subs2vec: Word embeddings from subtitles in 55 languages