An Evaluation of the Contextual Spelling Checker

Microsoft Office Word 2007 includes a “contextual spelling checker” that is intended to find misspellings that nonetheless form correctly spelled words. In an evaluation on 1400 examples, it is found to have high precision but low recall — that is, it fails to find most errors, but when it does flag a possible error, it is almost always correct. However, its performance in terms ofF is inferior to that of the trigrams-basedmethod ofMays, Damerau, and Mercer (1991). 1 Real-word spelling correction Most spelling checkers attempt to detect and correct only misspellings that result in a presumed nonword — a word that is not listed in the system’s dictionary — and they therefore cannot deal with an error that just happens to form a real word in the dictionary, albeit not the word that the user intended. There has been much research in recent years on methods for detecting and correcting such real-word errors or malapropisms (we use the two terms interchangeably); for a review, see Hirst and Budanitsky (2005) and Wilcox-O’Hearn et al (2008). The recently released Microsoft Office Word 2007 includes a “contextual spelling corrector” that attempts to detect and correct real-word errors (Microsoft 2006). A word that the system believes to be in error is flagged with a wavy blue underline, in contrast to Word’s regular red underline for non-word errors, and suggested corrections are available in a pop-up menu or in the ‘Spelling and Grammar’ window. This system operates not only on content words but also closed-class words (e.g., too and to). It can detect cases where a word has been wrongly split into two (e.g., through out for throughout), and missing or spurious apostrophes (e.g., corporations for corporation’s). It is not limited to a predefined set of frequently confounded words. Here we report an evaluation of this system that was carried out in order to compare it with the word-trigram method of Mays et al (1991) and the lexical cohesion method of Hirst and Budanitsky (2005). (A detailed evaluation of the Mays et al method and a comparison with the lexical cohesion method is given by Wilcox-O’Hearn et al (2008).) ∗This research was supported financially by the Natural Sciences and Engineering Research Council of Canada. I am grateful to Amber Wilcox-O’Hearn for comments and assistance.