Doctoral Dissertation Research: Neural Bases of Semantic Composition in the Verb Phrase

自然语言语义的丰富性源于将元素组合在一起并从这些组合中获得新含义的能力。理解这种能力一直是语言神经生物学和理论语言学的中心目标。然而，尽管研究的主体迅速增加，但人们对组成的大脑基础充其量也只是模糊地理解。过去十年的大部分进展仅限于对句子层面的一般性描述，确定了为组成句子意义而招募的大脑区域网络，但没有进一步具体说明句子组成的各个方面是如何编码的。考虑到形式语义学中的假设，即合成可以通过至少两种不同的规则发生，理解各种语言环境中的合成的非常基本的实例是表征句子解释的神经基础的关键的第一步。为此，本文采用自下而上的研究方法，以英语动词短语为研究对象，对小成分的神经基础进行研究。围绕动词的两种主要构成模式，即修饰(例如，画图+缓慢)和论元饱和(例如，画图+图片)，这项工作旨在回答复杂事件描述(例如，约翰不情愿地画了一幅画)的含义是如何由各种事件元素组成的问题，即约翰(施事)、勉强(方式)、绘制(动作)和图片(对象)。这项工作旨在为完整句子成分的计算详细表征奠定关键基础。本论文由三项研究组成，它们利用脑磁图的高时空分辨率来表征成分的动力学。第一个实验研究了修饰语在动词领域的神经特征，询问动词和它的修饰语的组合如何与大脑接触，以及这种整体模式与名词-修饰语组合的整体模式相比如何，之前的工作已经描述了稳定的结果。第二个实验考察了概念特异性在事件域中的神经表征。以前的文献已经表明，大脑的一个特定区域，即左前颞叶(LATL)，对对象领域(例如，船与独木舟)的概念特异性很敏感。第二个实验以同样的方式询问事件专一性是否也涉及相同的区域，第二个实验操纵了事件的概念性专用性(例如，“囤积燃料”与“储存燃料”)。最后一个实验探索了(可能的)语义合成的情况，在这种情况下，组合的单词不构成句法成分。总而言之，这些研究结合了理论语言学和心理学文献中关于概念和类别的见解，以阐明动词短语解释中事件组成的神经反射。这一奖项反映了NSF的法定使命，并通过使用基金会的智力价值和更广泛的影响审查标准进行评估，被认为值得支持。

The richness of natural language semantics stems from the ability to compose elements together and to derive novel meanings from these combinations. Understanding this ability has been a central goal in the neurobiology of language, as well as in theoretical linguistics. However, despite a rapidly growing body of research, the brain basis of composition has been vaguely understood at best. Most of the progress during the last decade has been limited to general descriptions of the sentence level, identifying a network of brain areas recruited for the composition of sentential meaning but not further specifying how the individual facets of sentence composition are encoded. Considering the assumption in formal semantics that composition can happen via at least two different rules, understanding very basic instances of composition in various linguistic environments is an essential first step for characterizing the neural basis of sentence interpretation. To this end, this dissertation takes a bottom-up approach, investigating neural substrates of small units of composition, with its focus on English verb phrases. Disentangling the two major modes of composition around a verb, namely modification (e.g., paint + slowly) and saturation of arguments (e,g., paint + a picture), this work aims to provide an answer to the question of how the meaning of a complex event description, e.g., "John reluctantly painted a picture," is composed from the various event elements, i.e., John (agent), reluctantly (manner), painted (action), and a picture (object). This work aims to lay a critical foundation for a computationally detailed characterization of full sentence composition. This dissertation consists of three studies that take advantage of the high spatiotemporal resolution of magnetoencephalography to characterize the dynamics of composition. The first experiment investigates the neural signatures of modification in the verbal domain, asking how the combination of a verb and its modifier engages the brain, and how this overall pattern compares to that of noun-modifier combination, for which stable results have been described in prior work. The second experiment investigates the neural representation of conceptual specificity in the event domain. Previous literature has shown that a particular brain region, the left anterior temporal lobe (LATL), is sensitive to conceptual specificity in the object domain (e.g., boat vs. canoe). Asking whether event specificity also involves the same area in the same way, the second experiment manipulates the conceptual specificity of events (e.g., "hoard fuel" vs "store fuel"). The last experiment explores situations of (possible) semantic composition in which the combining words do not form a syntactic constituent. In all, these studies combine insights from theoretical linguistics and from the psychological literature on concepts and categories to elucidate the neural reflexes of event composition as manifested in verb phrase interpretation.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.