[wtracy]

Most of the ones I got looked like they were generated from replies to other tweets. The lack of context made them boring, even if that makes it easier to make them comprehensible. (I would love to see this experiment run on a dataset that filtered out reply tweets.)

But then, it have me this gem:

"It's sad when people don't understand the value of a program that targets them. I mean I do see the value, I just don't like floppy pizza."