Hypernymy Relation in NLP: Tasks, Approaches, Resources, and Future Directions—A Systematic Literature Review

Hypernymy is a semantic relation between two terms, where a more specific term is entailed by a more general term—that is, the meaning of the more specific term is encompassed by the meaning of the more general term. This relation is crucial for many natural language processing (NLP) tasks, including textual entailment, search query expansion, and machine translation. In this systematic literature review, the hypernymy relation in the context of NLP is investigated, with a focus on identifying hypernymy-related tasks, employed approaches, available resources, and future research directions. The reviewed studies were extracted from five pre-defined databases, covering the period from 2018 to March 2023. The review process identified 75 primary studies that were analyzed to extract the targeted tasks, languages, approaches, representations, and datasets. The synthesized data were used to address the review questions. The review identifies the main hypernymy-related tasks, including hypernymy extraction, detection, directionality, graded lexical entailment, and discovery. The evaluation practices employed were summarized, including accuracy, F1, Mean Average Precision (MAP), and Spearman’s correlation coefficient. The targeted languages are highlighted, with English being the most studied; however, multilingual coverage is steadily growing. Several benchmark datasets for each task are presented, along with their statistics, characteristics, and construction techniques. Additionally, representation techniques are summarized, ranging from Word2Vec, FastText, and GloVe to hypernymy-specific representations. Finally, research gaps are discussed, and potential future directions are outlined. This review consolidates scattered findings and provides a practical map of tasks, resources, and techniques for researchers building hypernymy-aware NLP systems.