| 摘要: | 閱讀語句需要協調、統整多種語言處理的歷程，包括處理詞彙和複合語意，並確定單詞之間的結構和語法關係。 儘管已有許多研究探討和語句閱讀相關的神經因素及大腦區域，但這些大腦區域進行運算的確切法則和機制仍有待闡明。 來自印歐語言的神經心理學和神經影像學研究顯示，處理複合語意涉及大腦腹側結構（包括前顳葉皮層，anterior temporal cortex），處理語法則和大腦背側結構（包括左腦前額葉下回，left inferior frontal cortex）有關。 然而，來自不同語言類型的實徵研究結果對這些關聯是否在所有語言中都成立的普遍性提出挑戰。 此外，某些涉及不同認知範疇的因素（例如工作記憶），也已被觀察到會影響語言（特別是句法）的處理。
漢語具有中心在後（head-final）的特性，詞彙的語法類別也較為模糊，即動詞和名詞間的區分不太明顯，型態句法有限，和印歐語言大不相同。有鑑於此，本論文的主要目的是檢驗和閱讀語句相關之腦區的認知神經模型，是否適用於說明漢語的處理歷程。另外，本論文的第二個目的是研究和漢語句法加工相關的機制是否以及在多大程度上涉及其他認知領域、展現出跨認知範疇的特性。
為了探索漢語的語句閱讀和句法處理歷程，我們進行了兩個功能性核磁共振造影（FMRI）實驗（實驗一和三），自控節奏的閱讀行為實驗（實驗二）和腦電波儀（ERP）實驗（實驗四）。 來自實驗一和三的神經影像學結果顯示：位於大腦腹側區域的顳葉後部和前部，分別對由單字組成詞彙（compounding）和由詞彙組成語句（compositional semantics）的不同語義整合層次產生較大的反應，位於大腦背側區域的左腦前額葉下回和左腦頂葉下部則是和句法修正（syntactic revision）歷程有關，左腦頂葉下部的活化程度同時也和句法的複雜度有關。因為這些發現和根據語句處理的認知神經模型所形成的預測並不完全符合，因此我們認為漢語語句的閱讀比印歐語言更依賴由大腦腹側結構所支持的、由語意處理所驅動的歷程。
此外，來自本論文實驗四的腦電波結果顯示：在閱讀漢語中具有名詞性關係從句（prenominal relative clause）的語句時，出現在主語關係從句（subject relative clause）後的主語名詞（head noun）會比出現在賓語關係從句（object relative clause）後的主語名詞引發更強的P600效應。 這一發現與過往文獻中來自印歐語言(即主語名詞出現在關係從句之前)的研究結果有所不同，因此這些結果並不支持「句法處理乃由語法節點間的結構距離所決定」的理論觀點，但和「句法處理乃由具語法依賴性之詞彙間的線性距離所決定」的理論觀點一致。另外，雖然工作記憶表現的個別差異和數個測量語法處理的指標(例如：閱讀時間、功能性核磁共振造影的血氧濃度變化、腦電波訊號中的P600效應)間都沒有交互作用，但工作記憶的高低和某些腦電波訊號有關聯。因此，我們認為閱讀語句中的語法處理歷程，同時需要跨認知範疇及特屬於語言範疇的工作記憶資源。
總結來說，本論文的研究結果突顯了不同語言之句法處理歷程的特異性，並挑戰了假定有一具普遍性之句法結構偏好的語句處理理論； 這些發現也對跨語言共享的、和語句處理有關的相同神經網絡提出質疑。 另一方面，目前的實徵證據進一步強調：閱讀語句中的語法處理歷程同時需要跨認知範疇及特屬於語言範疇的認知資源，同時支持其他認知作業和只負責句法處理的工作記憶便同時參與在語句閱讀的歷程中。;Sentence reading requires orchestration of multiple linguistic operations including processing lexical and compositional semantics, and determining structural and grammatical relationships among words. Despite extensive investigations on the neural correlates of sentence reading, the exact computations of distinct brain regions associated with these operations remain to be elucidated. Neuropsychological and neuroimaging research in Indo-European languages has suggested involvement of a ventral stream (encompassing the anterior temporal cortex) in compositional semantics, and a dorsal stream (including the left inferior frontal gyrus) in the processing of syntax. However, whether these associations are general to all languages has been challenged by empirical findings from typologically different languages. Moreover, some domain-general factors (e.g., working memory) have been observed to affect language, particularly syntactic, processing.
The primary aim of the present dissertation is to test the predictions of the neurocognitive models of sentence processing concerning language-related brain regions in Mandarin Chinese, a head-final language with less transparent syntactic categories (i.e., less obvious verb-noun distinction) and with limited morphosyntax when compared to Indo-European languages. The secondary aim of the present dissertation is to examine whether and to what extent the correlates of syntactic processing in Chinese exhibit domain-generality.
Two FMRI experiments (Experiments 1 & 3), a self-paced reading experiment (Experiment 2) and an ERP experiment (Experiment 4) were carried out in order to explore sentential and syntactic processing in Chinese. The neuroimaging findings revealed a ventral stream spanning the posterior and anterior portions of the temporal lobe, which respond to lexical (compounding) and sentence-level (compositional semantics) integration, respectively, and a dorsal stream encompassing the left inferior frontal gyrus, which correlates with syntactic revision, and the left inferior parietal lobule, which is associated with both syntactic revision and complexity. These findings are not fully consistent with the predictions of the neurocognitive models of sentence processing. It is claimed that Chinese sentence processing is largely semantics-driven and more dependent on the ventral stream when compared to Indo-European languages.
Furthermore, it was found that when reading Chinese prenominal relative clauses, a stronger P600 effect was elicited by subject relative clauses than object relative clauses on the head noun. This finding contradicts with previous studies on processing of post-nominal relative clauses in Indo-European languages. It is claimed that a sentence processing account based on the linear distance between dependencies, rather than structural distance between syntactic nodes, can account for these findings. Moreover, while individual memory spans did not interact with most of the syntactic processing measures (reading time, BOLD signal, P600), they did modulate certain ERP correlates. Therefore, both domain-general and domain-specific verbal working memory resources underlying syntactic processing are proposed.
In conclusion, the current findings highlight language-specificity of syntactic processing, challenging sentence processing accounts which posit universal structural preferences. The findings also cast doubt on neurocognitive models which propose neural networks of sentence processing shared across languages. On the other hand, the present findings also underline both domain-specific and domain-general aspects of syntactic processing, which appears to be supplied by both domain-general and domain-specific working memory resources. |
