Sending emails can be a struggle, especially if it's too a professor. It's difficult to tell if you have the tone right, and I've found myself often asking my friends asking for feedback on email drafts.
What it does
MaiLing (Mail + Linguistics) uses computational sociolinguistic findings from Cornell and Stanford to both gauge how polite you are as well as to offer useful feedback on how to increase politeness and prevent conversations from going awry.
How we built it
We used the conversational analysis toolkit from Cornell CS to build an API that provides insights to politeness of text, using a variety of indicators. Then, we built a front-end Gmail plugin which presents those actionable results to the user.
Challenges we ran into
The politeness classifier was built with Python 2, then ported to Python 3, then revised for Python 2, so we had to move some code around in order to get automatic dependency parsing working for our Python 3 application. Then, because the model was outdated, the results tended to skew impolite. We didn't have time to retrain the model, so we had to find a way to rescale the output back to something more reasonable.
What we learned
A lot about language and social interaction.