Carniolan Provincial Assembly

Historical parliamentary corpora offer crucial evidence for studying political discourse over time, yet their usability is often limited by poor OCR quality and incomplete metadata. This paper presents the enhancement of the Kranjska 1.0 corpus, a collection of Carniolan Provincial Assembly proceedings (1861–1913) in Slovenian and German, through a two-phase process aimed at improving textual accuracy and enriching speaker metadata. First, we conducted a manual correction campaign on a representative sample of transcripts, involving trained historians proficient in Gothic script and 19th-century politics. The corrections addressed both structural and textual errors in TEI-encoded XML files, providing a gold-standard dataset for future model training. Error analysis revealed recurring OCR issues, including segmentation problems, misattributed speakers, and systematic character-level noise. Second, we harmonised and expanded speaker metadata using multiple historical sources to unify name variants, resolve ambiguities, and document parliamentary terms, factions, and attendance. The resulting metadata enhance corpus usability and interpretability. This work lays the foundation for the next project phase, which explores the automatic correction of transcripts and metadata using Multimodal Large Language Models (MLLMs). By combining historical expertise with computational methods, we contribute to more accurate processing of historical texts and promote transparency and reusability in digital humanities research.