Hierarchical Processing of Visual Word Forms

漢字是一種圖像文字，每個漢字都佔據固定的空間以做適當的安排，字形筆畫的結構是影響中文讀者在閱讀時的重要因素。本研究主要的目的是探討視覺系統對於漢字的處理歷程以及其神經機制。本研究以心理物理學的方式測量漢字的空間加乘功能、並採用功能性核磁共振造影的技術檢驗漢字在視覺區的表徵。在研究一中，我們操弄真字、假字、非字、甲骨文以及破碎字等刺激大小以及它們在視野的偏角。根據中文讀者對於這些漢字在偵測作業與區辨作業上的表現，我們使用空間加乘模型推估漢字在視覺系統中的機制。實驗結果顯示，在偵測作業中，參與者對於文字大小的偵測閾值不受文字類別影響，參與者對目標字的閾值會隨著刺激的變小而提高，且斜率接近-1/2。在區辨作業中，參與者對於目標字的閾值也是隨著刺激的變小而提高，且斜率接近-1，且參與者在區辨真字與假字或是非字的閾值高於區辨真字與甲骨文或是破碎字。再者，隨著視野偏角的增大，參與者對於文字的偵測與區辨力都會變差。這個結果顯示在偵測作業中，視覺系統主要受到漢字之細部特徵的影響；在區辨作業中，視覺系統則則是會受到漢字之熟悉特徵的影響。在研究二中，我們採取網膜拓樸對應派典找出視覺區相對應於漢字處理的特定表徵。我們比較大腦視覺區在"空白背景的旋轉楔形與擴大環狀之漢字"與"破碎字背景的旋轉楔形與擴大環狀之漢字"的激發狀況。實驗結果顯示，"空白背景的旋轉楔形與擴大環狀之漢字"會激發初級視覺皮質區與視覺顳葉區，這兩個區域都顯示了視覺區對漢字網膜拓樸的特性。"破碎字背景的旋轉楔形與擴大環狀之漢字"則只會激發視覺顳葉區，尤其是梭狀回對整字有網膜拓樸的特性；亦即梭狀回對於漢字在視網膜的位置具有特定的對應性。再者，相較於破碎字，梭狀回偏好那些呈現在視野中央的文字。在研究三與四中，我們以心理物理測量方法與功能性核磁共振造影比較參與者對於正立漢字與倒立漢字的表現。在行為結果上，我們發現參與者對於倒立漢字的配對表現顯著地比正立漢字差；而此我們稱之為倒立效果。我們同樣也在漢字部件上觀察到倒立效果，但在非字或是甲骨文上卻沒有這樣的倒立效果。這個結果顯示視覺系統會分析整字與部件的空間編碼訊息。配合神經造影的研究結果顯示，視覺系統是以階層的方式處理漢字，右腦的視覺頂葉區主要分析漢字的細部特徵，如，筆畫；右腦的視覺顳葉區處理漢字部件之間的空間關係；左腦的梭狀回則是處理漢字整體的空間訊息。總結，從視覺頂葉區到腹側視覺區，視覺系統會以階層的方式處理漢字。The present study is to investigate the processing of Chinese character in the visual system and its neural network. In study 1, we used a dual task paradigm with both detection and discrimination tasks in spatial summation to measure the spatial constraints for visual word form perception. Five types of stimuli (real, pseudo-, non-, Jiagu and scrambled characters) were used in the experiments. The detection thresholds for the same stimulus size and the same eccentricity were the same for all types of stimuli. When the target size was small, the detection threshold of a character decreased with the increase in its size, with a slope of -1/2 on log-log coordinates, up to a critical size at all eccentricities and for all stimulus types. Beyond the critical size, there was little, if any, improvement of visual word form detectability as target size further increased. That is, the detectability is based on local feature analysis regardless of character types. The contrast thresholds for discriminating real and pseudo- or non-characters were higher than for discriminating real characters and Jiagu characters or scrambled characters. The results suggested that the discrimination threshold is mediated by well-learned forms of components in a character. In study 2, we used a 2 (“rotating wedge of characters” vs. “expanding ring of characters”) by 2 (“with a blank background” vs. “with a scrambled background”) design with fMRI to define the visual cortical regions produce unique representations for Chinese characters. The ventral occipital region showed a retinotopic representation for character stimuli as we found the fusiform gyrus is selective for the contralateral visual fields and has a centre-to-periphery map with a preference for characters presented in fovea. This implies that the fusiform gyrus is selectively to the spatial location information of stimuli. In study 3 and 4, we examined the spatial configuration processing in the visual system for a character by comparing the visual performance for upright character with that for their inverted versions with psychophysics and fMRI. We showed a robust inversion effect for real character regardless of eccentricities. Such an inversion effect was absent for non-characters. The inversion effect was also found in radicals or components in a character. These results suggested that a visual analysis of characters and their components involved a spatial configuration processing. Our neuroimaging evidence showed that the left fusiform gyrus analyzes the spatial configuration of components in a character while the right occipitotemporal regions analyze the spatial configuration of strokes in a component. In addition, the right occipitoparietal region analyzes local features of a character. Together, we demonstrated a hierarchical processing of visual word forms in the visual cortex