語料庫於第二外語寫作之效益：後設分析

Abstract

本研究採用後設分析法檢驗語料庫在第二外語學生寫作表現中的整體成效。為達到總效果量， 本論文對30個自2000年至2022年間的研究 (總效應值56)進行後設分析。此外，另從四組資料（例如，出版資料、設計資料、人口資料和治療資料）中提取19個調節變項，進行了一系列調節效果分析，以探究導致觀察效果變異的因素。 後設分析的結果顯示，語料庫在第二外語的學生寫作表現中的整體擁有高度效果量（g = 0.95）。調節變量分析顯示，觀察到的效果變異可歸因於四組資料中的變量（例如，出版資料、設計資料、人口資料和實驗組資料）。本研究的主要發現是：（一）語料庫使用在協助第二外語寫作方面比傳統教學方法來得更有效；（二）尤其自從2015年起，語料庫在第二外語寫作中的整體效果提升，總效果量大幅度增加；（三）在研究設計上，抽樣方法和後測設計對效果量影響較大；（四）母語與英語為同一語系的學習者比不同語系的學習者，更能從使用語料庫中受益；（五）對不同程度的學習者，語料庫的使用都具高度成效，語言程度高者則擁有最大效果量；（六）語料庫使用對英語或非英語系的學習者都有高度效果，其中英語系的學習者的效果略大；（七）語料庫使用對擁有碩士學位的學習者有高度效果，但對於大學生的學習者僅有中度的效果量；（八）短至中時間的語料庫使用效果比長時間使用來得更有效；（九）小型語料庫（小於100萬個單詞）更能在第二外語寫作中提供最大的效果量，大型語料庫（1億至10億個單詞）和中型語料庫（100萬至小於1億個單詞）分別為大與中度的整體效果量，而龐大型的語料庫（超過10億個單詞）效果最低；（十）DIY（自建）語料庫為成效最佳的語料庫類型，並且整體成效相當於一般單語語料庫和專業語料庫；（十一）語料庫使用在學習詞彙語法/措辭層面上比學習篇章/結構層面時更有效果；（十二）教師在高度控制的環境（即教師同時涉及語料庫諮詢和寫作反饋）或自主學習環境（即教師無需涉及語料庫諮詢或寫作反饋）使用語料庫時成效最佳。但如果教師只涉及寫作反饋，效果為最低；（十三）第二外語寫作中的語料庫使用無論是實體或是非實體型的數據驅動學習，在基礎文句(local)或進階寫作(global)的預期成果，皆達高度效果量。總體而言，語料庫使用是一種能促進第二外語寫作的高效的學習方法。The present study performs a meta-analysis to examine the overall effectiveness of corpus use in ESL/EFL student writing performance. A total of 30 primary studies representing 56 effect sizes found from 2000 to 2022 are meta-analyzed to generate an overall effect size. Furthermore, a series of moderator analyses are conducted on 19 moderating variables extracted from four groups of data (e.g., publication data, design data, population data, and treatment data) to explore the factors contributing to the variability in the observed effects.The results of the meta-analysis indicate a large overall effect size (g = 0.95) of corpus use in ESL/EFL student writing performance. Analyses of moderators reveal that the variability in observed effects can be attributable to variables from all the four groups of data (e.g., publication data, design data, population data, and treatment data). The major findings of the present study suggest that: (1) corpus use is largely more effective than the traditional teaching methods in assisting ESL/EFL writing; (2) the overall effectiveness of corpus use in ESL/EFL writing has been increased over years, especially since 2015 when the overall effect sizes are from large to very large; (3) research design is also attributable to the variation in the observed effect sizes, especially regarding the sampling method and the post-test design; (4) learners whose L1s are in the same language family with English could benefit from corpus use more than those in the different language family group; (5) corpus use is largely effective to learners from different proficiency levels of which the largest effect size is found in the high proficiency group; (6) corpus use is also largely effective to learners from either English or non-English major of which a slightly larger effect size is observed in the former; (7) corpus use seems to provide a very large effect on learners at the postgraduate education level, but the effect is only medium to learners at the undergraduate education level; (8) short to medium treatment duration lead to a larger effect of corpus use than a long treatment duration; (9) corpora of small sizes (i.e., less than 1 million words, locally made corpora) are more likely to provide the largest effect in ESL/EFL writing, corpora of large sizes (i.e., 100 million to 1 billion words) and medium sizes (i.e., 1 million to less than 100 million words) provide the overall effects of large and medium respectively, but the effect of corpus use is lowest in case of corpora of excessively large sizes (i.e., more than 1 billion words); (10) DIY (do-it-yourself) corpora have been shown to be the most effective corpus type, and then the overall effect size powers are respective to general monolingual corpora and ready-made specialized corpora; (11) corpus use works more effectively when the lexico-grammatical/phraseological levels rather than the discourse/structural levels are the focus of learning; (12) corpus use is most effective in two cases of teacher role, either in a highly controlled environment (i.e., when the teacher involves with both corpus consultation and writing feedback) or in an autonomous learning environment (i.e., when the teacher involves with neither corpus consultation nor writing feedback), but the effect is lowest when the teacher only involves with writing feedback; (13) Corpus use in ESL/EFL writing could retain the large effect regardless of whether the data-driven learning type is hands-on or hands-off, and whether the target outcome is on the local or global aspects of writing. Overall, corpus use is a highly effective method and is recommended for facilitating ESL/EFL writing.