TIL eyewitness sketches aren't solely based on verbal descriptions. They are compiled using an "identikit", or a stock library of facial features (eyes, noses, etc.). The witness picks out the closest match for each feature which is then assembled into a more coherent image.