Scholarlink

The transcription of ancient earthquakes remains a significant challenge due to the ambiguity and uncertainty of historical descriptions. The vision is to document this information, found in books and manuscripts in various languages, in a structured, semantically rich, and interoperable manner, enabling a complete picture of past seismic activity. While ontologies like CRM-EQ offer semantically expressive models for capturing seismic events, the manual transcription of such information from historical sources remains time-consuming and requires expertise. In this paper we propose and evaluate a hybrid and collaborative workflow for transcribing earthquakes from historical texts. We compare five different methodologies from manual curation to LLM-based pipelines, assessing them in terms of effort, accuracy, and scalability. Our findings show that semi-automated approaches are practical, balancing effort and accuracy. We captured 248 earthquake events and produced over 24,500 RDF triples, demonstrating the feasibility of semi-automated transcription for historical seismic data.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

To Automate or Not to Automate the Transcription of Ancient Earthquakes: Toward a Global Knowledge Graph About Ancient Earthquakes

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sophia Sideri,
  • Emmanouil Patronakis,
  • Evangelos Nikiforos,
  • Isidoros Chatzichrysos,
  • Apostolos Baniotis,
  • Valantis Zervos,
  • Stavros Tzormpatzakis,
  • Michalis Saridakis,
  • Iosif Oikonomakis,
  • Michalis Mountantonakis,
  • Pavlos Fafalios,
  • Yannis Tzitzikas

摘要

The transcription of ancient earthquakes remains a significant challenge due to the ambiguity and uncertainty of historical descriptions. The vision is to document this information, found in books and manuscripts in various languages, in a structured, semantically rich, and interoperable manner, enabling a complete picture of past seismic activity. While ontologies like CRM-EQ offer semantically expressive models for capturing seismic events, the manual transcription of such information from historical sources remains time-consuming and requires expertise. In this paper we propose and evaluate a hybrid and collaborative workflow for transcribing earthquakes from historical texts. We compare five different methodologies from manual curation to LLM-based pipelines, assessing them in terms of effort, accuracy, and scalability. Our findings show that semi-automated approaches are practical, balancing effort and accuracy. We captured 248 earthquake events and produced over 24,500 RDF triples, demonstrating the feasibility of semi-automated transcription for historical seismic data.