Recognizing Textual Entailment
Mark Sammons · 2015
This chapter provides an overview of applied research into recognizing textual entailment. It identifies the fundamental challenges encountered so far, and surveys the models used to represent meaning and to determine entailment. The chapter situates and defines the recognizing textual entailment (RTE) task specification as it has developed over time, and addresses critiques and refinements of the task. It surveys the kinds of knowledge and capabilities required to perform the task. The chapter describes a straightforward application of a proof-theoretic model and a shallow lexical model to the entailment task and explains their limitations. It also introduces concepts from machine learning that are used by many RTE systems. The chapter presents two less constrained proof-theoretic models for textual inference. Section 6 surveys different models used by actual RTE systems. It finally summarizes the research situation and indicates promising lines of research to improve the state of the art.