The DBOX dialogue corpus was collected and annotated at the University of Saarland, in Saarbrücken, in the context of Eureka project 7152 “D-Box, A generic dialog box for multilingual conversational applications”. This project’s main goal is to develop and test an innovative architecture for conversational agents whose purpose is to support multilingual collaboration. The project develops interactive games based on spoken natural language human-computer dialogues, in three European languages: English, French and German. The first D-Box game scenario is a quiz game, in which a player may ask any type of question, such as “What are you famous for?”, in order to guess the name of a famous person. For this game situation, dialogues have been collected in a Wizard-of-Oz setup with a human Wizard who simulates the system’s behaviour by acting according to a pre-defined script. For further details see Petukhova et al. (2014).
Five annotated DBOX dialogues are included on the DialogBank. For more see the website of the D-Box project: see www.idiap.ch/project/d-box/
The annotations follow the ISO 24617-2 standard, making use of the possibility that this standard offers to add extra dimensions (with dimension-specific communicative functions); in particular, the dimension Contact Management has been inherited from the DIT++ annotation scheme, and an additional dimension called “Task Management” (also present in DAMSL) has been added for the annotation of utterances that discuss the rules of the quiz game. Moreover, three communicative functions are used that are not part of the ISO standard but that have been defined in DIT++, namely Dialogue Act Announcement (announcing the next dialogue act), Threat, and Pre-Closing (indicating the immanent closing of the dialogue), and the extra communicative function Congratulation has been added in order to account for those dialogue acts where a player is congratulated for correctly having guessed the identity of the famous person and thus having won the game.