Computers are increasingly able to mimic abilities we often think of as excusively human - memory, decision-making and now, speech. A new generation of speech recognition systems can make at least some attempt at understanding what is said to them and can respond accordingly. These systems are coming into daily use for home banking, for airline flights enquiries and for placing orders over the telephone and are fast becoming more powerful and more pervasive. Using data taken from a major European Union funded project on speech understanding, the SunDial project, this book shows how this data may be analyzed to yield important conclusions about the organisation of both human-human and human-computer information dialogues. It describes the Wizard-of-Oz method of collecting speech dialogues from people who believe they are interacting with a speech understanding system before that system has been fully designed or built and it shows how the resulting dialogues may be analyzed to guide further design. This book provides detailed and comparative empirical studies of human and human-computer speech dialogues, including analyses of opening and closing sequences, turn-taking, the organization of overlap and repair strategies to overcome troubles in verbal interaction. Humans, Computers and Wizards considers current perspectives on human-computer interaction and argues for the value of an approach taken from sociology which is based on conversation analysis. This breakes away from the individualistic, cognitivist approach of much HCI research and takes seriously the idea that a human-computer dialogue, like a human-human dialogue, is an instance of emergent social order.