The topic chain is an important concept used in narratives. However, its content is not very clear. This thesis mainly investigates a kind of topic chain in which the topic is introduced by Locative-Verb-Noun Phrase (LVNP) sentences. LVNP sentences have a locative phrase as the surface subject, followed by the main verb with the suffix zhe (the duration marker, DUR) or le (the perfect tense marker, PRT), and then the subject or the object. Following structural analysis and by surveying actual data--a novel, we examined and based on the distributions of ZA, PA, and NA which occur in topic chains to generalize three basic patterns and one derived pattern to explain their inner structures. The findings show that topic chains can be categorized according to the patterns they show. Besides, different types of topic chains show different inner structures. By describing the inner structures, the topic chain is not just a concept at all and we learn its actual content more clearly.

本論文首先從三個面向討論中文對話中的主題不連續現象。第一、當說話者改變主題的時候，停頓較長、話語的自我修正較多，且很少被打斷。第二、說話者可以直接進入新的主題，或者在談論新主題之前提供一些背景訊息。最常使用來引介新主題的是問句，因為說話者可以藉由提出問題邀請其他人一起討論新話題。第三、說話者在開啟話題時通常會利用各種情境資源，包括周遭的環境、說話者共有的背景知識以及在前文談論過的人、事、時、地、物。目前的語料顯示大多數的新主題都與前文有某種程度的關聯。 研究結果亦指出不同層次的對話主題之差異。當說話者引介最高階的對話主題時，他們傾向使用共有的背景知識，並在主題改變的地方提供背景訊息; 他們的話語有較長的停頓和較多自我修正的情境;。相較之下，在開啟其他階層的對話主題時，說話者較常利用前文出現過的訊息，且不傾向在話題一開始的時候提供背景訊息; 他們的話語也較流暢。然而，不同層次的對話主題也有一些共通點。例如，在各個階層中，說話者普遍使用問句引介新話題，而當他們利用前文的訊息改變話題時，新的主題通常和前一個主題有相同的指示對象。 / The present study aims at investigating topic discontinuity in Chinese daily conversations. This pragmatic phenomenon is discussed from three perspectives. First, since changing topics requires more cognitive effort, the speaker who initiates the change is likely to produce longer pauses and repairing utterances, and is fairly unlikely to be overlapped. Second, the speaker can move to a new topic directly, or orient a new topic in various ways. Questions are the most common backgrounded clauses at the topic-shift boundary because they function to invite other speakers to join the new subject. Third, the speaker most frequently draws on some contextual resource to guide other speakers into conversing about the new topic. The most common way is by recycling the prior text. Speech disfluency, grounding, and contextual resources are also found to distinguish various levels of conversational topics. The highest level of conversational topics are usually grounded in general background knowledge, produced with more disfluency, and tend to begin with background information. The other levels of conversational topics, however, are more likely to arise from prior text, more fluently brought up, and do not use background information more often than foreground information at the topic-initial position. Despite the above-mentioned differences, the various levels of conversational topics are similar in several aspects. For example, backgrounded clauses at the topic-shift boundary are mainly questions; topics grounded in prior text predominantly maintain referential continuity across the topical boundary

中文可說是個篇章導向的語言(discourse-oriented language)，以篇章為其基本單位，數個中文子句不藉連接詞即可連結成主題串(topic chain)。當如此龐大的篇章單位翻譯成主語顯著的英文時，譯者會面臨的困難是，如何將主題串切割成數個英文句子。然而，此議題卻鮮少前人研究。因此，本論文試圖探討中文主題串英譯時的翻譯策略，並著重於斷句的影響因素。分析的語料來自漢英對照版的台灣光華雜誌以及翻譯教科書。為了瞭解翻譯策略，本研究分析中文的語意分段標記與資訊順序，並與譯文比較。分析結果顯示譯者在翻譯時會採用三種策略：反映段落標記(Reflecting the Markers)、建立語句關係(Establishing Textuality)、重整資訊(Rearranging Information)。第一個策略是將中文段落標記作為英譯斷句的依據。段落標記包括三類：主題的形式、連接詞、標點符號。第二個策略，建立語句關係，表示英譯斷句依據中文主題串各句子之間的篇章關係。第三個策略，重整資訊，表示透過增加、刪除、調換順序等方式調整原文的資訊。 / Chinese is considered a discourse-oriented language. The basic unit of the Chinese language is discourse-based. Several Chinese clauses can be linked together without any connectives to form a topic chain. When such a large discourse is translated into English, a subject-prominent language, translators may have difficulty deciding how to segment a Chinese topic chain into English sentences. However, little research has been done on this topic. The present study aims to explore translation strategies used in translating Chinese topic chains into English. In particular, the demarcation mechanism will be the focus. Chinese-to-English translation data from Taiwan Panorama, a Chinese-English bilingual magazine, and from translation textbooks are collected for analysis. The demarcation markers and information flow in Chinese are analyzed and compared to understand how they are treated in the English translation. Three strategies have been found: Reflecting the Markers, Establishing Textuality, and Rearranging Information. Reflecting the Markers is to reflect the Chinese boundary markers as English demarcation points. Boundary markers contain nominal references of topic, connectives, and punctuation marks. Establishing Textuality is to organize the Chinese topic chain based on the internal textual relationships. Rearranging Information is to add, delete, or reorder the information.