This short presentation focuses on a reproducible Information Retrieval Q&A pipeline tailored for historiographical corpora, specifically the theses abstracts of our university (3,000+ texts in French). It addresses retrieval challenges by integrating vector-based indexing, semantic search, and LLMs, offering structured, contextualized responses to enhance research and exploration in large corpora.

This study introduces SentiAnno, a sentiment-annotated, topic-specific corpus of Austrian historical newspapers (1700–1938). Focusing on the topics of migration and minorities, SentiAnno enables fine-tuning of LLMs for sentiment analysis and topic classification. Annotation processes, tools, and inter-annotator agreement are described, with the final corpus to be published on Zenodo, supporting FAIR principles.

FLAME, using Leave-N-Out grams, detects formulaic language in medieval charters despite variations in wording and structure. It overcomes limitations of traditional n-gram and skip-gram approaches by flexibly capturing long-range dependencies and identifying functional equivalence across diverse expressions. FLAME facilitates analysis of formulaic language evolution, revealing flexible patterns in legal language.

We use BERTopic to analyse themes in the bilingual speeches of the Carniolan Provincial Assembly. We examine common topics discussed in the sessions and how they change over time to gain insight into the key societal issues at the regional level in the Austro-Hungarian Empire at the turn of the 19th century.