The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #9619

Title:Invisible Spelling Error Detection and Correction Using Statistical Methods
Date: 1996
Abstract:Computer spellcheckers often work by comparing words in a text with the words held in their dictionary, and if a word is not found then reporting the closest dictionary alternative as a correction. However, if a writer mistakenly uses a real word in place of the intended word, for example, form instead of from, this dictionary check will not detect the error. This project builds on work already started by a previous student to create various spellcheckers which use statistical methods to detect these kinds of errors. Choice of training data and language models are discussed. The spellcheckers are tested and show promising results, with an accuracy of up to 73 percent.

[Search These Pages] [DAI Home Page] [Comment]