Spell Checker

Hindi Spell checking using Norvigs code

• Created the unigram frequencies of all hindi corpus with smoothing as in the norvig code of reduction of 1/n for every unseen word and pruning to remove the errorneous words. Then we created a edit-distance code for hindi letters.

  ---

• We ran the norvigs code on the above unigram data and then on bigram data for the given error set.
• The accuracy obtained for unigram was 44.57%
• The accuracy obtained for unigram was 36.36%

  ---

• We added sound similarity and the file we used as output.txt is part of the zipped data. On adding that to the norvig code in place we obtained the accuracy at : 46.8%(unigram) and with 38.67%(bigram), on the given dataset.
• We ran the given text file, hi-spell.txt on the aspell application and on the code we created for the top three suggestions and that is also a part of the zipped data.

  ---

• The code can be found here : here