當我們在處理語句的訊息時，即使接收到的訊息是處於一種模糊不清的狀態之下，我們仍然可以藉由語境或者是儲存在長期記憶中的知識表徵來幫助我們理解語句。然而，在可以被視為一般語句處理的篇章中以及去除儲存在長期記憶的知識表徵的非篇章段落中的基本的言語的處理單位，甚少被提及。在本文中，我們將探討在兩種韻律結構的篇章語句以及非篇章為主的言語基本處理單位為何。 接著，我們將用具有心理實證的語句基本單位檢視詞彙學派以及範疇學派在句法分析上所使用的初始單位，探討哪一個學派的句法分析同時具有理論性以及實證性的證據支持。此外，經由探究韻律結構、詞語分斷之間的互動，來檢驗語言以及句法的獨立性假說。更進一步的，我們比較男女之間，對於語句提取的效率是否存有差異。分別有六男、六女，共十二名的大學生參與這次的實驗。實驗的材料是由實驗者將兩種韻律結構的篇章以及非篇章的段落分別錄音之後，讓受試者對所聽到的語句，做斷詞的工作。我們使用分斷派典以及統計考驗，來驗證本文中提出的各個假說。結果發現，在相似於我們一般說話語句的篇章中，基本處理的言語單位是詞，而在去除韻律結構以及分篇章為主的段落中，處理的基本言語單位是音節。這樣的結果符合工作記憶模式的預測。另外，詞彙學派的句法分析得到了理論上及實證上的支持。構詞、音韻、句法、以及長期記憶中的知識表徵，應該被視為是一個整體，而非句法、語言是獨立於整個認知系統之外的。至少，在本文中，利用語句產生證據來看，是反駁這個觀點的。大體上，男女之間對於語句處理的基本單位是沒有差別的，但似乎女孩子在語句產出時，更依賴韻律結構存在的與否。作者在文中提出了社會語言學的解釋。由本文可以得知，在語句產出的證據看來，詞彙句法學派得到了理論上及實證上的支持。語言是和整個認知系統互動的一種機制。而男女在提取言語單位上的效率並無不同。 / It is widely accepted that the routine of the perceptual processing of ongoing speech can be comprehended even in a degraded quality through the aid of the context or real-world knowledge stored in LTM. However, the basic units processed in the discourse-based text that can be recognized as the normal sentential input and those processed in the non-discourse-based text which removes the contribution of the knowledge representation stored in LTM are not yet investigated. In this study the basic units processed in these two textual patterns were discovered. The basic unit processed in normal prosodic, discourse-based text was employed to resolve the controversy between lexicalists’ and categorical syntacticians’ primitives of syntactic analyses. In addition, purposes of the interaction between prosodic patterns, namely the prosodic pattern, prosody-free pattern and the syntactic segmentation were examined to testify the autonomy hypotheses whether they gain the support from the speech production. Furthermore, the retrieval efficiency between genders were analyzed as well to investigate whether different genders employed different retrieval efficiency in different textual and prosodic patterns. 12 undergraduate students, six male and six female students participated in the two experiments. Discourse-based texts with two prosodic patterns and non-discourse-based pattern were recorded in the auditory form to conduct the experiments. Segmentation paradigm and statistical analysis were used to make the production analyses. Results indicated that a word is the basic unit processed in the prosodic, discourse-based text while a syllable is the basic unit processed in the prosody-free discourse-based text and non-discourse-based text which confirms to the prediction of WM model. The evidence favors lexicalists’ primitive of syntactic analysis psychologically. The cooperation between morphology, syntax, phonology and knowledge representations in general cognitive system argues against the autonomy hypotheses. Language should be recognized as a submechanism embedded in the cognitive system. The results suggested that, in general, there was no difference between genders but it seemed that female subjects tended more to rely on phonological cues. A plausible sociolinguistic reason was proposed. The results suggest that lexicalists’ primitive for syntactic analysis has theoretical as well as psychological support. In speech production, language seems to interact with other cognitive mechanisms rather than isolate to form an independent, self-contained domain. No retrieval difference exists between genders.

本文旨在從語法及語句處理的觀點探討中文關係子句中接應代詞(resumptive pronouns) 的現象。在語法上，本研究認為中文的接應代詞屬於詞基生成 (base-generated) 元素而非句法變項 (syntactic variables) 。接應代詞的出現受限於派生簡約原則 (derivational economy) 及合法度 (grammaticality) 的要求，唯有當間隙型關係子句 (gap relatives) 為非法時，接應代詞才是必要的；在其他情況下，接應代詞在中文裡為選擇性的策略之一。在語句處理上，本研究施測了三份合法度問卷並進行了兩項線上閱讀測驗。合法度的判別及線上的閱讀反應時間顯示，接應代詞之所以不被喜歡是因為接應代詞本身會引發花園路徑效果 (garden path effect) (在此為主要子句假象 (main-clause illusion) 的形成)。然而，接應代詞在間隙型關係子句為非法或引發花園路徑效果 (在此為重建困難 (difficult reconstruction) 所造成) 時卻變得較受喜愛。當間隙型關係子句涉及到複雜結構或複雜語句處理領域 (在此指的是重建複雜 (intricate reconstruction) ) 時，中文母語使用者並不會偏好使用間隙(gap)或是接應代詞來形成關係子句。這些研究結果意味著中文接應代詞不僅受限於合法度的要求，也受制於語句可處理性 (processability) 的程度，亦即有無受到花園路徑效果所影響。 / This thesis examines resumptive pronouns in Chinese relative clauses from the perspectives of grammar and processing. In the grammar, it is claimed in this research that resumptives in Chinese are base-generated pronouns rather than syntactic variables. Their occurrence can be predicted via derivational economy and grammaticality. Resumptives in Chinese are obligatory only when the gap relatives are ungrammatical. Otherwise, resumptives are optional. As for processing, three grammaticality judgments and two on-line self-paced reading tasks were conducted. The grammaticality judgments and the on-line reading times show that resumptives were dispreferred because they induced a garden-path effect (the main-clause illusion). However, resumptives became preferred when the gap relatives resulted in ungrammaticality or led to a garden path, thus inducing the difficulty of reconstruction. There was no preference for gaps or resumptives when the gap relatives involved complex structures or complicated processing domains (the intricate reconstruction). These results suggest that resumptives in Chinese are related not only to the issue of grammaticality but also to that of the processability (i.e. the garden-path effect).