限定詞組省略理論研究：從臺灣南島語觀點出發

本研究旨在探討南島文獻中較少觸及的限定詞組省略（DP ellipsis）、孤島效應和論旨分派間的互動機制，試圖在Chomsky (1993)最簡方案(Minimalist Program)的理論架構下，提出一致性解釋。省略現象涉及許多有趣的議題。不同階層的詞組可能因刪略而形成NP/N' ellipsis, VP ellipsis, PP ellipsis, IP-related ellipsis和CP-elliptical form (Chiu 2011; Gallego 2009; Gengel 2008) 等不同結構。空語類(empty category)的形成機制，文獻上存有許多精彩的論辯，主要可以分為PF-deletion (Merchant 1999) 和LF-copy (Chung et al. 1995) 兩派說法：前者認為，在句法運作一開始即存在特定語法成分，這些語法成分經過有關運作後，如拷貝位移(copy movement)，會在音韻介面上進行刪略(PF-deletion)，因而失去語音內容；而後者則認為在句法運作一開始，空語類所處的位置即不存在任何語法成分，其語意內容是透過語意拷貝(LF-copy)來滿足。此外，空語類是否要再詳細劃分？分幾類？仍莫衷一是。如Chomsky (1981)提出管轄約束理論(Government and Binding Theory)，將空論元分成PRO, pro, WH-trace和NP-trace；在最簡方案理論架構下，Chomsky (1993)進一步認為傳統PRO, WH-trace和NP-trace是透過拷貝位移(copy movement)生成，而pro則是基底衍生(base-generated)的成分；隨著理論本身和不同語言類型研究的持續發展，Huang (1984) 將漢語空論元分成代詞性(pronominal)和變項性(variable)兩類；Li (2005)則在黃正德的基礎上，認為還可以再分出第三類，即真空位(True Empty Category)；尤有甚者，Xu (1986) 則認為空語類在英語型語言依其屬性可再細分，然而，在漢語型語言均是自由空語類(Free Empty Category)。Huang (2010)則認為空賓語的代詞性用法是一種特殊的廣義控制規則(Generalized Control Rule or GCR)例子，可以透過Chomsky (1980)述謂結構(Predication)或強管轄(Strong Binding)的概念來解釋。本文考察泰雅、賽夏和阿美等三種臺灣南島語，並聚焦在主格、賓格和屬格限定詞組省略現象，進而提出融合性觀點(hybrid view)說明其生成機制。我認為根據句法分佈(syntactic distributions)和抽取條件(extraction conditions)，討論中的空限定詞組(null DP)主要可以劃分為兩種次類別，亦即代詞性(pronominal)和變項性(variable)。代詞性空論元基底生成後，受GCR管轄；變項性空論元是移位後的拷貝(copy)，因為音韻刪略，而失去語音內容。變項性空論元因為涉及位移，故受近距限制 (locality constraints) 規範，而有孤島效應(island effects)。語意上，代詞性空論元只能出現在從句主格位置，並以主句主語為先行詞(antecedent)；而變項性空論元可以出現在主/賓格，甚至屬格位置，並與句子主題(topic)或言談中對象(discourse referent)共指。臺灣南島語的例子中，與限定詞組省略相關的句子歧義(ambiguity)或非預期論旨分派(unexpected theta-role assignment)，都支持並說明了空論元有必要依其句法、語意特性，進一步細分。此外，我發現一些有趣且特殊的例子：有些例子涉及例外孤島效應(exceptional island effects)，或違反主、賓語不對稱(subject-object asymmetry)。事實上學者們針對這部分提出許多見解，如：自由空語位(Xu 1986)、異質孤島理論(A heterogeneous theory of island, Merchant 1999)、語意拷貝(Chung et al. 1995)、語用上的antilogophor (L. Liu 2004) 或anaphora (Huang 2009) ，真空位(Li 2005)，以及主語事件結構條件分析(Wei 2017)等，試圖從認證機制或空論元的句法語意特質，來解釋上述例外。我則認為南島語中的例子，可以透過兩步驟言談約束(2-step discourse binding, Huang 1984, 2010)，主格限定詞組限制(NDR)和主格位置競爭(NPC)等機制，獲得一致性解釋；另外，本文也探討了後向賓語控制結構，此結構共用論元涉及兩種論旨角色，且被控制的空論元處於句法高位，而控制論元則反在低位。我認為其衍生可以透過控制結構移動分析(Movement Theory of Control, Hornstein 1999)和鏈接縮減原則(Chain Reduction Principle, Nunes 2004)獲得合理解釋。最後，三種臺灣南島語在限定詞組省略的表現，略有不同，我亦進行類型學上的討論和初步分類。就語言類型而言，根據空論元的分佈和限制，三種目標語均可歸屬於言談導向語言，但可進一步細分，泰雅語為作格提取感應類(EEC-sensitive)，阿美和賽夏則是非作格提取感應類(non-EEC-sensitive)。簡要來說，在非主事語態結構中，屬格論元的省略在前者相對受限，但在後者相對自由。我認為這是由於泰雅語屬格論元主題特徵([TOP] feature)的值未指明，因此無法進行非論元移位去檢核特定核心語上的未譯的非論元特徵，例如：大句子上的主題特徵。透過南島語類型上的證據，驗證了不同類型的空語類，各具徵性，其空性(emptiness)程度有別。在光譜的兩端，嚴格者，如代詞性空論元，需在滿足特定句法條件下，才能被先行詞約束；其次，如變項性空論元，雖受近距限制，可以與語境對象共指，分佈相對自由；最後，少數例子最寬鬆，全然由言談語境決定共指對象，且不受近距限制規範。

關鍵字

限定詞組省略；作格提取感應；提取條件；孤島效應；臺灣南島語；空語類

並列摘要

This research aims to explore the interactions between DP ellipsis, island effects, and theta-role assignment. I try to provide a consistent explanation under the framework of the Minimalist Program (Chomsky 1993). The ellipsis phenomenon touches on many interesting issues. According to structural types, deletion operations can be divided into NP/N' ellipsis, VP ellipsis, PP ellipsis, IP-related ellipsis and CP-elliptical form (Chiu 2011; Gallego 2009; Gengel 2008). However, the derivation mechanism regarding empty categories is still under debate. In the literature, PF-deletion (Marchant 1999) and LF-copy (Lobeck 1995; Chung et al. 1995) are two representative approaches. The former argues that given grammar elements exist at the very beginning of the syntactic operation. After certain syntactic operations, such as copy movement, take place, some of their copies undergo deletion at the PF interface, which removes phonetic content. On the other hand, the latter advocates that nothing exists in the position where the empty category is located. The semantic content is supplied via LF-copy. Furthermore, the final conclusion about the subcategorization of empty categories has not yet been settled. The empty categories are divided into PRO, pro, Wh-trace and NP-trace in the framework of GB theory (Chomsky 1981). Chomsky (1993) further proposes that traditional PRO, Wh-trace and NP-trace are derived via copy movement while pro is a base-generated element. The subsequent development of grammatical theories and typological studies enriches this discussion. Huang (1984) divides Mandarin empty categories into pronominal and variable. Based on Huang’s research, Li (2005) proposes a third category as last resort, i.e., True Empty Category (TEC). Xu (1986) advocates that empty categories in English-type languages can be further divided according to their characteristics, while all empty categories in Mandarin-type languages belong to a single category, i.e., Free Empty Category (FEC). However, Huang (2010) argues that the pronominal construal of empty category is subject to the Generalized Control Rule (GCR), which can be explained by the concept of Predication or Strong Binding (Chomsky 1980). This research deals with DP ellipsis in three Formosan languages, including Atayal, Saisiyat and Amis. I focus on the characteristics of empty categories related to nominative, accusative and genitive position. According to syntactic distribution and extraction condition, I propose that the null DP in question can be divided into two types: pronominal and variable. The pronominal empty category is a based-generated element and is subject to GCR. However, the variable, a copy resulting from movement, has to conform to locality conditions. Semantically, only empty nominative DPs can have pronominal construal and take matrix subjects as their antecedents, whereas variable construal can occur in nominative and accusative position, or even in genitive position for some languages. This variable can be coindexed with a sentential topic or a discourse referent. According to Formosan languages, many facts, such as DP ellipsis-related ambiguity and unexpected theta-role assignment, support my argument that empty categories must be further classified depending on the corresponding syntactic and semantic characteristics. Furthermore, there exist some additional puzzles, such as the exceptional island effects or the violation of subject-object asymmetry. In fact, many linguists propose various kinds of solutions, which are based on identification mechanisms or syntactic and semantic properties of empty arguments to account for the above exception, such as FEC (Xu 1986), a heterogeneous theory of islands (Merchant 1999), LF-copy (Chung et al. 1995), antilogophor (L. Liu 2004), anaphora (Huang 2009), TEC (Li 2005), and Argument-Per-Sub-Event-Condition (Wei 2017). Instead, I argue that the Formosan examples can be explained via 2-step Discourse Binding (Huang 1984), the Nominative DP Restriction (NDR) and the Nominative Position Competition (NPC). Moreover, this thesis discusses backward object control constructions. In the constructions, the shared argument possesses dual theta-roles and controllee occurs in a higher position than controller. I argue that the derivation can be accounted for under the Movement Theory of Control (Horstein 1999) and Chain Reduction Principle (Nunes 2004). Lastly, I propose a preliminary typological classification of my target languages. First, all of my target languages are discourse-oriented. However, they can be further divided into EEC-sensitive type (Atayal) and non-EEC-sensitive type (Saisiyat and Amis), which indicates that genitive DP ellipsis is restricted in the former but relatively free in the latter. I argue that the value of [TOP] feature of Atayal genitive DPs is underspecified, which prevents it from undergoing movement to check the uninterpretable [TOP] feature of a given head. The Formosan materials support the claim that different kinds of empty arguments possess unique features.

並列關鍵字

DP ellipsis ； EEC-sensitivity ； extraction conditions ； island effects ； Formosan languages ； empty category

參考文獻

Aissen, Judith. 2017. Correlates of Ergativity in Mayan. In The Oxford Handbook of Ergativity, edited by Jessica Coon, Diane Massam, and Lisa Demena Travis, 737-758. doi: 10.1093/oxfordhb/9780198739371.013.30

Aboh, Enoch O. 2004. Topic and Focus within D. Linguistics in the Netherlands 21 (1): 1-12.