1 nota · última: 13 May 2016
The essay builds a toy spell corrector from a language model, an error model, and edit-distance search, then tests it and shows why better context, data, and error estimates are needed.
