Embodied Conversational Agents

argues that the sole criterion for excellent research is that the researcher produces "beauty." While seemingly ineffable and frustratingly imprecise, Hardy instead suggests that creating beauty is straightforward. First, the work must be accurate: erroneous results are useless. Second, one's peers must recognize the work to be interesting, exciting, elegant, and "cool." While this second criterion might seem arbitrary, there is generally good agreement between scholars in a given community about "interesting" work (see Cole and Cole 1973 for a discussion), so one need not survey numerous researchers to ensure research is beautiful; asking a couple is equivalent to asking them all.

This paper presents an analysis of the verbal and non-verbal cues of Conversational Agents, with a special focus on REA and GRETA, in order to allow further research aimed at correcting some traits of their performance still considered unnatural by their final users. Despite the striking performance of new generation ECA, some important features make these conversational agents unreliable to the users, who usually prefer interacting with a classical computer for information retrieval. The users’ preference can be due to several factors, such as the quality of speech synthesis, or the inevitable unnaturalness of the graphics animating the avatar. Apart from the unavoidable traits that can render ECAs unnatural to the ultimate users, instances of poor synchronization between verbal and non-verbal behaviour may contribute to unfavourable results. An instance of synchronization patterns between non-verbal cues and speech is here analysed and re-applied to the basic architecture of an ECA in order to improve the ECA’s verbal and non-verbal synchronization. A proposal for future inquiry aimed at creating alternative model for the ultimate Mp4 output is also proposed, for further development in this field.

How do you develop a game that is provably educational, that is engaging from pre-school to adult, and that independently targets different language ability and world experience/maturity? Our purpose is to teach students of German or English as a Second Language in the same way that a baby learns a First Language. Children acquire language automatically without overt teaching, without conjugating verbs, looking up dictionaries or taking complex classes in syntax and morphology. They acquire through being immersed in an environment in which they have to learn to communicate in the language in order to achieve a variety of social and personal goals. In this project we provide a mixed environment with real toys for the learner to play with, and a simulated environment and a simulated teacher/caregiver to interact with. Whilst young children may be happy building towers and bridges out of wooden blocks, older children and adults cannot be expected to find motivating goals from the same ...

The main goal of using non-verbal modalities together with the general text-to-speech (TTS) system is to better emulate human-like course of the interaction between users and the UMB-SmartTV platform. Namely, when human-TV interaction is supported by TTS only, the interactions tend to be still less functional and less human-like. In order to achieve more advanced interaction, and more human-human like, the virtual agent technology as a feedback interface has to be introduced. In this way more appropriate social responses from the UMB-SmartTV through personification of the TTS system, named PLATTOS, can be produced and close to human-human-like communicative behavior may be invoked. Verbal and co-verbal gestures are linked through complex mental processes. Understanding of attitude, emotion, together with how gestures (facial and hand) and body movements complement, or in some cases, override any verbal information produced by the TTS system, provides crucial information for modeling both the interaction and the embodied conversational agent's (ECA) socially-oriented responses. The social responses of the TTS system fused with ECA can then be presented to the user in a more human-like form, using not just audio but also facial expressions, such as: facial emotions, visual animation of synthesized speech, and synchronized head, hand, and body movements. In the paper a novel TTS-driven behavior generation system is proposed to be used for IPTV platforms. The behavior generation engine is implemented as a service and used by UMB-SmartTV in a service-oriented fashion. The behavior generation engine fuses both, speech and gesture production models, by using FSMs and HRG structures. Selecting the shape and alignment of co-verbal movement for embodied conversational avatar, named EVA, are based on several linguistic features (automatically extracted from the input text), and several prosodic features (symbolic and acoustic features produced within the TTS engine). Finally, the generated speech and co-verbal behavior are animated by embodied conversational agent's engine and represented to the user within the UMB-SmarTV user interface. In this way, personificated TTS system PLATTOS, integrated within the UMB-SmartTV system, enable more advanced, personalized, and more natural multimodal-output-based human-machine interface.