You don't need libraries to train word vectors in all languages, you can just load precomputed vectors. In order to measure similarity between two vectors you only need 3 lines of code (it's a simple sum of products for a couple hundred real values).