A natural language dictionary can be split into words that can be defined in terms of simpler words and words where the definitions end up cyclical. You can start simplifying a dictionary by repeatedly removing words that do not occur in any word definition in the remaining dictionary, until you can’t remove any more words. What you are left with is called a grounding kernel. This is relevant to the symbol grounding problem in AI, where language comprehension needs to bottom out to systems not expressed in the language being comprehended. Words in the grounding kernel are defined in terms of each other, so you need some way to bootstrap the understanding of them, but you can also single them out as the core problem area instead of trying to deal with the entire natural language lexicon.