Modelling High-level Cognitive Processes

Richard P. Cooper With Contributi, Peter G. Yule, John Fox, David W. Glasspool, Richard P. Cooper

• Publisher: Psychology Press
• Year: 2013
• Print ISBN: 9780805838831
• eBook ISBN: 9781135646813
• Edition: 1
• Topic: Psychology
• Subtopic: Cognitive Psychology & Cognition
• Index: Psychology

This book is a practical guide to building computational models of high-level cognitive processes and systems. High-level processes are those central cognitive processes involved in thinking, reasoning, planning, and so on. These processes appear to share representational and processing requirements, and it is for this reason that they are considered together in this text. The book is divided into three parts. Part I considers foundational and background issues. Part II provides a series of case studies spanning a range of cognitive domains. Part III reflects upon issues raised by the case studies. Teachers of cognitive modeling may use material from Part I to structure lectures and practical sessions, with chapters in Part II forming the basis of in-depth student projects. All models discussed in this book are developed within the COGENT environments. COGENT provides a graphical interface in which models may be sketched as "box and arrow" diagrams and is both a useful teaching tool and a productive research tool. As such, this book is designed to be of use to both students of cognitive modeling and active researchers. For students, the book provides essential background material plus an extensive set of example models, exercises and project material. Researchers of both symbolic and connectionist persuasions will find the book of interest for its approach to cognitive modeling, which emphasizes methodological issues. They will also find that the COGENT environment itself has much to offer.

Part II

Modelling in Specific Domains

Overview

The chapters in Part 1 have provided essential background material. Part 2 is concerned with developing and applying that material in domains ranging from sentence processing to executive processes. The purpose of this section is to provide a map of the domains and issues addressed in Part 2.

Two principal topics are addressed in Chapter 3: cognitive skill and production system models. Production systems are argued to provide a good framework for modelling cognitive skill (among other things), and this argument is backed up by production system models of two arithmetic tasks: multicolumn addition and multicolumn subtraction. Of particular interest is the model of multicolumn subtraction, which can account for many of children’s subtraction errors in a straight-forward way. Perhaps the most important issue that is left unresolved is that no account is given of the acquisition of cognitive skill. In fact, theories and production system models of skill acquisition do exist (e.g., Anderson, 1981, 1982, 1993; Newell, 1990). The lack of any account of skill acquisition in the chapter is due to space limitations, rather than any intrinsic conflict between the production system framework and the computational requirements of skill acquisition.

The primary focus of Chapter 4 is problem solving. The chapter is concerned more with the computational requirements of problem solving—the cyclic proposal, evaluation, selection, and application of operators or moves, and the need in some cases for a goal stack—than with the empirical evaluation of problem solving models. However, later chapters demonstrate that the basic processing cycle introduced when modelling problem solving behaviour has significant generality: it may also be applied in domains such as deductive reasoning (Chapter 5) and sentence processing (Chapter 7).

Chapter 5 presents two models of human deductive competence on syllogistic reasoning tasks. Both models generate similar behaviour on the full set of 64 syllogisms, but they achieve this by very different means: one through the construction and subsequent revision of a “mental model” consisting of tokens representing individuals and the other through the construction of a set-theoretic diagram. In this way the models illustrate two different algorithms for the same set of stimulus/response behaviours, and raise the question of indistinguishability: can behavioural evidence distinguish between model-based and set-theoretic approaches to deductive reasoning? The “mental model” model also illustrates the requirement of “computational completeness” introduced in Chapter 1. The intended workings of mental model construction and revision are easy to specify in informal terms, but specification at the level required by the computational model raises many difficulties. The result is a large number of rules relating to specific cases for both mental model construction and mental model revision.

Decision making and categorisation are the domains of interest in Chapter 6. The chapter has a subtext, however, comparative modelling. Three competing models of two medical diagnosis tasks are developed (based on Bayesian, associationist, and hypothesis testing approaches). Each of the models contains parameters that allow aspects of their behaviour to be tuned to fit empirical findings. It is suggested that an appropriate methodology for parameter estimation is to consider multiple dependent measures, and set parameters to fit only a subset of these, allowing other dependent measures to be predicted from the models. Although no single diagnosis model is found to yield a perfect account of the observed human behaviour, development of a range of models for the same tasks clarifies the strengths and weaknesses of the various approaches.

The domains considered in Chapters 3 to 6 are covered in most texts on thinking and reasoning. Chapter 7 considers a rather different domain: sentence processing. Sentence processing is different because it is usually automatic and effortless. When given a problem or syllogism it generally requires a deliberate act of will to solve it. When given a sentence it generally requires a deliberate act of will not to process it. Merely attending to a sentence will result in some attempt to process it. Given this qualitative difference, one might argue that sentence processing should not be considered alongside high-level cognitive processes such as problem solving, deductive reasoning, and decision making. Indeed, this is at the heart of the modularity hypothesis (Fodor, 1983). However, there are strong counters to this argument. Sentence processing has been shown to be sensitive to effects of knowledge (e.g., linguistic and visual context). Arguably the automaticity of sentence processing is a consequence of it being a highly practiced cognitive skill. From this perspective there is no qualitative difference between the automaticity with which an average person processes sentences and the way in which a chess Grand Master assesses the layout of pieces during a game of chess.

For the above reasons it is reasonable to at least consider the computational requirements of sentence processing. Such consideration reveals a role for the proposal, evaluation, selection, and application of operators and for stack-like data structures. Each of these is familiar from other high-level cognitive domains. The computational requirements of sentence processing therefore appear to be continuous with those of more standard high-level cognitive processes. This warrants the inclusion of sentence processing in this volume.

A further issue that is raised in Chapter 7 is the distinction between competence and performance. The basic idea is that competence refers to idealised knowledge of a domain. Performance results from the cognitive system’s use of this knowledge. It is claimed that this use may be sub-optimal (e.g., because of memory or processing limitations), so there may be a gap between actual performance and the idealised performance that would be predicted from competence.

The competence/performance distinction is also touched upon in Chapter 5 in relation to deductive reasoning. In this case the distinction is between deductive competence and deductive performance. The competence/performance distinction has some important lessons for modelling. It also raises some issues which cognitive modelling is well-placed to address. These lessons and issues are amongst those discussed in Part 3.

Chapter 8 is concerned with executive processes. These processes are distinct from other high-level cognitive processes because they operate on the cognitive system itself rather than on the external environment. Thus, executive processes are concerned with issues such as setting overall goals, resource allocation and control, co-ordination, and integration of sub-processes. As such, the issues raised by attempting to model executive processes differ from those raised by attempting to model behaviour in specific domains. Of particular concern is that a model of executive processes requires models of individual sub-processes on which to operate. Thus, a model of resource allocation cannot be evaluated without associated models that employ resources and whose behaviour is sensitive to resource allocation. In the context of Chapter 8, highly simplified models of subprocesses are employed to allow modelling of the executive process to progress. The extent to which simplifications made in modelling these sub-processes affect overall behaviour of the system is unclear, but the general problem suggests that careful attention to methodology is required when developing models of executive processes.

Chapter 3

Arithmetic: A Cognitive Skill

Richard P.Cooper

Overview: This chapter presents arithmetic as an acquired cognitive skill, and illustrates how aspects of this skill can be modelled within two production systems. The chapter’s primary aims are to provide an in-depth illustration of the workings of production systems, and to demonstrate one way in which models may be developed with the support of empirical data. The chapter concludes with a discussion of limitations of the simple production system approach.

3.1 Cognitive Skills

A cognitive skill is an acquired ability to perform some cognitive task with a high degree of fluency. Examples include mental arithmetic, chess playing, and reading mirror writing. Cognitive skills are interesting because anyone of normal intelligence can, with sufficient practice and dedication, become highly proficient in most cognitive skills. This was illustrated in the case of reading mirror writing by Kolers and Perkins (1975), who gave participants extensive training on reading text that was printed upside-down or mirrored. Initially participants were able to perform the task, but they were very slow. As the participants practiced, however, their reading grew faster. After extensive training (reading 200 pages) it was almost as fast as their reading of standard text.

Skill acquisition shows that the cognitive apparatus is highly flexible. It is able to adapt to a variety of complex tasks, although that adaptation can require substantial practice. It is common to assume that that adaptation involves a change to the knowledge or information used by the cognitive apparatus, rather than a fundamental change to the way in which the cognitive apparatus functions.

Consider the case of mental arithmetic. Anyone of normal intelligence can acquire the skills required to perform large addition, subtraction, multiplication, or division problems. These skills take the form of procedures or rules that should be followed in order to solve different types of arithmetic problems. Someone skilled at arithmetic must know how and when to apply these rules, and it is the r...

Table of contents

• Cover
• Half Title
• Title Page
• Copyright
• Contents
• Preface
• I: Background
• II: Modelling in Specific Domains
• III: Reflections
• Bibliography
• Name Index
• Subject Index