Integrating Croatian into Concepticon: a corpus-based frequency mapping of Croatian vocabulary

This thesis examines how a corpus-derived, frequency-based Croatian wordlist can be mapped to Concepticon concept sets, and what this reveals about the relationship between usage-based lexical evidence and concept-based infrastructures for cross-linguistic comparison. Concepticon is treated as a linking resource that provides stable, language-independent identifiers rather than as a language-specific lexicon. The study investigates what the resulting pattern of mapped and unmapped Croatian lemmas suggests about the compatibility of frequency-first and concept-based approaches, and what kinds of structured information the mapped subset offers for multilingual comparison and potential learner-oriented reuse. The dataset was compiled from the hrWaC web corpus by extracting the 1,000 most frequent nouns, verbs, and adjectives. After manual cleaning and English translation, the final list contained 2,668 items. Mapping combined pyconcepticon with manual verification. Of these, 1,262 items (47.3%) mapped to at least one Concepticon concept set, while 1,406 (52.7%) did not. Coverage varied by word class, with verbs mapping best and adjectives worst. Unmapped items show systematic patterns, especially derivational or part-of-speech mismatches, relational morphology, and frequent web-corpus vocabulary from contemporary institutional and public-life domains (e.g., politics, technology, media). Within the mapped subset, many-to-one mappings are common, and most mapped concepts are also represented in at least one additional Concepticon language. Overall, the results suggest that Concepticon captures broadly stable conceptual meanings well, but provides weaker coverage for institution-related and morphologically derived vocabulary typical of modern usage.