Humans, Computers and Wizards: Analysing Human (Simulated) Computer Interaction

218 Pages·2013·7.89 MB·English

by  Robin Wooffitt

HUMANS, COMPUTERS
AND WIZARDS
Computers are increasingly able to mimic abilities we often think of as
exclusively human - memory, decision-making, vision 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 in areas such as home banking, airline flight
enquiries and telephone shopping services, and are fast becoming more
powerful and more pervasive.
Using data from the SUNDIAL project, a major European Commission
funded project on speech understanding, this book shows how such data may
be analysed 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 spoken dialogues, which monitors people
who believe they are interacting with a speech understanding system before
that system has been fully designed or built, and shows how the resulting
dialogues may be analysed to guide further design. This book provides
detailed and comparative empirical studies of human-human and human
computer speech dialogues, including analyses of opening and closing se
quences, turn-taking, and the organisation of overlap and repair strategies to
overcome troubles in verbal interaction.
Humans, Computers and Wizards considers current perspectives on
human-computer interaction and stresses the value of a sociological approach
based on conversation analysis. This breaks 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.
Robin Wooffitt is Lecturer in Sociology at the University of Surrey,
Norman M. Fraser is Lecturer in Linguistics at the University of Surrey,
Nigel Gilbert is Professor of Sociology at the University of Surrey and Scott
McGlashan is Senior Research Scientist at the Swedish Institute of Computer
Science.
This page intentionally left blank
HUMANS,
COMPUTERS
AND WIZARDS
Analysing human (simulated)
.
computer InteractIon
Robin Wooffitt, Norman M. Fraser,
Nigel Gilbert and Scott McGlashan
1
~ Routledge
!
~
Taylor & Francis Group
LONDON AND NEW YORK
First published 1997
by Routledge
Published 2013
by Routledge
2 Park Square, Milton Park, Abingdon, Oxon OX14 4RN
Simultaneously published in the USA and Canada
by Routledge
711 Third Avenue, New York, NY, 10017, USA
Routledge is an imprint of the Taylor & Francis Group, an infomla business
© 1997 Robin Wooffitt, Norman M. Fraser, Nigel Gilbert
and Scott McGlashan
Typeset in Garamond by
Ponting-Green Publishing Services, Chesham,
Buckinghamshire
All rights reserved. No part of this book may be
reprinted or reproduced or utilized in any form or by
any electronic, mechanical, or other means, now
known or hereafter invented, including photocopying
and recording, or in any information storage or
retrieval system, without permission in writing
from the publishers.
British Library Cataloguing in Publication Data
A catalogue record for this book is available from the British Library
Library of Congress Cataloguing in Publication Data
A catalogue record for this book has been requested
ISBN 13: 978-0-415-06948-9 (hbk)
CONTENTS
List off igures and tables Vlll
Acknowledgements IX
1 SOCIOLOGY AND HUMAN-MACHINE DIALOGUE 1
Speech understanding systems 2
Perspectives on human-computer interaction 5
Sociological perspectives on human-computer interaction 8
The 'chicken and egg' problem 11
Overview of the book 12
2 INSPIRATION, OBSERVATION AND THE WIZARD
OFOZ 13
Design by inspiration 15
Design by observation 19
Design by simulation 22
Varieties of spoken WOZ experiments 30
3 THE SURREY WOZ SIMULATION PROCEDURE 36
The British Airways corpus 36
The Wizard of Oz simulation corpus 40
Questionnaire results 45
Some preliminary analysis from the WOZ corpus 50
4 CONVERSATION ANALYSIS 53
Conversation analysis and the turn-by-turn analysis of verbal
interaction 54
Recipient design 59
Sequences of actions 60
Language as a vehicle for social action 63
Conversation analysis and WOZ simulations 64
v
CONTENTS
5 GEITING STARTED: OPENING SEQUENCES IN THE BA
AND WOZ CORPORA 66
Response tokens in the BA corpus 68
Hesitation items in the BA corpus 70
Response tokens and hesitation items in the WOZ corpus 71
Request formulations in the BA corpus 74
Request formulations in the WOZ corpus 76
Sequential considerations 77
Request formulations are normatively prescribed 81
Problem work done through request formulations 83
'Doing' non-problematic requests 89
6 TURN-TAKING, OVERLAPS AND CLOSINGS 93
The turn-taking system for conversation 93
The turn-taking system in the British Airways corpus 97
Turn-taking in the WOZ corpus 102
Overlapping talk 106
Overlap types 108
Overlap in the WOZ simulation data 112
Closings 115
Closings in calls to the British Airways flight information
servIce 116
Closings in the WOZ corpus 121
7 SOME GENERAL FEATURES OF THE ORGANISATION
OF REPAIR 125
Analysing 'repair': some conceptual issues 125
Self-and other-initiation and execution of repair 129
The organisation of repair 130
The preference for self-repair 134
Some sequential properties of repair 137
Repair and interpersonal relations 139
8 SOME REPAIR STRATEGIES ANALYSED 142
The interpretation and repair of silence 142
Repair via alignment 152
Recycling 155
9 CONVERSATION ANALYSIS, SIMULATION AND
SYSTEM DESIGN 162
Wizards: implications for simulation studies 162
Machines: the use of conversation analysis and the design of
interactive systems 164
Humans; and their interactions with machines 166
VI
CONTENTS
APPENDIX A: RECOGNISING AND UNDERSTANDING
SPOKEN LANGUAGE 169
Representing speech 170
Speech technology 171
The front end 173
The linguistic processor 176
The dialogue manager 177
Speech synthesis 180
APPENDIX B: SCENARIOS IN THE WOZ SIMULATION
EXPERIMENT 183
APPENDIX C: TRANSCRIPTION SYMBOLS 187
Notes 189
References 197
Index 205
Vll
FIGURES AND TABLES
FIGURES
1 The architecture of a speech information system 172
2 An analogue speech signal 173
3 Structure of a four state Hidden Markov Model 174
4 A simple word lattice 176
5 A semantic interpretation of the utterance 'the arrival time of
BA296' 178
TABLES
3.1 Principal tasks in the human-human corpus 39
3.2 Subject/scenario matrix 44
3.3 Experience of using computers 46
3.4 Intelligibility of system's voice 47
3.5 Interpretability of content of system utterances 47
3.6 System's understanding accuracy 47
3.7 Satisfaction with system's answers 48
3.8 Average number of turns in the human-human and
human-computer corpora 50
3.9 Number of lexical tokens in the human-human and
human-computer corpora 50
3.10 Number of lexical types in the human-human and
human-computer corpora 50
3.11 Parameter order in scenarios and dialogues 51
5.1 Incidence of response tokens in the WOZ corpus 72
5.2 Incidence of second pair parts: explicit acceptances 73
5.3 Incidence of hesitation items in the WOZ corpus 74
Vlll
ACKNOWLEDGEMENTS
The authors would like to acknowledge the support and influence of all their
colleagues from the UK, France, Germany, Italy and Sweden who worked
on the SUNDIAL Project. Special thanks also to those colleagues who have
worked in the Social and Computer Sciences research group without whose
encouragement and advice this book would have been much harder to write:
Anne Ankrah, Sarah Buckland, Andrew Fordham, David Frohlich, Ian
Hutchby, Marina Jirotka, Carol Lee, Paul Luff, Catriona MacDermid, Alison
Mill-Ingen, Andrew Simpson and Ave Wrigley.
IX