Human children and adults know many things. They know how to speak a language (many people know several). They know how to read maps, and how to go from San Diego to London and Rome. Some of us know how to build cars, and others know how to solve partial differential equations.
Most people feel that knowledge comes from two kinds of sources: What is given us by virtue of our nature, and what we know as a consequence of our nurture. In reality, though, it is far from clear what is meant by either nature or nurture. Nature is usually understood to mean ``present in the genotype,'' and nurture usually means ``learned by experience.'' The difficulty is that when we look at the genome, we don't really see arms or legs (as the preformationists thought we might) and we certainly don't see complex behaviors. As we learn more about the basic mechanisms of genetics, we have come to understand that the distal effects of gene products are highly indirect, complicated, and most often dependent on interactions not only with other gene products but also with external events.
Learning is similarly problematic. We know that learning probably involves changes in synaptic connections, and it is now believed that these changes are effected by the products of specific genes which are expressed only under the conditions which give rise to learning.
The obvious conclusion is that the real answer to the question, Where does knowledge come from, is that it comes from the interaction between nature and nurture, or what has been called ``epigenesis.'' Genetic constraints interact with internal and external environmental influences, and they jointly give rise to the phenotype.
Unfortunately, as compelling and sensible as this claim seems, it is less a conclusion than a starting point. The problem does not go away, it is simply rephrased. In fact, epigenetic interactions must, if anything, be more complicated than the simpler more static view that x% of behavior comes from genes and y% comes from the environment. For this reason, the interactionist (or constructivist) approach has engendered a certain amount of skepticism on the part of developmentalists. To paraphrase David Klahr (whose complaint was about Piaget's concepts of assimilation and accommodation), nature and nurture are like the Batman and Robin of developmental theory: They hang around waiting in the wings, swoop in and solve a problem, and then disappear before they can be unmasked.
In fact, we believe that the interactionist view is not only the correct one, but that the field is now in a position where we can flesh this approach out in some detail. Our optimism springs from two sources. First, there has been extraordinary progress made in recent years in genetics, embryology, and developmental neuroscience. We are beginning to have an idea of how genes do their job. In addition, much has been discovered about the cortical basis for complex behavior. We also know more now than we did two decades ago about brain development; current evidence suggests a far higher degree of cortical plasticity than was anticipated, and this has obviously far-reaching consequences for theories of development. An impressive array of tools for studying brain processes has been developed, which permit non-invasive access to events in the brain and a spatial and temporal granularity that is quite remarkable.
Second, we have seen in recent years dramatic advances in a framework for computation which is particularly appropriate for understanding neural processing. This framework has been variously called parallel distributed processing, neural network modeling, or connectionism (a term introduced by Donald Hebb in the 1940's, and the name we adopt here). This approach has demonstrated that a great deal of information is latent in the environment and can be extracted using simple but powerful learning rules. But importantly, connectionist models also suggest new ways in which things can be innate. Furthermore, by using connectionist models together with genetic algorithms and artificial life models, it is possible to study within one and the same simulation evolutionary change at the level of populations of neural networks, maturation and learning in individual neural networks, and the interactions between the two.
This book offers our perspective on development, the nature/nurture controversy, and on the issue of innateness. The definition of innateness itself is controversial. We take the question to be essentially how to account for those behaviors which, given the normal experiences encountered during development, are universal across a species. This is a much broader perspective than many might adopt, but it lets us then ask what are the sources of constraint which lead to these universal outcomes.
We take a connectionist perspective, but we are very aware that ours is a specific and significantly enlarged conception of what connectionism is. In some ways, it is our view of what connectionism should (and hopefully, will) be. We are convinced that connectionism has a great deal to offer for understanding development. We also think that connectionists can only profit from the encounter with development. In a very deep sense, we believe that development is not just an accidental path on the way from being small to getting big. Rather, with Piaget, we are convinced that only by understanding the secrets of the process of development will we ever understand complex behaviors and adult forms; but our solution will be somewhat different from Piaget's.
It seems appropriate to say something about how we came to write this book and the process by which it was written. The reader should know that this book is a truly collaborative effort; thus, chapters are unsigned and each reflects our joint efforts. The collaboration began in the late 1980's. At that time, the John D. and Catherine T. MacArthur Foundation awarded a training grant to the Center for Research in Language at UCSD; this was part of the larger MacArthur Network in Transitions from Infancy to Early Childhood, directed by Bob Emde. The basic goal was to introduce developmentalists to the tools and methodology of connectionist modeling. But the training program was unusual in several respects, and reflects the open and innovative approach encouraged by Emde. There was a high degree of flexibility in the program. We were able to bring senior as well as junior fellows to UCSD, and for varying degrees of time, depending on the schedules and goals of the fellows. In some cases, the goal was simple literacy in connectionist modeling. In other cases, fellows developed computer simulations of data they had brought with them. An important component of the program was the set of simulation exercises which were created to illustrate properties of connectionist models that are especially relevant to developmental theory. These simulations have been extended and amplified and form the core of the companion volume to this book.
After several years of the program, a workshop was held at UCSD in 1991. Alumni of the training program returned for a four-day reunion and presented the work that they had done as a result of their participation in the program. This was an extremely exciting event, because it impressed upon us the extent to which connectionism not only provides a very natural computational framework for modeling many developmental phenomena, but also gives us concepts for rethinking some of the old chestnuts. Furthermore, we realized that a critical mass was building; as a group, we felt we had a great deal to say. Thus was born the idea of summarizing this work in book form.
About the same time, the organizers of the Society for Research in Child Development invited several of us to organize a special colloquium on connectionism and development for their 1992 meeting. Betty Stanton of Bradford Books/MIT Press was present at that symposium. Afterwards, she enthusiastically suggested that we turn the symposium contents into a book. Having just decided ourselves that the time was ripe to do this, we were pleased with Betty's excitement and support.
Primarily for logistical reasons, a subset of the original group proceeded to work on the book, with Jeff Elman being chiefly responsible for coordinating the joint efforts. We realized that we had several goals which would require more than one volume. First, we wanted to make a theoretical statement; this warranted a volume of its own. Second, we very much wanted to make the methodology accessible to as broad an audience as possible; so we designed a second volume with this pedagogical goal in mind. The second volume contains software and simulation exercises which allow the interested reader to replicate many of the simulations we describe in the first volume. The software is general purpose and can also be used by readers to carry out their own simulations. Finally, there is now a considerable body of literature in using connectionism to model development. Although we summarize and refer to much of this literature in the first two volumes, we felt it would be useful to collect some of the best work into a reader, the third volume. The first two volumes will be published almost simultaneously; our goal is for the third volume to appear within 18 months.
Our conception of the first volume has changed dramatically in the course of writing. Our original view was that this volume would bring together chapters written by us as individuals. Our discussions as a group proved so stimulating, however, that we soon moved to a very different model: A truly coauthored volume reflecting our joint ideas. Of course, this meant having to develop these joint ideas! As congenial as our viewpoints were and as great the overlap in our attitudes, we discovered that there were many areas around which we held different opinions, and very many more about which we held no opinions at all. The process of writing the book was thus highly constructive. Our meetings became seminars; the planning of chapter contents turned into lively discussions of theory. We all have found the process of working on this project to be enormously stimulating and we have learned much from each other. If we may be permitted a bit of self-appreciation, we are very grateful to each other for the forbearance, patience, and graciousness which have made it possible to forge a synthesis out of our different perspectives. This book is more than could have been produced by any one of us, and all of us feel that we ourselves have gained from the experience.
We owe a great deal to the MacArthur Foundation for their support. The far-sighted approach of Bob Emde, Mark Appelbaum, Kathryn Barnard, Marshall Haith, Jerry Kagan, Marion Radke-Yarrow, and Arnold Sameroff was critical in this effort. Without their support--both tangible as well as intellectual--we could not have written this book.
We also wish to make clear that although the cover of this volume bears the names of only six of us, the ideas within the book reflect an amalgam of insights and findings garnered from a much larger group. This includes the trainees in the program as well as other participants in the 1991 workshop: Dick Aslin, Alain Content, Judith Goodman, Marshall Haith, Roy Higginson, Claes von -Hofsten, Jean Mandler, Michael Maratsos, Brian MacWhinney, Bruce Pennington, Elena Pizzuto, Rob Roberts, Jim Russell, Richard Schwartz, Joan Stiles, David Swinney, and Richard Wagner. The trainers in the program, Virginia Marchman, Mary Hare, Arshavir Blackwell, and Cathy Harris, were more than trainers. They were colleagues and collaborators, and they played a pivotal role in the program and in our thinking. We are also grateful to a number of colleagues with whom we have interacted over the years. Some may still not agree with our arguments, while others will undoubtedly recognize some of their own ideas in the pages that follow. We owe to Jay McClelland the opening sentence of this Preface. We are grateful to these friends and hope that our translation of their ideas will not displease them.
In addition, we wish to thank those who read and commented on various sections of this book. Dorothy Bishop, Gergely Cisbra, Terry Deacon, Lucy Hadden, Francesca Happé, Henry Kennedy, Herb Killackey, Jean Mandler, Jay Moody, Yuko Munakata, Andrew Oliver, Adolfo Perinat, Paul Rodriguez, Marty Sereno, Jeff Shrager, Tassos Stevens, Joan Stiles, Faraneh Vargha-Khadem, and members of the UCSD DevLab have provided us with important and valuable feedback.
In December of 1994, the University of Higher Studies in the Republic of San Marino sponsored a two-day workshop entitled ``Rethinking Innateness,'' where three of our authors were able to air some of the ideas in this book and to listen and learn from some of the best cognitive neuroscientists in Italy. Herb Killackey joined us at the San Marino workshop, and we are immensely grateful to him for his input, and for extensive discussions about the topics covered in Chapter 5.
George Carnevale played a particularly important role in Chapter 4. Much of that chapter draws on his own work (Bates & -Carnevale, 1993), and George helped track down a number of errors in earlier drafts (of course, we reserve for ourselves the credit for those errors that remain). We also thank Jacqueline Johnson for making available the data from Johnson & Newport (1989), which we reanalyze in Chapter 4.
Meiti Opie not only read, commented, and proofed numerous drafts of Chapters 1, 5, and 7, but also served as general editorial assistant. She also tracked down and prepared the references. Meiti's persistence and attention to detail were extraordinary, and very much appreciated.
Betty and Harry Stanton of Bradford Books/MIT Press have been enthusiastic and eager in their support for this book from its beginning. We appreciate their faith in us, and their willingness to believe, as do we, that we have something important and exciting to say. Teri Mendelsohn, and later, Amy Pierce, of MIT Press have been an enormous help in the actual production of the first two volumes. A great deal of coordination was required, given the six co-authors, two books, and packaging and production of software. Teri and Amy made the job much easier, and their patience and encouragement is much appreciated.
When we began this effort, we did not fully appreciate the difficulties of producing a coauthored book with six authors who were located in San Diego, Pittsburgh, London, Oxford, and Rome. The logistics of travel, hotel accommodations, and arranging periodic meetings were formidable. Bob Buffington, Jan Corte, Miriam -Eduvala, Larry Juarez, John Staight, and Meiti Opie of the Center for Research in Language at UCSD, and Leslie Tucker at the Cognitive Development Unit, London, all played a critical role in arranging our meetings and making the time together as productive as possible. We are very much indebted to them for their help.
We acknowledge the financial support which has been provided to the authors and made the research described here possible. In addition to the funds from the MacArthur Foundation already mentioned, this includes support from the Office of Naval Research (contract N00014-93-1-0194) and National Science Foundation (grant DBS 92-09432) to Jeff Elman; the National Institutes of Health (grants NIH/NIDCD 2-R01-DC00216, NIH/NINDS P50 NS22343, NIH/NIDCD Program Project P50 DC01289-0351) to Elizabeth Bates; Carnegie Mellon University, the National Science Foundation (grant DBS 91-20433), and the Medical Research Council of the United Kingdom to Mark Johnson; a McDonnell/Pew Visiting Fellowship and the Medical Research Council of the United Kingdom to Annette Karmiloff-Smith; and the Science and Engineering Research Council, the Biological and Biotechnical Research Council, and the Economic and Social Sciences Research Council to Kim Plunkett.
Finally, we wish to thank Marta Kutas, who is a valued colleague and a treasured friend. She wrote a number of poems on the theme of innateness, and we are pleased and flattered that she allowed us to choose one to open the book.
While the central arguments and concepts of this book represent our collaborative efforts, in any enterprise involving several authors with very different backgrounds there are bound to be areas of disagreement that cannot be resolved. One of these concerned the title, with which MJ wishes to put on record his disagreement. In MJ's view the term ``innate'' is better dispensed with entirely, as opposed to being re-thought.