中文語意處理的神經有效性連結之階層式模型比較

Hierarchical Model Comparisons on Effective Connectivity in Semantic Judgments of Chinese Characters

本研究利用功能性磁振造影（fMRI）、動態因果推論模型（DCM）及階層式模型比較，探討中文語意處理的神經機制，並且以一個最適當的模型顯示不同腦區之間的神經動態連結。實驗作業為判斷視覺呈現的中文字對在意義上是否相關。本實驗採用兩種刺激字對，分別為語意相關組與語意無關組。在功能性磁振造影的結果顯示，處理語意相關字對時，位於左腦腹側下額葉（vIFG, BA 47），左腦背側下額葉（dIFG, BA 9），左腦顳中葉（MTG, BA 21）及左腦梭狀回（FG, BA 37）四個腦區會產生顯著的神經活化。進一步的階層式模型比較，針對語意相關組與無關組進行不同動態因果推論模型的檢驗，並嘗試找出一個最精簡且適當的模型，推論處理中文語意時的腦區有效性連結。結果發現從左腦下額葉到左腦顳中葉有顯著的調節效果（modulatory effects），推論左腦下額葉會透過由上而下的處理至左腦顳中葉提取語意表徵。同時，結果也發現從左腦梭狀回到左腦中顳葉的顯著調節效果，推論左腦梭狀回將字形訊息透過由下而上的傳送，至左腦顳中葉處理語意表徵的訊息。

In functional magnetic resonance imaging (fMRI), dynamic causal modelling (DCM) and hierarchical model comparisons were used to investigate the effective connectivity during semantic judgments to visual Chinese characters. Thirty-five participants were asked to indicate if character pairs were related in meaning. Experimental stimuli were character pairs that included semantically-related and semantically-unrelated pairs. In this study, DCM was used to examine the directional influences among brain regions, and hierarchical model comparisons were used to seek the optimal model with parsimonious connections in the semantic network. In conventional fMRI analysis, participants showed activations in left ventral inferior frontal gyrus (vIFG, BA 47), left dorsal inferior frontal gyrus (dIFG, BA 9), left posterior middle temporal gyrus (pMTG, BA 21) and left fusiform gyrus (FG, BA 37). Based on a fully connected model among these four regions, we built sixteen types of models that differed in the modulation of the “related” or “unrelated” conditions on specific connections. From hierarchical model comparisons, the results of the optimal model showed significant modulatory effects from vIFG to pMTG, suggesting top-down influences of the frontal cortex on retrieval of semantic representations. Significant modulatory effects was also found from FG to pMTG, suggesting bottom-up influences of orthographic representations on semantic representations.