CAREER: Logical Form Induction

近年来，人工智能（AI）系统的自然语言处理能力取得了令人瞩目的进步。除了众多的商业应用之外，这些进步表明，人工智能系统可能是加深我们对人类如何理解自然语言的理解的强大工具。将它们用于此目的的一个主要障碍是，虽然它们似乎可以很好地模拟类比推理的某些方面，但它们模拟复杂逻辑推理的能力仍有很大的改进空间。该项目开发了一个框架，用于将复杂的逻辑推理能力集成到人工智能系统的组件中，使其通过类比推理的能力成为可能。为了支持该框架的开发，该项目建立了一个大型数据集，通过使用人工智能系统来确定哪种逻辑关系对提高系统自身的逻辑推理能力最有用，从而捕获三种语言句子之间的逻辑关系。通过与研究生和本科生课程的整合，该项目作为一种工具，通过实际应用培训，提高编程和统计知识以及数据收集和数据管理技能。该框架将逻辑表示集成到人工智能系统中，对这些系统用于根据一些自然语言输入进行推理的数字表示类型施加限制。这些约束是根据从自然语言的系统数字表示到逻辑表示的映射来定义的。这种映射是从零开始学习的，并且它本身被约束(a)正确预测语言的实际使用者所做的推断——正如在项目下收集的大规模数据集所捕获的那样——以及(b)是组合的：某些语言的含义必须从其部分的含义中预测出来。该奖项反映了美国国家科学基金会的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Artificial intelligence (AI) systems’ natural language processing capabilities have made remarkable strides in recent years. Beyond their numerous commercial applications, these advances suggest that AI systems might be powerful tools for deepening our understanding of how humans comprehend natural language. A major obstacle to using them for this purpose is that, while they seem to simulate certain aspects of reasoning by analogy quite well, their capacity to simulate complex logical reasoning shows much room for improvement. This project develops a framework for integrating complex logical reasoning capabilities into the components of AI systems that make their ability to reason by analogy possible. To support the development of this framework, the project builds a large dataset capturing the logical relationships among sentences in three languages by using AI systems to determine which kinds of logical relationships are most useful for improving that system’s own logical reasoning capabilities. Through integration with graduate and undergraduate curricula, the project serves as a vehicle to enhance programming and statistical literacy as well as data collection and data management skills through training with hands-on applications. The framework integrates logical representations into AI systems by imposing constraints on the sorts of numeric representations that those systems use to make inferences on the basis of some natural language input. These constraints are defined in terms of a mapping from the system’s numeric representations of natural language to logical representations. This mapping is learned from scratch and itself constrained (a) to correctly predict inferences that actual speakers of a language make – as captured by the large-scale datasets collected under the project – and (b) to be compositional: the meaning of some piece of language must be predictable from the meanings of its parts.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.