知覺類別學習中的類別變異效果：以聽覺刺激為例

Category Variability Effect in Perceptual Category Learning with Auditory Stimuli

類別變異效果是指，人們傾向將介於兩類別中間的刺激，分類至變異程度較高的那個類別。過去研究指出範例模型對這個效果無法提供合理解釋，相對地，規則模型則可以，因此這個效果一直是這兩類模型比較的試金石。雖然過去研究時而觀察到此一效果，但由於各研究使用的實驗方法不盡相同，使得我們並不清楚究竟在何種條件下類別變異效果會出現。本研究使用單音刺激進行實驗，嘗試釐清類別變異效果的發生條件。刺激向度為單音頻率，並且透過心理物理函式轉換成心理量（mel）。三個實驗均以典型的回饋學習派典進行。實驗一嘗試瞭解是否可以透過指導語操弄，促使參與者覺察到類別變異程度的差距。結果發現不論有無提示參與者兩類別變異程度有所不同，均沒有發現類別變異效果。實驗二嘗試直接操弄類別變異的差距程度，縮小低變異類別的變異程度，使其成員間距縮小或僅有一個成員，結果均發現顯著的類別變異效果。實驗三則將加大高變異類別的變異程度，結果仍然支持類別變異效果。最後，本研究再根據三個實驗的結果以及前人研究的結果，歸納出一個量化指標D分數，並用於預測類別變異效果，當D大於1時，類別變異效果便會出現，反之則否。本研究結果將有助於未來進行相關研究的參考。

Category variability effect is referred to that the item at the mid position between two categories is more likely to be classified as the category with a larger variability. This effect is often regarded as evidence against the exemplar-based model, which would always classify the middle item as the low-variability category, due to a larger similarity between the mid-position item and that category. In contrast, the rule-based model is able to account for category variability effect, as it is sensitive to category distribution. Although category variability effect has been reported in several studies, it is still unknown under what circumstance this effect would occur, due to the diversity on methodology of the past studies. Thus, the purpose of this study is to examine in what condition, category variability effect would be induced. To this end, the stimuli adopted in this study were single tones of different frequencies, which were transferred to mel, an interval psychological scale of frequency. In Experiment 1, we examined category variability effect via giving the participants a hint or not about the difference on variability between two categories. The results showed no category variability effect, no matter a hint is given to participants or not. In Experiment 2, low-variability category was made more condensed or even having only one exemplar. The results showed a clear category variability effect in these two conditions. Also, in Experiment 3, this effect was observed, when the high-variability category was made more diverse. To summarize our findings, an index D was developed to display the ratio on logarithmized variability between two categories. According to D, when D > 1, category variability effect occurs and disappears otherwise. Also, this index can be applied to predicting which of the past studies would find this ef fect and which would not.