That would be nice, but at the moment the algo is at the mercy of the training dataset. The dataset doesn't distinguish adjectives from the ingredient, so this module doesn't :(