Supervised and unsupervised automatic spelling correction algorithms

We present two algorithms for automatically improving the quality of texts which contain a large number of spelling errors. A supervised algorithm, which automatically corrects unknown words that are generated primarily from typing errors, is presented first. The second algorithm is an unsupervised approach to automatically correcting typing errors, individual words that have been split, multiple words which have been concatenated, and a combination of these errors. The algorithms have been developed and tested on a large source of real-world, human- and machine-generated spelling errors.