Speakers
Description
This paper presents the theoretical and methodological foundations for an online dictionary of sixteenth-century Croatian. It utilizes a corpus of ten Gospel translations in Glagolitic, Cyrillic, and Latin scripts, en-compassing missals, lectionaries, Bible translations, and postils. This work is a primary objective of the project CroBiLex: Language Variation in Croatian Books of the Early Modern Period (Based on the Gospels Corpus) (Croatian Science Foundation, IP-2025-02-2567; 2026–2029) at the University of Zagreb’s Faculty of Humanities and Social Sciences. The paper outlines a conceptual lexicographic architecture for processing this linguistically heterogeneous, multi-script corpus. It emphasizes a Human-in-the-Loop approach where deterministic processing is complemented by machine-assisted transcription using handwritten text recognition and multimodal models. It also explores the experimental use of large language models (LLMs) for candidate lemma grouping, provisional semantic clustering, and preliminary entry-template preparation. Be-cause highly variable historical texts pose severe risks of historical inaccuracy and hallucination, LLM out-puts are treated strictly as non-authoritative proposals requiring expert verification. As CroBiLex is newly launched, this framework is conceptual rather than empirically evaluated. It seeks to balance computational assistance with scholarly accountability, providing a foundation for a sustainable digital model in historical Croatian lexicography.