Lately I've been spending quite a lot of time thinking about spellcheckers within the context of OLPC and Sugar. It initially started with my dissatisfaction that the way Sugar's Write Activity is configured makes the spellchecker it comes with pretty much useless. (Read this article if you're interested in learning how to hack the Write Activity and make the spellchecker useful.)

Only a week later I had an interesting conversation with a teacher who had just gotten back from spending 6 weeks starting a small OLPC project in Honduras and again the topic of the spellchecker came up. It made me realize that the current way spellchecking is implemented in the word processors I'm aware of isn't necessarily all that useful in a school or learning setting.

Good for business, bad for school

As a simple example I want to look at myself and how I haven't learned to correctly spell a number of words (endeviour, procastination, and independance among others) because using the spellchecker in my e-mail client (Thunderbird) and word processors (Word and Open Office Writer) is easier than actually remembering the correct way to spell them.

Similarly think back to your own days in school and how you learned how to write your mother tongue or any subsequent language. Particularly during exams I normally wrote a text, then went back to carefully read it for spelling and grammar errors and, if I had the time, finally reading the text backwards word by word in order to reduce the distraction of being focused on the content of the text.

I then handed in the exam, it was corrected by a teacher and in the end I had to re-write all the sentences where I had made mistakes. Similarly we were sometimes also encouraged to keep a list of words that we consistently misspelled and going through them before exams or while re-reading homework was certainly useful for me.

I know that at this point some people are going to scream bloody murder about antique teaching and learning styles. But you know what, I think it worked just fine for me. And I guess I'm not the only one.

Now the problem I see with existing word processors and their spellcheckers is that they're focused on reducing the risk of sending out texts with misspelled words to other people (which could be embarrassing) rather than teaching people how to spell correctly. Auto-correction is of course the worst-case example here because it teaches you that consistently misspelling something really isn't a problem.

As mentioned in the beginning having misspelled words underlined immediately without allowing users to access potentially correct suggestions is also barely a foul compromise. Adding the traditional suggestions is practical in many ways and will be very helpful for a lot of people but on the other hand lazy learners like me might not really benefit from it.

Hence I suggest that the concept of a spellchecker within a learning environment such as Sugar really has to be rethought. I don't have a clear picture yet what such a new spellchecker (I resist the temptation to call is spellchecker 2.0) would look like but let me offer some initial suggestions.

Reading one word at a time

First of all I think that collecting words that are commonly misspelled by a user could be a good basis for subsequent methods and tools to learn how they're actually spelled. Imagine a first or second language vocabulary learning Activity that specifically made you deal with words you repeatedly misspelled in the Write Activity.

Secondly being able to easily focus on single words rather than phrases or complete sentences should be enabled via a special view of a text. It's easy to imagine a black or white overlay that masks everything except for a single word on the screen that's shown for x-number of seconds,

Last but not least it would be great to have running a spellchecker result in a text that actually looked like it had been corrected by someone who knows the language. So rather than being presented with a perfectly spelled text that's ready to be sent to a business partner the spell-checked text should include some sort of markup that consists of the erroneous original as well as the correct spelling.

Again, a realize that I'm barely scratching the surface of what a learning-focused spellchecker could really look like. However I'm strongly convinced that the current ones aren't up to the task so we really need to rethink them.