关键词: 因果关系, 让步关系, 脑功能网络, 图论分析, 多变量模式分析

Abstract: Connectives signal semantic relations between discourse units and guide readers toward coherent mental representations. Causal connectives such as because and concessive connectives such as although instantiate fundamentally different coherence relations with distinct cognitive demands, yet previous neuroimaging studies have mainly focused on differences in regional brain activation or pairwise functional connectivity, leaving systematic examination of whole-brain functional network organization largely unexplored. This study employed fMRI combined with graph-theoretical network analysis and multivariate pattern analysis (MVPA), using a 2 (connective type: causal vs. concessive) × 2 (semantic consistency: consistent vs. conflicting) within-participant design, to systematically examine how coherence relation type modulates brain network organization during semantic conflict processing in native Mandarin speakers. Functional connectivity matrices were constructed from trial-wise beta estimates across 90 cortical and subcortical regions defined by the Automated Anatomical Labeling atlas. Graph-theoretical analysis quantified global network properties including global efficiency, modularity, and small-worldness, local nodal properties including degree, betweenness centrality, and nodal efficiency, and rich-club organization. Searchlight-based MVPA further examined whether neural representations of semantic conflict were distinguishable across sentence types and generalizable across coherence relation frameworks.
At the global network level, concessive conflict produced significantly lower global efficiency than causal conflict, while causal conflict maintained significantly higher modularity across multiple network sparsity thresholds. At the local nodal level, the left inferior frontal gyrus (IFG), middle temporal gyrus (MTG), and angular gyrus (ANG) exhibited significant connective-type-by-semantic-consistency interaction effects in nodal degree and other centrality measures, with nodal centrality values consistently higher for causal than concessive sentences under conflict conditions. Rich-club analysis revealed that concessive conflict caused substantially greater disruption to hub-to-hub (rich-club) connections and feeder connections than causal conflict, whereas peripheral (local) connections remained unaffected in both conditions. Within-condition MVPA successfully decoded causal conflict from concessive conflict trials, with peak classification accuracy in the left superior temporal gyrus, middle temporal gyrus, and inferior frontal gyrus. Critically, cross-relation-type generalization decoding failed: a classifier trained on one relation type showed no above-chance decoding accuracy in any brain region when tested on the other.
These findings indicate that the brain network mechanisms underlying semantic conflict processing are fundamentally shaped by the coherence relation framework established by connectives. Causal reasoning maintained a modular, high-efficiency network architecture under conflict, reflecting compartmentalized processing through selective inter-module communication. In contrast, concessive reasoning required cross-module integration that disrupted modular boundaries and weakened hub connectivity. The generalization failure further indicated that semantic conflict is encoded through coherence-relation-dependent neural representations rather than as purely abstract, domain-general signals. Taken together, these multi-level convergent findings advance coherence relation research from the level of regional activation to whole-brain network organization, providing novel network-level evidence for the neural basis of the syntax-semantics interaction observed in discourse comprehension.

Key words: causal relation, concessive relation, brain functional network, graph-theoretical analysis, multivariate pattern analysis