The central problem for extracting meaning from words is the conduit metaphor paradox (link: Reddy, 1979). Essentially, words by themselves do not produce meaning as any reaction to any word is arbitrary and both expression and reaction are arbitrary, ML has an insurmountable task that CS never took into account in developing LLMs and the like.
Arbitrary points can never be refined to specifics.
For a more recent take on the conduit metaphor and an appendix discussion, here is Rendall's seminal "What do animal signals mean?" from 2009
https://www.sciencedirect.com/science/article/abs/pii/S00033...