Chris Brockett and William B. Dolan
The problem of semantic equivalence, or paraphrase, is a fundamental one for applications that "understand" natural language. Learned approaches to this problem face a lack of colloquial training data with which to build models. This paper describes, Echo Chamber, a game aimed at collecting sentential paraphrases from web users. Much of our current focus is designing a framework that makes this potentially burdensome task engaging and challenging. The game draws on elements of enduring pen-and-paper games such as Battleship and Hangman, and incorporates a time component to impart a sense of urgency. In the final version, it is intended that automated validation of input will ensure that the game is scalable and can collect high quality data without editorial intervention. In this paper, we discuss design considerations and issues emerging from the prototype of this game.