以語意標籤及因素分析判別漢語文體差異研究—隨機偶發對比有備籌畫

Classifying Mandarin Genre Variation with Semantic Tags: Spontaneity vs. Preparedness

本文提出一組漢語語意標籤集，用於搭配因素分析方法以研究漢語文體差異，該語意標籤組合含三層結構：(1)第一層以名詞、謂語(動作)、修飾語三種詞類區隔；(2)第二層再區分為17組更細微之分類；(3)第三為最終層之129個語意特徵標籤。透過探索性因素分析，將語意標籤變數縮減並歸納出六個語意成分/面向(factor)，包含：表達性、事件、同理、移動、地點、及物件。使用常模化及標準化處理後的標籤頻率數值可計算出六個因素分數(factor score)，不同分數高低可呈現文體間差異，進而達成採用語意標籤辨別文體類別之功能。分析該語意標籤集顯示，即便均同屬口語或文字產出情況下，「隨機偶發」性質的訊息有別於「有備籌畫」後的語言產出，體現在使用更多的「表達性」成分，亦即，該性質文體的用詞含有更多「指示性、數字、事實性、及程度」等標籤，因此，本文提出除了表面上「口說-寫作」分類方法外，漢語文本類別可透過「隨機偶發」相對於「有備籌畫」二特徵得到進一步更細微的分類。

This paper proposes a newly constructed and developing semantic tagset for an EFA (exploratory factor analysis) of Mandarin genres. The semantic tagset contains a three-level structure: (1) the first layer covers the three main linguistic categories of“nominals, acts, and modifiers”; (2) the 17 level-two tags further divide the three categories with more detail; (3) and the third layer comprises 129 subordinate feature tags. The EFA variable reduction procedures generalized the level-three tags into six factors in Mandarin: exposition, events, affection, motion, places, and items. By comparing the six semantic factor scores derived from the normalized and standardized frequencies of the factor-related tags, register differences can be illustrated, and text types can be classified based on semantic annotations. The semantic tagset identified that spontaneous messages differ from prepared production in both speaking and writing scenarios by being more expository, i.e., having more“demonstrative, number, factual, and degree”tags. Instead of the surface spoken-written division, this paper argues that Mandarin text types could be further categorized with the“spontaneity vs. preparedness”contrasts.