Evaluating Word2Vec and WordNet for Semantic Similarity at Myanmar Word and Sentence Levels

One of the most important NLP processes is word embedding, which converts words into numerical vectors for various NLP tasks. Although Word2Vec has been widely used for word embeddings in the Myanmar language, WordNet is a more recent approach. This study evaluates Word2Vec and WordNet in assessing semantic similarity at both the word and sentence levels in the Myanmar language. Word2Vec, which leverages a neural network, is effective at capturing contextual relationships and sentence-level similarity. In contrast, WordNet, a structured semantic database, excels at word-level similarity by leveraging its ontological structure, though it lacks contextual depth for sentences. To address WordNet's lack of support for Myanmar, text is translated into English using Google Translate. The analysis shows that WordNet performs better at word-level tasks requiring semantic structure, while Word2Vec excels in capturing sentence-level context, providing complementary strengths for Myanmar NLP tasks. These findings provide valuable insights for NLP tasks in Myanmar.