以電腦輔助出題的台語詞彙測驗研究佮語法接觸調查
No Thumbnail Available
Date
2021
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
因為歷史佮政治因素,多數台灣人以華語做第一語言。台灣各語言的發展無平等,親像其他的本土語言,台語的資源相對較欠缺,而且受著強勢華語的影響,台語當咧發生變化,沓沓仔倚向華語,甚至可能會影響台語測驗的標準。為著拓展台語測驗的發展面向,以及了解語法變化對台語標準化可能產生的影響,本文進行三項研究。第一,建立一套自動出題的方法佮程序,產出選項是漢字佮羅馬字兩種題型的詞彙測驗。以台語現有的資料庫做基礎,利用電腦程式自動產生詞彙測驗選擇題的題身佮選項,產出的題目由人工略仔修改了,以線頂的方式實際施測。第二,驗證出題方法的可用性,閣分析試題的難度特徵。收集作答反應以後,以古典測驗理論(CTT)佮試題反應理論(IRT)的Rasch model方法分析,來驗證試題可用性,才閣進一步共兩種題型中會影響難度的試題特徵揣出來。結果發現,本研究設計的出題方法佮程序,確實會當產出有鑑別度、也符合測驗單向度要求的可用試題。難度特徵方面,漢字選項題型的各種客觀特徵,攏較輸專家主觀判斷的結果;羅馬字選項題型有「題目台語特色詞的數量」、「誘答選項佮正答選項的聲調差異數量」兩項客觀特徵會當預測難度,專家判斷顛倒無準確。第三項研究轉向關注可能影響測驗標準的語法變化。整理受華語影響的台語語法,設計做語法接受度量表,kap入去研究一的線頂測驗,利用測驗的形式做語言調查的替代方案,來評估台語語法受華語影響的情形。結果發現,台語語序佮華語無仝的語法特色,受影響的程度較低,語序仝款、語法詞無仝的語法特色,受影響的程度較懸,甚至有錯誤語法的接受度比正確語法的接受度較懸的例。本研究的成果,開創台語測驗自動命題的研究,會當以較低的成本增加台語語言能力認證考試的題目數量,也會當予一般大眾初步判斷家己的程度適合報佗一个等級的考卷;另外,因應中學階段本土語納入必修課,研究成果也會當發展做教師判斷學生台語程度的工具。
Due to historical and political reasons, the vast majority of population in Taiwan speak Mandarin as their first language. Taiwanese, like many other"native languages" in Taiwan, lack resources to maintain their native-speaking population size. As Taiwanese has been highly influenced by Mandarin, the standards for Taiwanese proficiency tests may need to be adjusted accordingly. To support the ongoing language revitalization efforts, this dissertation is devoted to the testing and understanding of Taiwanese proficiency in Taiwan, taking into account the fact that most test takers may be fluent in Mandarin, a language closely related to and partially sharing the writing system with Taiwanese.In this study, I developed two Taiwanese vocabulary tests using computer programs and Taiwanese databases. The two tests consisted of multiple-choice questions that were written in Chinese characters and Lomaji (a Latin-based orthography), respectively. Participants were recruited through social media and took the tests online. Participants' responses were analyzed with Classical Test Theory (CTT) and Rasch model from Item Response Theory (IRT). Results of the vocabulary tests show that theprocesses and procedures used to create the vocabulary tests are satisfactory in terms of both discrimination and unidimensionality. Item difficulty analysis reveals that the objective attributes of an item are not good predictors of its difficulty, but expert opinions are, when the item is written in Chinese characters. In contrast, when the item is written in Lomaji, its objective attributes - in particular, the number of words that are not present in Mandarin, and the number of tonal differences between the target and distractors - are good predictors of the item's difficulty, while expert opinions are not. Additionally, I created a Taiwanese grammar questionnaire that was administered along with the vocabulary tests. The questionnaire consisted of sentences with function words and sentence order that were either native sounding (traditionally grammatical) or highly influenced by Mandarin (traditionally ungrammatical). Participants were asked to judge the acceptability of each sentence with their gut instinct. I analyzed the correlation between a participant's performance on the vocabulary tests and their performance on the grammar questionnaire. Results of the grammar questionnaire shows that an ungrammatical sentence is relatively more acceptable if its sentence order is the same as how it would be in Mandarin, but its function words are not present in Mandarin. In contrast, an ungrammatical sentence is relatively less acceptable if its sentence order is different from how it would be in Mandarin, but its function words are present in Mandarin. Overall, the grammatical version of a sentence is more acceptable than the ungrammatical version of the sentence, although this is not true for every sentence in the questionnaire.This study is one of the first to programmatically generate items for Taiwanese proficiency tests. The test development method introduced by this study considerably lowers the cost to create new items for existing tests. An abundance of items will also enable test takers to estimate their proficiency beforehand and register for the appropriate difficulty levels. Furthermore, as mother tongue education is becoming mandatory in high schools, this study provides a convenient way for teachers to create Taiwanese exams for students at different proficiency levels.
Due to historical and political reasons, the vast majority of population in Taiwan speak Mandarin as their first language. Taiwanese, like many other"native languages" in Taiwan, lack resources to maintain their native-speaking population size. As Taiwanese has been highly influenced by Mandarin, the standards for Taiwanese proficiency tests may need to be adjusted accordingly. To support the ongoing language revitalization efforts, this dissertation is devoted to the testing and understanding of Taiwanese proficiency in Taiwan, taking into account the fact that most test takers may be fluent in Mandarin, a language closely related to and partially sharing the writing system with Taiwanese.In this study, I developed two Taiwanese vocabulary tests using computer programs and Taiwanese databases. The two tests consisted of multiple-choice questions that were written in Chinese characters and Lomaji (a Latin-based orthography), respectively. Participants were recruited through social media and took the tests online. Participants' responses were analyzed with Classical Test Theory (CTT) and Rasch model from Item Response Theory (IRT). Results of the vocabulary tests show that theprocesses and procedures used to create the vocabulary tests are satisfactory in terms of both discrimination and unidimensionality. Item difficulty analysis reveals that the objective attributes of an item are not good predictors of its difficulty, but expert opinions are, when the item is written in Chinese characters. In contrast, when the item is written in Lomaji, its objective attributes - in particular, the number of words that are not present in Mandarin, and the number of tonal differences between the target and distractors - are good predictors of the item's difficulty, while expert opinions are not. Additionally, I created a Taiwanese grammar questionnaire that was administered along with the vocabulary tests. The questionnaire consisted of sentences with function words and sentence order that were either native sounding (traditionally grammatical) or highly influenced by Mandarin (traditionally ungrammatical). Participants were asked to judge the acceptability of each sentence with their gut instinct. I analyzed the correlation between a participant's performance on the vocabulary tests and their performance on the grammar questionnaire. Results of the grammar questionnaire shows that an ungrammatical sentence is relatively more acceptable if its sentence order is the same as how it would be in Mandarin, but its function words are not present in Mandarin. In contrast, an ungrammatical sentence is relatively less acceptable if its sentence order is different from how it would be in Mandarin, but its function words are present in Mandarin. Overall, the grammatical version of a sentence is more acceptable than the ungrammatical version of the sentence, although this is not true for every sentence in the questionnaire.This study is one of the first to programmatically generate items for Taiwanese proficiency tests. The test development method introduced by this study considerably lowers the cost to create new items for existing tests. An abundance of items will also enable test takers to estimate their proficiency beforehand and register for the appropriate difficulty levels. Furthermore, as mother tongue education is becoming mandatory in high schools, this study provides a convenient way for teachers to create Taiwanese exams for students at different proficiency levels.
Description
Keywords
台語測驗, 自動命題, 難度特徵, 語言接觸, 語法變化, Taiwanese language fluency test, computer-assisted item generation, item difficulty, language contact, grammar change