Smoothing Techniques for Tree-k-Grammar-Based Natural Language Modeling