Fast Exact Inference with a Factored Model for Natural Language Parsing