語料情境限因對於中文關係子句理解歷程的影響：腦電波生理證據

Contextual Constraints on the Comprehension of Relative Clause Sentences in Chinese: ERPs Evidence

本論文報導兩個事件相關電位 (event-related potentials, ERPs) 之研究，用以探討中文關係子句處理歷程之認知機制及其神經認知基礎。本研究中我們操弄兩個變項：關係子句之種類（受試者中變項）及關係子句之語料情境（語境）限因（受試者間變項）。本實驗操弄關係子句之前置語料作為語境限因之變異基礎。實驗1a 中，關係子句中之前置語詞為不具語法限因之附屬詞句（例如：“今天一大早，…＂）。相對而言，實驗1b 則使用具有高限因語法表徵之主語語料（例如：用“名詞 (N)-動詞 (V)＂語序以表徵“主詞－－動詞－－＂之限制）作為關係子句的前置語料。兩實驗對於關係子句的語料處理之比較分析顯示語境限因因素在不同的實驗中引導出不同之事件相關電位之模式。實驗1a 及1b 之相關電位分析均顯示：“工作記憶＂(working memory) 這個認知機制在關係子句之語料的意義整合上有重要功能。相對而言，只有在具高度語境限因之實驗1b 中，事件相關電位之分析才會顯示語義 (N400) 及語法 (P600) 之句義整合效果。更進一步分析則指出，語境限因之效果對早期之中文語詞種類之辨識即有影響（約200 ms）。再者，此效果的分析亦發現一個複雜效果 (N400-P600)。本複雜效果對中文閱讀之特性有其獨特之意義。整體而言，本實驗之主要發現：語境制約之語詞種類效果以及句義整合之複雜效果，顯示中文句之處理歷程是依循訊息互動模式的理論基礎。中文篇章之理解歷程是依賴各種不同之語言訊息之交互支持作用。其中，中文篇章之理解歷程是高度依賴於語義；然而，本研究結果亦指出語料所具呈之語法表徵為語義衍生之基礎。這樣的現象符合一般語意學對中文篇章理解之看法。

This paper reports two event-related potentials (ERPs) studies designed to test the universal and language-specific characteristics of cognitive mechanisms and their neurocognitive substrates during the real-time processing of sentence integration in Chinese. Native Chinese (Mandarin) speakers read sentences with different kinds of relative clauses (RC, subject-/object-extracted) when the RC modified the grammatical subject of the matrix clause (Experiment 1a) and the grammatical object of the matrix clause (Experiment 1b). The between-subject Modifying Type (subject-/object-modifying) factor varies on the syntactic constraints of the initial contextual phrase. While Experiment 1a uses contextual phrase of optional adjunct that is constraint-free for syntactic mapping, Experiment 1b uses contextual phrase of an essential sentence mapping that provides strong processing expectation for sentence integration. Cross-experiment comparisons of convergent ERP results on the embedded RC words indicate distinctive morphology and scalp topography of related ERPs that vary with the contextual constraints. While both experiments found a working memory ERP (bilateral anterior negativity) for keeping the filler-gap dependency meaningfully tractable, the integration ERP effects (N400 & P600) were found only when the RCs were under strong contextual constraints in meaning in Experiment 1b. This context effect occurs as early as the processed RC words being encountered (~200ms) for the identification of word class. In addition, the results also indicate an ERP effect, the N400-P600 complex, that may be related to a language-specific property of Chinese reading. In a nutshell, the context-dependent lexical effect (~200ms) and the N400-P600 complex suggest an interactive model of sentence processing in Chinese. This model is compatible with the linguistic observation that the comprehension of a Chinese text relies heavily on semantically-derived information based on the initial guidance of syntactic mapping.