Shaping shared languages: human and large language models' inductive biases in emergent communication