From Fast Company:
Every time students take a writing exercise on Quill.org–a writing instruction platform for schools–their responses are logged by computers and analyzed for patterns. Algorithms take account of every false word they type, every misplaced comma, every inappropriate conjunction, deepening a sense of where the nation’s kids are succeeding in sentence-construction and where they need extra help.
The algorithms substitute for human intervention. Instead of teachers having to correct errors late at night with a red pen, the system does it automatically, suggesting corrections and concepts on its own. The goal, says Peter Gault, who founded Quill three years ago, is to reach more students than traditional teaching methods, including those who need support the most. About 400,000 students in 2,000 schools have used the (mostly free) writing-instruction platform so far.
. . . .
Kids today write all the time, perhaps more than previous generations. Whether it’s texts to their friends, or posting on Facebook, they’re constantly hitting the keys one way or another. But all this composition doesn’t necessarily make for better writing, at least not in the formal, academic sense. Just 24% of 8th- and 12th-grade students are “proficient” writers according to the Department of Education’s “The Nation’s Report Card: Writing 2011,” published in 2012. Teachers often complain they lack professional development to teach writing well. And, there’s a widespread acceptance in education circles that writing instruction is less developed and successful than, say, math or science teaching.
“Teachers just don’t have enough time in the day to offer feedback on everything students write, and that becomes a huge blocker to students moving forward,” Gault says in an interview. “Using machine learning to detect these patterns really unlocks a lot of options that allow us to bring this to thousands, or millions, of additional students in the coming years.”
The New York-based startup trains its algorithms with about 200 responses to each exercise, submitted by its programmers (it has about 300 exercises so far). As the students offer up thousands of their own responses, the code is then able to detect patterns without additional human intervention. When it prompts students to correct their sentences, it does so based on the collective trial-and-error of thousands of other users of the service.