Natural Language Inference by Tree-Based Convolution and Heuristic Matching

TL;DR

This paper introduces the TBCNN-pair model that uses tree-based convolutional neural networks and heuristic matching layers to improve natural language inference by better capturing sentence semantics and relationships.

Contribution

The paper presents a novel TBCNN-pair model combining tree-based convolution and heuristic matching for enhanced natural language inference performance.

Findings

Outperforms existing sentence encoding methods significantly

Effective in recognizing entailment and contradiction

Demonstrates the strength of tree-based convolution in NLI

Abstract

In this paper, we propose the TBCNN-pair model to recognize entailment and contradiction between two sentences. In our model, a tree-based convolutional neural network (TBCNN) captures sentence-level semantics; then heuristic matching layers like concatenation, element-wise product/difference combine the information in individual sentences. Experimental results show that our model outperforms existing sentence encoding-based approaches by a large margin.