Natural Human-Human-System Interaction

The importance of a vision can be that of providing a model within which we think and create. If the model is outdated, thinking becomes unduly constrained. The paper proposes to replace the paradigm of human-computer interaction (HCI) with a more comprehensive model for thinking about future systems and interfaces. Recent progress in speech technologies has managed to establish a powerful application paradigm, i.e. that of natural task-oriented spoken language dialogue systems. This application paradigm points towards the broader goal of natural humanhuman-system interaction (HHSI) in virtual, combined virtual and physical, and physical environments. On the backdrop of the natural HHSI model and the rapidly changing environment of advanced systems research, the types of research that are likely to be needed in the future are discussed. The discussion deliberately de-emphasises next-generation systems research in order to shift the focus to a range of equally important, existing or emerging research objectives which sometimes show a tendency to be overshadowed by the next-generation challenges. 1. From Single Word to Spoken Dialogue During the past 40 years or so, the field of speech technology has moved its focus from research on single word recognition to research on natural spoken human-system dialogue. The underlying tale of gradual progress in research is not the whole story, however. In those four decades, the environment in which research progress was made has changed dramatically, leading to entirely new perspectives for speech technology research. Taking a closer look at these developments may be helpful in trying to understand the roles and objectives of research in today’s Information Society as well as where we are, or should be, going in the 21st century. In 1960, promising speech recognition rates were reported for very small vocabulary (10 words), speaker-dependent, real-time recognition of isolated words [1]. Today, academic research in speech recognition seems about to reach the end of the road, being replaced by steady progress through competitive industrial development [2]. Medium-sized vocabulary (+5.000 words), speaker-independent, real-time recognition of continuous (or spontaneous) speech has become commercial reality, and very large vocabulary (+60.000 words) spoken dictation systems which only need a minimum of speaker-dependent training, can be purchased for less than 100 $US from companies such as IBM, Dragon Systems and Philips. Today, unlimited vocabulary, realtime speaker-independent continuous speech recognition is within reach and speech recognition technology has become a component technology which is finding its way into all sorts of interfaces to computer systems. In itself, speech recognition is a transformation of the acoustic signal into an uninterpreted string of words which may or may not make sense to a human but does not make any sense to the