Effects of Semantic Gaps on Arabic WordNet-Based Similarity Measures

The EnglishWordNet (EnWN) is extensively used by computational linguists as a lexical-semantic resource for Natural Language Processing and Knowledge Engineering. EnWN was extended with multilingual information in several non-English wordnet projects; including the Arabic WordNet (ArWN). ArWN is still limited in terms of semantic structure and coverage of the Arabic language. This paper focuses on the improvement of ArWN semantic structure, the Hypernymy relation. An interactive cross-lingual mapping approach is used to enrich ArWN with new synsets lexicalized in Arabic and mapped to the target synsets in EnWN. The paper also studies to which extent the richness of the semantic structure of the wordnets affects the semantic evidences, which can be utilized in wordnet-based applications. A set of experiments in the context of semantic similarity is conducted on Arabic word pairs dataset to examine the effectiveness and usability of the enriched ArWN resource. The experimental performance measures showed that filling in the semantic gaps significantly improves the quality of the similarity scores between concepts.