WIP| On the Capabilities and Limitations of Reasoning for Natural Language Understanding

[Under reading]

On the Capabilities and Limitations of Reasoning for Natural Language Understanding
Khashabi et al.
作者来自 University of Pennsylvania, Indiana University 和 AllennAI
题目: 论自然语言理解推理的能力与局限

Abstract

  • 一些NLU系统在克服查找 Style Reasoning 的语言可变性(linguistic variability)方面具有很强的实力, 但是他们的准确率会随着推理步数的增加而下降.
    • key: style reasoning; linguistic variability; reasoning step;
  • 本文基于上述观察第一次提出了一个正式框架, 旨在解决使用语言表示隐藏概念空间时引入的:
    • 1.模糊性(ambiguity)
    • 2.冗余性(redundancy)
    • 3.不完整性(incompleteness)
    • 4.不准确性(inaccuracy)
  • 模型使用了两个相互关联的(interrelated)空间:
    • conceptual meaning space: unambiguous and complete but hidden.
    • linguistic symbol space: captures a noisy grounding of the meaning space in the symbols or words of a language.
  • 本文引用此框架来研究无向图中的连通性(connectivity)问题: 是构成更复杂的多步推理的基础核心推理问题
    • 证明了构建高质量算法来检测 latent meaning graph 中的连通性是可能的, 前提: 基于一个可观察的 noisy symbol graph, 并且这些噪声低于我们定量的噪声等级.
    • 此外, 证明了一个不可能的结果: 如果一个query需要大量的推理步数, no reasoning system operating over the symbol graph is likely to recover any useful property of the meaning graph.
    • 这一点同时强调了对于推理问题和系统,需要的是限制两个空间的距离,而不是投入更多的推理步数(hops)

Introduction

  • Reasoning的定义: the process of combining facts(事实) and beliefs(信念), in order to make decisions.
**** END of This Post. Thank for Your READING ****
If you have any Question, welcome to Email me or leave your comments below.