Distributed Representations of Mongolian Words and Its Efficient Estimation