Interpretability of Hungarian embedding spaces using a knowledge base

While word embeddings have proven to be highly useful in many NLP tasks, they are difficult to interpret for humans. Sparse word embeddings are reminiscent of knowledge bases containing words that are already characterized in sparse forms. In our work, we investigate to what extent sparse word representations convey knowledge about the words in knowledge bases. We utilize Hungarian sparse word embeddings and ConceptNet, a knowledge base that supports Hungarian.