文本連貫效果在大學生閱讀科學文章之角色

大學是從課室學習邁向獨立自主終身學習的重要過渡階段，倘能在此時期找到有效解讀科學文章訊息的方法，將有助於提升現代公民必備的科學素養。文本連貫性（text cohesion）是影響閱讀理解的重要因素，本研究以文本連貫性對臺灣大學生閱讀科學說明文的理解為研究核心，而臺灣目前並沒有評量大學生說明文閱讀理解能力的測驗，因此本研究之主要目的有二：第一，因著研究需求，發展一套適用於測量大學生說明文閱讀理解能力的測驗；第二，探索文本特徵（即文本連貫性）、讀者特徵（即閱讀理解能力、先備知識）與任務經驗（即閱讀文章加上做測驗為一次任務週期）在科學文章閱讀理解的關係，透過眼動等線上（on-line）歷程資料、測驗等離線（offline）測量資料，以及結合線上與離線的提示回溯放聲思考原案，探討大學生閱讀高低連貫性科學文章時的閱讀理解表現和歷程。研究參與者為120名20—30歲之大學生（含研究生），隨機分派至四種實驗組別，每位參與者皆會閱讀兩篇各約1,500字的地球科學文章，閱讀順序採平衡設計，操弄條件為文本連貫性（高連貫性 vs. 低連貫性）和任務經驗（初次任務 vs. 再次任務）。參與者先進行特徵變項的測量，包括研究者自編《大專生說明文閱讀理解測驗》和《先備知識測驗》，隨後依照分派組別閱讀科學文章，同時錄製其眼動情況，閱讀後實施閱讀表現測驗，包括提示回憶測驗與理解表現測驗。第二次任務的閱讀和測驗結束後再進行提示回溯放聲思考（cued retrospective think-aloud），即播放參與者閱讀時的電腦螢幕畫面與眼動軌跡，請參與者解釋自己是如何進行閱讀的，做為閱讀歷程解釋的輔助。在資料分析的部分，閱讀表現和全文層次、主題層次眼動指標資料採線性混合模型（linear mixed-effects models, LMMs）進行分析。閱讀表現結果發現，在控制文本內容後，文本連貫性對大學生科學文章理解表現的效果會受到任務經驗調節，連貫效果（即高連貫文本比低連貫文本更有利）僅出現於再次任務者的理解表現，而初次任務者則否；另一方面，再次任務者閱讀高連貫文本時比初次任務者有更好的理解表現；從全文層次的眼動資料的確可以發現，再次任務者在高連貫文本的總凝視時間、總凝視次數皆顯著高於初次任務者，顯示有過任務經驗的讀者會調整自己的閱讀方式，投注更多心力在高連貫的新材料上，支持自我調整學習理論（self-regulated learning theory），但任務經驗對於低連貫文本的理解表現和全文層次眼動指標並無影響；至於提示回憶表現則未出現文本連貫性與任務經驗的交互作用。此外，不論在理解表現或提示回憶表現，連貫效果並未受到閱讀能力或先備知識調節，因此能力強化效果假說（ability-as-enhancer hypothesis）或補償假說（compensation hypothesis）並未獲得支持證據；但閱讀能力與先備知識皆產生主要效果，亦即高閱讀能力者與高先備知識者皆有較佳的理解表現和提示回憶表現。眼動資料顯示，閱讀能力越高者解碼科學文章訊息越容易（反映在較低的平均凝視時間），且閱讀能力有可能與主題段落較快的理解整合有關（反映在其中一篇文章的主題層次再次注視時間越少）；而先備知識越高，主題層次每字凝視時間與首次每字注視時間越少，表示處理科學文章訊息需花費的認知資源越少，尤其是初始處理。概念層次眼動指標資料採滯後序列分析（lag sequential analyses, LSA），結果顯示再次任務者閱讀高連貫文本會進行較為密集的概念參照，尤其是在因果集合、比較對照等結構性強的子主題內會有較多的概念間視覺轉移；初次任務者在閱讀低連貫文本時經常做的是跨主題的參照，再次任務者在閱讀低連貫文本反而增加的是局部參照，或許這也部分反映了任務經驗增加沒有帶來低連貫文本閱讀表現進步的原因。閱讀階段層次眼動指標資料使用三次成長模型（cubic growth curve analysis, GCA）進行分析，將凝視時間切割為總閱讀時間的每10%為一個波段，針對材料中各個別頁面形成縱貫資料，結果顯示科學文章第一頁在閱讀初期有較多凝視時間，但趨勢持續下降，隨後凝視又會在第八波段左右再度增加，顯示閱讀中後期讀者會回到首頁重讀，至於其他頁面的凝視時間通常呈現先升後降的趨勢，但依照頁面順序，由升轉降的反曲點位置會有波段遞移的情況，到了閱讀中後期又會緩升，但並非每一頁都出現這種斜率轉正的曲線，顯示到了閱讀中後期通常為選擇性的認知處理，帶有特定的閱讀目的。整體而言，本研究為Kintsch建構—整合模式中的表徵層次提供了實徵證據，以歷程資料和測量資料共同描繪大學生科學文章閱讀理解的樣貌，呈現文本特徵與讀者特徵交互影響的情況，強調理解模型需要考慮文本特徵（即文本連貫性）的潛在複雜性，同時涵蓋一些支持閱讀理解的其他歷程（例如：自我調整），以及通常需要納入的個別差異因子（即閱讀理解能力和先備知識）。最後提出理論與實務上的啟示、限制，以及未來可能的研究方向。

關鍵字

連貫效果；科學文章閱讀；自我調整；眼球追蹤；大學生

並列摘要

College marks an important transition from in-class learning to lifelong learning. Therefore, if students can develop an effective approach to acquire information from science text during this period, their scientific literacy, considered as a necessary ability for modern citizens, can be improved as a result. However, even though expository texts are common in scientific reading, no test is currently available for evaluating the expository reading comprehension skills of college students in Taiwan. Hence, the first aim of this study was to develop an acceptable reading comprehension ability test for college students. Besides, since text cohesion is regarded as an important factor of reading comprehension, the second aim focused on the role of cohesion effect in expository scientific reading for Taiwanese college students. That is, the relationship between text features (i.e., text cohesion), reader features (i.e., reading comprehension skills, prior knowledge) and task experience (i.e., text reading and test practicing as a task cycle) in science text reading comprehension was explored through online measures (i.e., eye movements), offline measurement (i.e., tests), and combined online and offline measures (i.e., cued retrospectively think-aloud, CRTA). One hundred and twenty college and graduate students aged 20-30 participated in the experiment, randomly assigned to four experimental groups. Each participant read two earth science articles containing about 1,500 words each, with a balanced reading order. Text cohesion (high cohesion vs. low cohesion) and task experience (cycle 1 vs. cycle 2) were manipulated. A task cycle involves the following phases. The demographic and characteristic variables of participants were first measured, including the self-derived “Reading Comprehension Test of Expository Text for College Students” and “Prior Knowledge Test,” and then they would read the assigned science text, with their eye movements recorded in the meantime. After reading, performance tests (including cued recalls and comprehension performance tests) were administrated. Once the two task cycles were done, the cued retrospective think-aloud would be performed, in which participants verbalized their thoughts while watching their eye-movement recordings. As for data analysis, three analysis methods were carried out. Firstly, the reading performance, full-text-level and topic-level eye-tracking indicators were analyzed with linear mixed-effect models (LMMs). The results of reading performance showed that after controlling for text content, the effect of text cohesion on the comprehension performance was moderated by task experience, and the cohesion effect (that is, high-cohesion texts were more favorable than low-cohesion texts) only appeared in Cycle 2. In the high-cohesion condition, the participants had better comprehension performance in Cycle 2 than those in Cycle 1. Besides, the results of the full-text-level eye-tracking indicators showed that the readers reading high-cohesion texts had significantly higher total fixation duration and total fixation count in Cycle 2 than those in Cycle 1, indicating that readers who had task experience would regulate their reading strategies and invest more effort in their next reading, which supports the self-regulated learning (SRL) theory. However, task experience had no effect on comprehension performance and full-text-level eye-tracking indicators in the low-cohesion condition. As for cued recall, there was no interaction between text cohesion and task experience. In addition, cohesion effects were not moderated by reading comprehension skills or prior knowledge, neither on comprehension performance nor on cued recall. Therefore, no supporting evidence was found for the ability-as-enhancer hypothesis or the compensation hypothesis. However, both reading skills and prior knowledge had main effects, that is, those with high reading skills and high prior knowledge had better comprehension and cued recall performance. Eye-tracking data showed that readers with high reading skills were better at decoding science text (reflected on shorter mean fixation duration), and that reading skills were likely to be associated with more efficient integration of information in each subtopic, indicated by the shorter topic-level second-pass fixation duration per character found in one of the reading articles. Also, the higher the prior knowledge, the shorter the total fixation duration per character and the first-pass fixation duration per character at the topic level, indicating that prior knowledge helped the cognitive processing of science text, especially in initial processing. In the second analysis method, the concept-level eye-tracking indicators were analyzed by lag sequential analyses (LSA). The results showed that readers who read high-cohesion texts in Cycle 2 would make more visual transitions between concepts, especially in highly structural subtopics made up of more causation and comparison. By contrast, readers who read low-cohesion texts in Cycle 1 often made cross-topic references, while the usage of local references in Cycle 2 increased instead, which might explain why increased task experience did not lead to improved reading performance in the low-cohesion condition. As for the last analysis method, the phase-level eye-tracking indicators was analyzed using cubic growth curve analysis (GCA), and the fixation duration was divided into ten waves (every 10% of the total reading time), forming longitudinal data for each individual page of the science text. The results showed that the fixation duration of the first page of science text was high in the early phase of reading but declined afterwards, and then increased again around in the eighth wave, indicating that readers were inclined to reread the first page in mid- and late-stage of reading. As for the fixation duration of other pages, the duration tended to increase then decline in each of the following pages, and the position of the first inflection point from rising to falling had a wave transition along with the page order. However, a second inflection point from falling to mildly rising was only found in some pages, indicating that readers would perform selective cognitive processing in the mid- and late-stages of reading in response to the specific reading purpose. Overall, this study provided empirical evidence for the representational level and the role of prior knowledge in Kintsch’s construction–integration model (CI model), and used process data and measurement data to jointly describe college students’ reading comprehension of science text. In addition, this study further highlighted the interaction between text characteristics and reader characteristics, emphasizing that comprehension models need to take the underlying complexity of text features (i.e., text cohesion) into account. Furthermore, several aspects were also considered in this research, including the processes that support reading comprehension (e.g., self-regulation), and individual differences, which also often need to be incorporated (i.e., reading comprehension skills and prior knowledge). Finally, the theoretical and practical implications, limitations, and possible future directions were addressed in the last section.

並列關鍵字

cohesion effect ； science text reading ； self-regulation ； eye-tracking ； college students

參考文獻

簡郁芩、吳昭容（2012）：〈以眼動型態和閱讀測驗表現探討箭頭在科學圖文閱讀中的圖示效果〉。《中華心理學刊》，54，385–402。[Jian, Y.-C., & Wu, C.-J. (2012). The effect of arrows in an illustration when reading scientific text: Evidence from eye movements and reading tests. Chinese Journal of Psychology, 54, 385–402.] https://doi.org/10.6129/cjp.2012.5403.07

Alemdag, E., & Cagiltay, K. (2018). A systematic review of eye tracking research on multimedia learning. Computers & Education, 125, 413–428. https://doi.org/10.1016/j.compedu.2018.06.023

Andrews, S., Miller, B., & Rayner, K. (2004). Eye movements and morphological segmentation of compound words: There is a mouse in mousetrap. European Journal of Cognitive Psychology, 16, 285–311. https://doi.org/10.1080/09541440340000123

Eitel, A. (2016). How repeated studying and testing affects multimedia learning: Evidence for adaptation to task demands. Learning and Instruction, 41, 70–84. https://doi.org/10.1016/j.learninstruc.2015.10.003

Höffler, T. N., Koć-Januchta, M., & Leutner, D. (2017). More evidence for three types of cognitive style: Validating the object-spatial imagery and verbal questionnaire using eye tracking when learning with texts and pictures. Applied Cognitive Psychology, 31, 109–115. https://doi.org/10.1002/acp.3300

主題瀏覽