A vision-grounded dataset for predicting typical locations for verbs