The STAC corpus contains 45 games involving three or more players. Each game is divided into dialogues, or sub-sections of the game that encompass one or more bargaining sessions. In each dialogue, individual discourse units form a weakly-connected graph, with the exception of those linguistic graphs that contain non-dialogue-initial chat moves which could not be connected to a previous chat move.


Game information in the form of Server/UI messages (EEUs) was added to the linguistic corpus, which contained only chat moves (EDUs), in order to create the situated version of the corpus. This changed the distribution of segments among the dialogues and the number of dialogues per game: in some cases the EDUs within one linguistic dialogue cannot be found in just one single corresponding situated dialogue. And, contrarily, in some other cases, a situated dialogue may contain the EDUs from multiple linguistic dialogues. In order to allow comparisons of EDUs in both corpus versions, we collected dialogues into "superdocs", or sets of dialogues which each EDU appears in both the linguistic and the situated context.

Ex. A linguistic structure which is preserved in the situated dialogue.


Typically, an EDU or EEU attaches to another discourse unit as the argument of a rhetorical relation. But in SDRT, it can also be added to a group of discourse units that collectively provide the argument to a discourse relation. This group of discourse units forms what is called a complex discourse unit or CDU. The CDUs are depicted in red, and they are attached to their constituents with a grey dotted line.

Note: When hovering over an EEU or EDU node, the turn number, player name and text appear. When hovering over a CDU, the global id of the CDU appears. (For further description of the dataset, see Modelling Structures for Situated Discourse.)
 Clarification Question
 Question-Answer Pair