Cognitive models of the time-varying lexicon

IBM沃森（Watson）或苹果Siri等现代人工智能系统已经显示出理解人类语言的希望。然而，人类语言的一个基本属性挑战了这些技术中的关键假设：自然语言系统通常依赖于固定的词典，而人类的词典则随着时间的推移而变化。特别是，随着时间的推移，词汇在意义上表现出了显著的流动性，例如，gay（“快乐的同性恋”），grasp（“行动的理解”）。是什么认知基础让这一切成为可能？解决这个问题对于理解语言的创造性本质至关重要，并为未来的人工智能研究设计类似人类的词典提供信息。与语言变化的其他方面相比，如语法或声音，这些方面有丰富的形式特征，历史意义的变化是最不容易理解的。一方面，意义变化的认知理论通常面临原则性形式化和大规模批判性评估的双重挑战。另一方面，尽管最近计算语言学的工作在新词义的检测方面显示出了希望，但这些工作通常没有具体说明产生新词义的认知机制。本研究计划旨在通过开发预测词义演变的认知基础模型来弥合这些领域。它将追求发展的长期目标，正式的解释词汇的演变，与短期的重点是理解历史意义变化的原则机制。这种方法将建立在我在机器学习和计算认知科学方面的背景，以及最近在这一领域的研究专长之上。具体来说，它将结合认知语言学理论和计算机科学的正式工具，并利用丰富的经验数据，从大型开放获取资源（包括历史语料库、词典和跨语言词汇）到行为实验。结果将有助于理论和建模的认知原则和机制，支持意义随时间的变化，在三个组的项目中探索单个单词，语义场和词汇的增量水平。该研究项目的跨学科性质将对认知科学、语言学和人工智能领域产生重大影响。因此，它将培养在学术界从事基础科学研究的学者，以及为工业界和其他领域做出贡献的计算科学家。

Modern artificial intelligent systems such as IBM Watson or Apple Siri have shown promise of understanding human language. However, a fundamental property of human language challenges the key assumptions in these technologies: A natural language system typically relies on a fixed lexicon, in contrast with the human lexicon that varies over time. In particular, words have shown remarkable fluidity in meaning over time, e.g., gay ('happy''homosexual'), grasp ('action''understanding'). What are the cognitive foundations that make this possible? Addressing this question will be centrally important to the understanding of the creative nature of language and inform future AI research in the design of a human-like lexicon.Compared to other aspects of language change such as in grammar or sound where there are rich formal characterizations, historical meaning change is least understood. On one hand, cognitive theories of meaning change typically suffer from the dual challenge of principled formalization and critical assessment at scale. On the other hand, although recent work in computational linguistics has shown promise in the detection of novel word meanings, such work does not typically specify the cognitive mechanisms that give rise to emerging meanings. This research program is situated at bridging these fields by developing cognitively grounded models that predict the evolution of word meanings. It will pursue the long term objective of developing a formal account of the evolution of the lexicon, with the short-term emphasis on understanding the principled mechanisms of historical meaning change.The approach will build upon my background in machine learning and computational cognitive science, and recent research expertise in this area. Specifically, it will draw on a combination of cognitive linguistic theories and formal tools from computer science, and engage rich empirical data ranging from large open-access resources including historical corpora, dictionaries and cross-linguistic lexicons, to behavioral experiments. The outcomes will contribute to theories and modeling of the cognitive principles and mechanisms that support meaning change over time, explored in three sets of proposed projects at the incremental levels of individual words, semantic fields, and the lexicon. The interdisciplinary nature of this research program will have a significant impact on the fields of cognitive science, linguistics, and artificial intelligence. As such, it will train both scholars for basic scientific research in academia and computational scientists who will contribute to industry and beyond.