Individuation in Neural Models with and without Visual Grounding