Towards a Gold Standard for Adjectival Hypernymy: Enriching the Open English WordNet with a Hybrid Approach

Adjectival hypernymy is an underexplored lexical-semantic relation essential for Natural Language Processing (NLP) and hierarchical semantic organization of the lexicon. While hypernymy in nouns and verbs has been extensively modeled in resources such as WordNet, adjectives remain largely unstructured due to their gradability and context-dependence. We present a hybrid Large Language Model (LLM)-Human approach towards the creation of a gold-standard dataset for adjectival hypernymy. Our method integrates three LLMs with systematic human evaluation, guided by a specifically developed theoretical framework ensuring consistency and linguistically-based principles, compiling a resource of 3,836 validated adjective hyponym-hypernym pairs. Results demonstrate high precision for consensus predictions (87%), confirming the utility of cross-model agreement as a proxy for semantic validity. This method highlights how LLMs can complement human effort and expertise to support the construction of lexical resources. The resulting dataset aims to enrich the Open English WordNet (OEWN) with explicit adjectival hierarchies and serves as a benchmark for hypernymy detection and lexical entailment evaluation.