Computer spellcheckers and dyslexics—a performance survey

Word-processors are widely used by dyslexics but the spellcheckers are often not appropriate to their needs. This paper examines a sample of around 3000 words of running text containing almost 600 errors typical of those made by dyslexics. It first describes the types of error and the ways in which they differ from the more conventional misspellings that spellcheckers are designed to deal with. It then presents the results of running four current spellcheckers over this sample and finds that the correct word was suggested for only around half of the errors. There were two types of failure: errors that were flagged but not corrected and errors that were not flagged at all. The former often bore a phonetic resemblance to the intended word and in many cases looked correctable. The latter were caused by misspellings that resulted in another valid word. Spellcheckers need to use syntactic and contextual information to improve performance in this area. This would also enable them to make more appropriate suggestions for the errors that are flagged.