Learning word embeddings efficiently with noise-contrastive estimation