機械翻訳の違和感を用いたCAPTCHAの提案

摘要

近年，既存のCAPTCHAにおける脆弱性が多くの研究者によって指摘されており，人間の「より高度な認知処理」に基づいた新たなCAPTCHAの導入が強く望まれている．そこで著者らは，「違和感の判別」をチューリングテストに用いることで，人間には容易で機械には困難な新しいCAPTCHAを提案する．本稿では特に，機械翻訳により生成された文章が有する違和感に注目し，機械卸訳された文章と人間が作った文章との切り分けを利用したCAPTCHAについて検討する．%Captcha is a technique to prevent automatic programs from being able to acquire free Email or online service accounts. However, as many researchers have reported, the conventional CAPTCHA could be defeated by recent malwares since the ability of PCs get closer to that of human. Therefore CAPTCHA should be based on an even more advanced human cognitive processing ability. In this study, to realize a new CAPTCHA, we propose to use a human ability to recognize "strangeness". As an example, this paper focuses on strangeness in machine translated sentences, and proposes a CAPTCHA which detects malwares by checking if a user can distinguish natural sentences created by human from machine translated sentences.