A companion robot that can tell stories Carole Adam1 and Lawrence Cavedon2 1
Grenoble Informatics Laboratory - Joseph Fourier University, Grenoble, France
[email protected] 2 RMIT University, Melbourne, Australia
[email protected]
Keywords: Interactive Storytelling, artificial companion, personalisation, engagement, believability. Introduction. Telling engaging stories is an interesting ability for an artificial companion for children. Three features can be made engaging: the agent itself (which is addressed by the field of Embodied Conversational Agents); the content of the story (e.g. interactive stories [1]); and the way the story is narrated. We designed and implemented engaging narrative strategies for Reeti, an affective expressive robot with a wide range of emotional facial expressions. Strategies. We used: a corpus of parent-child interactions [2] showing a tendency to use personalisation to engage children; guidebooks for human storytellers [3] highlighting the importance of interactivity and tailoring the story to the audience; and literature in HCI [4] and relational agents [5] revealing specific requirements such as customisation, interactivity, and user control. We thus identified the following engaging narrative strategies for Reeti: – Embody the different characters by changing voice, for livelier narration; – Adapt vocabulary to the child’s age, use simpler synonyms or definitions; – Show emotional intelligence: express emotions consistent with the story, and detect and react to the child’s emotions triggered by the story; – Make random changes in the text of the story to avoid boredom; – Make personal comments relating story to child’s profile and context; – Offer to play interactive games to favour engagement (quiz, guess...): – Offer multiple choices at some points of the story to give a feeling of agency; – Insert relevant diversions (jokes, anecdotes...) to prevent boredom; – Refrain from interrupting the story to focus on key moments (immersion); SMILE language. To perform these, the storyteller needs two types of information: about the user profile and context (already available to companions); and about the story, triggers for strategies and additional scripted content (provided as story annotations with our SMILE language [6]). For example the annotated snippet below tells the storyteller that wolf is an emotional word, and provides scripted comments to react to two emotions (as deduced from user profile). When
Little
Red Cap
Lucky
arrived in there
are
the no
woods, she wolves
met the
around
here
wolf right?
You like scary animals don’t you? . But she did not know it was a nasty animal and was not afraid.
Implementation. We implemented several modules in Java: a SMILE parser for the annotated stories; a GUI using Google speech recognition and/or text input to let the user interact with the robot during the narration; and a basic storytelling engine for Reeti, with only the ”change of voice” strategy so far. Pilot studies. To inform the implementation of our storytelling module in the Reeti expressive communicating robot, we conduced two pilot studies. We first had 22 visitors at the Innorobo robotic show (Lyon, France, March 2013) play a game with a robot and rate the acceptability of robots in different hypothetical roles with a child. Users explicitly stated that they could accept a robot only as a complement but not as a substitute; they would not trust the robot with responsibilities; and they were reluctant to letting it create a relationship with their child. The physical appearance of the robot was found to influence its perceived credibility in its role (e.g. too small to have authority). The storyteller and playing buddy roles were both considered as very acceptable. We later had 25 students and staff at Grenoble Informatics Laboratory rate our list of strategies on two criteria: believability (likeliness that a human storyteller would use it) and engagingness (efficiency to captivate a child). The users insisted on the importance of interactivity, in particular the storyteller’s ability to understand and answer the child’s questions, but also to itself ask questions about the child’s opinions and feelings. They found most strategies engaging, even when not human-like, except for changing the story (undesirable to modify the author’s work) and forcing focus (harsh to not let the user in control). Conclusion. The aim of our approach is to make it possible for an artificial companion to use strategies to modify a story (or another text), in order to really personalise its narration, not to a group or category of users, but to one specific user that it gets to know over time. More details can be found in [7].
References 1. Cavazza, M., Donokian, S., eds.: International Conference on Virtual Storytelling (ICVS). Volume 4871 of LNCS., Springer (2007) 2. Adam, C., Cavedon, L., Padgham, L.: ”Hello Emily” - Personalised dialogue in a toy to engage children. In: Companionable Dialogue Systems, ACL (2010) 3. Hostmeyer, P., Kinsella, M.A.: Storytelling & QAR Strategies. Libr. Ultd (2011) 4. Brandtzaeg, P.B., Folstad, A., Heim, J.: Enjoyment: Lessons from karasek. In: Funology - From Usability to Enjoyment. Volume 3 of HCI., Springer (2005) 55–65 5. Bickmore, T.W., Picard, R.W.: Establishing and maintaining long-term humancomputer relationships. ACM Trans. on Comp.-Human Interactions 12(2) (2005) 6. Adam, C.: Il ´etait une fois... un robot compagnon qui racontait des histoires. In: WACAI. Volume RR-LIG-039 of LIG research reports., LIG (2013) 7. Adam, C., Cavedon, L.: Once upon a time... a companion robot than can tell stories. Technical Report RR-LIG-??, LIG, Grenoble, France (2013)