A Decomposable Attention Model for Natural Language Inference

   Abstract