故事體文章的記憶研究：景觀模式的運作機制

Memory for Narrative Texts: How do Parts of the Landscape Model Work

理解的計算理論被視為是研究理解過程所發生的認知歷程與理解結束所形成的記憶表徵的一種有效的方式。本文旨在探討一個特定的理解的計算理論－景觀模式，試圖檢驗與建立其理論的效度。實驗1-a企圖尋找一組合適的參數組合得以有效預測參與者對短篇故事體文章的回憶量，進一步將這一組的參數值套用於實驗1-b，並且將模擬的結果再與參與者對另外的短篇故事體文章的回憶量比對，發現景觀模式預測力仍然良好，足證這個模式的效力並不完成依賴參數值的選定，而是該模式本身就有一定的效力。實驗2再以模擬的方式檢驗景觀模式對各種激發來源的預設值是否合理，結果發現如果改變原先預設的激發值的話，模式的預測力大為降低，足證預設的激發值具有合理性，也支持該模式假定之因果推論及指稱詞推論卻實是理解過程必要之推論機制。本文也討論諸多本研究的限制與未來的研究方向。

A computational approach is desirable for dealing with the complex nature of processes and memory representations in text comprehension. The study focused on a particular model of comprehension-the Landscape model-seeking to establish its validity. Study 1-a explored several conditions of the model and examined the model's ability to account for human recall data for short narratives. Parameter values were set for optimal conditions and cross-validated in Study 1-b for new story materials. The model showed high stability in accounting for these new texts, indicating its predictive power does not come from choices of particular parameter values. Several simulations in Study 2 revealed that the model's assumptions concerning the necessities of anaphoric and causal inferences were validated. The model's scheme of assigning activation values for different sources of activation was also supported because modifications of the current scheme reduced performance of the model. Limitations and several lines of future research are discussed.