A method and a system for interpreting information in a document are provided, with the system receiving an image of a document from a remote source and converting it into multiple sets of blocks of characters. Tags indicating likely meaning of blocks are assigned to them. At least some of the blocks have an associated score representing the probability that the characters in the block correctly represent the characters in the original image. The system selects one set from multiple sets based on the scores associated to certain blocks determined by accessing remote information over the Internet.