Event knowledge and language comprehension

事件知识和语言理解

• 批准号: RGPIN-2018-05652
• 负责人: McRae, Ken
• 金额: $ 2.4万
• 依托单位: University of Western Ontario
• 依托单位国家: 加拿大
• 项目类别: Discovery Grants Program - Individual
• 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=764083
• 关键词: Event; knowledge; language; comprehension

Our lives consist of a series of events. Over our lifetime, we experience a stream of activity that consists of a huge number of events and situations. As such, people have a tremendous amount of knowledge of everyday (and not-so-everyday) events, such as eating breakfast, washing dishes, going to a movie, and attending a wedding. Our knowledge of events plays a major role in our ability to understand what is going on around us, to predict what might happen next, to plan activities in the short and long term, and to navigate social situations. Knowledge of common events also plays a central role in our ability to understand language. It would not be possible to understand what people are saying to you if you lacked an understanding of how the world works.Because of the importance of event knowledge for how people think and communicate, there are fundamental scientific questions such as (1) How is event knowledge learned? (2) How is it represented in memory? (3) How is the order in which events occur represented in memory? (4) How do we use event knowledge to predict future activities? (5) When people are listening to a conversation or reading, how is event knowledge used to make inferences about things that have not been explicitly stated?Our research attempts to address these questions. We have built what is called a “neural network model” that learns about real-world events. The model learns to fill in missing components of events (when the model is told that “Bonnie cuts her steak”, it infers that she's using a knife). It also uses prediction-based learning. This is an elegant method for learning because the model (and presumably people) predicts what might happen next, and then learns by comparing its own predictions to what actually did happen next. In this way, knowledge of events “emerges” from prediction-based learning.Over the next 5 years, we will significantly extend our recent model of event knowledge. We will test the model's predictions by comparing its performance to that of people. We also will conduct studies regarding how people use their knowledge of events to understand what they hear and read, particularly in cases in which language is used to expand, contract, or spotlight time.Basic scientific understanding of how people represent events in memory, and how people use that knowledge to understand language, and to make their way around the world, is extremely important on a number of fronts. For example, we need to understand how memory systems like this work so that we can better confront various types of dementia, such as Alzheimer's. We need to understand how children learn (or fail to learn efficiently) this type of information. Constructing and testing models of how event knowledge is learned, stored in memory, and could be damaged, are part of the pathway to making headway on these issues.

我们的生活由一系列事件组成。在我们的一生中，我们经历了一系列由大量事件和情况组成的活动。因此，人们对日常（和非日常）事件有着大量的知识，比如吃早餐、洗碗、看电影和参加婚礼。我们对事件的了解在我们理解周围正在发生的事情、预测接下来可能发生的事情、计划短期和长期活动以及驾驭社会情境的能力中发挥着重要作用。对共同事件的了解在我们理解语言的能力中也起着核心作用。如果你不了解这个世界是如何运作的，你就不可能理解人们对你说的话。由于事件知识对人们如何思考和交流的重要性，有一些基本的科学问题，如（1）事件知识是如何学习的？(2)它在记忆中是如何表现的？(3)事件发生的顺序在记忆中是如何表现的？(4)我们如何使用事件知识来预测未来的活动？(5)当人们在听对话或阅读时，事件知识是如何被用来对没有被明确陈述的事情做出推论的？我们的研究试图解决这些问题。我们已经建立了一个所谓的“神经网络模型”，它可以学习真实世界的事件。模型学习填充事件中缺失的组件（当模型被告知“Bonnie cuts her steak”时，它推断她正在使用刀）。它还使用基于预测的学习。这是一种优雅的学习方法，因为模型（可能还有人）预测接下来可能发生的事情，然后通过将自己的预测与接下来实际发生的事情进行比较来学习。通过这种方式，事件的知识从基于预测的学习中“涌现”出来。在接下来的5年里，我们将显著扩展我们最近的事件知识模型。我们将通过将模型的表现与人的表现进行比较来测试模型的预测。我们还将研究人们如何利用他们对事件的知识来理解他们所听到和阅读的内容，特别是在语言被用来扩展，收缩或聚焦时间的情况下。人们如何在记忆中代表事件，以及人们如何使用这些知识来理解语言，并在世界各地行走，这在许多方面都非常重要。例如，我们需要了解像这样的记忆系统是如何工作的，这样我们就可以更好地面对各种类型的痴呆症，如阿尔茨海默氏症。我们需要了解儿童如何学习（或无法有效学习）这类信息。构建和测试事件知识如何学习，存储在记忆中以及可能被损坏的模型，是在这些问题上取得进展的途径的一部分。