Quantifying the information in the long-range order of words: Semantic structures and universal linguistic constraints