以語料庫為本探究多變項與雙及物構式的偏向分布

Abstract

雙及物構式的研究最初從形式結構分類以及雙賓動詞語義探討（朱德熙，1979；湯廷池，1979），之後學者採用Goldberg（1995）的構式語法理論討論雙及物構式語義，如張伯江（1999）、沈家煊（1999）等。除了以動詞語義、構式語法的角度分析雙及物構式，也需考量義項及語境因素如何影響雙及物構式選擇。 在漢語雙及物構式研究中，較少將動詞義項細分探討雙及物構式。因此，本文旨在以多義詞「送」為主題研究義項對句式選擇的偏向性影響以及其他多變項如何影響雙及物構式選擇。本研究以｢COCT書面語語料庫2019｣中2011至2018年的語料為研究範圍，利用顯著共現詞素分析法（Stefanowitsch& Gries, 2003）分析｢送｣的義項（給予義、傳送義）與雙及物構式的搭配關係，以及義項與哪些直接賓語搭配。由於雙及物構式受到多變項影響，本文以二元邏輯斯迴歸分析建立能夠預測雙及物構式的多變項模型，同時以實證研究為輔，視相同義項、尾重原則但不同的信息結構的條件下，中文為母語者是否受到信息結構制約。 本文的研究結果顯示（1）｢送｣的兩個義項句式偏好顯示傳送義偏好與格構式（PDC），給予義傾向雙賓構式（DOC）。（2）二元邏輯斯迴歸分析預測義項、信息結構、尾重原則為影響雙及物構式為DOC、PDC的變項。（3）實證研究結果與迴歸分析預測結果一致，整體的評分表現為PDC中，不符合信息結構、尾重原則時，PDC得分較高；DOC中，當符合信息結構、尾重原則時，DOC得分較高。論及兩句式的信息結構表現，本研究認為動詞語義、構式義仍影響句式的信息結構順序，信息結構需考量聽者信息結構。Mandarin ditransitive constructions are originally investigated from the classification of syntactic structures and the semantics of dative verbs (Zhu, 1979; Tang, 1979). Later, adopting Construction Grammar (Goldberg, 1995), scholars investigate construction meanings of ditransitive constructions (Zhang, 1999; Shen, 1999, etc.). In addition to verb semantics and Construction Grammar, how verb senses and contextual factors influence syntactic choices of ditransitive constructions should be considered. A sense-based study on Chinese ditransitive construction is rarely investigated. Therefore, taking a polysemous verb song ‘give’ as an example, this paper aims to investigate a sense-based verb bias and multiple variables toward sentence choices. The song ‘give’ instances selected from the collection of written COCT 2019 ranging from 2011 to 2018, a distinctive collexeme analysis (Stefanowitsch& Gries, 2003) is conducted to figure out the collocation pattern of two senses (give & deliver) of song ‘give’ with ditransitive constructions and the pattern of direct objects collocated with the two senses respectively. Due to the fact that ditransitive constructions are influenced by multiple variables, a binary logistic regression analysis helps develop a regression model with variables predicting syntactic choices of ditransitive constructions. The regression analysis is supplemented by anexperiment. This experiment manipulates information structures, controls senses and weight principles, to observe whether Mandarin native speakers are constrained by information structures. The results indicate (1) Each sense shows its syntactic bias, the deliver sense prefers PDC while the give sense tends to occur in DOC. (2) The binary logistic regression analysis selects sense, information structure, and end-weight principle as predictors in the syntactic choices of DOC and PDC. (3) The experiment result is consistent with the regression analysis. The overall rating of the experiment shows that PDC scores high when it disobeys given before new principle and end-weight principle. When the two principles are obeyed, DOC gets a higher score. In the discussion about the given-before-new principle that PDC and DOC differs in, this paper holds that verb semantics, construction meanings still influence the constituent order of information structure and that hearer-old/new information should be considered.