Enriching Language Models with Visually-grounded Word Vectors and the Lancaster Sensorimotor Norms