Paradigms, Citations, and Maps of Science: A Personal History Henry Small University City Science Center, Institute for Scientific Information, 3501 Market Street, Philadelphia, PA 19104. E-mail: hsmall@isinet.com Can maps of science tell us anything about paradigms? The author reviews his earlier work on this question, including Kuhn’s reaction to it. Kuhn’s view of the role of bibliometrics differs substantially from the kinds of rein- terpretations of paradigms that information scientists are currently advocating. But these reinterpretations are necessary if his theory will ever be empirically tested, and further progress is to be made in understanding the growth of scientific knowledge. A new Web tool is dis- cussed that highlights rapidly changing specialties that may lead to new ways of monitoring revolutionary change in real time. It is suggested that revolutionary and normal science be seen as extremes on a continuum of rates of change rather than, as Kuhn originally as- serted, as an all or none proposition. History of Science and Bibliometric Structure The deeper you delve into the history of any topic the more you begin to see the connections between ideas, peo- ple, and institutions captured in the historical context. I began thinking about mapping science as part of a project to document the history of nuclear physics (Small, 1972). The project was conducted at the American Institute of Physics, and its goal was to describe nuclear physics and its evolu- tion through time. I began wondering if methods could be found for graphically showing the evolution of historical connections within the field. As a doctoral candidate in the history of science, I was deeply influenced by Thomas Kuhn’s “structural” view of scientific development (Kuhn, 1970), and mapping a field seemed somehow related to his concept of the paradigm. Unfortunately, Kuhn never gave us clear empirical guidelines for defining exactly what a paradigm was (Mas- terman, 1970; Shapere, 1964), and his discussion remained mostly on a philosophic level. In the well-known Postscript to the second edition of his book (Kuhn, 1970), Kuhn emphasized that a paradigm was shared by members of a specialized community. But he did not explain how, given such a community, we could go about finding the paradigm shared by its members. Kuhn described paradigmatic ele- ments variously as theories, models, equations, tacit knowl- edge, beliefs, values, and “exemplary past achievements,” or exemplars, all components of his “disciplinary matrix.” He did note that specialty communities could be identified by studying communication patterns, such as referencing in the literature or informal communications among scientists, but was content to leave the details to others. Because of this vagueness in definition, paradigms have remained mostly a philosophic construct, and for most historians as elusive as unicorns. For the nuclear physics project I first tried to map the intellectual landscape of leading researchers in the field such as Ernest Rutherford. By an intensive reading of their papers, I constructed diagrams of the evolving models of the atomic nucleus. Ideas or hypotheses were represented as nodes that were linked together if they were part of a supporting argument or assertion. I could then show how these networks evolved with each successive paper, and the introduction of new concepts such as the neutron. More recently Paul Thagard (1992) developed a theory of “ex- planatory coherence” which is a more fully elaborated the- ory of what I was attempting. Thagard models each theory as a network of hypotheses, evidence, etc. that either sup- port or contradict each other. The computation of the win- ning theory is implemented in an AI program called ECHO. His model stresses that within any historical context the only guide we have is to find the most consistent collection of facts and theoretical statements. I soon realized, however, that if I continued coding the scientific literature in this way I would have to retire before my work was completed. Therefore, I turned to a simpler kind of analysis focusing on bibliographic elements in pa- pers that might function as concept surrogates. To explore the possibilities I coded all nuclear physics papers in the Physical Review for the 1920s and 1930s, recording authors, key words, affiliations, classification headings, and a sample of cited references for each paper. © 2003 Wiley Periodicals, Inc. JOURNAL OF THE AMERICAN SOCIETY FOR INFORMATION SCIENCE AND TECHNOLOGY, 54(5):394 –399, 2003