of social, psychological, linguistic, and discourse factors, each of which contributes to the pattern of variability to a greater or lesser extent. Only by examining multiple factors is it possible to determine the constraints exerted by different factors, how the constraints interact, and whether they operate in the same way at different levels of second language proficiency.
The Dynamic Paradigm
In the Dynamic Paradigm (Bickerton 1975), variation is accounted for in terms of the ‘varieties’ evident in different speech communities. Language variation arises as the result of the spread of a rule from one variety to another over time. A rule spreads through a given speech community in waves in two ways. First, the rule manifests itself in the variety of one group of speakers and then spreads to another group while—at the same time—the first group may have introduced a new rule which exists in competition with the first rule, resulting in variation. Second, the spread of a new rule is influenced by the linguistic environment, with some environments favouring the application of the new feature, and other environments continuing to favour the old feature. The Dynamic Paradigm, then, assumes that a language variety is systematic, but also dynamic and that variation is the precursor of change.
The Dynamic Paradigm evolved out of research investigating how creole languages develop over time. A creole has several varieties (or ‘lects’ as Bickerton 1975, called them). The lects can be arranged on a continuum from the basilect (the simplest, most basic variety) to the acrolect (the most complex variety) with the mesolect intermediate. A creole grammar is polylectal as it manifests features from different lects.
Bickerton was interested in the group varieties of creole speakers, but the Dynamic Paradigm can also account for variation in individual L2 learners—whose linguistic systems are more open to rapid change than the lects that comprise the creole continuum—and, thus, more likely to display variation. The Dynamic Paradigm also allows for the possibility of free variation (i.e. the non-systematic use of two or more variants) as it acknowledges that—for a short period of time when a learner acquires a new feature—this is likely to exist alongside a previously learned feature.
Whereas the Labovian approach employed variables rules to account for the factors that systematically induce variability, research based on the Dynamic Paradigm makes use of implicational scaling. This plots the variation that occurs at any one time (i.e. in cross-sectional data) to order the varieties of different learners according to how complex they are. It assumes that simpler varieties are developmentally prior to more complex varieties. Thus implicational scaling relates horizontal variability (i.e. the variability evident at a particular time) to vertical variability (i.e. stages of L2 development). An example of vertical scaling is provided below.
The Dynamic Paradigm also affords a way of investigating variability in form-function systems. In any language variety, specific linguistic forms are used to realize specific functions. For example, in standard English verb + -ing can be used to refer to an ongoing action (for example, ‘I’m eating at the moment’) or to planned future action (for example, ‘I’m flying to London tomorrow’). Learning a second language therefore involves learning the particular form-function mappings that characterize the variety the learner is targeting. Learners, however, develop their own unique form-function systems. The Dynamic paradigm offers a way of exploring how changes in form-function mappings occur over time.
SLA research based on the Dynamic Paradigm
We will examine two strands of research that have drawn on the Dynamic Paradigm. The first is a direct application of the Dynamic Paradigm using the same methodology as in the creole studies. The second involves the study of learners’ evolving form-function systems.
Gradual Diffusion Model
Gatbonton’s (1978) Gradual Diffusion Model proposes two broad stages of L2 development. In the ‘acquisition stage’, the learner begins by using a single form in all situations and contexts, and then subsequently introduces another form which is used alongside the first in all contexts (i.e. it is in free variation). In the ‘replacement phase’, each form is restricted to specific contexts o
f use as learners gradually and systematically eliminate competing forms from these contexts. Gatbonton developed this model to account for the patterns of variation, which she found in the production of three phonological features (/θ/, /ð/, and /h/) in the speech of 27 French-Canadian learners of English, elicited by means of reading-aloud and spontaneous-speaking tasks. Gatbonton found that in learners in the ‘acquisition stage’, a single variant or free variation was evident, whereas in the ‘replacement phase’, systematic variation in accordance with linguistic context occurred.
In a follow-up study, Trofimovich, Gatbonton, and Segalowitz (2007) collected data from 40 adult francophone speakers of L2 English. They reported that their production of /ð/ followed the dynamic and implicational pattern predicted by Gatbonton’s Gradual Diffusion Model. That is, target renditions of /ð/ replaced variable and non-target ones systematically, beginning in easy contexts and then spreading to difficult ones. This is shown in the implicational scaling in Figure 5.2. Trofimovich et al. found that the learners could be ordered in terms of the contexts in which they produced /ð/. Thus, the learners at stage one failed to produce it in any context; those at stage two produced it variably in one or more contexts; those at stage three began to produce it categorically in some contexts; and those at stage four were producing is categorically in all the contexts.