Evolutionary studies of cultural complexity often assume that group members select the best information available in the group, effectively diffusing the best innovations, whose advantages are subsequently passed on to the next generation. This would seem to describe the ideal of the scientific process — each cohort of papers in a field surfacing the best innovations, refining them and passing on to the next “layer” or cohort of scientific works. Here, we use academic journal databases to explore this “forking” (branching) process in the evolution of a scientific paradigm. We apply citation network visualization and Latent Dirichlet allocation topic analysis to three different paradigms defined pragmatically as the set of papers citing a highly influential paper in each respective case. Our three case studies indicate a founder effect in how the seminal paper is highly-embedded in the citation network, and yet peripheral to the evolution of topics in subsequent “layers” of publications within the paradigm. This and additional evidence suggest certain topics are selected and followed, while others are left behind. From these case studies, we discuss how hitherto undeveloped ideas of the past might be located in the topic space of seminal works of the same fruitful time period.

文化复杂性的进化研究通常假设群体成员选择群体中可用的最佳信息，有效地传播最好的创新，其优势随后传递给下一代。这似乎描述了科学过程的理想——一个领域中的每一组论文都呈现出最好的创新，对其进行提炼并传递到下一个“层”或科学著作的队列。在这里，我们使用学术期刊数据库来探索科学范式演变中的这种“分叉”（分支）过程。我们将引文网络可视化和 Latent Dirichlet 分配主题分析应用于三种不同的范例，这些范例在实用上定义为在每个案例中引用极具影响力的论文的一组论文。我们的三个案例研究表明，开创性论文如何高度嵌入引文网络，但与范式内后续出版物“层”中的主题演变无关。这个和其他证据表明某些主题被选择和遵循，而其他主题则被抛在后面。从这些案例研究中，我们讨论了迄今为止尚未发展的过去思想如何可能位于同一时期富有成果的开创性作品的主题空间中。