A Maximum Entropy Approach to Information Extraction from Semi-Structured and Free Text