Ranking Generated Summaries by Correctness: An Interesting but Challenging Application for Natural Language Inference