CAREER: Automaton Theories of Human Sentence Comprehension

人类用来理解一句话的心理过程，就像这样，可以用自动机来模拟。这些自动机是一种特定类型的计算机程序，当运行时，它们模仿认知科学家认为在人们阅读或听到单个单词的时间点到他们知道整个句子的含义之间发生的步骤。在这个项目中，研究团队正在构建自动机，其中通常单独研究三个因素。第一个因素是语言语法，它描述了人们在了解一种语言时所知道的东西。第二个是控制策略，它决定了理解者及时部署这些知识的特定顺序。第三个因素是单词和短语的记忆理论。研究人员预计，这些拥有三重天赋的自动机将在各种句子类型上与人类的表现相匹配。此外，由于自动机可以在数学上进行修改，以考虑到替代的语法、记忆和控制策略，因此它们可以用来获得从真人行为实验中不易获得的见解。例如，对老年人进行的行为实验可能会显示记忆容量降低对句子理解过程的影响。然而，使用自动机，研究人员不仅可以有选择地改变记忆，还可以有选择地改变控制和语法，以人类实验无法揭示的方式揭示每一个的作用。同样，例如，在数学上使机器人的语法更像西班牙语(而不是英语)，也可以更深入地理解这三个因素是如何相互作用的，以及更广泛地说，人们是如何理解彼此的。

The mental process humans use to comprehend a sentence, like this one,can be simulated by automata. These automata are specific kinds of computer programs that, when run, mimic the steps cognitive scientists think occur between the point at which people read or hear individual words and when they know what whole sentences mean.In this project, the research team is constructing automata in which three factors, usually studied singly, are combined. The first factor is the linguistic grammar, which characterizes what people know when they know a language. The second is the control strategy, which determines the particular order in which comprehenders deploy this knowledge in time. The third factor is the theory of memory for words and phrases. The researchers expect these triply-endowed automata to match human performance in a variety of sentence types. In addition, since automata can be mathematically altered to take into account alternative grammars, memories and control strategies, they can be used to gain insights that would not come easily from behavioral experiments with real people. For example, behavioral experiments with elderly adults might show the effects of reduced memory capacity on the sentence comprehension process. However, using automata allows the researchers to selectively alter not only memory but also control and grammar, revealing the role of each in ways that human experiments could not. Similarly, mathematically rendering the automaton's grammar more Spanish-like (as opposed to English-like), for example, could also yield a deeper understanding of how the three factors interact and, more broadly, how people understand one another.