Facebook Twitter Google Digg LinkedIn StumbleUpon Email
How the learning algorithms work, and what it can teach you about your writing strategy.

By Ben Feuer

Automated writing assessment (as a supplement to human writing assessment) has been used for years now in undergraduate university settings.  It is used by EdX MOOCs.  But how exactly does it work?

Those familiar with the technology will know exactly how it works -- it is based on machine learning algorithms, which consists of a computer studying human behavior and turning key components of it into patterns the computer can then recognize and spit out at will.  The more patterns the computer spots, the better it gets at reproducing the 'model' it has in its virtual head.  A much more detailed (and fascinating) explanation of the software can be found here.

All well and good in theory, but how does this affect how you take your GMAT?

The machine learning algorithm that governs computerized essay grading is going to focus on errors in grammar, passive use cases, ability to adhere to a central message or theme and the ability to support that message.  Obviously, in a context independent of human readers, it's easy to game that system, as MIT professor Les Perelman has been demonstrating for years.  But the GMAT uses the machines as a second reader, and (very hurried) human beings as first readers.

So the optimal strategy for the GMAT AWA would be to write a technically perfect essay that answers the central question well enough not to offend the sensibilities of the human 'first reader'.  Most likely, if the essay passes a sanity check and contains a few worthwhile insights, the human reader isn't going to give the essay a hard time, and the machine will recognize what it is trained to recognize -- good grammar and sentence structure and staying on topic.


Don't be shy! Schedule a consultation to find out how we can help you.