《全唐詩》的分析、探勘與應用──風格、對仗、社會網路與對聯

Textual Analysis of Complete Tang Poems for Discoveries and Applications － Style, Antitheses, Social Networks, and Couplets

唐詩是中國文學極重要的一部分，由清代官方力量所編纂、收錄兩千餘位詩人所著，內容四萬多首詩歌、包含超過三百萬字的《全唐詩》，無疑是研究唐詩最重要的資源之一。本文作者採取共現詞彙和distributional semantics的分析角度，利用計算語言學領域所發展的軟體工具，分析《全唐詩》的內容；就作者風格、詩歌內容，特別是唐詩中的顏色詞彙深入探索。同時我們也利用資訊技術，發掘唐詩內容所攜帶的唐代文人的社會網路，將研究成果擴大到歷史領域。另外，我們也藉由探勘唐詩中詞彙的共現、搭配、對仗關係，發展一個簡單的對對聯的應用。透過這一系列的工作，我們實踐了數位人文領域的初步理想，數位技術雖然尚且不足以直接被用來建立深度的人文論述，但是透過相關的資訊檢索、文本分析和資料整合的服務，數位技術讓專家可以比過去更加專注於深度議題的研究，而不需要花很多時間來蒐集基礎的研究資料。

The Complete Tang Poems (CTP) is the most important collection for studying Tang poetry, which in turn is arguably a very influential part of the Chinese literature. Our analyzing the CTP from the perspectives of antithesis2, collocation and distributional semantics offers some interesting overviews of the styles and imageries embedded in the works of some representative Tang poets. Our analyses include (1) a quantitative comparison of the uses of “wind” and “moon” in Li Bai's and Du Fu's works and (2) the functions of colors in Tang poems. In particular, we explored the appearances of “white” color, which is the most frequent color in Tang poems. Colors in static poems are like audios in motion pictures, so we thought the analyses could lead us to an important facet of the poems. In addition, we extracted social networks of poets from the poems, and built a simple couplet suggestion kit based on the textual analysis of the poems.