Toward an Architecture for Never-Ending Language Learning