A quick idea to improve coverage using word vectors, with my NLP library spaCy ( http://spacy.io http://spacy.io ): https://gist.github.com/honnibal/664cf54cb3

A quick idea to improve coverage using word vectors, with my NLP library spaCy ( http://spacy.io http://spacy.io ): https://gist.github.com/honnibal/664cf54cb30c48c0c977 https://gist.github.com/honnibal/664cf54cb30c48c0c977 Example output: aquamarine umber turquoise red-orange coppery malachite ochre russet blue-violet fuchsia butterscotch lilac tortoiseshell mother-of-pearl I thought it'd be fun to see whether I could get the shade of the color from the vectors, but so far no luck. Probably vectors trained for image captioning would work, though.