Does this system use the surrounding context of the conversation when doing its classification, or is each statement processed independently of what came before it?