The Tao of CHI: Towards Effective Human-Computer Interaction

The Tao of CHI: Towards Effective Human-Computer Interaction Robert Porzel Manja Baudis European Media Laboratory, GmbH Schloss-Wolfsbrunnenweg 33 D-69118 Heidelberg, Germany � robert.porzel,manja.baudis@eml-d.villa-bosch.de � Abstract End-to-end evaluations of conversational dia- logue systems with naive users are currently uncovering severe usability problems that re- sult in low task completion rates. Preliminary analyses suggest that these problems are related to the system’s dialogue management and turn- taking behavior. We present the results of ex- periments designed to take a detailed look at the effects of that behavior. Based on the result- ing findings, we spell out a set of criteria which lie orthogonal to dialogue quality, but neverthe- less constitute an integral part of a more com- prehensive view on dialogue felicity as a func- tion of dialogue quality and efficiency. 1 Introduction Research on dialogue systems in the past has fo- cused on engineering the various processing stages involved in dialogical human-computer interaction (HCI) - e. g., robust automatic speech recognition, intention recognition, natural language genera- tion or speech synthesis (cf. Allen et al. (1996), Cox et al. (2000) or Bailly et al. (2003)). Alongside these efforts the characteristics of computer-directed language have also been examined as a general phe- nomenon (cf. Zoeppritz (1985), Wooffitt et al. (1997) or Darves and Oviatt (2002)). The flip side, i. e., computer-human interaction (CHI), has received very little attention as a research question by itself. That is not to say that natural language generation and synthesis have not made vast improvements, but rather that the nature and design of the computer as an interlocutor itself, i. e., the effects of human-directed language, have not been scrutinized as such. Looking at broad levels of distinctions for dialogue systems, e. g., that of Allen et al. (2001) between con- trolled and conversational dialogue systems, we note the singular employment of human-based differentiae, i. e., the degree of the restriction of the human interactions. Differentiae stemming from the other communication partner, i. e., the computer, are not taken into account - neither on a practical nor on a theoretical level. In the past controlled and restricted interactions be- tween the user and the system increased recognition and understanding accuracies to a level that systems became reliable enough for deployment in various real world applications, e. g., transportation or cinema information systems (Aust et al., 1995; Gorin et al., 1997; Gall- witz et al., 1998). Today’s more conversational dia- logue systems, e. g., SMARTKOM (Wahlster et al., 2001) or MATCH (Johnston et al., 2002), are able to cope with much less predictable user utterances. Despite the fact that in these systems recognition and processing have become extremely difficult, the reliability thereof has been pushed towards acceptable degrees by employ- ing an array of highly sophisticated technological ad- vances - such as dynamic lexica for multi-domain speech recognition and flexible pronunciation models (Rapp et al., 2000), robust understanding and discourse modeling techniques (Johnston, 1998; Engel, 2002; Alexandersson and Becker, 2001) combined with ontological reasoning capabilities (Gurevych et al., 2003; …

Content truncated to 500 words