Atomian Papers

This paper presents a knowledge representation model designed to separate knowledge from the language used to express it. The model represents knowledge through a finite set of atomic primitives and their combinations, providing a common substrate for entities, facts, episodes, qualifications, relations, logical content, and dimensional information. This separation makes it possible to acquire and update knowledge independently of language, potentially allowing systems to incorporate new knowledge dynamically without retraining their language models. It also provides an explicit representation that can be inspected, manipulated, and reused across different linguistic interfaces.

The model is founded on the principles of orthogonality, compositionality, and closure, with meta-entitization allowing elements of the model itself to become explicit entities. Its expressive capacity is examined through mappings to established knowledge representation and logical formalisms and through progressively complex examples expressed in a visual modelling language called Atoms Modelling language (AML). The model has been implemented and is presented as a general-purpose representational foundation for systems in which knowledge, reasoning, and language can be treated as distinct but interacting components.