Given a certain language L, let's define it's language graph G as the graph where nodes correspond to words and a directed edge between node B and node A implies that the word referenced by B belongs to the definition of the word referenced by A.
My question is, can we construct a set of definitions that still reflect the semantic meaning of the words of L while also resulting in G being acyclic? If it is not the case, can we construct an artificial language that follows this property?
An acyclic definition graph would imply the existence of certain words without a definition but such a graph can result in a topological sort of words. Thus an ordered hierarchy for learning the words of the language.