Abstract:
Spelling error detection & correction techniques are used widely by word processing, machine
translation, information retrieval and natural language processing systems. Even though it is straight
forward to verify a misspelled word by looking up in a word dictionary, it is very hard to suggest the
best correction. For a morphologically rich and a complex Indic language like Sinhala a probabilistic
method is the best approach for qualifying the best correction for a detected misspelled word over the
other existing methods. This research intends to identify & analyze non-word spelling error patterns in
Sinhala. A word dictionary will be used to identify the errors and a special software tool will be
developed in order to record statistical data regarding the spelling errors of Sinhala documents. This
tool will be used by a Sinhala language expert to record data related to spelling errors in a selected
sample of documents. Errors will be categorized into different types along with statistical results and
will be analyzed. The reasons of language specific error patterns will be discussed and a weight based
decision tree format will be derived as an outcome which can be used to find the best correction from
a word dictionary to replace a misspelled word.
