Introduction

Research in cognitive science may focus on one or more "levels of explanation": the metaphorical level, drawing on common sense; the conceptual level, in which verbal concepts are defined as part of a network of relationships (neural, molecular or even biochemical level, etc.).

The research and applications presented in this book are situated on the conceptual level, where the human cognitive system is seen as a system manipulating symbols. The concepts used at this level are structured by a network of relationships, and the expected observables obtained from the symbol manipulation process may be relatively complex behaviors. A classic example would be the series of calculations produced by a high-school freshman solving a factorization problem in math.

We make no attempt to precisely define the term "symbol", the subject of intense debates throughout the 1970s and 1980s which remain unresolved to this day. We shall simply observe that at a "finer" level of explanation, for example neural level, cognitive processes are described using concepts of "activation", "inhibition", etc.; on the other hand, neither neurons nor neural areas "manipulate" anything. At the explanatory level chosen here, a symbol is a representation of a thing. For example, the word "hand" represents a part of the human body; this body part may also be represented by a drawing, a sign, etc.

The level of symbol manipulation considered in the works of this series comes after the treatment of sensory data "of environmental or linguistic origin" (Richard, 2004). In other terms, the data that we consider here is already the result of manipulations of symbols at a more elementary level. For example, consider the perceptive data that makes up the two patterns of rods shown in Figure I.1.

Figure I.1. Perceptive data

These may be apprehended, by a baby, as a configuration of lines; the reader of this book, however, will see them as two French words, "viens" and "ici". These words result from several stages of processing; to mention just two, these include 1) the process in which a configuration of lines is interpreted as a configuration of letters, and 2) the process in which these words are interpreted to give the information "viens" and "ici". These two words may then be interpreted as a phrase, and, depending on the context, as a phrase signifying an instruction ("come here"), etc. This places us on the symbol manipulation level qualified by Richard (2004) as "mental activities".

Note that this example suggests that any cognitive activity, at a given moment, requires a certain level of prior knowledge, whether acquired by the individual or the species. As we shall see1, no knowledge is acquired ex nihilo.

The mechanisms and cognitive processes that we shall use to describe observable behaviors draw on a metaphor in which the human cognitive system is considered as an information processing system (IPS). In this book, no attempt will be made to take account of two crucial factors at play in all human cognitive activity, namely the social environment and the state of motivation of the system during a cognitive activity. Cognitive psychologists highlight the essential role of these factors, and extensive research has been (and continues to be) carried out on these themes2, but the work in question has yet to reach the stage of formalization.

We cannot observe the operation of an IPS itself, but only the system "input" and "output", as shown below:

Input [observable data originating in the external environment]

==> [INTERNAL PROCESSING SYSTEM, not directly observable]

==> Output [observable data].

Example: Observable data: Andrew hears: "What's four plus seven?"

==> [Non-observable data: Andrew looks for the answer to the question] ==>

Observable data: Andrew answers: "Eleven".

The purpose of cognitive psychology is to propose operating mechanisms and processes3 for the INTERNAL PROCESSING SYSTEM which would result in the observed behaviors. In the case given above, we would aim to describe how Andrew goes about obtaining the answer "eleven". The internal process might, for example, consist of consulting a personal "addition table" stored in the memory, or of activating a mental calculation procedure, for example "seven and three make ten, ten plus one: eleven".

This simple example suggests that, a minima, we need to know what we mean by the "memory" in which a person's addition table is stored; how to describe the knowledge which a person possesses in order to activate these internal processes; what conditions are necessary to mobilize this knowledge, etc.

Before we address the central question of learning, we shall present a certain number of essential concepts, such as "information", "comprehension", "problem", "long-term memory", "episodic memory", etc. These verbal terms all have meanings in everyday parlance, but may be understood in different ways depending on the specific knowledge of different individuals. To prevent misunderstanding, it is important that we define what is meant by the terms in question in the context of this work.

We shall begin by presenting learning situations studied in a laboratory setting. These learning experiences are necessarily limited to highly compartmentalized knowledge in a particular domain, and may even involve "artificial" knowledge. The advantage of these situations is that they can be closely controlled (and, where possible, repeated), enabling the detection of relatively fine mechanisms that researchers may then generalize - with the necessary precautions - to real-world fields of knowledge. However, despite the precautions taken in defining the concepts used to describe observable behaviors as the result of non-observable mental processes, verbal descriptions are always polysemic to some extent. Formalization provides the means of defining concepts and operating processes in an unequivocal manner. From the 1970s onwards, computer modeling (see Nguyen-Xuan, 1986) came to replace statistical models (factorial analysis, regression analysis, stochastic models, etc.) in the domain of psychology, offering a greater level of flexibility4.

The first computer models used in cognitive psychology were created by researchers working across multiple domains, notably combining informatics and psychology with philosophy or linguistics. The earliest work on artificial intelligence in relation to learning was inspired by research in psychology. However, not all of the models of learning mechanisms developed as a result of laboratory research are used in AI.

In humans, the acquisition of vast domains of knowledge, featuring a wide variety of skills, must necessarily draw on multiple learning mechanisms, and more importantly, will involve multiple learning situations.

The aim of this book is to establish conceptual frameworks and outline the main theories of knowledge acquisition based on the IPS metaphor. These theories first emerged in the 1950s and 1960s, and have been the subject of extensive research in the intervening years. Many of the results of this research are now considered as established knowledge in cognitive psychology and are no longer debated. The principal aims of current experimental work include exploring established theories in greater detail5, and developing educational tools in which these theories form a reference framework. We shall not provide an exhaustive bibliography covering the last 50 years of research here; instead, we shall focus on a number of publications which we consider particularly interesting, representative works on specific questions, and publications concerning work discussed here in some detail.

At this juncture, it is important to state our own position as researchers. The ever-increasing interdisciplinarity, richness and rigor of research in cognitive psychology should not be underestimated; the domain provides a wide basis for useful (and widely used) knowledge for designing settings for learning, new teaching technologies, etc.

Cognitive psychology based on the IPS metaphor is an experimental science, in that it makes use of the hypothetico-deductive approach and "controlled" experimentation. These experiments can be viewed as in vitro tests. In certain cases, the hypotheses formulated from laboratory situations can be transposed into "real-world" (in vivo?) situations. However, in many cases, this transposition is limited by deontological or even legal considerations.

Consider, for example, a comparison of decisions made by two groups, A and B, differentiated on the basis of socio-cultural levels. The groups were placed in a situation similar to the "prisoner dilemma" proposed in game theory, which is widely used across the humanities and social sciences (economics, psychology, sociology, etc.). The levels of cooperation observed in group A were significantly higher than those for group B. However, in this type of experiment, the gain/loss for the participants involved (e.g. small sums of money) is necessarily minimal in comparison with a 1 - or 8 - year difference in a prison sentence. This highlights the...