複數指涉對象的指涉歧義處理

Abstract

指稱性成分，如代詞，係用以連接語言使用者的心理表徵以及外部實體、概念的載體，說話者及聽話者分別透過使用／解讀指稱性成分以指示／獲得所指（referent）。指稱性成分在篇章中可能帶有歧義（ambiguity），即語境中有超過一個前行語（antecedent）的存在而無法建立正確的指涉關係。當語言使用者處理這類歧義時，在事件相關電位技術（Event-Related Potential, ERP）中，經常觀察到一稱之為Nref 的腦波成分。前人認為Nref 效應反映的是處理歧義指稱性成分的機制，並提出兩種可能的解釋，分別是表徵維持假說（representation maintenance account）及歧義覺察假說（ambiguity detection account）。前者認為語言的處理機制會將篇章中所有可能的前行語保留在工作記憶中以待後續的去歧義處理，而後者則是認為Nref 效應反映的是覺察到指稱性成分有歧義的歷程。為了探究Nref效應反映的認知歷程較趨近於上述哪一種解釋，以瞭解Nref 效應的本質，我們在實驗中操弄代詞的前行語數量，使代詞在篇章中有一至三個可能的前行語，並以此觀察Nref 振幅大小與代詞前行語數量之間的關係。根據表徵維持假說，在歧義消除前（disambiguate），所有可能的前行語需保留在工作記憶中，因此Nref 效應的振幅應隨前行語數量增加而遞增；反之，若Nref 效應是由覺察歧義的歷程所引發，則Nref 效應的振幅應只與歧義的存在與否有關，不論前行語的數量，只要該數量與代詞所需的所指數量不一致，則皆會引發振幅大小類似的Nref 效應。 除了探討 Nref 效應的本質之外，為探討一般認為較與語言處理無關的認知功能，如視覺工作記憶（visual working memory, VWM）與解析歧義指稱是否共用工作記憶資源，我們測量了受試者的視覺工作記憶廣度（VWM capacity），並以此來檢驗Nref 效應與其他負波成分（negativity）是否因共享工作記憶資源而反映相似的神經機制。 實驗結果顯示，當解析有歧義的代詞，即代詞有兩或三個可能的前行語時，皆引發一Nref 效應。此外，在代詞有兩個可能的前行語時引發的Nref 效應振幅大於代詞有三個可能前行語時所引發的Nref 效應振幅，這個結果與基於上述兩個解釋的預期皆不同，因此此實驗結果並無直接支持任一種解釋。然而，雖然歧義代詞所引發的效果不如預期，我們卻在後續的歧義消除名詞上發現了類似Nref 效應的負波成分。在前行語有三個的篇章中，歧義消除名詞出現時會使得歧義代詞的可指涉對象由三個減少為兩個；而在前行語有兩個的篇章中，歧義消除名詞的出現則是會使得代詞的可指涉對象由兩個減少為一個，即完全去歧義。在這兩種篇章條件下，前者的歧義消除名詞會相較於後者的有一個類似Nref 效應的負波。雖然該名詞本身的指涉關係並無歧義，然而深層的篇章情境中代詞仍然可指涉到兩個前行語，因此仍然觀察到一個負向的腦波成分。此項結果說明Nref 效應反映的較可能是保留在工作記憶中與歧義有關的心理表徵，而且能夠成功維持在工作記憶中的表徵數量上限可能只有兩個。此一結果較支持表徵維持假說。至於Nref效應振幅大小與視覺工作記憶廣度的相關性分析，則沒有發現任何顯著的正相關性，這可能是由於語言與視覺實驗作業的差異所導致。 總的來說，本研究試圖透過增加前行語數量的方式來探討Nref 效應的功能性機制，並且初步地檢驗了Nref 效應與其他負波成分的可能關聯性。同時本文亦針對Nref 效應所反映的神經機制提出新的觀點，即Nref 效應並不僅代表單一認知處理歷程的此一可能性。我們認為解析歧義的歷程，或許是由多個階段的認知活動組成，而Nref 效應可能包含了不同階段的認知活動所引發的腦波成分。因此，未來研究應針對解析歧義歷程所涉及的功能與認知活動加以釐清。

Constructing coherent mental representations from referential processing is critical for language comprehension. However, it is not uncommon in natural language use when multiple antecedents can be eligibly taken as the referent for an anaphor, leading to processing cost reflected in a sustained anterior negativity—the Nref. The present study investigated the properties and the functional nature of the Nref effect. We used the Event-Related Potential (ERP) technique to examine the ambiguity detection account and the representation maintenance account. Participants read stories containing one, two, or three characters that were all suitable referents for a critical pronoun in the subsequent sentence, yielding unambiguous, 2-referent, and 3-referent conditions respectively. The referential ambiguity was then disambiguated with character names (e.g., ’他覺得家俊與國維明天. . . ’). This experimental design allowed us to test the Nref effect’s sensitivity to increasing referential load and to observe brain responses subsequent to the ambiguity. In addition to examining the underlying processing reflected by the Nref effect, another goal of present study was to explore the possibility of general resources subserving the working memory operations across modalities. In particular, we examined whether the neural mechanisms of the Nref effect were associated with other sustained negativities through utilizing similar working memory resources. To that end, a change detection task was incorporated to assess each participant’s visual working memory (VWM) capacity and its associations with the Nref effect were investigated. ERP data time-locked to the critical pronouns showed that ambiguous pronouns in both 2-referent and 3-referent conditions elicited the Nref effect, which was larger in the 2-referent condition than in the 3-referent condition. The results of the preliminary analyses reported in this thesis do not provide support for either account, and further validation and examination of the provided tentative interpretations are needed. Despite the unclear data time-locked to the pronouns, ERP data time-locked to the first disambiguation names revealed a Nref-like effect in the 3-referent condition, in which the already-seen pronouns were still ambiguous in the situation model, relative to the 2-referent condition, in which the ambiguity was resolved by the current disambiguation name. These findings suggest that the Nref effect may reflect the ambiguity-associated mental representations that are held in the working memory, and that the limit of the number of referents to be kept in the working memory may be restricted to be no larger than 2. These views are consistent with the maintenance hypothesis but require further verification. As for the correlational results, however, we did not find positive correlations between the Nref effect and the VWM capacity. These findings could be due to the differences in experimental approaches between language and visual tasks. In sum, our findings yielded further insight into the functional interpretations of the Nref component and provided an initial investigation into the potential associations between sustained negativities. The present study has raised important questions about the sequential processing of the Nref effect. Future study is required to better characterize and identify the likely distinct functional aspects that underlies the ambiguity processing.