Hacker News Reader

What about sending them fake bank account info after that point? You could probably get them to try 3 or 4 different accounts etc if you kept saying that there were typos, wrong bank, etc.

Could human input help? How about a site that lets visitors answer questions that are then used to improve the responses to spammers? The result could be smarter than AI and more efficient than humans.

Social or game aspects could make it interesting for users.

a one day project related to natural language quantification?

You might have more luck if you try to datamine the incoming emails, they start to read your replies after a couple of mails.