Understanding Second Language Acquisition (2nd ed)

by Rod Ellis

  Symbolism

  Symbolist accounts seek to explain acquisition in terms of a set of abstract constructs—i.e. symbols—and the relationships between these constructs. They draw on a range of constructs, different psycholinguistic and linguistic theories.

  Examples of psycholinguistic constructs that figure in symbolist models are those introduced in the preceding chapter—input, comprehension, noticing, working memory, intake, acquisition, and output. Symbolist models are based on an information-processing model of language learning: we perceive features in our environment; we process them in our working memory and sometimes store them in long-term memory; and then use them in output. Information-processing models underscore the research on input and interaction, whilst linguistic constructs—which are derived from linguistic theory—are used to label the architecture of language. In the case of Universal Grammar (UG) they take the form of a set of highly abstract principles which constitute ‘the mechanisms made available by the brain for building mental grammars for specific languages’ (Hawkins 2001: 2). How language is represented in Universal Grammar is considered below.

  Connectionism

  Connectionist accounts of language and L2 acquisition differ from both information-processing theories and UG. They have come to the fore in SLA in recent years, driven in particular by the work of N. Ellis. For connectionists such as Ellis there are no rules, only strengths of associations which merely give the impression of rules underlying behaviour. In connectionist accounts, language is represented not in terms of symbols and rules, but as associations of varying strengths, derived from elements encountered in the input. Knowledge of an L2 consists of the stored memories of previously experienced elements. Grammar emerges gradually out of the labyrinth of these stored associations when certain connections become well-established, leading to rule-like linguistic behaviour.

  Information-processing theories, Universal Grammar, and connectionist theories all address how knowledge of a second language is represented in the human mind. I will begin, therefore, by examining how these cognitive theories account for L2 representation, including the crucial issue of how the mind stores two languages. I will then move on to consider the role of attention before turning to examine—more broadly—general cognitive theories of L2 acquisition. I will conclude the chapter with a brief look at the research methods that have been used to investigate cognitive aspects of L2 acquisition.

  The representation of L2 knowledge

  It is generally accepted that there are two types of linguistic knowledge—implicit and explicit. Both symbolist and connectionist accounts view implicit knowledge as primary as it is the type of knowledge that is acquired naturally during L1 acquisition and is needed for fluent, easy communication. However, no cognitive account of L2 knowledge can be complete without also considering the learners’ explicit knowledge.

  Implicit and explicit knowledge

  The fundamental difference between implicit and explicit knowledge lies in whether learners are aware of what they know: in the case of explicit knowledge, learners have this awareness, but in the case of their implicit knowledge, they do not, even though their language behaviour may demonstrate that they have knowledge. Table 8.1 below provides a more detailed specification of the two types of knowledge.

  Characteristics Implicit knowledge Explicit knowledge

  Consciousness We are not conscious of what we know implicitly; implicit knowledge is only evident in communicative language use. We have conscious knowledge about the ‘facts’ of language (e.g. the meanings of words and grammatical rules).

  Accessibility Implicit knowledge can be accessed effortlessly and rapidly; it is available for automatic processing. Explicit knowledge requires controlled processing and thus can typically only be accessed slowly and applied with difficulty. However, with practice, access can be automatized.

  Verbalization Implicit knowledge cannot be verbalized unless it is made explicit; learners cannot tell what they know implicitly. Explicit knowledge is often verbalizable; learners can report what they know. This calls for knowledge of the metalanguage needed to talk about language.

  Orientation Implicit knowledge is called upon when learners are oriented towards encoding or decoding the meaning of messages in communication. Explicit knowledge is called upon when learners are formulating and monitoring sentences to ensure they conform to target language norms or because they lack implicit knowledge.

  Table 8.1 Distinguishing implicit and explicit knowledge

  Our knowledge of how to ride a bicycle or tie a shoe lace is implicit. We know how to do these, but would struggle to explain how. In contrast, our knowledge of history dates or of mathematical formulae is explicit: we can tell what we know. Likewise, in the case of language, we may be able to automatically add an -s to third person verbs in the present simple tense without any awareness that we are doing so, or we may consciously draw on a rule to remind ourselves that an -s is needed.

  Our actual use of language—whether we are native speakers or L2 learners—can draw on both types of knowledge, but variably so, depending on the type of language use we are engaged in. In everyday conversation, we rely more or less exclusively on our implicit knowledge, but in academic writing, we will probably also need to access our explicit knowledge. The kind of knowledge we utilize depends also on the extent of our implicit and explicit knowledge. L2 learners who have only experienced traditional form-focused instruction may have to rely on their explicit knowledge. In contrast, L2 learners who have picked up a language naturalistically may have very limited explicit knowledge and draw primarily on their implicit knowledge.

  Representation in information-processing models

  The information-processing model that has proved most influential in SLA is Anderson’s Adaptive Control of Thought (ACT) (Anderson 1980; 2005). At the level of representation, this model distinguishes ‘declarative’ and ‘procedural’ knowledge, which closely resemble the implicit/explicit distinctionNOTE 1. Anderson saw the differences between L1 and foreign language speakers in terms of the type of knowledge typically developed. He described foreign language learning in this way:

  We speak the learned language (i.e. the second language) by using general rule-following procedures applied to the rules we have learned, rather than speaking directly, as we do in our native language. Not surprisingly, applying this knowledge is a much slower and more painful process than applying the procedurally encoded knowledge of our own language.

  (1980: 224)

  In other words, native speakers develop full procedural knowledge, whereas L2 learners rely more on declarative knowledge and may not attain the same level of procedural ability as native speakers. However, Anderson acknowledged that foreign language learners are able to use L2 rules without awareness.

  Information-processing models draw on the idea of a limited ‘processing capacity’. McLaughlin’s Information-Processing Model of L2 acquisition (McLaughlin and Heredia 1996) proposes that learners are limited in how much information they are able to process at any one time. Like Anderson, he views language learning as skill-learning. Initially, a skill may be available only through controlled processing, which is demanding on the learner’s processing capacity. Practice results in qualitative changes by enabling learners to assemble discrete elements into chunks, thereby facilitating automatic processing. In so doing, it reduces the burden on the learners’ information-processing capacity.

  These models are symbolic in that they conceive of language representation primarily in terms of rules that differ depending on whether they are subject to controlled or automatic processing. Skehan (1998), however, drew on the idea of limited processing capacity in a somewhat different way. He proposed that L2 learners possess two separate systems that coexist—a rule-based system and an exemplar-based system consisting of ready-made chunks. The rule-based system consists of powerful ‘generative’ rules and is utilized to compute complex and well-formed sentences. The exemplar-based system is requ
ired for fast, fluent, language use. Skehan argued that ‘language users can move between these systems, and do so quite naturally’ (p. 54) depending on the demands placed on their information-processing capacity.

  These information-processing models differ in one key respect. Both Anderson and McLaughlin view the two types of knowledge as connected; that is, declarative knowledge requiring controlled processing transforms through practice into procedural knowledge, which is available for automatic processing. Skehan, however, views the two systems as distinct and disconnectedNOTE 2. These alternative positions regarding representation have implications for how learning takes place. Is procedural knowledge the same as implicit knowledge or is qualitatively different but functionally equivalent as suggested by Husltijn (2002)? Is explicit knowledge the starting point for acquiring implicit knowledge or do the two types of knowledge involve fundamentally different learning processes? We will revisit these questions later in the chapter.

  Representation in Universal Grammar

  Universal Grammar was a term coined by Chomsky (1981) to refer to the abstract knowledge of language which children are claimed to bring to the task of learning their native language and which constrains the shape of the particular grammar they are trying to learnNOTE 3.

  The theoretical case for Universal Grammar is based on a number of assumptions. First, it is argued that the grammar of a language is so complex that it is impossible for anyone to learn it simply through exposure to input and, thus, it is necessary to posit an innate capacity for learning language. The claim is that input alone cannot explain how children acquire their mother tongue and therefore the child must be equipped with knowledge that enables the deficiencies of the input to be overcome. Second, this capacity is highly specific in nature. That is, although it is clearly cognitive, it is a specialized cognitive resource, separate from the general cognitive apparatus involved in other types of learning. Third, Universal Grammar is biologically determined—i.e. it is only found in humans.

  It should be clear that these assumptions are in direct opposition to those of information-processing models, which assume that input and practice suffice for learning a language; that the cognitive mechanisms involved are the same as those involved in any other kind of learning; and—that to some extent at least—these mechanisms can be found in other species—for example, chimpanzees.

  What then is the nature of the abstract knowledge that comprises UG? The specification of this knowledge has changed as linguistic theory has developed over the years. According to one specification, UG consists of linguistic principles and parameters. Principles are the highly abstract properties that underlie the grammatical rules of all languages. Parameters define the restricted ways in which different languages vary. For example, all languages manifest the grammatical category of ‘subject’. However, the ‘null subject parameter’ allows two settings—a plus and a minus setting—for this principle. In some languages, such as Spanish, the subject can be supressed—for example, ‘Es el Presidente de los Estados Unidos’—whereas in other languages, such as English, it is required—‘He is the President of the United States’.

  According to UG, then, such principles and parameters are part of our innate knowledge about the forms that a specific grammar can take. In L1 acquisition, the task facing children is simply to discover which parameter settings apply in the language they are learning. The child ‘knows’ what the possible settings are. In SLA, one of the key questions is whether L2 learners have continued access to Universal Grammar. A number of different views have been advanced; L2 learners have (1) complete access; (2) no access; (3) partial access; and (4) dual access—i.e. adult learners have continued access to UG, but also make use of a general problem-solving module. More recently, however, SLA researchers have moved away from asking whether L2 learners have ‘access’ to UG and focused instead on ‘the nature of the representations that L2 learners achieve’ (White 2003: 27).

  UG is a symbolist theory of implicit knowledge. It has nothing to say about explicit, metalinguistic knowledge. For this reason alone, it can provide only a very limited account of how knowledge is represented in the mind of the L2 learner. It is also limited in another way: it is a property theory and aims to provide an explanation of the ideal speaker’s linguistic competence; it says nothing about how this competence is used in performance or how it differs in individuals. By distinguishing competence from performance in this way—and dismissing performance as of little or no relevance—it limits the role that experience of the L2 plays in the representation of L2 knowledge.

  Representation in connectionism

  Like information-processing models, connectionist accounts of L2 representation make a clear distinction between implicit and explicit knowledge. They assume that ‘humans have separate implicit and explicit memory systems, that there are different types of knowledge of and about language, that these are stored in different areas of the brain, and that different educational experiences generate different types of knowledge’ (N. Ellis 2007: 18).

  Connectionist accounts of language representation treat implicit and explicit knowledge as disassociated. Evidence for this comes from studies of aphasia and amnesia (Paradis 2004). In some cases, people lose access to their explicit memories, in others their implicit memories. An implication of this dissociation for SLA is that L2 learners will differ in the type of knowledge they acquire depending on their learning experiences. Learners with ample exposure to L2 input will be able to develop implicit knowledge. Those whose experiences consist primarily of explicit language instruction in a classroom context will develop explicit knowledge although—as Hulstijn (2002) pointed out—the learning processes responsible for implicit knowledge are unstoppable and so, even in a traditional classroom, some implicit knowledge is acquired. Some learners, of course, develop rich repositories of both types of knowledge.

  Evidence concerning how different learning conditions—for example, education and access to English in the broader community—shape the type of knowledge L2 learners develop comes from a study of different groups of L2 learners by Philp (2009). Using a battery of tests designed to provide relatively separate measures of the two types of knowledge, Philp showed that a group of Malaysian students had developed high levels of both implicit and explicit knowledge, whereas many of the Chinese and Korean learners in her study demonstrated little implicit knowledge, but had high scores on the tests of explicit knowledge.

  In connectionist accounts, implicit knowledge is conceptualized as a complex adaptive system that is in continual flux. It takes the form of an elaborate neural network comprised of connections between elements that do not correspond to any of the symbols or rules that figure in linguistic descriptions of language. To picture this, imagine a series of dots scattered over a page and a computer equipped with software for drawing lines between these dots. Over time, the software programme generates a maze of lines: some of them are faint—suggesting a low weight of connection—others are heavier—suggesting more firmly established connections. Some dots will be connected to only a few other dots, whilst others will be connected to a large number of other dots—some with faint lines, others with stronger lines. As the software continues to run, so the network of dots grows, gradually becoming more complex; new connections emerge, but also some old ones fade and eventually disappear. Of course, the software will need some input that will determine what lines to draw, where to draw them, how faint or heavy they should be, and when some of them should be eliminated. In the case of language, the input consists of the language learner’s encounters with a particular language. The software corresponds to the human capacity to constantly and unconsciously register what has been selectively attended to in the input as the network develops.

  It is common to talk of this kind of network as a ‘system’. However, implicit knowledge can only be thought of as a ‘system’ in the sense that the patterns of connections that emerge become so well-established that they reflect (rather than r
epresent) the categories and rules found in linguistic descriptions. What people learn and store in their implicit memories are ‘memorized sequences’ (N. Ellis 1996)—for example, the formulaic sequences which we noted are so characteristic of learner language in Chapter 4. Memorized sequences exist at every level: at the level of word (as a sequence of sounds); at the level of discourse (as sequences of ready-made lexical phrases); and at the level of grammar (as pre-set sequences of words). However, connectionist theories of language also allow for the emergence of abstract schema that are not exactly ‘rules’, but are ‘rule-like’. The human capacity for language includes the ability to extract abstract categories from memorized sequences. These arise through the unconscious ‘positional analysis of word order’ (N. Ellis 1996: 100). That is, the emerging network recognizes that a certain type of element occurs in a specific slot in a set of formulaic sequences. This element then takes on an abstract value. So something resembling what we refer to as a ‘verb’ emerges when the network recognizes that words such as ‘know’, ‘understand’, ‘want’ are of the same type because they can all fill the empty slot in the sequence beginning ‘I don’t …’. Similarly, something resembling a ‘noun’ emerges from the recognition that ‘pencil’, ‘book’, ‘crayon’, etc. can all complete the sequence ‘Can I have a …?’. As N. Ellis (1996) put it: ‘the abstraction of these regularities is the acquisition of grammar’ and formulas are ‘the concrete seeds of abstract trees’ (p. 111).

  The process of abstraction is primarily unconscious. Learners, however, may choose to dwell on its products and formulate explicit rules to account for them. Thus, implicit knowledge—or rather the language usage that results from it—can serve as a basis for developing explicit representations. In this way, explicit knowledge of a language can arise inductively through the analysis of output derived from implicit knowledge. It can also be taught deductively—as is the case in much traditional language teaching. Explicit knowledge—whether arrived at inductively or deductively—is dependent on a different set of cognitive processes from those involved in implicit knowledge. It arises through the conscious identification of patterns which are stored as declarative representations.