The notion of reducing all human concepts to combinations of a few "semantic primitives" (or a few simple classes of primitive concepts) is one of those things that's extremely appealing at first and then gets thornier and thornier as one digs in further.
To me is seems as if the mistake is to presume that the English expression defines the primitive rather than "is an attempt to express". That there are exceptions to the English definition does not imply a problem with the primitive concept, but rather with the definition. It's as if they were centrifugal force. The momentum is there, and the force away from the center of rotation is there, but the centrifugal force is not a primitive. Consider the problems quantum physicists have when they attempt to express their theories in English rather than in math.
I can’t comment on whether all human concepts can be reduced to simple primitives.
But it appears to me that all of human conscious experience can be reduced to a very few simple primitives.
There are a set of immutable facets of the present moment that every human is constantly immersed within:
• Gravity (compression)
• Balance (mobility)
• Relationship (connection)
• Precision (alignment)
• Effortlessness (ease)
• Flow (transformation)
In the present moment, there is nothing outside of these facets that human consciousness is ever experiencing/interacting with. (And we are only ever interacting with the present moment.)
Then, there are only three ways in which we interact with these facets. Human consciousness can turn against or away from the experience of these facets - two halves of one coin, that generate stress in our experience. Against and Away (aggression and avoidance) are the mind experiencing reactivity, reacting to thought, and we experience this as the endless complication of our intellect.
The third option for human consciousness is turning towards the experiences of these facets. Turning towards (resting into, savoring this living moment, embracing, participating with) is the experience often explored through meditation. Resting into is the state of flow, where we experience maximum creativity (music and arts), athletic achievement, connection, etc.
When the mind is authentically turning towards experience, this is when profound spiritual insight unfolds, where insightful creativity unfolds - where our innate genius expresses.
The interaction of the three possibilities for conscious experience (against/away/towards), with the immutable facets of the present moment, are the absolute reduction to the primitives of human experience.
Here, another interesting paper I just found (you could probably just scan Hestenes' researchgate):
I challenge you to reduce entropy, or energy for that matter, to a combination of primitives! Physicists are famous for saying, "We don't even know what entropy (energy) is?" These are just properties of the physical systems we isolate and study, patterns discovered in their mathematical models.
I would highly recommend that you read David Hestenes on this subject. Hestenes is better known for his work in math and physics education research than he is for his work in mathematical physics. He took a grad student in math and physics education and they researched this subject in depth, what you are discussing here. They developed the Force Concept Inventory, the very first Concept Inventory. They found that very few students, 15%, understood the concept of Newtonian force at the end of an introductory class on Newtonian dynamics: Newtonian force is a momentum exchange but our common sense "understanding" gets in the way. Anyway, he developed a Model Theory of Mind and a Model Theory of Math and Physics Education. It's great stuff!
Here's a paper on the FCI: http://ptc.weizmann.ac.il/_Uploads/dbsAttachedFiles/1852FCI.pdf
Here's a paper on Model Theory: http://modeling.asu.edu/R&E/Hestenes-ModelingTheory2007.pdf
There's more stuff out there, most from the 1980's and 1990's. By the way, did you read those Uli Klein papers I sent you via email, the Quantum Foundations papers? Most essentially, he shows that quantum mechanics is a substructure in an extended version of classical probabilistic mechanics. The classical probabilistic mechanics he calls HLLK theory and its dynamical equation, which models flows on phase space, is of the same form as Schroedinger's equation; it reduces to Schroedinger's equation when he projects away momentum with a simple projection to configuration space. Unlike normal classical probabilistic theories, in HLLK the probability density and the action modeling the deterministic evolution of the systems are coupled; it is this coupling and the desire for linearity which necessitates the complex algebra. Finally, Planck's constant only becomes meaningful upon the projection to configuration space, which strongly supports Hestenes' Zitter model. Presently he is working on a more complex projection which includes momentum fields and produces spin 1/2 particles.
You might want to check out the Basic Formal Ontology, as it is the most successful attempt so far to reduce all human concepts to combinations of a few semantic primitives. https://basic-formal-ontology.org/
The search for semantic primitives first became popular in the late 1960s, with Roger Schank's Conceptual Dependency. In the 1970s, Yorick Wilks and Ray Jackendoff also developed their own sets of primitives, and Stu Shapiro created SNePS, which has a very small number of primitives at a kind of intermediate level between logic and semantics, in which most "semantic primitives" are "nominalist", or perhaps "structuralist", by which I mean their semantics are defined not in the inference engine, but by the set of propositions they're used in.
Obviously semantic primitives can work; neurons have just one semantic primitive, which is the spike train. But this is at a representational level far below what we think of as "semantic"--even below what we think of as "logic".
I don't think it's possible to have semantic primitives on the high, abstract level where people want to have them. The reason symbolic AI failed was that it is too high-level--the fuzziness, flexibility, and generalizability which symbolic AI fails so dramatically and consistently at are all due to the representation being based on neuronal representations. You just can't get the flexibility if you remove the sub-symbolic level; I think this has been definitively demonstrated over the past 3 decades in which statistical learning has climbed higher and higher while symbolic AI has not.
At the neural level, semantic primitives make no sense, because everything is grounded in the senses, and all of the senses--and also motor output--are extremely idiosyncratic, depending on the sensors and effectors they connect to. The brain uses general learning mechanisms whose semantics are defined externally to the brain itself, in the physical properties of sensors and effectors, which the brain has no knowledge of.