서울대학교 AI디지털인문학센터와 철학사상연구소에서는 AI를 활용한 서양 고전문헌 연구 플랫폼을 구축하고 있는 일본 휴마니텍스트 프로젝트의 참여 학자들을 초청하여 “AI로 만들어가는 인문학의 미래”라는 주제로 강연회를 개최합니다. 연구자와 학생분들의 많은 관심과 참여 부탁드립니다.
휴마니텍스트: AI로 만들어가는 인문학의 미래
Humanitext: Shaping the Future of Humanities with AI
장소: 서울대학교 신양학술정보관[4동] 302호
시간: 3월 9일 오후 2시-4시
사회: 서울대학교 철학과 이상엽 교수
신청링크 : https://docs.google.com/forms/d/e/1FAIpQLSchMFwLlCCIScUq4ze6Z06XV8EE0Uyqc7AWQuHfnuaeX2HFJA/viewform
강연자 1: Naoya Iwata (Nagoya University; National Institute of Informatics)
강연 제목:Scaling Humanitext Antiqua: Advances in RAG Retrieval Accuracy and Corpus Expansion for a Western Classical Literature Platform
강연 초록: This talk presents recent progress on Humanitext Antiqua, an AI-powered platform for Western classical literature. We report on three developments: improvements to RAG retrieval accuracy through context-aware query reformulation; an ongoing corpus expansion from approximately 1,000 to 2,500 works; and a metadata enrichment workflow that combines automated Wikidata linking with human-in-the-loop review as a scalable annotation strategy.
강연자 2: Ikko Tanaka (J. F. Oberlin University)
강연 제목: Extending the Horizons of Western Classical Texts toward Networked Interpretation — Linking Texts, References, and Spatiotemporal Context with LLMs
강연 초록: This lecture presents two systems developed within the Humanitext project as extensions of its core vectorized corpus of Western Classical texts. Humanitext Reader develops a structured reference framework linking primary texts with ancient commentaries and modern research, addressing variation in citation scope and differences between textual editions. Humanitext GEO integrates the vectorized corpus with temporal and geographic frameworks through Linked Open Data, enabling city-centered synchronic and diachronic exploration. Together, these systems outline the development of an AI-assisted interpretive infrastructure connecting texts, references, and spatiotemporal context.
강연자 3: Jun Ogawa (University of Tokyo)
강연 제목: Humanitext Schema: Struggles to Connect Text Sources and Integrate Broad Knowledge
강연 초록: Humanitext Antiqua generates answers grounded in accurate primary sources, yet its major limitation is the insufficient integration of extra-textual context, such as related text fragments, commentaries, and secondary scholarship. Because semantic connections between sources rely on expertized knowledge, they cannot be captured through vector similarity alone. We therefore propose linking texts and contextual information through a knowledge graph and incorporating it into the generation process. Our work explores methods for modeling text–commentary relationships as a first step, including a TEI-based DTS, character-level RDF representation, and property graphs. This talk outlines our experiments and the trial-and-error process behind this approach.
