We don't have words like v1agra. As I mentioned in the README, we took vectors pretrained on wikipedia. One of the possible improvements can be to train the vectors on our own dataset.