The authors of word2vec (Mikolov et al.) extended it to create fastText at Facebook. It works on character n-grams instead of entire words. Character n-grams are effective in languages with specific morphological properties.
We can create our own fastText embeddings, which can handle OOV tokens as well.