Page 115 - Socially Intelligent Agents Creating Relationships with Computers and Robots
P. 115

98                                             Socially Intelligent Agents

                             ordinary knowledge about liquor, and they are not members of the mailing list.
                             Each subject was shown the VEs’ conversation about brandy for three minutes
                             and could comment on a point made by a VE when s/he feels an interest in the
                             speech anytime.
                               The following is part of a conversation between VEs A, B, C and D and a

                             subject .
                             A: “Let’s talk about brandy.”
                             B: “We have more prejudice against brandy than other liquor.”
                             A: “I often chat over drinking.”
                             C: “A lot of bars don’t serve brandy because it is too expensive.”
                             B: “Yes, certainly.”
                             A: “Sure.”
                             Subject: “Convenience stores don’t deal in brandy either.”

                             C: “Shouchu is cheaper than brandy.”
                               In the conversation, line 4 follows line 3 mainly because of the coherence
                             criterion for storytelling. Both messages have been humanly labeled with the
                             same key word “shouchu” in advance since these two messages originally ap-
                             peared in the context of the advantages of shouchu. And the short responses
                             “Yes, certainly” and “Sure” are randomly inserted in place of other messages
                             from the personal memories to make the conversation rhythmic.
                               After the conversation, the subjects were asked whether they thought it was
                             easier to comment on the messages of other persons using text-based mailing
                             lists or the EgoChat system.

                             4.2     Results and Discussion
                               Two subjects answered that the EgoChat system was easier and more casual
                             than a mailing list because a chat-style conversation with voices facilitated
                             their interaction with VEs. This suggests that using storytelling agents with
                             voices may facilitate interaction between humans and agents. For our previous
                             system, CoMeMo-Community, we evaluated to what extent people shared their
                             knowledgebythevirtualexchangeofmessagesbytext-basedwordsandimages.
                             The results for CoMeMo-Community were satisfactory [4]. Hence, we expect
                             that the EgoChat, owing to its use of more human-like modalities such as voice
                             and gestures, will bring community members more knowledge than CoMeMo-
                             Community.
                               One subject answered that interacting with VEs is too unnatural for smooth
                             communication because VEs can’t answer to his questions. Certainly, the sto-
                             rytelling method on EgoChat is not powerful enough to give concise responses
                             to users. On the other hand, EgoChat can easily store and process a great
                             number of messages because the representation of personal memories and the
                             methods of ordering messages are very simple. We are planning to practically
   110   111   112   113   114   115   116   117   118   119   120