Natural Language Satisfiability: Exploring the Problem Distribution and Evaluating Transformer-based Language Models