Universal knowledge engines for lexical semantic creativity in humans and machines

用于人类和机器词汇语义创造力的通用知识引擎

基本信息

批准号：
576149-2022
负责人：
Xu, YangYX
金额：
$ 1.82万
依托单位：
University of Toronto
依托单位国家：
加拿大
项目类别：
Alliance Grants
财政年份：
2022
资助国家：
加拿大
起止时间：
2022-01-01 至 2023-12-31
项目状态：
已结题

来源：
https://www.nserc-crsng.gc.ca/ase-oro/Details-Detailles_eng.asp?id=750011
关键词：
Universal knowledge engines lexical semantic

项目摘要

This proposed research investigates knowledge structures that are shared in the lexicons across the world's languages and thereby aims to develop universal knowledge engines that will enhance machine interpretation of human languages.A key aspect that distinguishes human language users from artificial natural language systems is our ability to interpret and create novel word usages, and this creative character is reflected in the lexical semantic structures of the world's languages. This phenomenon, known as word meaning extension, poses significant challenges for machine understanding of the most fundamental units of natural languages: words. This collaborative research program will launch a new effort in bridging the gap between humans and machines in lexical semantic creativity, by taking two main steps. 1) We will characterize at a large scale knowledge representations in word meaning extension that are shared across the world's languages and through child language acquisition, drawing data and resources from a variety of publicly accessible venues including cross-linguistic databases, psychological experiments, and deep learning models. 2) We will develop new methodologies that harness and integrate different knowledge modalities toward constructing universal knowledge engines to support the flexible and adaptive interpretation and inference of novel word meanings across languages.

这项拟议的研究调查了世界语言中词典中共享的知识结构，从而旨在开发通用知识工程，以增强机器对人类语言的解释。这是区分人类语言使用者与人工自然语言系统的关键方面，是我们解释和创建新颖的单词用法的能力，并且在世界语言的词汇结构中反映了这种创造性的角色。这种现象被称为单词含义扩展，对机器对自然语言最基本单位的理解构成了重大挑战：单词。该协作研究计划将通过采取两个主要步骤来弥合词汇语义创造力的人与机器之间的差距。 1）我们将以单词含义扩展的大规模知识表示形式来表征，这些表示含义扩展，并通过掌握儿童语言，从包括跨语言数据库，心理学实验和深度学习模型在内的各种公共访问场所中获取数据和资源。 2）我们将开发新的方法论，这些方法可以利用和整合不同的知识方式，以构建通用知识引擎，以支持跨语言的新单词含义的灵活和适应性解释和推断。