OCR post-processing for low density languages

Abstract

We present a lexicon-free post-processing method for optical character recognition (OCR), implemented using weighted finite state machines. We evaluate the technique in a number of scenarios relevant for natural language processing, including creation of new OCR capabilities for low density languages, improvement of OCR performance for a native commercial system, acquisition of knowledge from a foreign-language dictionary, creation of a parallel text, and machine translation from OCR output.