類別結構的亂度因素、刺激向度個數對分類學習行為的影響

Abstract

Sloutsky (2010; Kloos與Sloutsky, 2008) 操弄不同的類別結構亂度 (categorical entropy) 進行類別學習作業，藉此提出了雙系統理論，認為人們會啟動不同的系統，濃縮式系統 (compression-based system)或選擇式系統 (selection-based system)，以適應不同的類別結構組成之刺激材料。本研究回顧了Sloutsky的研究證據與過去類別學習領域的相關文獻，認為此雙系統理論可能只適用在向度數目較多的情境之下，因此設計了三個實驗，使用和Kloos與Sloutsky (2008) 相同的實驗派典，欲說明刺激材料的向度個數確實會影響到人們的類別學習行為。實驗一發現，Sloutsky所預測的類別結構與學習方式之交互作用只出現在向度個數較多的情境，向度個數少時則無此交互作用。實驗二得到與實驗一相同的結果，並排除了刺激材料本身特性（幾何圖形或類自然類別材料）此一混淆變項。實驗三採用特別設計的依變項，直接觀察受試者採用相似性(similarity)或規則(rule)的方式進行分類判斷，集群分析的結果顯示在向度數目少的情境時，不管何種類別結構受試者均傾向使用以規則為基礎的選擇式系統學習。因此，綜合以上發現，本研究認為Sloutsky的雙系統理論必須考慮到向度數目此一變項，才能更廣泛的應用於各種類別學習情境之中。

The goal of this research is to point out that the dimensions of experimental materials can influence human category learning, which is neglected by traditional models of category learning. Three experiments in this research examined the effect of stimuli complexity by following the paradigms of Kloos and Sloutsky (2008). In Experiment 1, the prediction of Sloutsky’s theory (2010) on the interaction effect between category structures and learning conditions succeeds only at high complexity of materials, but fails in the low complexity condition. Experiment 2 was conducted by the same experimental setting as Experiment 1, but the natural-like stimuli were replaced by well-defined artificial geometrics. The result of Experiment 2 is the same as Experiment 1, suggesting that the complexity of materials plays a critical role in category learning no matter what kind of stimuli are used. Experiment 3 found that various materials complexity had distinct effects on human category representations. Namely, when experimental stimuli are relatively complex, people would use the corresponding category learning system to represent stimuli to learn dense categories or sparse ones. In contrast, when the stimuli are relatively simple, participants would represent the stimuli in a rule-based manner both in dense and sparse category structures.