以量化分析方法進行漢譯佛典譯者風格研究

Abstract

佛教傳入漢地之時，記載佛教義理的佛經主要使用印度的梵語或是其他的中亞語言。在東漢至唐中葉的數百年間，佛教思想大興，進而出現了大規模的譯經活動，不僅產生了巨量的漢譯佛典，也進一步影響了中國的文化以及語言的發展。然而，早期的佛經翻譯受到傳抄、戰亂、偽託譯者等現象影響，使得譯者記錄中出現許多記載不一致或錯誤登錄的現象，這也連帶造成相關研究者的困擾。為找出正確的佛典翻譯者，許多佛學研究者使用傳統文獻學之方式，提出新的證據與看法。然而傳統文獻學之研究方式十分倚賴人工判斷與處理，不僅耗時費工，且無法處理大量文獻資料。 在現今資訊科學的幫助下，以數位化資料及統計量化方法進行資料的比對分析，已是現今人文資訊學的一大趨勢。使用量化分析方法不僅可以進行大量資料比對，也能夠找出譯者風格的潛在因子。然此方面之研究，雖在西方文獻研究中行之有年，但是運用在漢文典籍的研究卻相當少見。因此，本研究提出利用古籍虛字的常見字，結合主元素分析法（Principal Components Analysis），建立譯者風格判別模型，來嘗試解決佛典譯者記錄不清楚的問題。本研究的前半部，利用多個不同的實驗設定對模型進行測試，證明模型對於區分譯者風格確實有效。後半部以一個典型的一經多譯作品──《華嚴經》〈入法界品〉為例，探討當作品主題內容相近時，如何透過模型參數的調校，過濾出具有關鍵影響力的特徵值，以達到更佳的辨別效果，最後於結果中得到可供未來研究參考的線索。

The Taishō edition of the Chinese Buddhist canon (1924-1932) collects ca. 1000 Indian texts that were translated into Chinese between the 2nd and the 11th century CE. 153 of these texts are marked as 失譯 indicating that the name(s) of the translator(s) are unknown. For the texts translated between the 2nd and the late 6th century, however, we have to confront the dilemma that many attributions are uncertain, problematic or simply wrong. Over the years Buddhist scholars have leveraged traditional text-critical methods to corroborate or dispute traditional attributions. Although these methods method can produce high quality results, they often rely heavily on the intuition of a single scholar honed over many years of research. Information technology offers an alternative vector of inquiry that aims to complement rather than supersede more traditional approaches. For this we will adopt statistical, quantitative methods and artificial intelligence algorithms to analyze ancient Buddhist texts translated into Chinese in order to discover new evidence to address the translator attribution. The major advantage of stylometrics and quantitative authorship attribution is being able to discover hidden patterns, which cannot be discerned by traditional approaches. In the past four decades, considerable attention has been paid to quantitative authorship attribution of literature in western languages; however, there have been only few attempts focusing on texts written in classical Chinese, much less in the particular form of ‘Indian Buddhist Chinese’ of early translated texts. In this paper, our main focus will be on grammatical particles (xuci 虛詞) that are widely used in classical Chinese to express grammatical relations. After measuring their occurrence in Indian Buddhist Chinese we use Principle Component Analysis (PCA) to discuss how their use reflects on the authorship of some selected sutras. Our analysis explores different scenarios that have to be accounted for such as a translator changing his style in the course of his career, how to understand commonalities between of contemporaneous translations, and how to quantify the difference between different translations of the same sutra. Also, in latter part of this thesis, we apply our analysis model with the three different translations of Gandhavyūha from different dynasties. We have conduct a series of experiments with different values of arguments. Through the experiments, we demonstrates that the T.278 was translated three to four hundred years earlier than T.279 and T.293 and shows which of its features can identify it as an earlier text.