LLM Comparative Assessment: Zero-shot NLG Evaluation through Pairwise Comparisons using Large Language Models