From Spelling Correction to Text Cleaning - Using Context Information

Spelling correction is the task of correcting words in texts. Most of the available spelling correction tools only work on isolated words and compute a list of spelling suggestions ranked by edit-distance, letter-n-gram similarity or comparable measures. Although the probability of the best ranked suggestion being correct in the current context is high, user intervention is usually necessary to choose the most appropriate suggestion (Kukich, 1992). Based on preliminary work by Sabsch (2006), we developed an efficient context sensitive spelling correction system dcClean by combining two approaches: the edit distance based ranking of an open source spelling corrector and neighbour cooccurrence statistics computed from a domain specific corpus. In combination with domain specific replacement and abbreviation lists we are able to significantly improve the correction precision compared to edit distance or context based spelling correctors applied on their own.