Let me start by explaining what I mean by an isolated concept cluster. It is often remarked that you can't define any moral term without using other moral terms. For example, you can define obligation in terms of right and wrong, or you can define right and wrong in terms of obligation, but you need to use moral terms to define moral terms. Morality is an example of an isolated concept cluster. It is a cluster of concepts which cannot be defined or understood entirely in terms of concepts outside of the cluster.

I used to think that there were just four or five isolated concept clusters, but it turns out that there are a lot. Here are some proposed examples:

mental states/qualia
cause/laws of nature
states of affairs

Let me also try to distinguish this notion of an isolated concept cluster from the notion of an ontological category, because some of the things I listed are also ontological categories. In this question, I'm not concerned with whether one category can be reduced to another category. All such reductions lose part of the concept. For example, many philosophers try to reduce morality to some form of practicality. I'm not going to argue against that, but just point out that you have lost the fundamental concept, understood by most people, of moral obligation, and turned it into something else like pragmatic obligation.

Or, for perhaps a less contentious example, consider the common practice in mathematics of reducing a straight line to a set of points. You can define what features a set of points must have to be a straight line, but you can't communicate the concept of a straight line without drawing a straight line or referring to some related concept such as curvature or the shortest distance between two points.

Has anyone investigated this area? Are there any lists of what I'm calling isolated conceptual clusters? And what are they called in the literature?

  • the terms have this problem, but words are not terms/used not only in term's role. words are names for some experience, do not confuse names with nominative meaning. Name is a form for the experience, nominative is a label for the form. Self language, or natural language of every man consisted from named experience. All other language systems are only for translation self natural language to common words. Commented Apr 30, 2023 at 0:37
  • The term sometimes used is definitional circle, Quine identified one at the bottom of modal logic (possibility, necessity, analyticity, synonymy, etc.). The failures of non-circular conceptual analyses of interesting concepts (as illustrated by the case of knowledge) suggest that there is a definitional circle at the bottom of every subject field. We can no more define all our terms than we can prove all our statements (attributed to Hilbert). Axiomatic approach is founded on that: formalizations introduce undefined terms and relate them through axioms.
    – Conifold
    Commented Apr 30, 2023 at 0:37
  • I might be speaking out of my ass here, but isn't that why we need science, to define a cause effect structure, and effectively an ontology (categorization of stuff) for the world? Life, mental states, energy, aren't they all subject to cause-effect?
    – Hex Heager
    Commented Apr 30, 2023 at 3:22

1 Answer 1


Modern linguists and philosophers of language have a number of theories of semantics that take concepts as clusters of primitives. The work I'm most familiar with is Wierzbicka's work. I have a couple of her works. She is known for her work on natural semantic metalanguage which is the notion there are conceptual primes that make appearances in specific languages in a lexicon, including those outside of PIE. It reinforces notions of compositionality of language and might be taken as supportive of notions of universal generative grammar.

In computer science, there's an interest and there are attempts to codify these for the use of setting up formal systems to reason across a knowledge base, and is often spoken about in the context of data models. Core ontologies, upper ontologies, and ontological frameworks like RDF and OWL nourish the hope that the hypertext architecture can be enriched semantically using various RFCs to enrich the use of the http protocol under such frameworks as Roy Fielding's model of resource constraints given in his doctoral thesis as central to understanding RESTful API implementation.

Of particular note, such organizations as Cyc, who I have been made to understand hire PhDs in philosophy to help codify these ontologies are working to codify human knowledge symbolically across a vast range of domains with the hope to upsell semantically enriched AI systems that process concepts in the spirit of Alan and Newell's PSSH. Filmmore's frame semantics is an example of a linguistic theory that underlies semantic databases and the construction of AI frames to map out these sorts of clusters of conceptual primes, a strategy enshrined with Minsky's philosophical speculations on the computational composition of mind.

  • There is certainly a relationship between isolated concept clusters and semantic primitives, but they aren't the same thing. For one thing, semantic primitives are single concepts that are primitive, but isolated concept clusters don't privilege one concept in the cluster over others. For another, isolated concept clusters can include some very sophisticated concepts that probably aren't part of all languages. Still, thanks for the link. It's worth looking into. Commented Apr 30, 2023 at 5:18
  • I see what you mean about differences. My sense, and I'm by no means involved in NLP day-to-day is that some of the actual work being done partially is done to drive theory development, so you might be in new territory. The most famous projects I'm aware of FrameNet which is at Berkeley and CLICS which is outside the US. They're both fun to use because you can click through the semantic web. CLICS represents a structure that looks across more than 100 languages.
    – J D
    Commented Apr 30, 2023 at 14:50
  • They changed the website quite substantially, since I visited last a couple of years ago. The graphs are here: framenet.icsi.berkeley.edu/fndrupal/FrameGrapher
    – J D
    Commented Apr 30, 2023 at 14:55
  • I think on reflection, that if one were to try to preserve the analytic synthetic divide, than the notion of "in and out of the predicate" might be given by "in and out of the semantic frame". It seems to me FrameNet is driving in the direction of providing a graph theoretic appraisal of the nearness of concepts that might be considered relatively "isolated", though webs of belief and meaning ultimately are all interconnected, even if through irrational or intuitional manners.
    – J D
    Commented Apr 30, 2023 at 15:03
  • @polcott It occurred to me that your goal of validating the physical symbol systems hypothesis by somehow trying to salvage the analytic-synthetic distinction might benefit from the use of semantic frames to justify the approximate use of the analytic-synthetic dichotomy where one could use the graph-theoretic distance of concepts to gauge a degree to which one term is "in the predicate of another". In essence, there might be a metric on compositions of predicates such that C1 predicates C2 predicates C3 (though I think Jackendoff's conceptual semantics makes this theoretically unviable). 0.02
    – J D
    Commented Apr 30, 2023 at 17:55

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .