Natural Language Inference over Interaction Space

ICLR 2018 Yichen GongHeng LuoJian Zhang

Natural Language Inference (NLI) task requires an agent to determine the logical relationship between a natural language premise and a natural language hypothesis. We introduce Interactive Inference Network (IIN), a novel class of neural network architectures that is able to achieve high-level understanding of the sentence pair by hierarchically extracting semantic features from interaction space... (read more)

PDF Abstract

Evaluation results from the paper


Task Dataset Model Metric name Metric value Global rank Compare
Paraphrase Identification Quora Question Pairs DIIN Accuracy 89.06 # 3
Natural Language Inference SNLI 448D Densely Interactive Inference Network (DIIN, code) % Test Accuracy 88.0 # 16
Natural Language Inference SNLI 448D Densely Interactive Inference Network (DIIN, code) % Train Accuracy 91.2 # 27
Natural Language Inference SNLI 448D Densely Interactive Inference Network (DIIN, code) Parameters 4.4m # 1
Natural Language Inference SNLI 448D Densely Interactive Inference Network (DIIN, code) Ensemble % Test Accuracy 88.9 # 9
Natural Language Inference SNLI 448D Densely Interactive Inference Network (DIIN, code) Ensemble % Train Accuracy 92.3 # 22
Natural Language Inference SNLI 448D Densely Interactive Inference Network (DIIN, code) Ensemble Parameters 17m # 1