Indirect Isomorphisms in Mirror Domains

Imagine that you introduce an LLM to a totally new domain, say the fighters in a fictional fighting tournament in a fictional universe, or the algae ecosystem on an alien planet. Say you describe this domain such that its maps perfectly onto some other structure, e.g., Pokemon or different kinds of Prokaryotes. Would LLMs be better at modeling domains that map onto something they knew, even though all the words are different, would they have some special intuition that would guide them to repeat the mapping? And if you trained a probe, would see some remnant of the mirror domain in its features?