SOAP

IRIS Wiki - IS Systems - SOAP

Availability

Currently the system is not directly available for download, but parts of the system architecture , such as the Virtual Beergarden[1] and SceneMaker[2] are available for download. The developers are working on a version which is available for download. The tool in the recent finished version is available on request by first contacting the developers. Please contact the main author Birgit Endrass [3] for more information.

Technical Description

In the soap-like story, the user can interact with a set of virtual characters. Through dialog interactions, the progress and outcome of the story can be influenced. For the realization of the SOAP scenario, several components were needed such as language understanding that parses the user's text input into abstract dialog utterances, a dialog model that controls the narrative flow of the story and a graphical representation holding the virtual characters as well as the user avatar.

Result Description (end user perspective)

In the scenario, the characters are involved in a romantic conflict. The user, who is represented by an avatar can approach the different characters, listen to their conversations and interact with them. In that manner, the user can either approach a group of girls, a group of guys, or a waitress that is working in a Virtual Beergarden. In order to save resources, dialogs take place only if the user is near the characters, otherwise idle dialog behavior is presented. Through observation and interaction, the user will learn that there is a love story secretly going on. Dependent on the user's interactions, the characters will reveal their love, ask for help and follow the user's advice. Different scenario endings were modeled, while the user can make a match between two persons of his choice.

Utterances of all parties in the conversation can influence parameter states in the game, two of which are the most important: The ‘Killer Phrase Level’ as kind of a stress level of the whole debate, and an ‘Agreement Level’. In the best case, the player can influence the stress level in a negative way so that the chances for constructive arguments are higher. Certain arguments increase the ‘Agreement Level’, and the game can end successfully for the moderator. In the worst case, the increasing ‘Killer Phrase Level’ reaches a threshold value that makes the discussion end with an escalation.

Authoring Description

Authoring of the conversations is supported by a graphical interface, and can also be accomplished by directly writing AIML code (an XML-dialect, see www.alicebot.org). Single utterances are defined on 2 levels: an abstract dialogue act, such as "PARKING COMPLAINT", and a concrete wording of the utterance, such as "You should try coming home in the afternoons, you'd be hard pressed to find a parking space." Possible abstract dialogue acts of one character (bot) are then arranged to form either sequences or reaction possibilities to utterances coming from either another bot or the user. In order to accomplish sequences of turn-taking between 2 bots, the output of one utterance of Bot A has to be input as a pattern for another utterance of Bot B.

Dialogue sequences / utterances are groups within dialogue graphs, which are associated to an actor in a scene. Several scenes can be combined as a high-level plot graph, defining the overall structure of the conversation.

Strong Points

Scenejo enables the definition of multi-partner conversations with high-frequency interaction possibilities for users. A strong point is its accessibility for authors stemming from non-programming areas, who can quickly experiment with creating small conversations that can be directly experienced, without knowledge in programming. Therefore, it is well suited for people making their first steps in experiencing typical authoring issues in Interactive Storytelling.

Limitations

Scenejo is currently limited to the design of verbal interaction with "talking heads" by typing text-chat. As it is based on the ALICE chatbot philosophy, combined with a low-level dialogue manager, recognition of user utterances is limited to author-defined patterns, the design of which is challenging.

Main Publications

• Birgit Endrass, Christoph Klimmt, Gregor Mehlmann, Elisabeth André, and Christian Roth, Exploration of User Reactions to Different Dialog-based Interaction Style, Proceedings of the 4th International Conference on Interactive Digital Storytelling (ICIDS 2011), 2011
  
• Gregor Mehlmann, Birgit Endrass, Elisabeth André, Modeling and Interpretation of Multithreaded and Multimodal Dialogue, Proceedings of the 13th International Conference on Multimodal Interaction (ICMI 2011), 2011.

Supporting Narrative Theories

The tool was designed without any background of narrative theories.

Computational Model

• Virtual Beergarden: The graphical representation is done in the Virtual Beergarden application which holds the characters and the virtual scenario.
  
• SceneMaker: Dialog modeling and the flow of the scenario is done using the SceneMaker tool.
  
• SPIN: The user's typed text input is parsed into abstract dialog utterances using the semantic parser SPIN.
  

Type of Interaction

• Interaction devices: keyboard, mouse
• Input modalities: typed-text
• Output modalities: text-to-speech & nonverbal behavior of virtual characters