A distributional perspective on the gavagai problem in early word learning