how to compute similarity of multi word tokens in pre-trained GLoVE word embeddings? Example: similarity("new york","delhi") OR similarity("united states of america","united kingdom")??

—