NSF-BSF: Bridging encoding and retrieval perspectives on sentence processing errors: Comparing Hebrew and English

为了理解语言，人们需要在相隔很远的单词之间建立联系。例如，这句话“有着非常闪亮和健康的黑色皮毛的狗通常不会吠叫”要求听者或读者将狗和吠叫联系起来，即使这两个词相差甚远。要做到这一点，语言使用者需要依靠记忆来连接单词和概念。然而，人类的记忆是出了名的容易出错：人类经常忘记、记错、混淆自己经历的方方面面。在语言理解的背景下，这些记忆失败会导致对句子的错误解释。这个项目旨在通过观察记忆错误如何影响英语和希伯来语这两种截然不同的语言的使用者，来理解记忆如何以及为什么会扭曲阅读和语言理解。英语和希伯来语的不同之处在于它们如何组织句子中的单词，以及它们是否给名词分配性别；希伯来语将阳性和阴性赋予名词，类似于西班牙语和法语等语言，但与英语不同。研究人员将研究希伯来语和英语之间的这些语言差异是如何影响这两种语言的使用者产生口译错误的。在此过程中，研究人员将试图揭示记忆错误的特征，这些特征对跨语言理解和特定语言的理解有相同的影响。这个项目的结果将用于理解人类记忆系统如何支持实时语言理解。关于这个问题的研究表明，当一个句子需要读者记住多个单词时，两种过程会破坏语言理解。当较新单词的特征意外地覆盖较早单词的部分时，就会发生一个过程。这种“编码错误”意味着读者错误地理解了一个由两个不同单词的特征组合而成的单词。例如，在“通往山的路被堵住了”这样的句子中，他们可能会把“路”误认为是“路”，因为他们把“路”的单数形式和“山”的复数形式结合在一起。当试图从记忆中检索一个特定的单词时，可能会出现另一种类型的错误。例如，在理解一个句子的关键时刻，读者或听者可能会从记忆中挑出错误的单词，认为上面的句子被山挡住了，而不是被路挡住了（“检索”错误）。原则上，这两种过程中的一种或两种都会破坏语言理解。研究人员将跟踪人们在阅读英语和希伯来语时的眼球运动，并收集快速可接受性判断。总之，这些措施应反映两种语言的编码和检索错误的相对贡献。在阅读中产生的理解错误的特定模式，将在人类语言处理工作记忆的计算模型中进行测试。这两种研究方法的结合将有助于解释为什么人们在大多数时候如此容易地理解句子，以及为什么误解可能而且确实在其他时候出现。理解解释错误是如何以及何时出现的，也可以帮助我们更好地理解各种非典型语言和阅读模式，比如阅读障碍。该项目将促进美国和以色列语言研究人员之间的合作，并将为不同职业阶段的研究人员提供高级STEM培训，从本科生到博士后学者。该项目由PAC，语言学和EHR核心研究共同资助。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。

To understand language, people need to form links between words that are far apart. For example, the sentence "The dog with the very shiny and healthy black fur doesn’t usually bark" requires the listener or reader to associate the dog with bark, even though those words are quite far apart. To do this, language users need to rely on memory to link words and concepts. However, human memory is famously prone to error: Humans routinely forget, misremember, and conflate aspects of their experience. In the context of language understanding, these memory failures can lead to incorrect interpretations of sentences. This project aims to understand how and why memory can distort reading and language comprehension by looking at how memory errors impact users of two very different languages, English and Hebrew. English and Hebrew differ in how they organize the words within sentences and whether they assign gender to nouns; Hebrew assigns masculine and feminine genders to nouns, similar to languages like Spanish and French but unlike English. The researchers will study how these linguistic differences between Hebrew and English influence when interpretation errors will arise in users of these two languages. In doing so, the researchers will try to uncover characteristics of memory errors that have the same effect on understanding across languages and those that are language-specific. The results of this project will be used to understand how human memory systems support real-time language comprehension.Research on this question suggests that two kinds of processes can disrupt language comprehension when a sentence requires the reader to hold multiple words in memory. One process occurs when the features of more recent words accidentally overwrite parts of earlier words. This type of ‘encoding error’ means that the reader erroneously perceives a word that recombines the features of two different words. For example, in a sentence like “The road to the mountains was blocked,” they may misremember "road" as "roads" by combining the singular “road” with the plural feature of “mountains.” Another type of error can arise when trying to retrieve a particular word from memory. For example, a reader or listener might pick out the wrong word from memory at the critical moment in understanding a sentence, thinking that the mountains were blocked in the sentence above, rather than the road (a ‘retrieval’ error). Language comprehension can in principle be disrupted by either or both of these processes. The investigators will track eye movements of people while they are reading in English and Hebrew and collect speeded acceptability judgments. Together these measures should reflect the relative contribution of encoding and retrieval errors in both languages. The particular pattern of comprehension errors that arises in reading will then be tested against computational models of human working memory in language processing. The combination of these two research methods will help account for how people understand sentences so easily most of the time, and why misinterpretations can and do arise at other times. Understanding how and when interpretation errors arise can also help us better understand various atypical language and reading patterns, such as shown in dyslexia. This project will advance collaboration between American and Israeli language researchers and will involve advanced STEM training for researchers at different career stages, from undergraduates to post-doctoral scholars.This project was co-funded by PAC, Linguistics, and EHR Core Research.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.