While adversarial games have been well studied in various board games andelectronic sports games, etc., such adversarial games remain a nearly blankfield in natural language processing. As natural language is inherently aninteractive game, we propose a challenging pragmatics game called AdversarialTaboo, in which an attacker and a defender compete with each other throughsequential natural language interactions. The attacker is tasked with inducingthe defender to speak a target word invisible to the defender, while thedefender is tasked with detecting the target word before being induced by theattacker. In Adversarial Taboo, a successful attacker must hide its intentionand subtly induce the defender, while a competitive defender must be cautiouswith its utterances and infer the intention of the attacker. To instantiate thegame, we create a game environment and a competition platform. Sufficient pilotexperiments and empirical studies on several baseline attack and defensestrategies show promising and interesting results. Based on the analysis on thegame and experiments, we discuss multiple promising directions for futureresearch.