Can We Utilize Pre-trained Language Models within Causal Discovery Algorithms?