Domain-oriented dialogue systems are often faced with users that try to cross the limits of their knowledge, by unawareness of its domain limitations or simply to test their capacity. These interactions are considered to be Out-Of-Domain and several strategies can be found in the literature to deal with some specific situations. Since if a certain input appears once, it has a non-zero probability of being entered later, the idea of taking advantage of real human interactions to feed these dialogue systems emerges, thus, naturally. In this paper, we introduce the SubTle Corpus, a corpus of Interaction-Response pairs extracted from subtitles files, created to help dialogue systems to deal with Out-of-Domain interactions.