An adaptive model of lexical repairs in language production

理解人类如何学习、产生和理解语言是认知科学的基本工作之一。这个项目研究了语音处理中一个经常被忽视的方面，即我们如何在说话时发现并修复我们所犯的错误。尽管以母语为母语的成年人不会犯很多语言错误，但儿童、第二语言使用者和脑损伤的人会，而且他们中的许多人表现出发现和修复自己错误的非凡能力。修复错误很重要，因为不进行修复，沟通很容易中断。此外，在一个新的学习者中，如儿童或第二语言学习者中，培养修复错误的能力是学习者取得进步的一个很好的信号。然而，我们对大脑如何捕捉和修复语言错误知之甚少。我们也不知道说不同语言的人这种能力是否不同。这个项目揭示了这些问题。利用现有的语言生成模型，我们构建了一个模拟大脑如何检测和修复语言错误的计算模型。然后，我们使用一个大型数据库来测试这个模型的预测，这个数据库是我们在项目的整个生命周期中建立的，来自700多名母语为英语、法语、荷兰语和波斯语的人。利用这些数据，我们进一步完善和微调计算模型，以获得一个最能捕捉说话者如何修复语音错误的模型。该项目有两个主要成果：(1)最大的公开的多语言语音错误和修复数据库。(2)第一个开源的语音错误修复计算模型。世界上任何对理解或研究人类如何产生语言感兴趣的人都可以免费访问这两个网站。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。

Understanding how humans learn, produce and understand speech is one of the fundamental endeavors of cognitive science. This project investigates an often-overlooked aspect of speech processing, namely, how we can catch and repair the errors we make when we speak. Even though native adult speakers do not make many speech errors, children, second-language speakers, and individuals with brain injury do, and many of them show a remarkable ability to detect and repair their own errors. Repairing errors is important, because without repairs, communication can easily break down. Also, developing the ability to repair one’s errors in a new learner, like a child or a second-language learner, is a good signal that the learner is making progress. Yet, we know very little about how the brain can catch and repair speech errors. We also do not know if this ability differs between speakers of different languages. This project sheds light on these issues. Taking advantage of existing models of language production, we construct a computational model which mimics how the brain may detect and repair speech errors. We then test the predictions of this model using a large database that we build over the lifetime of the project from over 700 native speakers of four languages, English, French, Dutch, and Farsi. Using these data, we further refine and fine-tune the computational model to arrive at a model that best captures how speakers repair their speech errors. The project has two major outcomes: (1) The largest publicly available multi-lingual database of speech errors and repairs. (2) The first open-source computational model of speech error repairs. Both of these will be made freely accessible to anyone in the world who is interested in understanding or researching how humans produce language.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.