CAREER: The Grammar Matrix: Computational Linguistic Typology

语言学家通过建立语法模型来研究自然语言的结构，语法模型将单词串与其语义表示联系起来。由于自然语言的结构非常复杂，因此将这些模型构建为计算机软件有很大的好处，无论是在管理模型子部分的复杂交互方面，还是在对大量自然语言文本的模型进行自动化测试方面。然而，这种计算机化模型既昂贵又难以建立，因此大多数语言假设测试都是离线进行的。同时，语言之间有很大的相似之处，这表明针对一些研究得很好的语言的计算机化模型的现有工作可以被用来为其他语言快速启动模型的创建。Emily Bender博士的目标是开发一个语法定制系统，该系统将跨语言核心语法与一系列“库”相结合，指定实现不同语言的不同语言子系统的替代方法（例如，时态和aspect的表达，协调或否定）。这个系统将允许语言学家轻松地为特定语言定制一个小的工作语法，然后他们可以将其用作进一步语言学研究的试验台。构建系统本身的过程将涉及对计算语言类型学的前所未有的探索，探索跨语言样本的各种语言现象的精确分析，这些现象显示了变化的边界。这个项目更广泛的影响包括可能更快、更精确地记录濒危语言，以及在资源不足的语言中更便宜地开发自然语言技术（包括机器翻译、语法检查和计算机辅助语言学习程序）。本项目由社会行为经济科学理事会行为与认知科学部语言学项目和计算机信息科学与工程理事会信息与智能系统部健壮智能集群共同资助。

Linguists investigate the structure of natural languages by building models called grammars which associate strings of words with their semantic representations. As natural languages are quite complex in their structure, there are great benefits to building these models as computer software, both in managing the complex interactions of subparts of the model and in automating the testing of the model against large collections of natural language text. However, such computerized models are expensive and difficult to build, so most linguistic hypothesis testing takes place off-line. At the same time, there are great similarities among languages, suggesting that existing work on computerized models for a few well-studied languages can be leveraged to jump start the creation of models for other languages.With this CAREER award, Dr. Emily Bender aims to develop a grammar customization system that combines a cross-linguistic core grammar with a series of 'libraries' specifying alternate ways of realizing different linguistic subsystems that vary across languages (e.g., the expression of tense and aspect, coordination, or negation). This system will allow linguists to easily customize a small working grammar for a particular language, which they can then use as a test-bed for further linguistic investigation. The process of building the system itself will involve an unprecedented exploration of computational linguistic typology, exploring precise analyses of diverse linguistic phenomena across a sample of languages which show the boundaries of variation. The broader impacts of this project include the potential for quicker, more precise documentation of endangered languages and cheaper development of natural language technology (including machine translation, grammar checkers and computer-assisted language learning programs) in under-resourced languages. This project is cofunded by the Linguistics Program in the Behavioral & Cognitive Sciences Division of the Social, Behavioral & Economic Sciences Directorate and the Robust Intelligence Cluster in the Information & Intelligent Systems Division of the Computer & Information Science & Engineering Directorate.