DR-BiLSTM: Dependent Reading Bidirectional LSTM for Natural Language Inference

  • 2018-04-11 02:45:25
  • Reza Ghaeini, Sadid A. Hasan, Vivek Datla, Joey Liu, Kathy Lee, Ashequl Qadir, Yuan Ling, Aaditya Prakash, Xiaoli Z. Fern, Oladimeji Farri
  • 0

Abstract

We present a novel deep learning architecture to address the natural languageinference (NLI) task. Existing approaches mostly rely on simple readingmechanisms for independent encoding of the premise and hypothesis. Instead, wepropose a novel dependent reading bidirectional LSTM network (DR-BiLSTM) toefficiently model the relationship between a premise and a hypothesis duringencoding and inference. We also introduce a sophisticated ensemble strategy tocombine our proposed models, which noticeably improves final predictions.Finally, we demonstrate how the results can be improved further with anadditional preprocessing step. Our evaluation shows that DR-BiLSTM obtains thebest single model and ensemble model results achieving the new state-of-the-artscores on the Stanford NLI dataset.

 

Quick Read (beta)

loading the full paper ...