A Categorical Analysis of Large Language Models and Why LLMs Circumvent the Symbol Grounding Problem