지워진 연결: 5･18 광주항쟁 보도검열 기사 네트워크 분석

초록

This study examines 445 news articles on the May 18 Gwangju Uprising that were deleted under the new military regime’s press censorship between May 19 and December 9, 1980―a period of roughly seven months―analyzing the co-occurrence networks among the individuals appearing in these censored articles in order to map, from a holistic perspective, what the regime’s censorship apparatus blocked and how. To uncover censorship mechanisms that single-actor frequency analysis cannot capture, the study draws on Bourdieu’s (1991) reconceptualization of censorship not as “the deletion of words” but as “the control of utterable combinations,” and accordingly adopts the actor pair as its unit of analysis. Individuals were extracted through a hybrid procedure combining LLM-based extraction with human verification and were classified into twelve role types; the analysis then combined four techniques―centrality measurement, Louvain community detection, MRQAP (Multiple Regression Quadratic Assignment Procedure) regression, and role-label permutation testing. Three principal findings emerge. First, the actor network of the censored materials exhibits a strongly partitioned, multi-domain structure of 78 communities (modularity＝0.752), with three religious leaders―Yun Kong-hi, Kim Sou-hwan, and Song Wol-ju―ranking within the top ten in betweenness centrality, occupying brokerage positions comparable or superior to those of core political figures. Moreover, the most densely connected core (k-core) consisted of civilian victims grouped together as a list of the dead, indicating that one of censorship’s central targets was the very visibility of the scale of casualties. Second, in the MRQAP analysis the media effect yielded the largest coefficient across all period pairs, confirming that the primary operational unit of censorship was the individual news outlet; significant negative autocorrelations at the T2 (civic self-governance)-T3 (final suppression) transition and the T1 (precursor)-T5 (political trials) period pair further revealed that censorship targets were systematically reconfigured as the phases of the uprising shifted. Third, role combinations in two domains—civil society linkages (religious leader-dissident activist, dissident activist-scholar, and civilian militia-religious leader) and political discourse (government- politician)―were significantly over-represented in the censored materials. This study constitutes the first quantitative examination of how the new military regime’s press censorship system operated at the level of actor combinations within May 18 Gwangju Uprising research, and it carries methodological significance in demonstrating the integrative potential of semantic network analysis and LLM-based historical document processing.

키워드