Multi-turn Evaluation of Anthropomorphic Behaviours in Large Language Models