Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Neural Natural Language Inference Models Enhanced with External Knowledge

Qian Chen, Xiaodan Zhu, Zhen-Hua Ling, Diana Inkpen, Si Wei

TL;DR

This work addresses the limits of purely data driven natural language inference by integrating external lexical knowledge into neural NLI models. It introduces KIM, a framework that injects WordNet based knowledge into co attention, local inference collection, and inference composition within a BiLSTM based architecture. Empirical results show state of the art performance on SNLI and MultiNLI, with particularly large gains when training data are scarce, and strong generalization on a lexical inference test set. The findings highlight the value of combining external knowledge with neural models and suggest broader applicability to other NLP tasks requiring nuanced lexical inferences.

Abstract

Modeling natural language inference is a very challenging task. With the availability of large annotated data, it has recently become feasible to train complex models such as neural-network-based inference models, which have shown to achieve the state-of-the-art performance. Although there exist relatively large annotated data, can machines learn all knowledge needed to perform natural language inference (NLI) from these data? If not, how can neural-network-based NLI models benefit from external knowledge and how to build NLI models to leverage it? In this paper, we enrich the state-of-the-art neural natural language inference models with external knowledge. We demonstrate that the proposed models improve neural NLI models to achieve the state-of-the-art performance on the SNLI and MultiNLI datasets.

Neural Natural Language Inference Models Enhanced with External Knowledge

TL;DR

This work addresses the limits of purely data driven natural language inference by integrating external lexical knowledge into neural NLI models. It introduces KIM, a framework that injects WordNet based knowledge into co attention, local inference collection, and inference composition within a BiLSTM based architecture. Empirical results show state of the art performance on SNLI and MultiNLI, with particularly large gains when training data are scarce, and strong generalization on a lexical inference test set. The findings highlight the value of combining external knowledge with neural models and suggest broader applicability to other NLP tasks requiring nuanced lexical inferences.

Abstract

Modeling natural language inference is a very challenging task. With the availability of large annotated data, it has recently become feasible to train complex models such as neural-network-based inference models, which have shown to achieve the state-of-the-art performance. Although there exist relatively large annotated data, can machines learn all knowledge needed to perform natural language inference (NLI) from these data? If not, how can neural-network-based NLI models benefit from external knowledge and how to build NLI models to leverage it? In this paper, we enrich the state-of-the-art neural natural language inference models with external knowledge. We demonstrate that the proposed models improve neural NLI models to achieve the state-of-the-art performance on the SNLI and MultiNLI datasets.

Paper Structure

This paper contains 21 sections, 7 equations, 3 figures, 7 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Neural-Network-Based NLI Models with External Knowledge
  4. External Knowledge
  5. Encoding Premise and Hypothesis
  6. Knowledge-Enriched Co-Attention
  7. Local Inference Collection with External Knowledge
  8. Knowledge-Enhanced Inference Composition
  9. Experiment Set-Up
  10. Representation of External Knowledge
  11. Lexical Semantic Relations
  12. Relation Embeddings
  13. NLI Datasets
  14. Training Details
  15. Experimental Results
  16. ...and 6 more sections

Figures (3)

  • Figure 1: A high-level view of neural-network-based NLI models enriched with external knowledge in co-attention, local inference collection, and inference composition.
  • Figure 2: Accuracies of models of incorporating external knowledge into different NLI components, under different sizes of training data (0.8%, 4%, 20%, and the entire training data).
  • Figure 3: Accuracies of models under different sizes of external knowledge. More external knowledge corresponds to higher accuracies.