Has anything been done in this regard?

I am looking for research that has been conducted for spell checking of Romanized Devanagari text.

I suppose well researched English spell checking algorithms would fail in this scenario as they would not work in this context. Or are they similar?

As an example correction for user input Aallu would be Aalu or for user input Naafrat would be Nafrat.

I am sorry if this question is too vague but I am posting it here after lots of google searches without any lead.

  • You haven't mentioned the fact that Romanized Devanagari doesn't have standard spellings at all. That would make spell-checking really hard, I think. (Also, I'm wondering who would ever need a Romanized Hindi spellchecker; apart from corporate slogans, it's only ever used in informal contexts, where correct spelling isn't usually expected.)
    – Anubhav C
    May 14, 2014 at 14:25
  • Yes there is that too. System users will be not proficient in English so would use Roman to interact with the system. Its in Nepali, we use Devanagari too :)
    – meadhikari
    May 14, 2014 at 14:47
  • Are you really looking for a spellchecker, or just something that converts Romanized Nepali → Devanagari Nepali? If it's the latter, Google Input Tools does that already. (Try downloading the Nepali version.)
    – Anubhav C
    May 14, 2014 at 14:59

1 Answer 1


Modern spelling correction algorithms work independent of language. However, their statistical models have to be trained for each individual target language. This blog post by Peter Norvig is a good place to start exploring.

  • 2
    Well yes and no. The older isepll was found not to work too well for all scenarios so newer tools were build to handle peculiarities of Hebrew and Hungarian. I believe the Hungarian one went on to become the new default spell tool, taking the position of ispell. It originally had an identical API to make it easy to switch in, I believe. May 13, 2014 at 2:02
  • 1
    yep, most Linux distributions come today preloaded with hunspell.sourceforge.net rather than ispell. It's mostly backwards compatible with ispell plus it can better handle agglutinative languages like Magyar and compound words in German - so I guess it'd be ideal for Sanskrit or something based off of it.
    – Joe Pineda
    May 13, 2014 at 3:15
  • 1
    @hippietrail: Thanks for the historical note. I have amended my answer.
    – prash
    May 13, 2014 at 9:51
  • @JoePineda: Sanskrit isn't agglutinative to the best of my knowledge. All of the Dravidian languages are though. Sanskrit might have compound words like German though, I'm not sure, but it does seem long words occur in all languages of India... May 14, 2014 at 4:30
  • 1
    @hippietrail: Sanskrit is much werse than German when it comes to compounding. See this blog post for an example. (Found via this Ling SE answer).
    – Anubhav C
    May 14, 2014 at 14:20

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.