Skip to main content

Spelling out Microsoft Word’s weakness

There have been a number of add-ons to Microsoft Word that attempt to refine its editing capabilities, but they merely suggest that Microsoft's work is not done when it comes to improving the program.

You know the state of the word processor is pathetic when I can get away with writing the following sentence:

I think it is goo that I kin write a sentence like this and no spell checker kin manager to fund anywhere wrong with it.

This is a disaster. There are wrong word choices, spelling errors, and grammatical gaffs, yet Word detects nothing. No red squiggles, no green squiggles, nothing. The software sees this and says: "It's good to go!"

Check the calendar, people. It is 2012 – almost 2013, in fact. Word processing on the desktop has been around for almost forty years and spell check has been around for almost as long.

If the above sentence can still survive as is, something needs to change. We need computerised contextual proofing. Everything is lined up to make this happen, if only someone would create the database in the sky to do it.

What I'm about to describe would be the Manhattan Project of proofreading. But with everyone going to the cloud for intense cross-referencing, there is no reason it cannot be done.

Right now, Autodesk is developing a cloud process that will allow designers to do complex calculations in the cloud, returning the results to the desktop for display or further tweaking. The company believes that the cloud can be used for world-class number crunching and I agree.

It is this world-class number crunching that will be required for the kind of re-engineering of the word processor based on a massive database of the written language. Both Google and Microsoft have the resources to do this and given the magnitude of language poured into their search engines every day, it would be easy to make a contextual analysis tool.

In the unacceptable example above, a contextual analysis tool would look at the relationship between the words. It would, for example, read "fund anywhere" and determine if this is valid in the wild. Does it work in context? Yes, it does; there is a possibility that "fund anywhere" could be used. But then it would consider "fund anywhere wrong." This is an unlikely combination so it should know something is amiss.

Google already has a "Did you mean" feature that could attack this sentence to determine that "fund" could variously be "fund," "find," "fend," or "fond," among other twists such as "font” or even "gun."

With so much crowd-sourced data out there, network analysis tools should be able to examine the whole structure and the meaning of the phrase and actually turn out "find anything wrong."

It's true that few people would unknowingly write such an error-ridden sentence, but this mess could easily result from voice recognition software.

If this can be done – and it eventually must and will be done – then the next step is language translation. But I'll save that hybrid problem for another day and another column.