Probing LLMs for Joint Encoding of Linguistic Categories