本研究以認知神經心理學的方法，一方面探討各種不同的訊息如何被用來解決句法結構所造成的歧義情況，一方面探討「經驗和策略」等非語言因素對於句子結構處理的影響，企圖捕捉語言處理中的動態歷程。
過去無論是語言學或心理語言學的理論多半是以英文為主，跨語文變異性為句子處理的理論提供了良好的檢驗方式，本研究採用了三個中文句型為實驗材料，這些句子裡包括關係子句(RC)、從屬詞組(GnP)和複合名詞組(CJ)，這三個句型符合句法歧義句的條件，因為它們具有相同的表面形式，但背後的句法結構並不相同，受試者必須解析局部結構上的岐異，方能達成正確的理解。並且我們可以在「功能詞」和「內容詞」分別誘發受試者進行「句義解歧處理」，從而檢驗相應的腦波活動。我們要求受試者進行逐字閱讀作業，並且同時記錄腦波活動，進而分析出事件關聯電位(Event-related potentials, ERPs)。
實驗一的目標是找出和中文句子處理有關的ERP指標，我們並且依據這些指標的型態來檢驗「句法獨立性假說」，亦即句子的結構複雜度是否為P600的唯一預測因子？抑或有其他非句法因素介入？結果發現，在解歧歷程中，功能詞僅誘發P600，而內容詞則誘發N400和P600。中文句子閱讀確認了ERP指標的基本發現，即N400對語義敏感；而P600對句法敏感。更重要的是，證據顯示：中文受試者在歧義解析歷程中採用了「延後處理策略」，而且他們的P600幅度並非句法結構複雜度的函數，這兩項結果並不支持純粹句法的理論（例如，Garden path theory (Frazier & Rayner, 1982)），必然有非句法因素介入。
實驗二關注於究竟是否存在著專屬於語言處理的神經機制？這個機制是否獨立於非句法訊息處理和其他認知運作？我們透過檢驗語言表現因素對於P600的影響效果來探討這個議題。我們採用了區段設計，其中，每個區段只包含了兩種句型（而非一次出現三種句型），配合區段的先後次序，藉由實驗情境來誘發斷句策略，從而檢驗P600的型態是否會隨著策略而發生系統性的調整。基本的想法是：在第一個區段中，受試者並未覺察到另一型歧義句的存在，亦即不需採用「比較分析」的策略（如實驗一的情境），因此，較為接近自然情境的語言處理模式；反之，受試者在接受第二個區段實驗時，已經經歷了第一階段句型的處理經驗，因而影響第二階段所包含的句型的結構處理。結果發現，即使在排除「比較策略」之後，「延遲處理」現象仍然發生，顯示此現象反映出受試者在實驗室外、長期的語言經驗。此外，P600並不獨立於策略的影響，相反的，P600隨著情境有系統的變化，這些調整現象發生在「內容詞」，而非「功能詞」。
綜合實驗一和實驗二的結果，肯定了結構知識的存在，並且顯示：在不同層次的語言表現因素下，語言處理的相應腦波隨之做策略性的調整。語言表現需要衡量情境並且嫻熟的運用語言知識，語言行為雖然衍生出高度抽象化的規則，當情境要求的策略與經驗累積而來的結構知識抵觸時，語言處理歷程保留了相當的彈性，並且顯現了問題解決的特徵，即使是過去認為是自主系統的句法處理也不例外。我們的結果可以整合到一個更為功能取向的理論架構，不僅捕捉到了句子處理的動態本質，也兼顧到漢語語言的特殊性，如此，有助於發展一個兼容並蓄的跨語言的理論。

A central issue about language and cognition is the distinction between language competence and language performance. We approach the issue by studying the manner how various sources of information are used during sentence processing. Cross-linguistic variations at the sentential level provide another test ground for theories in sentence processing. Three types of Chinese sentences, the relative clause, the genitive noun phrase, and the conjoined noun phrase, were used as materials for the online measure of sentence processing. These sentences are ambiguous because they share the same surface forms but they have different syntactic structures. Subjects need to resolve the local syntactic ambiguity to allow comprehension. Event-related brain potentials (ERPs) were recorded while subjects reading sentences in a word-by-word passive reading task.
In the first experiment, we aim at examining the language-related ERP indexes in Chinese sentence processing. We test the independency thesis of syntax by examining whether or not syntactic complexity is the solely predictor for the index of syntactic processing (P600). The results reveal that: Serving as disambiguation words, content words (nouns) elicited both N400 and P600 while function words elicited the P600 component only. ERP responses to Chinese sentence reading confirm the basic findings that N400 is sensitive to semantic processing and P600 is sensitive to syntactic processing. More importantly, Chinese readers adopted a delayed processing strategy for ambiguity resolution and the P600 amplitude is not simply the function of structure complexity. That is, our results did not support the purely structure-driven model such as Garden-path theory (Frazier & Rayner, 1982). Instead, performance factors such as working memory or strategy may also play important roles in online language processing.
In the second experiment, we address the issue of language competence and language performance. The debate among researchers concerns whether or not there exists a language-specific brain process that is independent of non-syntactic information processing and other cognitive operations. We pose the question by examining ERP patterns under a variety of experimental conditions. We adopted a block design to elicit parsing strategies. In each block, two out of three types of sentences were presented to subjects. We analysis the ERPs elicited by the keywords in types of sentences and the ERP patterns changed systematically under conditions that elicited different parsing strategies. Results show that “Delayed processing” remains even when comparison strategy was excluded by assigning GnP and RC to separate blocks. This suggests that “delayed processing” is a stable phenomena derived from language experience outside laboratory situation. Processing bias produced effect on P600 pattern and this was found in content word but not in function word. These results suggest that P600 is not independent of processing experience.
Language performance requires sophisticated use of language knowledge and it exhibits the characteristics of problem solving. We suggest that these findings can be integrated into a functional framework to capture the dynamic properties of sentence processing.