A Survey on Textual Entailment: Benchmarks, Approaches and Applications

Abstract

Textual Entailment Recognition (TER), also known as natural language inference, is a crucial task in natural language processing that combines many fundamental aspects of language understanding. TER focuses on predicting the inference relationship between text fragments. Given two sentences (known as premise and hypothesis), the goal is to determine if the meaning of the hypothesis can be entailed/inferred from the premise. Understanding this relationship between two texts can be helpful in several tasks, such as information retrieval, semantic parsing, and common-sense reasoning. This survey paper provides an overview of TER and its variants and applications. We then highlighted TER benchmark datasets for various languages and the main approaches that have been proposed to tackle the problem for a better understanding of the progress this task has reached.