Synthetic Dataset for Evaluating Complex Compositional Knowledge for Natural Language Inference