漢語空語類的相關腦區探究

Abstract

傳統句法理論將空語類(empty categories)分為四類：名詞軌跡（NP-trace）、疑問詞軌跡(wh-trace)、小隱性代詞(pro)與大隱性代詞(PRO)。空語類各受制於不同的句法限制，也有不同的分布，但這些類別只是理論需要，抑或是人類在句子處理時在行為上、心理上、腦神經上能有相關之反應，包含辨識其空位(gap)及填充語(filler)，並且在空位還原填充語等，都有極大的爭議。雖然行為實驗、事件相關電位實驗，甚至功能性磁振造影實驗對填充語－空位依存關係已有了一定程度的了解，然而包含空語類之間的腦神經機制比較，以及空語類最根本的神經特質為何，卻缺乏全面性的探討。本論文透過三個功能性磁振造影實驗探討形式句法理論中空語類的腦神經基底，了解大腦如何辨識與處理空語類，進而理解大腦處理句法移位的相關腦神經機制。實驗一比較漢語主題結構與關係子句，希望能從填充語－空位的相對順序了解大腦在處理移位方向的差異。實驗結果發現「填充語先行」與「空位先行」的兩種順序皆會引起大腦左額葉下迴(left inferior frontal gyrus) 以及左顳葉上／中迴 (left superior/middle temporal gyrus)的反應，表示該兩個腦區對移位處理至關重要，但對於移動方向則不敏感。然而關係子句的空位先行卻額外誘發了左前顳葉的反應(LaTL)，我們主張該腦區所扮演之角色為借助窺探動詞語意，以推敲空位之語意與詞彙內容，暫行完成動詞片語結構，才能繼續句子處理，預測填充語之存在。第二個實驗比較主題結構與被字句結構，進而了解帶有空位的不同結構是否有類似的處理。過去理論研究認為漢語被字句結構並非如英文一般的派生方式(A-movement)，留下名詞軌跡；但漢語的長被字句被認為帶有一個原生(base-generated)的受事者主詞，伴隨結構後方的隱性空算符移動 (null operator movement)。然而本實驗卻發現被字句處理與主題結構處理極為相似，同樣引起左額葉下迴與左顳葉中迴的反應，因此並無法支持「隱性移動」或者「不移動」的理論假設。但也因此，本實驗無法直接證實名詞軌跡與疑問詞軌跡之間有處理歷程上的差異。第三個實驗比較漢語當中希望類動詞與嘗試類動詞所引導的空主詞子句，其中前者帶有小隱性代詞，後者帶有大隱性代詞，兩者依空語類理論有不同的分布。而根據實驗結果我們發現，小隱性代詞能激發左額葉下迴前端，大隱性代詞則沒有顯著的腦區反應，也反應了兩者處理的難易與差異。左額葉下迴前端的反應可能牽涉到處理小隱性代詞需汲取先行詞的訊息，然而由於小隱性代詞的處理過程並不透過移位，因此未能誘發出前兩個實驗的左額葉下迴的活動。此外，被字句結構與希望類動詞結構同樣都引起右腦前額葉的反應，此反應或許與結構本身有關，因為兩種句構的空語類位子皆可以有空語類、顯性代名詞、名詞等可能性，但三者的最自然指涉又恰好不同，因此引起受試者額外的判斷與決定，這也是本論文較為意外的發現。三個實驗結果似乎部分支持空語類框架，因為隱性代詞彼此之間、隱性代詞與軌跡之間處理同中有異；但也因此無法全面支持空語類框架的四分法，例如本實驗無法找出名詞軌跡與疑問詞軌跡之腦神經處理機制的差異。整體來說，我們認為左額葉下迴在句子理解的主要功能為工作記憶，在辨識填充語或空位時，能以額外的工作記憶儲存，保持，並予以釋出；左顳葉上／中迴主要功能為整合：整合填充語與空位，空位與動詞，以利句子建構。但左額葉下迴與左顳葉上／中迴的反覆出現，反映了空語類除了「空」以外最重要的基底：依存性。依存性需要透過兩個腦區的共同處理：左額葉下迴處理儲存與汲取結構中被依賴的成分（無論能否被預先辨識出），而左顳葉上／中迴進一步整合空語類所需之語意、語類、句法訊息對於先行詞／填充詞之依賴。此推測能夠進一步解釋兩個腦區對於其他具有依存性的結構的反應，包含代名詞，反身代名詞，以及主詞－動詞呼應等。

This dissertation explored the neural correlates of empty categories (ECs) in Generative Grammar. Traditionally, the ECs are categorized as wh-trace, NP-trace, pro, and PRO under different syntactic constraints with complementary distributions. However, whether ECs are postulated only to suffice theoretical needs is intensively debated. Also, some researchers even challenged the filler-gap dependencyby associating the filler directly with the subcategorizer. If ECs are more than theoretical assumptions, the behavioral or neural foundations of ECs should be found. Despite the fruitful findings in previous behavioral and neuroimaging studies on filler-gap dependencies, comprehensive research on the neural basis of ECs is lacking in understanding how the brain recognizes and differentiates the ECs and how they are processed via networks of multiple regions. The dissertation thus comprised three fMRI experiments aimed at investigating how the human brain identified and processed ECs, as well as examining the neural underpinnings of syntactic movements.Experiment 1 compared Chinese topic constructions and relative clauses to see how the brain responded to gapped structures with different filler-gap orders (or movement directions). The result showed that both constructions elicited activities in the left inferior frontal gyrus (L-IFG) and the left superior/middle temporal gyrus (L-S/MTG). However, it also indicated that these two core regions, both playing a crucial role in movements, were insensitive to the movement direction. The gap-first structure, i.e., relative clauses, additionally triggered activation in the left anterior temporal lobe (L-aTL). It is hypothesized that the L-aTL was responsible for exploring the semantic content of the verb to predict the categorical and semantic information of the gap and the filler to merge the verb and the gap for a VP temporarily.Experiment 2 compared topic and passive constructions in Chinese to see if the brain processed gaps contrastively in various gapped structures. The passive construction in Chinese was analyzed differently from the traditional view of A-movement, where the (covert) null operator movement was proposed for the long passives. In this study, however, we found that the brain activity pattern for passive constructions resembled that for topic constructions, where the co-activation of the L-IFG and the L-MTG did not support the covert movement and the non-movement (complementation) analyses. However, the similarities also failed to either postulate an NP-trace in the gap position or distinguish an NP-trace from a wh-trace (in the topic construction).Experiment 3 investigated the neural bases of the two types of non-gapped ECs (pro and PRO) in embedded clauses selected by Chinese hope-verbs (with a pro) and try-verbs (with a PRO), respectively. The two ECs were argued to be distributed complementarily due to their syntactic properties. The results revealed that, while pro elicited the anterior portion of the L-IFG (pars orbitalis, POrb) and the L-MTG, the PRO elicited no significant brain regions, which was an unexpected finding. The distinct brain activation results also indicated the differences in brain activation and processing difficulties. The implication of the POrb may be accounted for by the fact that the construction with a pro did not involve movements, which required a filler to be stored with extra working memory. However, the retrieval of the antecedent was still necessary.The activation of the right prefrontal cortex (RPFC) for the passive construction and the construction with a hope-verb (with a pro) was an unexpected finding because the R-PFC was generally assumed for more domain-general or non-language-specific functions. By careful comparison, it was found that the two constructions contained a position allowing a gap or an NP (and sometimes an overt pronoun). Interestingly, the most natural interpretations of the three were distinct. It was also argued that the activation of the R-PFC was an extension of its decision-making function to (1) identify the type of noun and (2) determine the most natural interpretation of its referent.The results of the three experiments partially supported the EC framework for their similarities in neural activities. However, the results failed to support the four-way distinction because, at least, the NP-trace and the wh-trace cannot be differentiated. Also, the weak results for PRO impeded us from understanding how the brain processed it.Overall, it is argued that the main function of the L-IFG was storing and retrieving necessary information (including the filler) for the gap, while the L-S/MTG was mainly responsible for filler/gap and verb/gap integration. The resembling results also led to the main argument of this dissertation that, in addition to emptiness, the core characteristic of the empty categories is dependency. However, processing the dependency of ECs requires the collaboration of the L-IFG and the L-S/MTG, which also allows us to explain the processing of constructions involving other dependencies, including pronouns, reflexive binding, and subject-verb agreement.