Initial Results with Structured Queries and Language Models on Half a Terabyte of Text