Implicit and Explicit Learning, Knowledge and Instruction

ROD ELLIS

Introduction

The distinctions relating to implicit/explicit learning and knowledge originated in cognitive psychology, so it is appropriate to begin our examination of them with reference to this field of enquiry. Cognitive psychologists distinguish implicit and explicit learning in two principal ways:

(1) Implicit learning proceeds without making demands on central attentional resources. As N. Ellis (2008: 125) puts it, 'generalizations arise from conspiracies of memorized utterances collaborating in productive schematic linguistic productions'. Thus, the resulting knowledge is subsymbolic, reflecting statistical sensitivity to the structure of the learned material. In contrast, explicit learning typically involves memorizing a series of successive facts and thus makes heavy demands on working memory. As a result, it takes place consciously and results in knowledge that is symbolic in nature (i.e. it is represented in explicit form).

(2) In the case of implicit learning, learners remain unaware of the learning that has taken place, although it is evident in the behavioral responses they make. Thus, learners cannot verbalize what they have learned. In the case of explicit learning, learners are aware that they have learned something and can verbalize what they have learned.

The focus of research in cognitive psychology has been on whether implicit learning can take place, and, if it does, how it can best be explained. However, since Reber's (1976) seminal study of implicit learning, there has been an ongoing debate about the validity of his 'multiple learning systems' view of human cognition. Many researchers dispute the existence of multiple systems and argue in favor of a single system that is capable of achieving different learning outcomes.

This controversy within cognitive psychology is very clearly evident in a collection of papers addressing the role of consciousness in learning (Jimenez, 2003). In the opening paper, Shanks (2003) critiqued the research that used a technique known as 'sequential reaction time' to stake out the claim for multiple, differentiated learning systems. In studies using this technique, the time it takes for people to respond to an array of predictable visual information is compared to the time it takes when this array is suddenly disturbed. The claim here is that a difference in response times demonstrates that some learning must have taken place implicitly prior to the disturbance, even though the participants involved were unable to verbalize what they had learned. Shanks (2003: 38) argued that 'previous research has failed to demonstrate convincingly that above-chance sequence knowledge can be accompanied by null awareness when the latter is indexed by objective measures such as recognition'. He concluded that there was no convincing evidence that implicit learning is functionally or neurally separate from explicit learning and that it was misguided to look for such dissociation. He advanced the alternative view that there is a single knowledge source that underlies performance and that apparent differences in performance are due to 'subtle differences between the retrieval processes recruited by the tests' (p. 36).

In contrast, other papers in the same collection argued strongly for distinguishing the two types of learning. Wallach and Lebiere (2003), for example, developed a strong argument for a dual learning system based on the central concepts of ACT-R cognitive architecture (Anderson & Lebiere, 1998). This proposes a hybrid learning system consisting of a permanent procedural memory and a permanent declarative memory. The former consists of condition-action rules called 'productions' that enable a certain action to be performed provided that specific conditions have been met. Such 'productions' operate automatically. Declarative knowledge consists of factual knowledge stored as chunks organized into schemas. It operates in a more controlled fashion and with awareness. Wallach and Lebiere claimed that these two 'architectural mechanisms' could account for implicit and explicit learning and, crucially, the interplay between the two systems. They went on to demonstrate how they can account for the findings of a number of previous studies of implicit/explicit learning. The ACT-R model has also proved influential in second language acquisition (SLA) studies (see, e.g. DeKeyser, 2007).

In the same collection, Hazeltine and Ivry (2003) mustered neuropsychological evidence to support the existence of distinct learning systems. They reviewed studies of the neural activity when people are engaged in sequence learning. They noted that although such activity has been observed in regions across the whole brain, differences in task conditions result in distinct sets of neural regions becoming activated. When the learning task is complex (i.e. involves dual-task conditions) and thus favors implicit learning mechanisms, the medial supplementary motor area, parietal regions and the basal ganglia are involved. In contrast, when the task is simpler (i.e. involving single-task conditions), the prefrontal and premotor cortex are activated.

The controversy evident in cognitive psychology is mirrored in SLA. The clearest example of this can be...