Dictionary-based Debiasing of Pre-trained Word Embeddings