Understanding Second Language Acquisition (2nd ed)

by Rod Ellis

  There is, however, another interface position—the strong interface position – that draws on a symbolist rather than connectionist model of language representation and perceives of language learning as a type of skill learning.

  Skill-learning theory

  Skill-learning theory draws on the distinction between declarative and procedural knowledge and between controlled and automatic processes discussed earlier in this chapter. It is a symbolist theory. It has provided a justification for the presentation-practice-production (PPP) so popular in language teaching (see Chapter 10).

  The theory claims that language learning is like any other form of learning. It involves a progression from an initial declarative knowledge stage involving controlled processing, to a final stage where knowledge is automatic. L2 learners proceduralize their declarative knowledge through extensive practice. DeKeyser (2007) defined practice as ‘specific activities in the second language, engaged in systematically, deliberately, with the goals of developing knowledge of and skills in the second language’ (p.1). However, DeKeyser (1998) argued that the transformation of declarative knowledge into procedural knowledge is only likely to occur when the cognitive operations involved in the practice activity match those in natural communication. That is, it requires practice activities that involve ‘real operating conditions’ and not just decontextualized mechanical practice.

  Skill-learning theory, as explained by DeKeyser, is in part compatible with the idea of Transfer Appropriate Processing (TAP)—the fundamental tenet of which is ‘that we can use what we have learned if the cognitive processes that are active during learning are similar to those that are active during retrieval’ (Lightbown 2008: 27). Lightbown suggests that this can explain why linguistic forms learned in isolation are not available for use in communicative interaction. When learners experience having to use linguistic forms in communication, deeper processing occurs than when attention is purely on surface form. TAP serves as a rationale for communicative and task-based language teaching. However, it does not constitute a clear justification for PPP, as it makes no claims about a direct link between declarative and procedural knowledge and, in this respect, differs from skill-learning theory.

  In another respect, however, skill-learning theory and TAP are more clearly in line. According to skill-learning theory, the knowledge that characterizes the later stages of development—i.e. the procedural stage—is highly specific and so does not transfer to tasks that are dissimilar from those used to develop the knowledge. In other words, the effects of practice are skill specific: practice in listening improves listening, but does not assist speaking and vice versa. This claim directly contradicts VanPatten’s (1996) views about input-processing, according to which practice in comprehending the meaning of specific grammatical structures leads to restructuring of the interlanguage system which can then be drawn on for both reception and production. Skill-learning theory predicts that there is no single knowledge store, but rather different knowledge stores that support receptive and productive language skills. VanPatten’s theory—and, one might add, connectionist accounts of L2 representation—suggest a single store, drawn on irrespective of the language skill involved.

  Skill-learning theory is perhaps best seen as explaining how declarative knowledge becomes automatized rather than how explicit knowledge transforms into implicit. It is difficult to see how skill learning can account for the sequences of acquisition we discussed in Chapter 4 if the starting point is symbolic knowledge of linguistic rules, and if there is no allowance for the automatic tallying of information in implicit learningNOTE 5. DeKeyser (2009) appears to acknowledge this when he comments that it may be the case that ‘highly automatized knowledge is still not necessarily qualitatively the same as implicit knowledge’ (p. 127). He does suggest, however, that such knowledge is functionally equivalent to implicit knowledge—i.e. it enables learners to perform in a near-native way.

  Processability Theory

  A brief account of Processability Theory was provided in Chapter 4. Central to the theory is the claim that the processing strategies are hierarchical in nature and mastered one at a time. As Pienemann (2005: 13) put it ‘it is hypothesized that processing devices will be acquired in their sequence of activation in the production process’. Thus, the failure to master a low-level procedure blocks access to higher-level procedures and makes it impossible for the learner to acquire those grammatical features that depend on them. The theory, then, focuses on production, but has nothing to say about how L2 forms enter the interlanguage system. It was designed to explain the sequence of acquisition of grammatical structures in learner language (see Table 4.5 in Chapter 4). Pienemann (2005) claimed ‘once we can spell out the sequence in which language processing routines develop we can delineate those grammars that are processable at different points of development’ (p. 2).

  Following extensive research into the developmental sequences of different languages, Pienemann (2005) proposed the hierarchical processing routines shown in Table 8.4. The strength of Processability Theory is that it provides not only an explanation for attested sequences of acquisition, but also has predictive power. The cognitive dimension afforded by the identification of underlying processing strategies and operations allows researchers to form hypotheses regarding which grammatical structures will be acquired at different stages of development.

  Level Routine Structures

  1 word/lemma Production based entirely on words/ formulaic chunks which are invariant in form.

  2 category procedure (lexical category) Production based on lexical entries that are now annotated with a number of diacritic features (e.g. ‘possessive’ and ‘number’).

  3 phrasal procedures (head) Production of grammatical phrases that involve matching one element with another now occurs. At this level, learners can handle such structures as articles, plural agreement, and ‘do’ fronting (e.g. ‘Do he like it?’).

  4 s-procedure and word order rule Production involving the exchange of information between structural phrases is now possible. The learner can exchange information between a noun phrase and a verb phrase as required for subject-verb agreement (e.g. ‘Mary lives in London’).

  5 matrix/subordinate clause The final procedure to be acquired enables learners to process the word order of subordinate structures such as that found in embedded questions in English (e.g. ‘He asked where I lived’).

  Table 8.4 Processing routines involved in the production of different grammatical features

  The theory claims that the processing operations are specific to language—rather than of a general cognitive nature—and in this respect, differs from connectionist theories. Pienemann (2011) argued that grammatical processing takes place in a ‘grammatical memory store’. He also proposed that the grammatical memory store was part of procedural—rather than declarative—memory. In other words, Processability Theory is a theory of implicit rather than explicit learning.

  Processability Theory is limited in a number of ways. It relies on emergence as the criterion of acquisition—i.e. a feature is considered acquired when it first appears in learner production—and thus does not account for how learners achieve gradual control over a grammatical structure. Also, it does not account for how learners develop receptive knowledge of grammatical structures. The major limitation, however, is that it tells us nothing about how learners obtain intake from input, and how this is then used to construct and restructure internal grammars. However, the theory is useful in that it complements other theories—such as Van Patten’s Input Processing Principles—that focus exclusively on input-driven learning.

  Complex adaptive systems and L2 acquisition

  Finally we turn to a group of theories: the Competition Model, Complexity Theory, usage-based theories, Dynamic Systems Theory. As these have in common the idea that language is a complex adaptive system I will not consider each of these theories separately but instead consider the general principles that underlie all of them. The theories al
l draw heavily on connectionist and emergentist models. That is, they reject the idea of a language acquisition device and top-down, rule-symbolic processing and instead assert that linguistic systems emerge gradually, driven by the exemplars people are exposed to in social interaction, which are processed by domain-general cognitive mechanisms such as those that regulate attention. I will consider what is meant by ‘complex’ and ‘adaptive’ and in what sense a ‘system’ arises.

  L2 systems are complex

  Like any complex adaptive system, everything in an L2 system is interconnected:

  Complexity theory aims to account for how the interacting parts of a complex system give rise to the system’s collective behaviour and how such a system simultaneously interacts with its environment. (Larsen-Freeman and Cameron 2008: 1).

  Larsen-Freeman and Cameron noted that the term ‘complex’ does not just mean ‘complicated’, but also refers to the idea that ‘its behaviour emerges from the interactions of its components’ (p. 2). A complex adaptive system is complex in the sense that it includes social, psychological, and linguistic components all interacting with each other. An implication for the study of L2 acquisition is that one component of the complex system (say grammar) cannot be understood in isolation from other aspects.

  The components of a complex system are in competition with each other. Beckner et al. (2009) suggest that there is ‘a tug-of-war of conflicting interests between speakers and listeners’ (p. 18). Speakers prize production economy whereas listeners want clarity and explicitness. Beginner L2 speakers necessarily simplify (see Chapter 4), but are under pressure to make the meanings of their messages more explicit to meet the needs of their listeners. Thus there is competition between the need to simplify and the need to elaborate the L2 system.

  There is also competition within the L2 system. The Competition Model (MacWhinney 2001) takes its name from the ‘competition’ that arises from the different devices or cues that signal a particular language function. For example, in a sentence like ‘Mary bit the dog’, there is competition between ‘Mary’ and ‘dog’ for the role of agent. Semantic expectancy would suggest that ‘dog’ is the agent but word order indicates that ‘Mary’ is the agent. The crucial clue in English is word order, which overrides semantic expectancy, but in other languages—for example, Japanese—semantic expectancy or morphological features constitute the stronger cue. L2 learners are influenced by their L1 processing strategies. For Japanese learners to overcome their natural tendency to assign agency on the basis of semantic expectancy, they need to attend to the word order cue to identify ‘Mary’ as the agent of the sentence. Research has shown that the processing strategies utilized by L2 learners are located somewhere on a continuum between the strategies required to process the input and the strategies of the L1 (Harrington 1987).

  A final note: the complexity of a system makes prediction difficult, if not impossible. Just as it is impossible to make reliable predictions about what the weather will be like at any one moment in a particular location, so it is impossible to predict how the interlanguage of an individual learner will develop over time. As Larsen-Freeman (1997) put it ‘the best we can do is to explain the occurrence of change a posteriori, not actually look at the language and make exact predictions of what change will transpire next’ (p. 148).

  Systems are adaptable

  Any linguistic system is open rather than closed. It is constantly changing as a result of the interactions among the components. This is what makes precise predictions difficult as it is not possible to foretell with accuracy how the individual L2 learner will respond to the configuration of elements in the system. A good example of a system’s adaptability is Dörnyei and Ottos’ (1998) process model of motivation (see Chapter 3). L2 learners’ motivation undergoes constant change in response to a variety of influences—for example, their subjective values and norms; their learning experiences; and their perceptions of their progress.

  The adaptability of a system is never-ending. Beckner et al. (2009) noted that ‘even adult grammars are not fixed and static but have the potential to change with experience’ (p. 7). L2 grammars are inherently unstable. Thus, from the perspective of a complex adaptive system, fossilization is not possible, although periods of stabilization can arise. The human brain never ceases responding to the linguistic environment and some change—however small—in the language network will occur.

  Evidence for the adaptability of systems comes from usage-based theories of grammar. These propose that the organization of a grammar is based directly on the learner’s experience with the language. From this perspective, grammar is a process rather than a product. At any one moment in its evolution, it contains memories of how words co-occurred in the interactions the learner participated in and, also, of the probabilities of their occurrence and co-occurrence. Subsequent usage brings about change in these memories.

  A central claim of complex adaptive systems theory is that higher-level properties arise from the interaction of lower-level properties and these higher-level properties are gradable rather than categorical. Consider, for example, how learners learn to attach an -s to the third person singular of present tense verbs in English. Initially, they develop memories of combinations of third-person pronouns and verb + -s (for example, ‘he comes’, ‘she works’). However, at this point the learner may still fail to produce third-person -s verbs following noun subjects (for example, * ‘John come’, * ‘Mary work’). Pronouns belong to a closed class, whereas nouns are members of an open class. As a result of differences in input frequency and contingency, the learner more readily establishes an association of pronoun + verb + -s than one between noun + verb + -s. A higher order category emerges out of the memories of pronoun and specific verb combinations—i.e. ‘he/she’ + verb + -s—leading to regular use of -s following a pronoun. However, this falls short of a still higher-order category that can account for all cases—i.e. subject + verb + -s—which in some L2 learners never emergesNOTE 6.

  Unstable systems

  In what sense, then, can a complex adaptable system claim to be a ‘system’? The essence of a system is that its parts are constituted in such a way that the behaviour that results from the system is systematic. However, as complex adaptable system theories make clear, much of the behaviour that results from the system is non-systematic and chaotic. In Chapter 5—where I introduced Dynamic Systems Theory—we saw that chaotic variation—what I called ‘free-variation—is endemic. But if variation is entirely chaotic, there would be no basis for claiming that there is a ‘system’. The solution to this problem is to propose that—although systems are in flux resulting in unpredictable behaviour—from time to time, they settle into attractor states. De Bot and Larsen-Freeman (2011) defined an attractor state as ‘the state the system prefers to be in over other states at a particular point in time’ (p. 14). Thus, when a complex system evolves into an attractor state, systematic behaviour—probabilistic rather than categorical—will arise until further experience leads to further change. The Basic Variety—i.e. the highly simplified system that beginner learners construct—can be seen as an attractor state. For some learners, it constitutes a stable state, but they can escape from it if the learning conditions change and promote greater attention to formal accuracy. No state is fixed.

  An implication of complex adaptive systems theory is that the systems that individual learners build are different. Learners develop idiolects that reflect their own specific language learning experiences. This raises two key issues. First, if learners’ systems are all different, how does a communal language—i.e. a shared linguistic code—develop? Second, if learners’ systems are idiolects, to what extent is it possible to claim that interlanguages manifest universal properties?

  Beckner et al. (2009) provide an answer to the first question. They argue that while there will be linguistic differences in the connectionist networks of individual learners, reflecting the particular interactions they have participated in, commonalit
ies will develop because to a considerable extent their experiences will be similar—for example, in terms of the frequency with which specific linguistic forms appear in the input—and also because of the inbuilt human tendency to converge towards, rather than diverge from, the usage of our interlocutors (Giles 1971). In other words, we need to talk about both ‘interlanguages’—the systems of individual learners—and ‘interlanguage’—the general properties of the systems of all learners.

  Of course, it is only possible to talk about ‘interlanguage’ as opposed to ‘interlanguages’ if there are some properties that are universal. In Chapter 4, we provided evidence that there are indeed universal properties: for example, in the progression from the Pre-Basic to a Post-Basic variety and in the stages of acquisition that learners pass through in acquiring L2 negation. Complex adaptive systems theory acknowledges these universal properties. They can be explained in terms of the attractor states that arise in the course of acquisition. However, stages of development are not neat, encapsulated affairs. Learners do not progress abruptly by abandoning an earlier stage in favour of a later one. Development is gradual and dynamic, characterized at each stage by variability, some of it chaotic. Also, care needs to be taken not to overstate the universality of developmental patterns. As de Bot, Lowie, and Verspoor (2007) noted ‘it is very well possible that if we look close enough, the general developmental stages individuals go through are much less similar than we have assumed so far’ (p. 19).