PyTorchKR Papers

최신 논문

총 2편
태그·기간 등 상세 필터

카테고리 탐색

로그인 후 이 태그 구독

arXiv:2601.12164v5cs.CY

질문하는 언어: 논쟁적인 정치 문서에 대한 LLM 분석에서의 언어 조건 이념적 분기

The Language You Ask In: Language-Conditioned Ideological Divergence in LLM Analysis of Contested Political Documents

ChatGPT 5.2와 Claude Opus 4.5를 대상으로 동일한 우크라이나 시민사회 문서를 러시아어·우크라이나어 프롬프트로 분석하게 한 결과, 두 모델 모두 프롬프트 언어에 따라 해당 문서의 서명자를 각각 정당성 부정(러시아어) 또는 정당성 인정(우크라이나어) 방향으로 해석하는 언어 조건부 이념적 편향을 일관되게 보였다. 이는 대규모 언어 모델이 정치적으로 논쟁적인 문서를 분석할 때 중립적 입장을 취하지 않고 프롬프트 언어의 지배적 해석 틀을 암묵적으로 채택함을 시사하며, 다국어 환경에서의 다원주의적 정렬(pluralistic alignment) 필요성을 제기한다.

arXiv:2606.28325v1cs.CY

The Digital Afterlife of Empires: Four Language Models Converge on the Same Imperial Cartography of Writing

Large language models process the world's writing systems with radical inequality. We constructed the Digital Script Representation Index (DSRI), a seven-axis measure of digital support, and applied it to the 300 writing systems of the Global Script Database (Fukui, 2026). Only 29 scripts (9.7%) are fully supported by contemporary digital infrastructure; among 158 living scripts, 60 (38.0%) lack complete support. Tokenizer efficiency varies by a factor of 31.7 across 45 scripts measured with parallel text. A serial mediation model -- imperial intervention to speaker population to web corpus to tokenizer efficiency -- is consistent with full mediation, with the direct effect of empire indistinguishable from zero (beta = -0.22, p = 0.39) and structural equation model fit indices indistinguishable from saturation at n = 45; the bias-corrected bootstrap CI grazes zero, and we treat the mediation as suggestive rather than confirmatory. Across four independent LLM families (Claude, GPT-4o, Grok, DeepSeek; 12,000 API calls), base-rate-deviation error patterns converge at Spearman rho = 0.85-0.98 (all p < 0.002). 172 script-feature items are answered identically wrong by all four models; over-attribution outnumbers under-recognition 3.9:1, and "used for religion" alone concentrates 43.6% of convergent errors (enrichment 4.1x). With religion excluded as a sensitivity check, the cross-architecture convergence is preserved (mean rho = 0.87 on nine features) and the over-attribution asymmetry persists at 1.77:1 (n = 97, binomial p = 0.008), indicating multi-channeled rather than single-channeled bias. The findings are consistent with an interpretation in which the structural inequalities historical empires inflicted on script communities persist in contemporary language models through the shared training corpus rather than through any individual model's design choices.