We present a novel deep learning architecture to address the natural language
inference (NLI) task. Existing approaches mostly rely on simple reading
mechanisms for independent encoding of the premise and hypothesis. Instead, we
propose a novel dependent reading bidirectional LSTM network (DR-BiLSTM) to
efficiently model the relationship between a premise and a hypothesis during
encoding and inference. We also introduce a sophisticated ensemble strategy to
combine our proposed models, which noticeably improves final predictions.
Finally, we demonstrate how the results can be improved further with an
additional preprocessing step. Our evaluation shows that DR-BiLSTM obtains the
best single model and ensemble model results achieving the new state-of-the-art
scores on the Stanford NLI dataset.
Description
[1802.05577] DR-BiLSTM: Dependent Reading Bidirectional LSTM for Natural Language Inference
%0 Generic
%1 ghaeini2018drbilstm
%A Ghaeini, Reza
%A Hasan, Sadid A.
%A Datla, Vivek
%A Liu, Joey
%A Lee, Kathy
%A Qadir, Ashequl
%A Ling, Yuan
%A Prakash, Aaditya
%A Fern, Xiaoli Z.
%A Farri, Oladimeji
%D 2018
%K naacl2018 neuralnet nli rnn session9
%T DR-BiLSTM: Dependent Reading Bidirectional LSTM for Natural Language
Inference
%U http://arxiv.org/abs/1802.05577
%X We present a novel deep learning architecture to address the natural language
inference (NLI) task. Existing approaches mostly rely on simple reading
mechanisms for independent encoding of the premise and hypothesis. Instead, we
propose a novel dependent reading bidirectional LSTM network (DR-BiLSTM) to
efficiently model the relationship between a premise and a hypothesis during
encoding and inference. We also introduce a sophisticated ensemble strategy to
combine our proposed models, which noticeably improves final predictions.
Finally, we demonstrate how the results can be improved further with an
additional preprocessing step. Our evaluation shows that DR-BiLSTM obtains the
best single model and ensemble model results achieving the new state-of-the-art
scores on the Stanford NLI dataset.
@misc{ghaeini2018drbilstm,
abstract = {We present a novel deep learning architecture to address the natural language
inference (NLI) task. Existing approaches mostly rely on simple reading
mechanisms for independent encoding of the premise and hypothesis. Instead, we
propose a novel dependent reading bidirectional LSTM network (DR-BiLSTM) to
efficiently model the relationship between a premise and a hypothesis during
encoding and inference. We also introduce a sophisticated ensemble strategy to
combine our proposed models, which noticeably improves final predictions.
Finally, we demonstrate how the results can be improved further with an
additional preprocessing step. Our evaluation shows that DR-BiLSTM obtains the
best single model and ensemble model results achieving the new state-of-the-art
scores on the Stanford NLI dataset.},
added-at = {2018-06-04T18:02:02.000+0200},
author = {Ghaeini, Reza and Hasan, Sadid A. and Datla, Vivek and Liu, Joey and Lee, Kathy and Qadir, Ashequl and Ling, Yuan and Prakash, Aaditya and Fern, Xiaoli Z. and Farri, Oladimeji},
biburl = {https://www.bibsonomy.org/bibtex/2d50a0c2ed17f16e0c08cdd71a16da0e7/albinzehe},
description = {[1802.05577] DR-BiLSTM: Dependent Reading Bidirectional LSTM for Natural Language Inference},
interhash = {a502a70c01c38ad26ad8e81c36f21b22},
intrahash = {d50a0c2ed17f16e0c08cdd71a16da0e7},
keywords = {naacl2018 neuralnet nli rnn session9},
note = {cite arxiv:1802.05577Comment: 18 pages, Accepted as a long paper at NAACL HLT 2018},
timestamp = {2018-06-04T18:02:02.000+0200},
title = {DR-BiLSTM: Dependent Reading Bidirectional LSTM for Natural Language
Inference},
url = {http://arxiv.org/abs/1802.05577},
year = 2018
}