The Genesis of a Graph: From Library Catalogs to Linked Data Dreams

For millennia, humanity has wrestled with a paradox: we create more knowledge than we can hold, yet we desperately need to connect that knowledge to make sense of our world. From the clay tablets of ancient Mesopotamia to the digital clouds of today, the eternal quest to organize information has shaped civilizations. This quest is not merely practical; it is philosophical, deeply embedded in how we understand meaning itself. Our story begins not with code, but with this very human struggle—a struggle that eventually led to the birth of knowledge graphs. It is a tale of dreams and frustrations, of systems that imposed order and those that embraced chaos, of a vision that knowledge, at its best, is not a static collection but a living tapestry of interconnected truths.

The Dewey Decimal System: Ordering the Unruly

Let us step back to 1876, to a small library in Amherst, Massachusetts. A young librarian named Melvil Dewey, frustrated by the inefficiency of existing cataloging systems, introduced a decimal-based classification that would become the Dewey Decimal Classification (DDC). Dewey’s vision was audacious: to encapsulate all human knowledge into ten main classes, each subdivided by decimals. It was a triumph of order—a hierarchical tree where every book had its place. For over a century, DDC and its cousin, the Library of Congress Classification, reigned supreme. They gave libraries a universal language, allowing books on ‘500 Natural sciences’ to be found in Tokyo or Toronto. But this order came at a cost.

The Dewey system is fundamentally hierarchical and *a priori*. It imposes a single, fixed structure on knowledge, assuming that subjects naturally nest within broader categories. But knowledge is not so obedient. Consider ‘Artificial Intelligence.’ In DDC, it might be squeezed under ‘006.3 Special computer methods’ or ‘005.1 Artificial intelligence’ in some editions, but what about its ethical dimensions? Its historical roots in philosophy? Its applications in medicine? The hierarchical tree forces choices, privileging one perspective over others. Moreover, Dewey’s categories reflect a 19th-century Western worldview—Christian theology gets its own main class (200), while other religions are subsumed. This is not merely a historical artifact; it is a philosophical statement about what knowledge is worth organizing centrally.

Critics like S.R. Ranganathan, father of the colon classification, argued that any fixed hierarchy is inherently flawed because subjects are multidimensional. A book on ‘Climate Change’ might belong to ‘551.6 Climatology’ but also to ‘363.7 Environmental policy’ and ‘304.2 Human ecology.’ In a card catalog, you might assign multiple subject headings, but the underlying classification forces a primary location. This tension between order and flexibility persists today in digital libraries and content management systems. The lesson is clear: any system that tries to pre-categorize all knowledge will inevitably be both over-simplistic and culturally biased. What we need is not a better hierarchy, but a way to represent knowledge that embraces its inherent connectivity.

The Semiotic Foundation: Signs, Meaning, and the Triad of Representation

To understand why these systems falter, we must dig deeper into the nature of representation itself. How do we signify? How do we connect a word, a concept, and the thing it denotes? This is the domain of semiotics, the study of signs. In the early 20th century, Ferdinand de Saussure revolutionized linguistics with his dyadic model: the signifier (the sound or written mark) and the signified (the mental concept). For Saussure, meaning arises from differences within a system—’cat’ means what it does because it is not ‘bat’ or ‘hat.’ But knowledge representation requires more than differences; it requires connection to reality.

Charles Sanders Peirce extended this to a triad: the sign (or *representamen*), the object, and the interpretant. The sign stands for the object to some interpretant, which creates another sign, and so on. This endless chain of interpretation is crucial: meaning is not fixed but is generated through use and context. Peirce further classified signs into icons (resembling their object), indexes (causally linked), and symbols (conventional). Consider a photograph of a tree (icon), smoke indicating fire (index), or the word “tree” (symbol). Each connects to its object differently.

In a knowledge graph, an entity like ‘Steve Jobs’ is a sign. It stands for the actual person (object) through

Peirce’s triad does more than describe how signs work; it provides a blueprint for representing knowledge in a way that captures context and multiplicity. In the signifier-signified model, meaning is static and internal. In Peirce’s framework, meaning is a *process*—a chain of interpretations that never ends. This is the critical insight: no entity, no fact, exists in isolation. A sign (representamen) points to an object through an interpretant, which is itself a sign. The “Steve Jobs” entry in a database is not a final truth; it is a sign that invites interpretation. Who is the “Steve Jobs” here? The co-founder of Apple? The biological father who abandoned his child? The Buddhist seeker? Each interpretation generates new signs, new connections. A knowledge graph embraces this by allowing multiple, even contradictory, relationships to coexist. It does not force a single “correct” classification; it maps the terrain of interpretation.

Consider the simple statement “Steve Jobs founded Apple.” In a traditional database, this might be a single record: `(Steve Jobs, founded, Apple)`. But Peirce would ask: What is the *object* here? The act of founding in 1976? The legal incorporation? The cultural mythos? What is the *interpretant*? A historian might see it as a pivotal moment in the personal computing revolution; a labor historian might focus on the working conditions in the early Apple garage; a business scholar might analyze the strategic partnership with Microsoft. Each interpretant generates a new sign—a new node and edge in the graph. The power of a knowledge graph is not in storing the statement, but in enabling the *reasoning* about it from multiple perspectives. It makes explicit the implicit web of context that Dewey’s decimal number could never capture.

This semiotic lens reveals the fundamental flaw in hierarchical classification: it assumes a single, authoritative interpretant—the classifier. Dewey and his successors played the role of the ultimate interpretant, deciding once and for all where a subject “belongs.” But knowledge is dialogic, not monologic. It is a conversation across time, cultures, and disciplines. The sign “Artificial Intelligence” means something different to a computer scientist in 2023 than to Alan Turing in 1950, or to a philosopher worried about consciousness, or to a policymaker regulating autonomous weapons. A knowledge graph, by allowing multiple types of edges and contexts, can represent this evolution of meaning. It can show that “AI” is *related to* “machine learning” *technically*, *influenced by* “cognitive psychology” *historically*, and *regulated by* “the EU AI Act” *juridically*. These are not subcategories; they are different interpretive lenses, each valid in its domain.

The transition from semiotics to information science was not direct, but the DNA is unmistakable. When Ted Nelson coined “hypertext” in 1965, he was dreaming of a “docuverse” where any piece of information could be linked to any other, creating a “non-sequential” writing space. His Project Xanadu, though never fully realized, was a direct descendant of Bush’s Memex and Peirce’s interpretant chain: meaning is created through links, not through linear order. Meanwhile, in libraries, the “faceted catalog” movement, inspired by Ranganathan’s colon classification, began to experiment with multiple, orthogonal access points to resources. A user could start with “su