關係子句於英語為第二語言之即時處理研究

Abstract

本論文研究主旨為探討不同程度之台灣英語學習者理解英語關係子句之即時處理歷程。本研究劃分為兩大部份：第一部份（實驗一與實驗二）觀察英語學習者如何處理帶有「先行詞（filler）－間隙（gap）結構依存關係」之主／賓語提出關係子句，並檢視非結構因素（詞頭之名詞生物性）於其處理過程中之角色；第二部份（實驗三與實驗四）觀察英語學習者如何處理帶有「主動詞－過去分詞結構歧義性」之省略式關係子句，並檢視非結構因素（名詞生物性和動詞做為過去分詞之頻率）於消歧過程中之角色。研究方法皆以自訂步調（self-paced reading）與眼動追蹤（eye-tracking）二種線上閱讀實驗法來了解結構與非結構因素於學習者處理語句的歷程中所扮演的角色，並檢視其是否如淺層結構模型（Shallow Structure Hypothesis, Clahsen& Felser (2006a)）所假設之與母語人士的語句處理歷程有著本質上的不同。本研究中各實驗之受試者皆為二組不同英語程度之台灣英語學習者（分別為符合《歐洲共同語文參考架構》所定義之中級與高級程度）和一組以英語為母語之外籍人士。 第一部份的研究結果顯示，不管對英語人士或學習者而言，主語提出的關係子句較賓語提出的關係子句容易理解，而帶非生物性詞頭之賓語提出關係子句又較帶生物性詞頭之賓語提出關係子句容易理解。上述之「主語優勢」以及名詞生物性造成關係子句處理難易度的不同顯示出結構與非結構因素皆影響了處理關係子句之即時歷程，也證明了學習者於語句處理過程中可運用的資訊與英語人士並無本質上的不同，唯一的差異在於如何運用名詞生物性於依存關係的建構或修改方式。本研究主張英語人士處理主／賓語提出關係子句的模式支持了以語句結構為本的「模組假設」（module-based hypothesis, Frazier (1987)），而英語學習者的處理模式則支持了以功能為本的「約束假設」（constraint-based hypothesis, McDonald et al. (1994)），因為自訂步調的實驗結果顯示在非結構因素所提供之訊息彼此產生衝突時，英語學習者傾向不做立即的句法分析決策與指派句法角色給關係代名詞，因此上述的「主語優勢」以及名詞生物性對語句處理的影響只反應在主動詞區。此「延後決定」之假設也於眼動追蹤實驗裡得到證實：上述效應只反應於語句再加工的二項眼動指標（regression path time與total reading time）但未反應於初始語句處理的二項眼動指標（first pass time與first pass regression ratio）。學習者英語程度的差別只反應在處理語句的速度，但並無證據顯示使用的處理策略有所不同。 第二部份的研究結果顯示，不管對英語人士或學習者而言，消解「主動詞－過去分詞結構歧義」的方式與詞彙歧義消解的機制類似，支持了以功能為本的「約束假設」（constraint-based hypothesis, McDonald et al. (1994)），也證明了學習者於語句處理過程中可運用的資訊與英語人士並無本質上的不同：於語句處理的初始階段中，名詞生物性與辭彙頻率等非結構因素皆與最小連接處理原則（Minimal Attachment）所造成的歧義產生立即性的交互作用，因而造成消解「主動詞－過去分詞結構歧義」之難易度有所不同。其間的差異在於如何運用非結構因素來消解歧義的方式：對英語人士而言，當主詞為非生物性名詞時，不管動詞辭彙頻率高低，皆能避免花園路徑句（garden path sentence）的產生，因此消歧介系詞片語（by phrase）的出現並不會造成更長的閱讀時間，而辭彙頻率的優勢只反應於介系詞片語後的主動詞區內；對學習者而言，當訊息產生衝突時（如辭彙頻率高但主詞為生物性名詞），交互作用則提早反應於歧義動詞區內的處理時間，此處理成本能有效避免花園路徑句的產生。上述的處理成本交換機制（trade-off mechanism）也透過眼動追蹤實驗的first pass time與regression path time二項指標獲得支持。學習者英語程度的差別則反應了其對何種非結構資訊的敏感度。 綜合上述結果，英語學習者於關係子句的處理歷程中，可運用的資訊與英語人士並無本質上的不同。然而學習者傾向評估訊息的一致性，當訊息彼此衝突時，可能造成「延後決定」的策略運用或處理成本交換機制的產生，而對訊息的敏感度又與學習者本身的語言程度有關。因此，本研究主張第二語言的語句處理模式支持了以功能為本的「約束假設」（constraint-based hypothesis, McDonald et al. (1994)）。This dissertation aims at investigating (a) how Chinese-speaking learners of English as their second language (L2) process relative clauses (RC), (b) what information may get involved during the real-time processing, and (c) whether linguistic proficiency modulates the parsing procedure. In particular, two grammatical aspects of RCs were examined: filler-gap dependency as shown in (1) and local syntactic ambiguity between main verb (MV) vs. reduced RC (RRC) interpretations, as shown in (2); in addition, noun animacy of RC heads (animate vs. inanimate in both (1) and (2)) and lexical frequency of the RC verbs being used as participles (High-PP vs. Low-PP in (2)) were also configured as a way of understanding how and when the syntactic and non-syntactic information have their effects. Experiment 1 (Exp. 1.1: self-paced reading task& Exp. 1.2: gated sentence completion task) and Experiment 2 (eye-tracking task) were conducted to examine whether the filler-gap dependency was formed and modulated by both/either syntactic (e.g., Active Filler Strategy: Frazier, 1987) and/or lexical-semantic information (e.g., animacy of sentential subjects: musician vs. accident) during L2 processing. Participants for each experiment included one group of native English readers and two groups of advanced and intermediate Chinese-speaking learners. The results showed that the L2 learners were able to employ both syntactic and non-syntactic information during RC processing just like the native readers, but the ways of how the information was employed by the native and non-native readers may differ. To be specific, although L2 learners showed a hierarchy of processing difficulty similar to native readers’, i.e., animate ORC (1b) >> inanimate ORC (1d) = animate SRC (1a) = inanimate SRC (1c), where “>>” means “more difficult than”, there was a crucial difference in where the patterns were observed, i.e., the RC and MV region by native readers vs. the MV region by L2 learners. In addition, although more alternative semantic interpretations were indeed associated with animate ORCs than inanimate ORCs as shown in Experiment 1.2, the semantic indeterminacy in the sense of Gennari and MacDonald (2008) did not modulate the on-line RC comprehension by L2 learners. It was thus argued that L2 learners tended not to make syntactic commitment during their initial parsing within the RC region if the information available was incongruent. Such a late assignment strategy applies to the cases of inanimate SRC (1c) and inanimate ORC (1d), in which the inanimate RC heads contra their topichoods in terms of the subject interpretation. On the other hand, when lexical information and its topichood was in accordance, the syntactic decisions were readily made without hesitation as in the cases of (1a) and (1b). It was during the re-reading stage where the wrong SRC parse for an animate ORC (1b) was more difficult to erase, thus resulting in longer RTs in the MV region in both the self-paced reading and eye-tracking tasks. Experiment 3 (self-paced reading task) and Experiment 4 (eye-tracking task) were carried out to examine whether the local syntactic ambiguity is resolved and moderated by both syntactic (e.g., Minimal Attachment Strategy: Frazier, 1987) and non-syntactic information (e.g., animacy of sentential subjects: defendant vs. evidence& PP-frequency of RC verbs: Low-PP “examined” vs. High-PP “accepted”) during L2 processing. Participants for each experiment included one group of native English speakers and two groups of advanced and intermediate Chinese-speaking learners. The results suggested that local syntactic ambiguity resolution between MV vs. RRC in L2 processing is lexically conditioned, showing that L2 learners had access to both syntactic and contextual information as native speakers did. However, unlike native speakers, L2 learners showed their early sensitivity to the contextual information in the RC verb region. In particular, when information was incongruent, e.g., an animate NP& High-PP in (2c) or an inanimate NP & Low-PP in (2b), L2 learners spent more time in evaluating their congruency. The evaluation process was argued to correlate with the processing difficulty associated with the occurrence of the disambiguating by-phrase. In addition, linguistic proficiency modulated the evaluation procedure. Taken together, the results refuted the view that L2 parsing is fundamentally different from L1 parsing (Shallow Structure Hypothesis, Clahsen and Felser, 2006a), and theevaluation procedure observed in our L2 participants should be accommodated under a constraint-based L2 processing model in the sense of MacDonald et al. (1994).

(1)Filler-Gap Dependency in Full RC a.Animate RC Head_Subject-extracted RC (SRC) The musiciani [that ei witnessed the accident] angered the policeman a lot. b.Animate RC Head_Object-extracted RC (ORC) The musiciani [that the accident terrified ei] angered the policeman a lot. c.Inanimate RC Head_ Subject-extracted RC (SRC) The accidenti [that ei terrified the musician] angered the policeman a lot. d.Inanimate RC Head_Object-extracted RC (ORC) The accidenti [that the musician witnessed ei] angered the policeman a lot. (2)Syntactic Ambiguity between RRC and Main Verb (MV/RRC) a.Animate RC Head_Low-PP The teacher (that was) loved by the students was informative. b.Inanimate RC Head_Low-PP The textbook (that was) loved by the students was informative. c.Animate RC Head_High-PP The friend (that was) accepted by the man was very impressed. d.Inanimate RC Head_High-PP The award (that was) accepted by the man was very impressive.