Preface
At first glance, calling this book Knowledge Representation seems to remove all suspense as to what it is about, but actually I think it heightens the tension. Cognitive scientists have seldom agreed about what knowledge representation is (or even whether there really are representations). Furthermore, many people have their own favorite way to think about knowledge representation and regard anyone who thinks otherwise with guarded skepticism.
As I wrote it, I thought of this book as a Michelin guide to knowledge representation. (I would call it a Fodor’s guide, but there seems to be a Fodor who has had a few things to say about knowledge representation and who has apparently written his own guide.) The typical travel guide presents an overview of a city, country, or region. It defines boundaries, describes languages and currencies, and mentions museums, monuments, and other sights. The descriptions are never as rich as the sights themselves (or people would never travel). Invariably, there are suggestions of things to see that you would not have considered, and sights your mother’s best friend said not to miss that are not even in the guidebook.
I hope that this book serves as a good guide as well. In chapter 1, I define representation and discuss some foundational issues. Then, in chapters 2 to 7, I discuss kinds of representations (akin to regions of a country), ways that people have thought about representations in the context of psychological, computational, or linguistic models. By organizing the book around types of representations, I have tried to bring together things that I think are deeply similar, although investigators may traditionally not have considered them to be similar. Chapters 8 and 9 focus on the role of specific content in representation, and chapter 10 draws some general conclusions about the uses of representation in cognitive models.
By design, I have written this book to accommodate people just entering the field (like the advanced undergraduates and graduate students in my classes on knowledge representation). That means that some approaches have been sketched to give the flavor of the representational scheme. I have tried to provide enough references to other sources of information for people who want to learn more about the topics covered here. Like a traveler, if you see something interesting, go there.
Any travel guide has biases. The authors of the Let’s Go series are college undergraduates who place a premium on inexpensive places to eat and drink. Other guides that cater to wealthier clientele include a different list of culinary delights. I also have my biases. My own research has focused on similarity, analogy, and categorization, and this research focus has had two effects on this book. First, I have drawn many examples from work on similarity and analogy. Second, the work on similarity and analogy assumes structured relational representations. I have tried to be evenhanded in my approach to representation in this book, but some biases seep through from my own research. My deepest bias about knowledge representation is that there is no single right way to think about the topic. Different problems require different representational decisions. For this reason, I think it is important to be conversant with many different techniques of representation and to know their strengths and weaknesses. Thus, this book is also a bit like a field artillery guide: It provides information about the weapons available to attack various problems in cognitive science.
In an effort to draw parallels between models with similar sets of representational assumptions, I have sometimes ignored familiar distinctions that are made in cognitive science. Perhaps the most obvious of these concerns connectionist models. Readers may expect an entire chapter on connectionist models in a book on knowledge representation, but a glance at the contents shows that there is none. I have included at least a chapter’s worth of material on connectionism in the book, but I present specific connectionist techniques along with other models that make similar representational assumptions. I describe distributed connectionist models along with spatial models of representation, parallel constraint satisfaction networks following a discussion of spreading activation, and techniques for role-argument binding in connectionist models in the chapter on structured representations. Although this approach is nonstandard, I think it ultimately provides a good indication of how these connectionist tools work.
For the past few years, I have taught a course in knowledge representation in which we have read articles about different types of knowledge representation and have talked about their strengths and weaknesses. We have also focused on how different representational assumptions bias the way people think about different psychological processes. If you are interested in using this book as a class text, I would recommend a similar approach. To facilitate class discussion, I have numbered all the examples throughout the book. The chapters of the book must be supplemented with readings that provide more details than I can include here.
As I said, I wrote this book with graduate students seeking an introduction to knowledge representation in mind. Because knowledge representation is a crucial topic for anyone who has an interest in cognitive science, however, I recommend the book to psychologists in general (except perhaps those whose names already appear in more than three of the references). I think the book is particularly useful for people in the cognitive science community outside psychology (such as those working in philosophy of mind or cognitive anthropology) who want to know how psychologists have thought about representation.
Finally, the preface of every academic book has an obligatory paragraph that tells readers how to read it. My recommendation is to read it straight through. I have tried to provide pointers from one chapter to another for discussions of related material, but I think the book reads better front to back. Readers with an intense need to skip around may start with chapter 1; chapters 5 to 7 should be read as a group in that order; chapter 8 should probably be read after chapters 2 to 7; chapter 10 makes the most sense if it is read last.
Bon voyage.
ACKNOWLEDGMENTS
Since I started studying cognition, I have had the good fortune to be in contact with people who have had an abiding interest in knowledge representation. Foremost on the list is Dedre Gentner, who was once my graduate advisor and is now a colleague and friend. I based my knowledge representation course on hers, and the rest followed from there. Ken Forbus was also an important influence. His guidance in thinking about the computational aspects of analogy have been invaluable. Finally, I would be remiss to leave out Jim Anderson and Doug Medin, who were also role models in my formative years.
Over the past few years, I have also had the benefit of knowing many extremely bright colleagues, friends, and students who have been willing to talk about representation for hours, among them Larry Barsalou, Curt Burgess, Rob Goldstone, Tory Higgins, Robert Hoffman, Keith Holyoak, John Hummel, Mark Keane, Bob Krauss, Patricia Lindemann, Valerie Makin, Gary Marcus, Tomislav Pavlicic, Robert Remez, Brian Ross, Michael Schober, Colleen Seifert, Yung-Cheng Shen, Bobbie Spellman, Ed Wisniewski, Takashi Yamauchi, and Shi Zhang. Thanks also to the students in my knowledge representation classes, whose ideas have contributed to this book in ways known and unknown. Special thanks to Frank Riebli, who read drafts of some chapters in their early form.
I want to acknowledge the help of Eric Dietrich for many discussions about representation and for letting me steal from our joint paper for chapter 10. Thanks also to Terry Regier for answering some questions about prepositions. Philip Johnson-Laird provided thorough and thoroughly useful comments on the manuscript, and was also kind enough to supply me with a copy of a computer program that generates spatial mental models. Gregory Murphy also gave me extensive feedback and an ink-filled manuscript. His thoughts greatly improved the clarity of the arguments here. David Krantz was a great help, particularly in walking me through the intricacies of measurement theory. Thanks also to David Leake, who sent me a copy of microSWALE to play with.
I’d like to thank Judi Amsel for encouraging me to write this book after I mentioned the idea to her at Psychonomics in Los Angeles. Thanks also go to Barbara Wieghaus, who kept the process of book production moving along on schedule.
During the time that this book was being written, my research was generously supported by National Science Foundation CAREER Award SBR-95-10924.
Finally, I want to thank my wife Betsy for putting up with my work hours and for willingly sharing the life and times of an assistant professor without going crazy. Because she has been a constant source of support as well as a great mother to our son Lucas, this book is dedicated to her.
—Arthur B. Markman