There are good theoretical and educational reasons to place matters of
implicit and explicit learning high on the agenda for SLA research. As for
theoretical motivations, perhaps the most central issue in SLA theory
construction in need of explanation is the differential success in
one's first language (L1) and in one's second language (L2).
Although acquisition of an L1 results in full mastery of the language
(provided that children are exposed to sufficient quantities of input and
do not suffer from mental disabilities), learners of an L2—even
after many years of L2 exposure—differ widely in level of
attainment. How can we explain universal success in the case of L1
acquisition and differential success in the case of L2 acquisition? Among
the many explanations that have been proposed, including brain maturation
and brain adaptation processes (critical period), access to Universal
Grammar, L1 interference, and sociopsychological factors (see Hyltenstam & Abrahamsson, 2003, for a review), one
finds explanations that involve the notions of implicit and explicit
learning. Scholars working in different disciplines, in different
theoretical schools, and sometimes using different terminology have argued
that L1 acquisition (or at least the acquisition of L1 grammar) relies
principally on processes of what we might now call implicit learning,
whereas the acquisition of an L2 often relies on both implicit and
explicit learning (Bley-Vroman, 1991; DeKeyser, 2003; N. Ellis, this issue; R. Ellis, 2004; Krashen,
1981; Reber & Allen, 2000).
As concerns educational motivations, the extent to which implicit and
explicit learning can be shown to explain the differential success of SLA
is likely to determine their relevance for L2 instruction. Curriculum
planners, material designers, teachers, and learners all have a vested
interest in knowing in which linguistic domains L2 learning might best
benefit from implicit or explicit learning modes.
DEFINITIONS
To put the papers brought together in this thematic issue into
perspective, I will make some introductory remarks concerning key
concepts. Following Schmidt (1994a), I will
distinguish between implicit and explicit memory, implicit and
explicit knowledge, implicit and explicit learning,
implicit and explicit instruction, inductive and
deductive learning, and incidental and
intentional learning. The definitions given here are in line with
most, but not all, of the literature. As it is more practical to define
the implicit member of a pair in relation to the explicit member in some
cases, I will define the latter first.
Implicit and Explicit Memory
Explicit and implicit memory is memory of a past event with or without
conscious awareness, respectively (Schacter,
1987). The constructs are usually operationalized in terms of
behavior exhibited in an information-retrieval task. The tasks are
distinguished operationally by the instructions given to the participants.
On explicit memory tasks, participants are explicitly asked to recall past
events or to recognize previously studied events. On implicit memory
tasks, no reference is made to past events; participants are simply asked
to perform the task as accurately and quickly as possible. In general,
studies have found that participants' responses are affected by the
absence or presence of previous events (“priming”), with no
awareness of the effect on the part of the participants. Evidence for the
dissociation between explicit and implicit memory comes from experiments
with patients suffering from retrograde amnesia who perform very poorly on
explicit tasks but almost equally well as normal subjects on implicit
tasks (Jacoby, 1983; Schacter; Roediger, 1990; for a discussion, see Buchner & Wippich, 1998).
Implicit and Explicit Knowledge
Explicit and implicit knowledge differ in the extent to which one has
or has not (respectively) an awareness of the regularities underlying the
information one has knowledge of, and to what extent one can or cannot
(respectively) verbalize these regularities (Anderson
& Lebiere, 1998, p. 5; Bialystok,
1982; R. Ellis, 2004). Explicit and
implicit knowledge are often associated with effortful and automatic
processing, respectively (Hasher & Zacks,
1979; Segalowitz, 2003; Segalowitz & Hulstijn, 2005). Declarative
knowledge is sometimes used as a synonym for explicit knowledge (Anderson
& Lebiere). Knowledge is declarative when subjects can explicitly
declare or verbalize their knowledge. Episodic knowledge (Tulving, 1983) is knowing “when and
where.” (Episodic memory is the behavioral manifestation of episodic
knowledge. They are often used as virtually synonymous.) Episodic
knowledge is a form of autobiographical memory. L2 learners sometimes have
episodic knowledge of new, recently encountered, L2 words or expressions.
This episodic knowledge might also be regarded as a form of explicit
knowledge.
In terms of brain activation, long-term memory of explicit,
declarative, and episodic facts—including many aspects of vocabulary
knowledge—has been claimed to reside in various areas of the
neocortex (especially the frontal and temporal lobes but also the parietal
and occipital lobes). During initial stages of its acquisition (i.e.,
during the consolidation phase), disparate cortical sites associated with
a memory are bound together by the hippocampus in the medial temporal
lobe, which is part of the limbic system. The consolidation phase might
last several weeks or months (Byrnes, 2001, p.
71; Eichenbaum, 2001; Meeter
& Murre, 2004; Squire & Knowlton,
2000; Ullman, 2001, 2004). Implicit knowledge also resides in various
regions of the neocortex (especially the basal forebrain, striatum,
amygdala, and cerebellum) but is not subserved by the hippocampus (Byrnes, 2001, p. 71; Paradis,
1994; Reber, Allen, & Reber, 1999;
Ullman, 2001, 2004).1
Ullman (2001, 2004) did not speak of
explicit and implicit memory or knowledge but of declarative and
procedural memory/knowledge. The declarative system, depending on
medial temporal lobe structures including the hippocampal region,
subserves knowledge of facts and events including word-specific knowledge
(the mental lexicon). The procedural system, depending on
“frontal/basal-ganglia circuits, with a likely role for portions
of parietal cortex, superior temporal cortex, and the cerebellum”
(Ullman, 2004, p. 238), subserves procedural
memory, supporting the learning and execution of motor and cognitive
skills, especially those involving sequences, including the rule-governed
combination of lexical items into complex representations (the mental
grammar). An essential feature of Ullman's declarative/procedural
model is its specific claims concerning the lexical and grammatical types
of linguistic knowledge being subserved by the declarative and procedural
systems, respectively.
Implicit and Explicit Learning
Of all key concepts dealt with in this introduction, explicit and
implicit learning are the two for which the least consensus exists. As I
will explain subsequently in more detail, there are several reasons for
this lack of consensus. For the moment, let me give a definition,
notwithstanding the issues that continue to be debated. Explicit learning
is input processing with the conscious intention to find out whether the
input information contains regularities and, if so, to work out the
concepts and rules with which these regularities can be captured. Implicit
learning is input processing without such an intention, taking place
unconsciously. According to Reber et al. (1999):
Implicit learning (a) operates largely independent of
awareness, (b) is subsumed by neuroanatomical structures distinct from
those that serve explicit, declarative processes, (c) yields memorial
representations that can be either abstract or concrete, (d) is a
relatively robust system that survives psychological, psychiatric, and
neuroanatomical injury, (e) shows relatively little interindividual
variability, and (f) is relatively unaffected by ontogenetic factors. (p.
504)
Note that learning is often defined with reference to the nature of
the knowledge learned. Explicit and implicit learning is then the learning
of explicit and implicit knowledge, respectively. In the pedagogical
literature, explicit and implicit L2 learning are sometimes rather loosely
defined as learning with or without the aid of grammar rules,
respectively.
Implicit and Explicit Instruction
Instruction is explicit or implicit when learners do or do not receive
information concerning rules underlying the input, respectively (R. Ellis, 1994, p. 642; Norris
& Ortega, 2000).2
The definitions
of explicit and implicit instructional treatment, given by Norris and
Ortega (2000), actually comprise, respectively,
a combination of what is here teased apart as implicit instruction and
implicit learning, on the one hand, and explicit instruction and explicit
learning on the other.
Inductive and Deductive Learning
Deductive learning takes place when rules are presented before
examples are provided; inductive learning takes place when examples are
given before rules are presented (DeKeyser,
1995, p. 380). The terms deductive and inductive
learning are used in an instructional context. By definition, deductive
and inductive learning are part of explicit instruction because the
correct rule is always given at some point.
Incidental and Intentional Learning
Intentional learning refers to the learning mode in which participants
are informed, prior to their engagement in a learning task, that they will
be tested afterward on their retention of a particular type of
information. Incidental learning refers to the mode in which participants
are not forewarned of an upcoming retention test for a particular type of
information. Incidental learning is also given a more general
definition—not limited to experimental situations—as the
unintentional picking up of information (see Hulstijn,
2003, for a detailed discussion).
THREE FACTORS THAT CAUSE CONFUSION
Not everyone will agree with these definitions. Especially in the case
of the two labels that form the umbrella for this special
issue—implicit and explicit learning—there exist many
definitions. SLA-oriented reviews of the vast literature on implicit and
explicit learning have been provided by DeKeyser (2003), N. Ellis (1994, this
issue), Paradis (1994), and Schmidt (1994a, 1994b, 2001). Rather than attempting to summarize the
information conveyed in these review papers, I will first focus on three
factors that must be taken into account in theories of explicit and
implicit L2 learning and then argue that it is especially the first of
these (concerned with the regularity or irregularity of the linguistic
phenomenon to be learned and closely related to the differences between
symbolic and subsymbolic accounts of language knowledge) that might have
caused confusion in the explicit-implicit debate. Let us first consider
the following two learning tasks.
In task A, the input information can easily be described in symbolic
terms with noncompeting rules operating on nonoverlapping categories or
features. In task B, the input information cannot be described with
noncompeting, large-scope rules but only with fuzzy categories resulting
from the competition between many cues, differing in availability,
strength, and validity (in terms of the competition model of Bates & MacWhinney, 1989). Explicit
learning—defined as input processing with the intention to find out
whether the data can be described with rules and, if so, to discover the
rules—is possible in task A provided that the rules are not too
complex but is bound to fail in task B. Implicit learning—defined as
an absence of the intentions just mentioned—is a possibility in task
A. However, to the extent that the data are presented in a way that their
underlying structure becomes salient and to the extent that learners have
acquired (in school and elsewhere over time) metalinguistic knowledge and
metacognitive problem solving strategies that facilitate the discovery of
the underlying regularities, input processing in task A might
spontaneously invoke an explicit learning mode.3
The debate in the psychological literature concerning the
question of whether implicit memory does or does not result from a process
of implicit learning is plagued, in my opinion, by the same confusion. In
Reber's implicit learning experiments (Reber &
Allen, 2000; Reber et al., 1991, 1999), the stimuli could be represented by
noncompeting, categorical rules representing a miniature artificial
grammar (as in task A). This was not the case, however, in the
implicit-memory experiments of Jacoby (1983) and
Schacter (1987). These experiments (as in task
B) were concerned with the recall or recognition of individual words or
pictures, whose interrelationship (if there was any) could not be
represented with categorical, syntactic rules.
What this example
aims to illustrate is that definitions of learning—whether implicit
or explicit—as a process (
how) can easily become
contaminated with the object of learning (
what). Different views
on the object of learning will easily lead to different views on its
process. Thus, when making claims about the effect or feasibility of
implicit and explicit learning modes, we must take into account the
possible interaction of at least the following three factors: (a) the
regularity and complexity of the system underlying the data (see N. Ellis,
this issue; Williams, this issue); (b) the frequency and salience with
which any underlying regularity of the data is represented in the input to
which learners are exposed (see N. Ellis, this issue; Williams, this
issue); and (c) learners' individual differences in knowledge,
skills, and information processing styles, which might be beneficial or
detrimental to discovering underlying regularities (Robinson, this issue;
see also
Reber & Allen, 2000;
Reber, Walkenfeld, & Hernstadt, 1991).
The interaction of these factors might determine how learners process
and categorize the input data to which they are exposed. The psychological
literature on input categorization might be highly relevant for the study
of L2 learning. Learners' categorization of the input data might take
place (a) by computing similarities to previously encountered exemplars,
(b) by computing similarities to prototypes, (c) by computing the
frequency of relevant features, (d) by applying rules, or (e) by a
combination of these mechanisms (see Ashby &
Maddox, 1998, for an overview; and Zaki,
Nosofsky, Stanton, & Cohen, 2003, for a recent contribution to
an ongoing debate).
In conclusion, a systematic investigation into the ways in which the
three factors mentioned in this section might interact under explicit and
implicit learning modes should be part of our research agenda, which calls
for collaboration between linguists and psychologists.
ATTENTION
The three factors mentioned in the previous section also play a role,
I believe, in discussions of whether implicit learning requires attention,
consciousness, and awareness (DeKeyser, 2003; N.
Ellis 1994, this issue; Schmidt, 1994a). In the field of SLA, it has now
become customary to accept the noticing hypothesis of Schmidt (2001), which claims that at least some attention to
and awareness of elements of the surface structure of utterances in the
input is necessary for learning to take place. There is currently a debate
in cognitive psychology over the question of whether implicit learning
runs independently of attention or might rely on the same type of
attentional mediation that is often considered to govern explicit learning
processes. In the introduction to a volume dedicated to the controversial
role of attention in implicit learning, Jiménez (2003) suggested that the analysis of the effects of
attention on implicit and explicit learning could “tell us something
very important about which kind of regularities our cognitive systems are
prepared to capture immediately, and which other contingencies can be
grasped exclusively by relying on a series of strategic,
resource-demanding, and conscious recoding operations” (pp.
6–7). Hence, the question of attention still stands as an essential
topic. However, its investigation must take into account the factors
mentioned in the previous section.
SYMBOLIC AND SUBSYMBOLIC ACCOUNTS OF
REGULARITY AND IRREGULARITY IN LANGUAGE
Factor one, which concerns regularity and complexity in form-meaning
relationships of the input data, perhaps contributes the most to the lack
of consensus in the literature on implicit and explicit learning. Natural
languages are particularly intriguing insofar as they are multifaceted
phenomena that defy simple definitions. A machine language could consist
of (a) a set of well-defined form-meaning units (lexical items) and (b) a
set of syntactic rules, with which grammatically well-formed and
semantically unambiguous strings of units can be formed. Natural
languages, however, are characterized by the absence of a one-to-one
relationship between form and meaning, from the morpheme level all the way
up to the text level. Both the lexicon and the grammar of natural
languages contain, on the one hand, too many irregular form-meaning
phenomena to allow a comprehensive characterization by means of rules
operating on categories and, on the other hand, too many regular
form-meaning phenomena to represent them simply as a large unstructured
set of items. Thus, on the one hand, grammars can be seen as governed by
abstract principles with great generality (e.g., the structure-dependency
principle, the projection principle, and the subjacency principle in
generative linguistics), but, on the other hand, grammar rules without
exceptions hardly exist (Givón, 1999).
Theories of the representation of linguistic knowledge must reflect this
competition between regularity and irregularity in one way or another.
Theories of the symbolist school attempt to do this with rules that
operate on categories (the grammar) and an inventory of categories and
category members (items). Theories of the subsymbolic school represent
knowledge in the form of small, meaningless units interconnected in an
enormous network that is in a state of permanent flux, the activation and
inhibition of internode connections resulting from verbal communication
(Daelemans & de Smedt, 1996; Hulstijn, 2002; see also Williams, this issue).
It is useful to keep the symbolic and subsymbolic approaches to
knowledge representation in mind when reading the literature on implicit
and explicit language learning. Of course, when starting to think about
implicit and explicit learning, one would begin, as did Schmidt (1994a), to distinguish between the product and the
process of learning. One could then distinguish between implicit and
explicit knowledge and between implicit and explicit learning and then
argue that—in principle—these are orthogonal dimensions.
However, in the view that linguistic cognition is at least partially
subsymbolic, for an individual to have explicit knowledge of the
architecture of that subsymbolic portion of his or her cognition would
amount to being a neurocognitive scientist having advanced skills in
programming parallel-distributed processing systems. Explicit
learning—defined as an intentional effort to uncover the rules of
the system underlying the input data—must then be seen as a
fruitless process that is bound to fail. On the other hand, in the view
that linguistic cognition is symbolic (at least partially), it would make
sense to argue that an individual can have either implicit or explicit
knowledge of the rules and categories of the system and, hence, that the
adoption of an explicit learning mode is, in principle, a viable
option.4
For a discussion of the question of
how regularities and irregularities are distributed in the syntax and
lexicon of a mental grammar, see, for instance, Jackendoff (2002, p. 57 and Chapter 6). For some fascinating
speculations as to how differential representations of symbolic and
subsymbolic knowledge might interface in early and later interlanguages,
see Towell (2003).
When we look at this issue from a less principled but more practical
perspective, a picture emerges characteristic of L2 pedagogical grammars,
L2 learners, and L2 classrooms all over the world: Although natural
languages can only insufficiently be described with categorical rules,
teachers and textbooks provide learners with categorical rules
(“rules of thumb,” as Krashen (1981,
p. 114) called them). L2 learners, to the extent that they know these
rules, can be said to have explicit, metalinguistic knowledge of the L2,
which might coexist with implicit knowledge (Hulstijn,
2002; Macnamara, 1973).
In conclusion, given that there are different views regarding the type
of architecture to best represent linguistic knowledge, we should not be
surprised to see definitions of implicit and explicit learning (the
process) that are influenced by views on knowledge representation (the
object or product of learning).
INDIVIDUAL DIFFERENCES IN IMPLICIT AND
EXPLICIT LEARNING
Robinson (this issue) examines the influence of individual differences
in language learning aptitude, intelligence, and working memory on rule
and instance learning for learners in what he calls incidental and
implicit conditions, finding areas of both similarity and difference.
Robinson concluded, on the basis of his results, that individual
differences are influential on both instance-based and rule-based implicit
and incidental learning. Furthermore, this study suggested that these two
learning conditions and input stimuli draw on related but separable
learning processes. Investigations like the one conducted by Robinson have
the potential to substantially deepen our understanding of L1 and L2
learning. Most of the SLA literature has treated individual differences as
additional, mediating variables rather than as intrinsically associated to
the fundamental issue of the learnability of language. If language
aptitude, intelligence, and working memory can be conceptually related to
the constructs of implicit and explicit learning and knowledge, their
status might change from peripheral and correlational to central and
causal. However, to make this enterprise successful, the rather general
notions of aptitude and intelligence will need to be broken down into
their components. The role of differences in working memory in implicit
and explicit learning is also addressed by N. Ellis (this issue).
Furthermore, according to R. Ellis (personal communication, December 26,
2003), individual differences in learner orientation must be separated
from developmental factors that influence learners' ability to
process specific information as implicit or explicit knowledge.
THEORY CONSTRUCTION AND DEVELOPMENT:
TOP-DOWN AND BOTTOM-UP
The titles and abstracts of the papers in this thematic issue speak
for themselves. There is no need to summarize their contents here.
However, it is appropriate to end this introduction with some remarks on
potential pitfalls concerning theory construction and development in the
years to come. There are lessons to be learned from history.
Almost 20 years ago, Krashen (1981)
formulated his Monitor Theory of SLA, centered around two central
constructs: acquisition and learning. One of the criticisms leveled
against Monitor Theory was that it failed to provide precise definitions
of its two main constructs (acquisition and learning), thus preventing
researchers from operationalizing them and putting the hypotheses of
Monitor Theory to the test (Gregg, 1984; McLaughlin, 1978). In principle, this criticism was
correct. However, as Jordan (2004) noticed
(referring to Nicola, 1991), when we look at the
history of science, it appears that many scientific theories, in their
formative years, used poorly defined constructs in rather vague laws.
Newton, when he launched his theory that explained why the Moon travels
around the Earth, was not able to provide a precise definition of the
proposed construct of gravity. Thus, initially, his theory violated the
principle of falsifiability. Only in later years were Newton's
followers able to develop the construct of gravity conceptually and
empirically. Unfortunately, Krashen's constructs of acquisition and
learning have fared less well. Since the launch of Monitor Theory, which,
at the time, had a high potential for success and fruitfulness, nobody has
been able to give definitions of the notions of acquisition and learning
that would render Monitor Theory testable (with the exception, perhaps, of
Schwartz (1993) and Paradis (1994), although Paradis treated acquisition and
learning as implicit and explicit knowledge, respectively). To avoid a
similar fate in the case of the notions of implicit and explicit learning,
with which we are concerned here, it is crucial that we make simultaneous
progress on two fronts: theory development and empirical testing. If we
continue to focus on the conceptual and speculative aspects of theory
construction, neglecting measurement issues, theories of implicit and
explicit L2 learning will not survive. For example, the potentially
important issue of an interface between implicit and explicit knowledge
(or, in Krashen's terms, between acquired and learned knowledge),
known in the literature as the strong interface, weak interface, and no
interface positions (R. Ellis, 1993; Hulstijn & De Graaff, 1994; see N. Ellis, this
issue, for a fresh look at this issue), might then become untestable (as
suggested by Hulstijn, 2002). On the other hand,
however, the history of science also shows that theories, and the
definitions of their key constructs, are adapted as new empirical findings
are being produced. In summary, progress in our understanding of SLA is
best served by a cycle of top-down theoretical and bottom-up empirical
work, avoiding not only overly strong forms of (exclusively
concept-driven) rationalism but also overly strong forms of (exclusively
data-driven) empiricism.
For the moment, it appears that we should first be concerned with the
empirical side of implicit and explicit learning and knowledge. To reach
this goal, the papers in this issue make an important contribution. R.
Ellis presents concrete proposals of how to operationalize implicit and
explicit knowledge by means of various tests. We might disagree with his
operationalizations or with his interpretation of the outcome of his
factor analyses, but, regardless of future empirical or theoretical work,
Ellis's paper signals a crucial moment in rendering theories of
implicit and explicit knowledge and learning testable. It is not unlikely
that the SLA field will now enter a phase marked by questions of validity,
reminiscent of the debate concerning the definition and testing of the
notion of intelligence. We could soon witness discussions of
construct-definition claims such as “Implicit knowledge of a L2 is
what task X measures.” I would welcome such discussions as part and
parcel of normal science.
The advent of technologies with which we can look into the brain (such
as the measurement of event-related brain potentials, or ERPs) will add
depth to the measurement issue, as we might now be able to compare the
data obtained with behavioral measures (such as the tasks used in the
contributions by R. Ellis, De Jong, Robinson, and Williams) with those
elicited with neurophysiological measures, as in the contribution of
Tokowitz and MacWhinney. The latter authors argue that comparing ERP and
behavioral data might provide a sensitive method for disentangling
implicit and explicit knowledge. At the same time, Tokowitz and
MacWhinney's study shows that there are difficulties to overcome with
respect to data elicitation techniques and statistical analysis in ERP
studies comparing the processing of L1 and L2 stimuli. R. Ellis sees a
need for cross-validating behavioral measures with neurophysiological
measures. Meanwhile, his study attempted to explore to what extent it is
possible to distinguish implicit from explicit knowledge on the basis of
behavioral measures. Thus, the present papers by R. Ellis and Tokowitz and
MacWhinney already provide empirical data for the debate alluded to in the
previous paragraph.
The papers in this issue illustrate how the construction and
development of a theory of implicit and explicit learning and knowledge
that attempts to explain the differential success of L1 and L2 acquisition
can move forward on both the conceptual and the empirical fronts, avoiding
the pitfalls of extreme rationalism and extreme empiricism while producing
ideas and data on which we can base further research.
