Narrative Editing of Web Contexts on Online Community System with Avatar-like Agents Toru Takahashi, & Hideaki Takeda*, Graduate School of Information Science, Nara Institute of Science and Technology 8916-5 Takayama-cho, Ikoma-city, Nara 630-0101, JAPAN ATR Media Integration & Communications Research Labs 2-2-2 Hikaridai, Seika-cho, Soraku-gun, Kyoto 619-0228, JAPAN * National Institute of Informatics 2-1-2 Hitotsubashi, Chiyoda-ku, Tokyo 101-8430, JAPAN tooru-t@mic.atr.co.jp, ABSTRACT In this paper, we introduce our online community system named TelMeA and a general outline of our social information system. Asynchronous online community system takes a capital part of our social information system because the community system brings out four important functions towards information: information building, information filtering, information editing, and information archiving. Current text-based community systems, however, are insufficient for the social information system because the text-based representation has limits. TelMeA employs scriptable embodied agents for multimodal conversational interface instead of text interface. By means of these agents, TelMeA enhance abilities of online communities as a social information system. DESIGNING SOCIAL INFORMATION SYSTEM What Is Needed for Social Intelligence Since information floods in the worldwide network have begun, it has become difficult for us to find adequate information which satisfies our needs on the network. To solve this problem, many retrieval and recommender systems are researched and developed. Although these systems are useful to extract some portion of information from a huge amount of network information by specifying keywords or user profiles, they lack abilities to associate or integrate the extracted portion of information. The process of associating or integrating information involves intelligence. It is difficult for computers to associate or integrate information intelligently because they need to understand meaning of information, to become interested in its content, and to raise awareness in its context. Therefore, engagement upon human wonderment or concerns is indispensable in design of social intelligence. To bring in human interest or concerns, engagement of online community provides effective alternatives. In many cases, online communities have their special themes or purposes and participants of an online community hold shared interests or intentions. Furthermore, many of these takeda@nii.ac.jp participants have wide experience and knowledge in their shared topics or problems and various viewpoints toward subjects of conversation. Since participants of these communities have such extensive experience and knowledge, the communities themselves can be seen as an information source of experience and knowledge. In these communities, especially in asynchronous ones, participants experience and knowledge are built with members notions through a process of message description. The built messages are shared among the community, evaluated by each other, and archived as conversation logs. Online Communities as Social Information Systems We believe this voluntary activity of online communities is indeed a base of social intelligence. From our viewpoint, an online community can be regarded as a social information system that has four functions: 1) Information Builder: Communications causes information. Participants concepts are realized as storable representations through communication in the community. 2) Information Filter: Information related to the special themes or purposes is collected and distributed. 3) Information Editor: Information is analysed, evaluated, and redressed to compose new information on the special themes or purposes. 4) Information Archive: Information is stored and served as abundant information resources for specialized fields. Figure 1 is a conceptual diagram of our social information system. An online community system is located at the left side in this diagram. Participants have conversations on this system asynchronously, and the conversations are stored in the server as log files. These log files include all processes and results of information building, filtering, and editing in the community. In order to increase availability of these processes and results, the social information system
Open to Any Participant Well, the origin of this church is Creating a script for the personal agent behavior contribute A TelMeA Community with Avatar-like Agents Scripts enact Analyze Make available to the Internet Autonomous Guide retrieve Knowledge Knowledge of the Community the Internet analyses restructures stored data as knowledge of the community. This knowledge would be made available to this community itself, to other communities, and to the Internet as knowledge databases associated with the community s special theme. WHAT IS IN NEED FOR ONLINE COMMUNITY SYSTEM IN OUR SOCIAL INFORMATION SYSTEM For our social information systems, we believe that asynchronous community systems such as ML (Mailing List), BBS (Bulletin Board System), and Usenet newsgroup are suitable. Social information systems need welldeliberated and well-organized information. Asynchronous community systems are ideal, because they give their participants time to read others contributions carefully, to retrieve information about the current topics, and to improve their writing as they compose messages. Current most asynchronous community systems, however, are insufficient for the social information system. The main reason is that the current community systems are simply designed with traditional communication technologies, i.e., most of them are text-based communication systems. The text-based conversation is conventional and familiar, but lacks following abilities. Awareness of Human Environment In text-based community, identifying participants is not easy because the existence of each participant is represented by not her or his corporeal aspects but her or his textual name and messages. In actual online communities, although participants seem to identify each other, spectators of text-based communities (especially newcomers) incur much effort for identification of each participant. If a spectator could not identify each participant, he or she could not also learn each participant personality and human environment of the community. Therefore, the misidentification often causes an obstacle to community perspective and understanding his standpoint in the community and hinders spectators from contributing message. Fig. 1 Conceptual diagram of our social information system Various Modalities of Communication In face-to-face conversations, people express themselves with physical non-verbal cues such as facial expressions, postures, gestures, gazes, finger pointing, and so on. People use these non-verbal cues in a way that is complementary to vocal speech in order to convey messages quickly, clearly, effectively and safely. For example, suppose a person wants to give someone a message related to an object near her. She may smile and simply say pronouns like this or that while turning her gaze and pointing her finger to the object. This expression is not prolix but instead economical. If she represents the same message only by text, the description may be long-winded and tedious. Text-based conversations can hardly express physical cues like smiling, gazing, or pointing. People are therefore often forced to use complicated phrasings to avoid misunderstanding emotional expressions or referred terms. Such complicated phrasing can cause much trouble for not only writers and readers of massages to make asynchronous conversation sequences awkward, but also automatic analysis by our social information system to build the communities knowledge. Seamless Integration with WWW Information Source Nowadays, WWW is the most valuable information source. As a matter of course, web pages are referred in text-based communication such as e-mail and BBS in the form of describing their URL. This method of referring to web pages, however, does not play enough roles because this description does not denote whole context of the page. If people wish to denote certain context in a part of referred web page, they must describe this context in words separately from the URL annotation. In face-to-face conversation, people can visibly refer to objects around them with gaze, finger pointing, or so on. If people refer to contents of a web page visibly in online community, people can easily present the context of this web page. Furthermore, if people refer to multiple web pages visibly, plenty of web pages and web contents are
Fig. 3 TelMeA authoring page Fig. 2 Screenshot of TelMeA linked by a narrative representation. Moreover, in the case of online communities, vast numbers of web pages are interrelated through conversations and built narrative network by communities special theme, while conventional web pages make the WWW network by arbitrary linkages by each web page s authors. Well-structured Description Many news groups have long histories and they have a huge amount of contribution archives. These archives are often regarded as an abundant knowledge database and many researchers quest for the method to abstract information from the archives. However, noteworthy results are not efficiently collected because of difficulties of naturallanguage semantics analysis. If people put marks to demonstrate semantic structures when they compose their contributions, the contributions would be easily analyzed and restructured to be reused as community knowledge. In a text-based conversation, however, putting marks is unnatural. We think that natural ways to put such marks are in need. To resolve these issues, we developed an asynchronous community system named TelMeA [Takahashi 2001]. Instead of text-base communication, TelMeA employs avatar-like agent, or scriptable embodied agent, as communication interface. TelMeA: ASYNCHRONOUS COMMUNITY SYSTEM WITH AVATAR-LIKE AGENT Figure 2 is a screenshot of TelMeA. In this figure, three anthropomorphized embodied agents are shown and one of the three agents, penguin is explaining a figure on a web page while pointing at it. TelMeA is an asynchronous community system as a web application like BBS on the web. However, participants of TelMeA communities contribute their remarks mediated by their personal scriptable embodied agents (avatar-like agent) instead of text interface. By means of these agents, TelMeA resolves the four issues of text-based community systems. Awareness of Human Environment In TelMeA communities, all contributed remarks are represented through avatar-like agents. As a result, spectators of conversations can easily identify each participant by their agents and associate their remarks to their agents corporeal aspects. Therefore, spectators easily recognize personalities of each participant and human environment of the community. Accordingly, we believe that the spectators of conversation made by avatar-like agents easily contribute new remarks to the conversation compared to the spectators of text-based conversation. The spectators of the conversation made by avatar-like agents easily capture conversations and decide their own standpoint in the community. We performed a psychological experiment to clarify such effects of avatar-like agents in [Takahashi 2001]. Various Modalities of Communication TelMeA supports following modalities: A) Voice by text-to-speech synthesizer B) Textual utterances in balloons C) Facial expressions by animations D) Gestures by animation E) Face rounding by animation F) Finger pointing by animation G) Moving on the display in connection with animation By combining these representation modalities, avatar-like agents can represent four types of representation: Verbal representation balloon Physical representation Gestures, behaviors, and facial expressions Interpersonal space Interpersonal distances among participants Joint attention Refer to an object with gaze or pointing and engage other participants attentions
designated a user s avatar-like agent let move near by Make smile by animation Open a web page Let move near and refer to it <#actor>penguin <#approach>bear <#play>smile <#speak>i found such a web page. <#open>http://www.mic.atr.co.jp <#speak>this is so cool! <#refer>img3@http://www.mic.atr.co.jp <#speak>this picture is nice, isn t it? Seamless Integration with WWW Information Source In TelMeA, participants can script their agents to open web pages and refer to images on the web pages in their utterances. Participants can also upload image files that are stored in their local machine to share among their community and script their agents to refer to these images. By combination of opening web pages and referring to contents on the opened web pages, TelMeA realizes seamless integration of online community systems and WWW information service. Well-structured Description When participants contribute remarks to a community, TelMeA turn into editing mode by clicking an edit button in the main page of each TelMeA community. Under the editing mode, the authoring page (Fig. 3) appears. The down below area of the authoring page is an editing place. Participants compose ALAScript (Fig. 4) in this editing place with graphical user interface, for example: When a sentence is written in the editing place, the sentence is marked by a speak tag. The utterer s agent speaks the sentence with synthesized voice and text balloon. When a name of gesture or facial experiment is selected from an animation list place in the authoring page, the name of selected animation is added in the editing place with a play tag and the agent plays the animation. When a URL of a web page is put in a specific place in the authoring page, the URL is added in the editing place with an open tag and a web page represented by the URL opens as a new browser window. When an avatar-like agent other than the utterer s agent is clicked, the clicked agent s name is added in the editing place with an approach tag and the utterer s agent approaches closes to the clicked agent. When an image on an opened web page is clicked, an ID of the clicked image assigned by TelMeA is added in the editing place with a refer tag. The agent moves and refers to the image with a pointing finger. By means of such graphical user interface, participants can easily compose and confirm their multimodal representation Fig. 4 Example for ALAScript on ALAScript description. Moreover, the description is marked up with ALAScript tags at the same time. As shown in the example of ALAScript in figure 4, the spectators of the avatar-like agent naturally assume the semantics such as "smile" to indicate positive feelings, and "such" and "this" to indicate the following web contexts. It is not only easy for spectators to assume, but also for machine analysis to predict the actions because the description is well structured by tags. CONCLUSIONS In this paper, we regarded asynchronous online communities as a base of social intelligent design and introduced our community system named TelMeA as a capital part of our social community system. By employing avatar-like agents as conversational interface, 1) TelMeA provides awareness of human environments of communities to facilitate contributions, 2) TelMeA provides various modalities for natural and abundant communication, 3) TelMeA realizes seamless integration of online community systems and WWW information service for valuable references to web pages, and 4) TelMeA produces wellstructured multimodal descriptions as a result of conversations. Therefore, TelMeA can effectively abstract and store knowledge from the conversations. (More detail information for TelMeA is described in [Takahashi 2001].) In future work, we will consider tag set of ALAScript by seeking more effective model of both avatar-like agents representation and semantics structure of the conversations. At the same time, we will make TelMeA public on the WWW to analyse usage and availability of the TelMeA. In addition, design of the authoring page would also be an important future issues. ACKNOWLEDGMENTS Avatar-like agent characters that are used in figures in this paper are designed as a part of a project on personal agent at ATR MIC. REFERENCES [Takahashi 2001] Takahashi, T, Takeda, H: TelMeA: An Asynchronous Community System with Avatar-like Agents, In Proceedings of Eighth IFIP TC.13 Conference on Human-Computer Interaction (INTERACT2001), July, 2001 (in preparation).