Spelling checkers, spelling correctors and the misspellings of poor spellers

摘要

A large corpus of spelling errors taken from free writing is analyzed to assess how great a challenge such errors present for automatic checking and correction. The analysis reveals a high proportion of errors that match dictionary words; these would necessitate the use of context in error detection. Some of these errors are caused by incorrect word-division, a type of error difficult to spot since it calls into question the placing of word boundaries. Misspellings tend to differ from the correct words more than mistypings do. Some knowledge of pronunciation would help in correcting many of the errors, but misspellings do not always reflect pronunciation in a simple way.