Enriching WordNet lexical database without using external resources

WordNet is a widely-used lexical database with significant implications for Natural Language Processing (NLP) tasks. Despite its popularity, some researchers have expressed concerns regarding WordNet's incompleteness and have sought to enrich the database to improve downstream NLP tasks. To address this issue, researchers have relied on external resources to supplement WordNet. In this paper, we have proposed a novel approach to developing WordNet without using any external resources. Instead, we have leveraged the text within each synset to inject new relations into the WordNet graph. The proposed method is evaluated through knowledge-based UKB word sense disambiguation (WSD), which utilizes the entire WordNet graph. The results indicate a significant increase in F۱-score after injecting a specific number of relations into the structural WordNet graph based on the similarity between pairs of synsets. These findings suggest that utilizing the text within synsets can be an effective way to enrich WordNet and improve the quality of NLP tasks without relying on external resources.

Keywords:

WordNet , Natural Language Processing , Embeddings , Similarity.