FluTCHA: Using Fluency to Distinguish Humans from Computers

Authors

Kotaro Hara, Mohammad T. Hajiaghayi, Benjamin B. Bederson

Abstract

Improvements in image understanding technologies are making it possible for computers to pass traditional CAPTCHA tests with high probability. This suggests the need for new kinds of tasks that are easy to accomplish for humans but remain difficult for computers. In this paper, we introduce Fluency CAPTCHA (FluTCHA), a novel method to distinguish humans from computers using the fact that humans are better than machines at improving the fluency of sentences. We propose a way to let users work on FluTCHA tests and simultaneously complete useful linguistic tasks. Evaluation demonstrates the feasibility of using FluTCHA to distinguish humans from computers.

[pdf]