USF Libraries
USF Digital Collections

Toward a social-cognitive psychology of speech technology


Material Information

Toward a social-cognitive psychology of speech technology affective responses to speech-based e-service
Physical Description:
Polkosky, Melanie Diane
University of South Florida
Place of Publication:
Tampa, Fla.
Publication Date:


Subjects / Keywords:
Human computer interaction
Dialogue design
Customer satisfaction
Speech user interface design
Dissertations, Academic -- Psychology -- Doctoral -- USF   ( lcsh )
government publication (state, provincial, terriorial, dependent)   ( marcgt )
bibliography   ( marcgt )
theses   ( marcgt )
non-fiction   ( marcgt )


ABSTRACT: Speech technologies, or technologies that recognize and respond to human speech, have recently emerged as a ubiquitous and cost-effective form of customer self-service (e-service). Although customer satisfaction is regarded as an important outcome of e-service interactions, little is known about users affective responses to conversational interactions with technology. Using a theoretical foundation derived from research in social cognition, interpersonal communication, psycholinguistics, human factors, and services marketing, two studies develop items for a speech interface usability scale, which is then used to examine interrelationships among individual differences (e.g., self-monitoring, need for interaction with a service provider, inherent novelty seeking), usability, comfort, and customer satisfaction.
Thesis (Ph.D.)--University of South Florida, 2005.
Includes bibliographical references.
System Details:
System requirements: World Wide Web browser and PDF reader.
System Details:
Mode of access: World Wide Web.
Statement of Responsibility:
by Melanie Diane Polkosky.
General Note:
Title from PDF of title page.
General Note:
Document formatted into pages; contains 160 pages.
General Note:
Includes vita.

Record Information

Source Institution:
University of South Florida Library
Holding Location:
University of South Florida
Rights Management:
All applicable rights reserved by the source institution and holding location.
Resource Identifier:
aleph - 001680963
oclc - 62458476
usfldc doi - E14-SFE0001158
usfldc handle - e14.1158
System ID:

This item is only available as the following downloads:

Full Text


f Speech Technology: Affective Responses to Speech-Based e-Service elanie Diane Polkosky artial fulfillment of the requirements for the degree of Doctor of Philosophy Department of Psychology ences South Florida Major Professor: Judith Becker Bryant, Ph.D. rannick, Ph.D. Mark Pezzo, Ph.D. Sandra Schneider-Wright, Ph.D. Date of Approval: February 28, 2005 Keywords: Human computer interaction, dialogue design, customer satisfaction, e-service, speech user interface design Copyright 2005, Melanie D. Polkosky Toward a Social-Cognitive Psychology o by M A disertation in p s College of Arts and Sci University of Michael B Paul Spector, Ph.D.


signed speech system and those who rely on communication technologies and know that conversation is a most precious gift the majority of us take for granted. May we continue to find ways to use technology to And for Dr. Tom Tighe, who showed a treme ndous love of knowledge and concern for his students during the short time I knew him this paper would have greatly benefited from his guidance. Dedication For the users those who have ever had their day ruined by a poorly de make life better.


There are so many people to thank for all the support and assistance I received while working on my degree and this paper. First a nd most importantly, my wonderful husband Steve, e when I dont, makes me gratef ul for everything I have been blessed with, and giv biggest cheerleader, who To the teachers who helped me see a future when I needed direction and strength Marianne Sullivan, Dr. Steve Miller, and Dr. Jim Jen kins. Special additional thanks to Dr. Judy ough some of the models and friends, for fighting the fight to the dastardly dissertati on and daunting degree with me. John Pruitt, who suggested that I would make a better human f actors practitioner than computational linguist. stand what I do the yin to my yang on this particular speech technology top inking about it in the first place. Tom Capone, who helped me recover my lost proposal draft after my computers hard drive fi nally refused to cooperate for the last time. Vanessa Michelini, Brent Davis, Tami Harris, and Marv Silver, who have known the right thing to say at the right moment to make me keep believing. I am truly blessed. Acknowledgements who believes in m es me a reason to smile every day. To Lynd sey, my niece, friend, and brings joy to my heart. Bryant for patiently picking me up, dusting me o ff, and sending me on my way thr most trying times. Maria Rosa Brea-Spahn, Laurie Diamond, and Lin Smith, great role And to all my colleagues at IBM, who cha llenge me and help me under better than I ever could alone. Jim Lewis ic, who got me th


Table of Contents les iv List of Figures vi vii Induct1 Reiew o8 S 11 mmunication is Goal-Directed Social Beh12 es 14 Humans Automatically Infer Traits from a Partners Social Cues 17 irected Interpersonal 19 ction out Interactive Technolog20 Psychology for Speech Based e-Servic23 24 25 26 tisfaction 27 28 Point of View 30 ality and Behavior of Service Providers 32 35 Comfort 35 Use and Usability 36 ences of Customer Satisfaction 37 f Potential Speech-Based e-Service Constructs 38 Overview of the Proposed Study 38 Rationale 40 Research Questions 42 Study 1: Speech Interface Expert Ratings 43 Method 44 Participants 44 Measures 44 Stimuli 45 List of Tab Abstract troion vf Literature ocial-Cognitive Theoretical Framework Interpersonal Coavior Speech and Language Behavior are Social Cu Service Delivery is a Special Case of Goal-D Intera Human Users Have Social Cognitions abies Toward a Social-Cognitive e Service Quality Customer Satisfaction User Satisfaction Antecedents of Customer Sa Individual Differences P erson Trust Ease of Consequ Summary o i


Procedu re 46 47 tistics 47 Reliability 47 49 tive Responses to Speech Interfaces 49 50 50 51 52 Questions 52 ing 52 on with a Service Employee 53 Self-Monitoring 53 Speech Interface Usability 54 55 56 56 57 57 ould be measured in assessing ty of 59 s related to comfort 71 Question 3: What speech user interface characteristics best predict customer satisfaction? 73 r interface characteristics be74 with speech satisfaction? 75 76 77 77 e 78 Comparison Between the Usability Scale and Other Measures of Speech Technology and e-Service 80 asure Speech System Usability 83 Use of Third-Party Observers to Measure Affective Outcomes 85 Theoretical Implications 87 Explanatory Role of Pragmatic and Social-Cognitive Theory 87 Expectations and Social Perception of Speech-Based e-Service 90 Refining Constructs to Accommodate New Applications of Theory 93 Future Research 96 References 100 Results Descriptive Sta Stimuli Selection Study 2: Observers Affe c Method Participants Procedure Measures Demographic Inhere nt Novelty Seek Need for Interacti Comfort Customer Sati sfaction Stimuli Results Manipulation Checks Question 1: What items shthe usabili speech interfaces? Question 2: How are individual difference variable and customer satisfaction? redict Question 4: What speech usest p comfort? Question 5: Do individual difference variables interact interface quality to create customer Discussion M ethodological Implications The Usability Scale for Speech Interfaces Psychometric Quality of the Usability Scal Use of Experts and Students to Me ii


Ap120 ing Transcripts 121 140 141 ar and Bagozzi (2002) Inherent Novelty See 142 nteraction with a Service 143 olfe () ing Scale 144 145 Appendix H: Modified Spake, Beatty, Brockman, and Crutchfield (2003) omfort Scale 148 Appendix I: Eroglu and Machleit (1990) Customer Satisfaction Scale 149 About the Author End page pendices Appendix A: Audio Record Appendix B: Participant Instructions Appendix C: Participant Demographic Questions Appendix D: Dabholkeking Scal Appendix E: Dabholkar and Bagozzi (2002) Need for I Employee Scale 1984 Appendix F: OCass (2003) Revision of the Lennox and W Self-Monitor Appendix G: Items for Usability Scale C iii


List of Tables le 1 eech-based e-service articles in leading journa 1990-2003 4 le 2 45 le 3 48 le 4 ponses 58 Table 5 Means, standard deviations and correlations for hypothesized Customer Service 60 le 6 esized Pragmatics 60 le 7 esized Recognition factor 61 Table 8 Means, standard deviations and correlations for hypothesized Affective 61 Table 9 Means, standard deviations and correlations for hypothesized User Goal 61 tions and correlations for hypothesizedcy factor 62 e 11 elations for hypothesized Prompt Wording factor 62 Table 12 Means, standard deviations and correlations for hypothesized Usefulness factor 62 Table 13 Means, standard deviations and correlations for hypothesized Speech Impression factor 63 Table 14 Summary of reliability analysis for usability scale 65 Table 15 Initial factor analysis loadings for usability scale 67 TabSummary of spls TabSpeech interface stimuli TabExperts overall ratings of quality for 16 speech interfaces TabOpen ended and multiple choice manipulation check res Expectations factor TabMeans, standard deviations and correlations for hypoth factor TabMeans, standard deviations and correlations for hypoth Response factor Orientation factor Table 10 Means, standard devia Accura TablMeans, standard deviations and corr iv


Table 16 Second factor analysis loadings for usability scale 68 e 17 70 Table 18 Descriptive statistics and bivariate correlations for individual differences, usability factors, comfort, and customer satisfaction 72 TablPost-hoc results for interface quality v


List of Figures Figure 1. Speech user interface quality and customer satisfaction 59 Figure 2. Scree plot for usability scale factor analysis 66 Figure 3. Usability factor scores for the different interfaces 71 vi


Toward a Social-Cognitive Psychology of Speech Technology: ASpeech-Based e-Service Melanie Diane Polkosky ABSTRACT an speech, have ce (e-service). Although customer satisfaction is regarded as an important outcome of e-service interactions, little is known about users affective responses to conversational interactions with technology. Using a l communication, p items for a speech g individual differences (e.g., self-monitoring, need for interaction with a service provider, inherent novelty seeking), usability, comfort, and customer satisfaction. In the first study, speech and language e second study that ts listen to the six f interest. Results indicated that speech interface usability consists of four factors (User Goal Orientation, Customer Service Behavior, Verbosity, and Speech Characteristics). Usability items and individual differences predict affective responses to speech-based e-service. Implications of these findings for psychological and communication research and applied speech technology are described. ffective Responses to Speech technologies, or technologies that recognize and respond to humrecently emerged as a ubiquitous and cost-effective form of customer self-servitheoretical foundation derived from research in social cognition, interpersonapsycholinguistics, human factors, and services marketing, two studies develointerface usability scale, which is then used to examine interrelationships amonexperts ratings of sixteen speech interfaces are used to identify six stimuli for threpresent high, average, and low quality. In the second study, participaninterfaces and provide ratings for the primary variables o vii


f Speech Technology: e In their influential text, Sanders and McCormick (1993) assert that the two major objectives of the field of human factors are to: (1) increase the effectiveness and efficiency with ble human values, er user acceptance, en a problem-focused field, the scope and boundaries of it continually change in response to technological advances (Sanders & McCormick, 1993; Wickens, Gordon, & Liu, 1998). Sanders rs beyond the 20th actors experts: We hod for its contribution to the quality of life and work, contributions that go beyond issues of productivity and safety and embrace more intangible criteria such as satisfaction, happiness, and dignity (p. 9). tomer service via al notions of customer ibe the emerging uild customized service offerings, counting on knowledge about the customer to build strong customer relationships. Profitability is based on revenue expansion more than cost reduction, with revenues driven by enhanced service and higher levels of customer satisfaction. One of the natural outcomes of the network-based economy is the emergence of customers as the focal point of all businesses (p. 36). Toward a Social-Cognitive Psychology o Affective Responses to Speech-Based e-Servicwhich work and other activities are carried out and (2) enhance certain desiraincluding improved safety, reduced fatigue and stress, increased comfort, greatincreased job satisfaction, and improved quality of life (p. 4). Because human factors has beand McCormick (1993) recognized that part of the advancement in human factocentury would probably involve variables rarely considered by past human f pe that in the future human factors will become more involved and recognizeIn the early 21st century, the emergence of technologies that provide cuselectronic networks, known as e-service, marks a paradigm shift in traditionservice and human interaction with technology. Rust and Kannan (2003) descre-service paradigm: This service-focused paradigm uses two-way dialogue to b 1


Speech user interfaces are a means of human-computer interaction in which both the input esentation) use speech portant part of y allow highly personalized, easy to use, and cost-effective interaction between an enterprise and its customers (Balentine & Morgan, 1999; Kotelly, 2003; Rust & Kannan, 2002). This form of e-service requires theimary mechanism of irical attention over the past several years but the applied literature has only focused minimally on the interactive aspects of speech technology. Instead, the majority of empirical work has been conducted with hey enable human clude the g, telephone, and augmentative and alternative communication for individuals with communication disabilities (Bedrosian, Hoag, Calculator, & Molineux, 1992; Coleman, Paternite, & Sherman, 1999; Fussell & Walther, most forms of ed research and etween human partners, the application of social-cognitive and interpersonal communication theory to computer-mediated communication is clearly reasonable (for a review, see Spears, Lea, & Postmes, 2001): both partners are human and can be assumed to engage intentionally in their communicative interactions. In addition, humans have a wide range of social-cognitive behaviors, including attributions, to the system (user response) and the output from the system (machine pr(Balentine & Morgan, 1999). Speech interfaces are rapidly becoming an ime-service, often replacing or supplementing customer call centers, because the customer (user) to converse with a self-service technology system as the prcustomer service. Humans interaction with technology has received some limited empcommunication technologies that serve as the medium of communication; tpartners to converse. Technologies that allow human-human communication innow-familiar forms of email, chat, videoconferencing, instant messaginBenimoff, 1995; Savicki, Kelley, & Oesterreich, 1999; Storck & Sproull, 1995;1996,1997; Wellens, 1993). These forms of technology disrupt or obscurenonverbal and extralinguistic communication, which has been the focus of applisocial-psychological theory development. Because the interaction occurs b 2


judgments, and inferences about their communicative exchanges (Fiske & Taylor, 1994; h are thought to be markedly different in computer-mediated contexts (Sonversational exchanges. When speech interfaces are used in customer service, they generally replace a human customer service representative or operator. Speech interfaces may use speech recognition to may use speech l voice talent to respond ternet webpages, speech interfaces generally do not provide significant visual input to the user, but they allow ubiquitous access between business and its customers via the common telephone or other wireless cause the mode of interaction er skills and needs may isual interfaces (Sanders & McCormick, 1993). In addition, human social-communicative skills and conversational expectations form an important part of the usefulness and ease of use of sp speech user interfaces ations, stock and able, or constrained Perhaps due to the recent emergence of both speech technologies and e-service, the empirical literature to date has been fragmented. A number of disparate disciplines have handled aspects of interpersonal communication, social interaction, customer service delivery, and speech technology, but these literatures have had minimal influence on the characteristics of speech Holtgraves, 2002), whic pears, Lea, & Postmes, 2001). In contrast, speech user interface systems replace a human partner in cunderstand a human users utterances and synthetic speech to respond, or theyrecognition for comprehension and the recorded utterances of a professionato the user (Balentine & Morgan, 1999; Kotelly, 2003). Unlike e-service via Intechnologies (Balentine & Morgan, 1999; Rust & Kannan, 2002). Bebetween speech interface and a user is almost exclusively auditory, usdictate numerous aspects of design in a fundamentally different way than do v eech-based e-service interfaces. Functions that are currently handled byinclude banking and financial transactions, information retrieval, airline reservmutual fund inquiries, directory assistance, and other relatively simple, predictcustomer service transactions (Balentine & Morgan, 1999). 3


technology in the customer service arena. As shown in Table 1, a review of the articles published in uman factors, and social words speech, Table 1. Summaryf Speech-Based e-Service Articles in LNumber of Articles several major and specialized journals in marketing, communication, h psychology revealed no articles that simultaneously used the broad key o eading Journals, 1990-2004 Keyword 1 1995 ng Customer service Speech technolog Self-service technology Human-computer Interpers Social intera Speech AND tech s e 10 0 0 0 2 5 0 Custom Speech technology Self-service techn Human-c Interpersonal comm Social interaction Speech s e * * * ation Customer service or e-ser Research Speech technolog Self-service Human-compute Interpersonal com Social i Speech AND technology s e 1 0 0 2 41 9 0 Journal of AppliPsychology ed or e-serviceology nology interaction sonal communication Speech AND technology AND service 2 1 0 1 2 7 0 3 0 0 1 1 14 0 Customer se rvice Speech techn Self-service tech Human-computer Interper Social interaction 6 0 0 2 2 3 0 4 Journal Title (s) 99019961999 20002004 Journal of Marketi or e-servicey interaction onal communication ction nology ANDervic 9 0 0 2 1 2 0 14 0 1 0 1 1 0 Journal of Service Research er service or e-serviceology omputer interaction unication AND technology ANDervic11 0 0 0 1 0 0 39 0 2 0 1 1 0 Human Communicvicey technology r interaction munication nteraction ANDervic 0 0 0 4 31 8 0 0 0 0 3 20 17 0 Journal of Personality and Social Psychology Customer service or e-service Speech technology Self-service technology Human-computer interaction Interpersonal communication Social interaction Speech AND technology AND service 0 0 0 0 19 146 0 0 0 0 0 18 110 0 0 0 0 1 12 86 0


Personality an d SoPsychology Bulletirvice or e-servicey ology -computer interaction tion ND technology ANDervic 0 0 0 0 13 45 0 0 0 0 1 12 64 0 cial Customer se n Speech technolog Self-service techn Human Interpersonal communica Social interaction Speech A s e 0 0 0 1 11 63 0 Customer service Speech technolog Self-ser Human-computer interac Interpersonal com Social intera Speech AND tec s e 0 0 0 6 2 1 0 al of Customer service Speech Self-service technology Humancomputer Interpersona Social interaction Speech AND tech s e 0 2 0 2 0 0 0 al of Custom ology * Human Factors or e-servicey vice technology tion munication ction hnology ANDervic0 1 0 10 2 2 0 0 2 0 22 1 1 0 International JournHuman Computer Interaction or e-service technology interaction l communication nology ANDervic0 3 0 9 1 0 0 1 4 0 57 1 0 1 International Journer service or e-servicechnology Human-computer interaction Social interaction * 0 5 0 0 0 0 0 0 46 0 1 0 0 3 Speech Technology Speech technSelf-service te Interpersonal communication Speech AND technology AND service * (*)=journal began after 1995 (JSR 1999, IJST 1997) common (and broad) keywords showed that these journals remain devoted toaspects of speech-based e-service, consistent with their traditional scope. For exof Marketing and more specialized Journal of Service Research focused on cu technology, and service in their titles or abstracts during the past 13 years. Searches for other discipline-specific ample, the Journal stomer service (as indicated by publication of 83 articles over the 13 year period) but were very limited in their treatment of technology and communication. Similarly, social psychology and communication journals also excluded extensive treatment of technology, in preference to articles concerned with human-human communication or broader social interaction topics. As an example of the limited treatment of communication technologies in these disciplines, all 10 articles using 5


human-computer interaction as a keyword in Human Communication Research were concerned l of Personality and ocial Psychology Bulletin published any applied articles coSpeech technology has also had very limited impact in applied psychology. Human Factors and the International Journal of Human-Computer Interaction have only tangentially l Journal of Speech ted to user-system nism. Its published articles focused on the technology itself, as opposed to user issues related to speech interface design. Overall, this brief review of journal contents supports the contention that interpersonal customers is largely ignored based e-service, human factors experts who work in this field have a limited theoretical basis and few empirically-supported guidelines for designing speech user interfaces. This situation is inconistinct field. Sanders and Mrines of human factors: e humans and must ys be designed with the user in mind Recognition of individual differences in human capabilities and limitations and an appreciation for their design implications Conviction that the design of things, procedures, etc. influences human behavior and well-being with computer-mediated communication between humans. Neither the JournaSocial Psychology nor Personality and S ncerned with technologies for social interaction. included speech technology over the past decade. The specialized InternationaTechnology was initiated in 1997, but it largely excluded topics closely relainteraction and aspects of the technology as a customer service mecha communication between e-service speech user interfaces and humanby leading journals in each of the relevant disciplines. As a consequence of the limited research and applied nature of speechsistent with the fundamental principles that define human factors as a dcCormick (1993) identified six doct Commitment to the idea that things, machines, etc. are built to servalwa 6


Emphasis on empirical data and evaluation in the design process od and the use of objective data to test hypotheses and recognition that things, procedures, environments, and people do not exist in isolation (p. 5). Although these principles have provided grounding for the discipline of human factors, they have r service requires a firmer rfaces impact their human users. In addition, better understanding of the interactive effects of human social-communicative skills and speech interface characteristics is needed. Although the empirical earch streams have considered munication fields have uences person perception, impression formation, expectations, and partner behavior within a normatively-structured conversation (Baron, Byrne, & Johnson, 1998; Fiske & Taylor, 1991; rovides a powerful raves, 2002; y. Social interaction sence of social information (Fiske & Taylor, 1991), which suggests that a speaking technology interface may causally influence users cognitions. Finally, researchers in service marketing have identified customer service as a specialized form of interpersonal social interaction (Solomon, Surprenant, Czepiel, & Gutman, 1985), linking findings from social-cognitive psychology to the applied setting Reliance on the scientific meth generate basic data about human behavior Commitment to a systems orientation and a not guided speech interface research and design. At the present time, speech technology design for custometheoretical basis and a comprehensive understanding of the ways that these interesearch has not specifically filled this void to date, several resaspects of the problem. First, the social psychology and interpersonal comdemonstrated that communication is a goal-oriented social behavior that inflHoltgraves, 2002). In addition, a speakers speech and language behavior pform of social information that can influence a communicative partner (HoltgNeumann & Strack, 2001), even when the speaker is a non-animate technologresults in automatic trait inference, or the assumption of speaker traits in the pre 7


of customer service delivery. A missing piece of this theoretical puzzle is an understanding of the among perceptual and affective variables when customer service is provided by a spindicates that human users of speech recognition systems perceive these systems and interact with them much as they would a human conversational partner. The speech and language cues provided by a speech for human partners. onships among affective k, I review the social psychological literature concerned with interpersonal communication and social cognition (specifically, impression formation and person perception), followed by a summary of this t have been logy, and service ing disciplines to identify an initial set of constructs that may illuminate users social-cognitive responses to interactive speech systems. From this foundation, several research questions are derived for an initial foray into an applied social-cognitive psychology of speech logy: some researchers assume that human social behaviors are not relevant to the design of interactive dialogue systems. In such systems, the speaker is an abstraction and does not appear to warrant the same social consideration as a human conversational partner. The system does not have intentionality, as humans are presumed to have. Balentine and Morgan (1999) argue that interrelationships eech user interface. In this introduction, I review a broad, multidisciplinary literature that interface activate affective responses that would appear to be reserved only The primary goal of the current study was to identify and elucidate relatioutcomes of speech-based e-service. To establish a theoretical frameworliteratures impact on customer service research. Then, I review variables thaconsidered in the interpersonal communication, human factors, social psychomarket technology. Review of Literature A controversy exists in the field of speech techno 8


humans have no precedent for verbal interaction with non-sentient devices that are not l set of social f the interaction. This assumption collapses when the partner is a machine. The result of this expectation of social awareness is that structured and goal-oriented protocols become necessary to steer d, task-oriented so as not to be social, which will subsequently lead to greater efficiency in reaching user goals. Efficiency and sociability are seen as opposites: an increase in one leads to a decrease in the other. Balentine and cteristics to non-human ech interfaces. They rs tendency toward anthropomorphism. Their discussion indicates that they view personification negatively and caution designers to make informed choices in design: The reasons for designing a personified be based on nified design ociety (p. 217). ligent technologies (Russell & Norvig, 1995), other researchers have embraced the application of social psychology to human-computer interaction. One of the most developed programs of research in this realm has been undertaken by Nass and his colleagues. Nass and Moon (2000) describe humans as mindlessly applying social rules to computers, despite their explicit knowledge that self-aware. Indeed, much human speech is dependent on a powerfutechniques that derive from a fundamental assumption when a human being talks to another, that other human presumably has some stake in the outcome o the user away from social speech behaviors and toward work-orienteinteractions (p. 12). This perspective suggests that a speech-based interface can be designedMorgan (1999) identify anthropomorphism, the attribution of human charabeings or things (p. 218), as the underlying reason for social responses to spesuggest that a designer may choose to personify an interface, exploiting use interface may be marketing, design, or aesthetic in origin. They are not likely to productivity arguments or ergonomic principles. In effect, choosing a persorepresents a philosophical statement about the role of machines in human sIn contrast to views that broadly reject the human model as a foundation for intel 9


computers are not humans. Further, they argue that users are aware that social behaviors such as h they are frequently mindlessness provides an e rejecting anthropomorphism, experimenter demand, and personification of the computer programmer as explanations of their findings. For example, Nass and Lee (2001) examined whether listeners o voices that xtrovert). The consistent with their own personality and recommended that designers measure user personality and provide matching synthetic voices in their applications. Other empirical work in this line has been concerned with intley, Mortati, Sloan, on, & Skageby, Although Nass and colleagues rely on Langers mindlessness theory as a basis for their work, it appears to be a limited theoretical foundation for social human-computer interaction. Fiske o their environment d is a cognitive state appropriate for routinized, overlearned tasks. Abrformance (p. 723). ma theory, may offer a more intuitively reasonable explanation of Nass findings (Abelson, 1981; Fiske & Taylor, 1991; Glover, 1995; Shank & Abelson, 1977). At the present time, the few empirical studies that apply social psychological theory to speech interfaces have fueled the controversy over the validity of a social approach to speech politeness and reciprocity should not be used in these interactions, even thougobserved. Nass and Moon (2000) state that Langers (1989) theory ofexplanation of the behaviors exhibited by participants in their research, whilwould identify acoustic voice cues in synthetic text-to-speech and be attracted tseemed similar to their own measured personality characteristics (introvert or eauthors interpreted their results as indicating that users prefer synthetic voices eraction between prompt style and interface organization (Vanhoucke, Nee& Nass, 2001) and foreign accent perception (Dahlback, Swamy, Nass, Arvidss2001). and Taylor (1991) assert that mindlessness makes individuals less responsive t(as compared to mindfulness) an elson (1981) asserted that mindlessness involves total automaticity of peIn contrast, more broadly accepted theories of social cognition, such as sche 10


technology. On one side of the debate, researchers suggest that social behaviors and cognitions dividuals suggest that on is somewhat robust theoretical foundation for a social-cognitive approach to speech-based e-service design? Indeed it does. Social-Cognitive Theoretical Framework design guidelines have y the literatures that offer sers. Social psychological findings have had relatively limited application in human factors and human-computer interaction research, except in the areas of workplace or environmental design 999). ement of speech rsonal communication, service marketing and management, and social-cognitive psychology suggest that speech and language skills are intimately associated with social interaction. Further, humans in any interaction, eaking, interactive users is based on four streams of research that demonstrate: a) interpersonal communication is goal-directed social behavior, b) humans use speech and language behaviors to obtain social information, c) humans automatically infer personality traits from social information, and d) customer service is a special case of interpersonal communication. I now summarize each of these literatures in turn. have no relevance to interaction with a speech interface. On the other side, insocial-cognitive behaviors are relevant but the theoretical support for this positiweak. Does the previous literature offer a more To date, a social psychology of speech technology and interface emerged in a piecemeal fashion, typically without being informed bsignificant theoretical insight into the social-communicative skills of human u and team interaction (Sanders & McCormick, 1993; Wickens, Gordon, & Liu, 1Why might social-cognitive psychology be important to the advanctechnology theory and interface design? Distinct streams of research in interpeautomatically infer personality traits from speech and language cues contained even one with a speaking and listening technology. The premise that a sptechnology elicits social behavior and cognition from its human 11


Interpersonal Communication is Goal-Directed Social Behavior d and social tical approaches to interpersonal communication assert that golanguage can also be viewed as a tool, a tool that is used for accomplishing particular ends. To use language is to perform an action, and it is a meaningful action, with consequences be a speaker. And context is critical. be derived with some reference to a context (Holtgraves, 2002, p. 5). In this sense, language is the interpersonal means for accomplishing a particular goal in a defined ssarily implies social nged by social context (Whitney, 1998, p. 34). Pragmatics and speech act theory conceptualizes linguistic meaning as use of language: meaning is derived from the purpose an utterance is put to (Holtgraves, t model of language, sides in speaker intentions. However, listeners have a distinct role in language usis sense, then, rs goals, through the use of language. This perspective is based primarily on the theoretical work of philosophers Austin, Searle, and Grice. Austins (1962) speech act theory contradicted the prevailing view of language known as logical positivism, which claimed that all utterances were to be evaluated based on their truth value. Although some speech technology researchers suggest that goal-directebehavior are mutually exclusive, theore al-directed communication is social behavior: for the speaker, hearer, and the conversation of which it is a part. This is a very different view of language. To understand meaning there must What a speaker means with an utterance (what he intends to accomplish) can only social context, and is inseparable from social behavior. Language use neceintent (Austin, 1962; Searle, 1969; Holtgraves, 2002). Pragmatics may be defined as rules for how literal meaning can be cha2002). Krauss and Fussell (1996) describe this perspective as the intentionalisin which meaning re age: they construct the speakers meaning and goal through inference. In thconversational interactants accomplish their social goals, and derive their partne 12


Instead, Austin (1962) defined an utterance as an action or performative (e.g., Im sorry), which stepped on your hearer may or may not s, 2002). Austin (1962) also defined three dimensions of speech acts: 1) locutionary, or meaning of an utterance (propositional content); 2) illocutionary, or the speakers intent; and 3) perlocutionary, or the n, Austin (1962) described a conventional procedure appropriate context, and a conventional effect. He indicated that violations of the felicity conditions would result in a misfire or misapplication of the message. Thus, Austins contribution incf the unique roles of the 1969) elucidated Speech act theorists recognized that there were multiple ways of expressing a single meaning and that the hearers recognition of the speakers intention occurs through inferencing. cerned with how oth speakers and n such as is he talk exchange in which you are engaged (p. 45). He refined this cooperative principle with four maxims: 1) Quantity the contribution should be as informative, but not more informative, than necessary, 2) Quality the contribution should be true (do not say false utterances, or those that lack evidence), 3) Manner the contribution should be clear, unambiguous, brief, and orderly, and 4) Relation the he contrasted with a constative utterance, or message that conveys truth (e.g., Ifoot). He observed that performatives may or may not be successful (e.g., a accept an apology), but these utterances do not contain truth value (Holtgraveachieved consequences of an utterance (effect on the hearer). In additiofelicity conditions for performing a speech act: these conditions include executed correctly and completely by an intentional speaker, an appropriate hearer in the luded the initial view of language as an action and early specification ospeaker and hearer to communicative interaction, contributions which Searle (further. Grice (1975) proposed a theory of conversational implicature that was concommunicative partners infer speakers intended meanings. He observed that bhearers abide by a cooperative principle: make your conversational contributiorequired, at the stage at which it occurs, by the accepted purpose or direction of t 13


contribution should be relevant. ations iolate or flout these overstatement). Speakers then expect the hearer to infer the intended meaning. Grice (1969) provided an example of flouting the Quantity maxim: a professor might write a very short recommendation letter when a n interviewer, instead of refusing). ly to have so little knowledge of her own pupil. Through inference, the interviewer might further conclude that the recommendation includes information that is not explicitly written that the student should not be listeners have a vital isteners must assume that speakers are intentional. In addition, they must use the context of a conversation to piece this meaning together. Empirical work confirms that speakers speech and language characteristics tion. e Behaviors are Social Cues the intentionality of on to determine the partners goals and meaning. The previous literature has identified speech and language as a major source of social information. It is well known in the social psychological and communication literatures that linguistic (e.g., lexical choice, language style), nonverbal (e.g., body language, eye gaze, facial expression), Grices maxims and the cooperative principle are the normative expectconversational partners rely on during conversation. However, speakers can vnorms, usually to convey a particular meaning (e.g., metaphor, understatement,student applies for employment. The individual receiving the letter, probably aassumes that the professor is being cooperative by agreeing to write the letter (However, because the letter is short, she may reason that the professor is unlike considered for employment. According to speech act theory, language is used to convey meaning;and collaborative role in creating this meaning. To infer meaning, l serve as social cues for listeners (and vice versa) during communicative interacSpeech and Languag Pragmatic research demonstrates that humans make assumptions abouttheir conversational partners and actively use observed behaviors in an interacti 14


and extralinguistic (e.g., tone of voice, intonation, emphasis, rate of speech) communication ferences about raves (2002) notes biguous, and variable for any speaker, under even the most ideal interaction conditions. The previous literature has examined the effects of both speech and linguistic behaviors on social perception. eyed by individual sing, pitch, pitch n interaction, research has indicated that speech and extralinguistic behavior are some of the mechanisms that speakers manipulate strategically to control their listeners impressions of them (Bradac, Cargile, & teners automatically infer are of the source of their 000). DePaulo (1992) reviewed the literature on the role of nonverbal behavior in self-presentation (which can be interpreted to include extralinguistic aspects of speech) and d less accessible to actors aviors minimally bout a speakers ic, amount of stress, or current emotional state. Dialectal variants and nonstandard accents also have social consequences, are usually linked to the status (competence, power) and solidarity (friendliness, generosity) of the speaker (Holtgraves, 2002), and can even account for unique variance in listeners attributions about a speakers behavior (Dixon, Mahoney, & Cocks, 2002). In general, variables are sources of information about speakers that listeners use to make inintended meaning and speaker traits (DePaulo, 1992; Holtgraves, 2002). Holtgthat, in human-human conversation, these cues are context dependent, am Speech is the physical, acoustic representation of language as convspeech sounds (e.g., vowels and consonants), intonation, stress, loudness, pauvariability, and other extralinguistic behaviors (Cairns, 1999). In human-huma Hallett, 2002; DePaulo, 1992; Holtgraves, 2002). In addition, lischaracteristics of the speaker from these vocal cues, and may even be unawinferences (Bargh & Ferguson, 2000; Holtgraves, 2002; Neumann & Strack, 2asserted that nonverbal behavior is irrepressible, linked to emotion, anthan observers. Krauss, Chen, and Chawla (1996) showed that nonverbal behassist in conveying a speakers message, but instead reveal information aintrapersonal cognitive processes to the listener, such as attitude toward the top 15


the voice is used to make judgments of personal characteristics such as gender, age, socioeconomic up, personality, and status (Aronovitch, 1976; Kappas, Hess, & Scherer, 19ce characteristics and emotional effects, or the role of speech in conveying affect (for reviews, see Kappas, Hess & Scherer, 1991; Murray & Arnott, 1993). This research suggests that the voice is intimately teristics are n, and stress (Kappas, s produce similar acoustic configurations to convey emotion, which listeners use to detect their affective state. Apple and Hecht (1982) showed that listeners generally could recognize anger, sadness, surprise, ough accuracy ed, happiness least Strack (2000) provided an empirical demonstration of mood contagion, in which simply listening to emotional speech automatically induced a similar mood state in listeners, even when they were under and unmotivated to share n, and Gruenfeld particularly important for impressions of partners in mixed-sex conversations (e.g., liking, aloofness, self-centeredness, interest in partner, liking for partner). Two experiments revealed gender effects, indicating that more positive impressions occurred when males used elaboration and females used question reciprocation. Additional research on normative expectations of communication suggest background, ethnic gro 91; Murray & Arnot, 1993). Another vein of research has explored the relationship between voiassociated with the affective state of the speaker, and specific acoustic characassociated with such affective states as indifference, boredom, joy, depressioHess, & Scherer, 1991). Cosmides (1983) demonstrated that different speakerand happiness in content-filtered recordings with above-chance accuracy, althdepended on the specific emotion being expressed (sadness best recognizrecognized). In an especially provocative series of experiments, Neumann and cognitive load, unaware that vocal cues may have influenced their moodthe speakers mood state. In their study of the language-based aspects of impressions, Wyer, Swa(1995) found that linguistic behavior (question reciprocation, elaboration) was 16


that males and females violations of their expected linguistic style can have positive (e.g., when ful style) effects on ueller, and Hiller al items, and verb tense accounted for more variance in listener perception than ratings of nonverbal expressiveness, attractiveness, and facial maturity. Speakers and listeners also use speech and language cues in tners knowledge, which is iable messages (Bavelas, ll, 1991). In general, the theoretical models of social interaction clearly implicate speech and pragmatic language variables as a major source of speaker information during interaction (Bradac, 4; Patterson, 1996, 2001). They fuhearers information Humans Automatically Infer Personality Traits from a Partners Social Cues Recent researchers have posited that automaticity of frequently used social processes dent automaticity, of the stimulus ted by Bargh & which one is not aware of the stimulus or ones ensuing cognitive processes, and postconscious automaticity, in which one is aware of the stimulus but not its effects on cognitions. Bargh and Ferguson (2000) reviewed the literature on automaticity and asserted social behaviors in the external environment often if not usually access their corresponding mental representations in an immediate and direct manner, males use a nurturing style) or negative (e.g., when females use assertive/powerimpressions (Burgoon, 1990; Burgoon & Klingle, 1998). Berry, Pennebaker, M(1997) found that categories of emotion words, self-reference, unique lexictheir partners behavior to make attributions about the conversational par mportant to establishing mutual knowledge and constructing understandCoates, & Johnson, 2000; Clark, 1996; Fussell & Krauss, 1992; Krauss & Fusse Cargile, & Hallett, 2001; Cargile, Giles, Ryan, & Bradac, 199 rther suggest that a speakers speech and pragmatic variables influence a processing, responses, affective state, attitudes, attention, and social judgments.improves cognitive efficiency (Fiske, 1993). Bargh (1989) described goal-depenin which an individuals ability to identify an activating stimulus and the effectsdepend on his perceptual goal, implying some level of cognitive control (as ciFerguson, 2000). He contrasted this type with preconscious automaticity, in 17


without conscious and effortful processes of categorization and interpretation being necessary (p. e stimulus (within behavior (Bargh & hat social information is processed rapidly and automatically, often outside the individuals conscious awareness, and affects subsequent cognition and behavior. ediate and beyond ses may lead to ference of personality traits when an individual is exposed to behavior (Brown & Bassili, 2002). This form of automaticity is thought to help individuals make sense of each other and promote cognitive pontaneous trait owronski, Carlston, Mae, & Cr2002). Thus, following this line of reasoning, because any communication behavior has trait-implying properties, it is likely that users would infer traits from a conversational speech interface. eption-behavior link itate smooth and make rapid attributions (Holtgraves, 2002). Clark (1996) postulated that these attributions are based on heuristics, suggesting that speakers assume that the common ground of a conversation consists of items in the immediate physical environment, information that has been previously introduced in the conversation, and inferences about a partners social memberships (e.g., gender, race, ethnicity). A series of experiments has 929). Perception of social cues results in immediate evaluative judgments of th250 msec of the cue), and the resulting judgments also affect subsequent social Ferguson, 2000). Thus, current research in automaticity supports the notion t In addition to the notion that social perception and judgment are immconscious awareness, several researchers have shown that these procescounterintuitive effects. Spontaneous trait inference is the unconscious inefficiency (Fiske & Taylor, 1993). Several researchers have demonstrated that sinference occurs for speakers communicating about others (Sk awford, 1998) as well as inanimate objects (Bassili, 1976; Brown & Bassili, The interpersonal communication literature has also recognized a percin conversational exchanges, in which automatic inferences about a partner facilsuccessful interaction. The research suggests that conversational partnersabout a partners knowledge in order to construct appropriate messages 18


demonstrated that individuals construct messages differently for self, friends, and strangers in a 91, 1992), nversational behavior. e tenuous, probabilistic hypotheses that are continuously modified throughout interaction and are also subject to systematic errors and biases. Clark and Shaefer (1989) postulated that conversational progress rs acceptance. Although n knowledge attributions about listeners, lisrt of successful communication (Bavelas, Coates, & Johnson, 2000; Clark, 1996). Thus, the literature indicates that speech and language information contribute to the auation subsequently ion Solomon, Surprenant, Czepiel, and Gutman (1985) used a foundation of social-cognitive theory to argue that service encounters are a specialized form of interpersonal interaction. They rovider must adopt ollow a ritualized tomers learn these rgue that customer satisfaction is determined by the amount of congruence between expectations of the service encounter and the actual encounter itself, a definition that has been operationalized as the most common measure of service quality (Parasuraman, Zeithamal, & Berry, 1985, 1988, 1994). A body of empirical work suggests that a service providers behavior and personality influence customer referential communication task (Fussell & Krauss, 1992; Krauss & Fussell, 19demonstrating that assumptions about a partner do impact subsequent coKrauss and Fussell (1991) suggest that assumptions about mutual knowledge aroccurs via utterance pairs, or a speakers initiation followed by the listenespeakers do appear to modify their messages based o tener feedback and co-construction of messages remains an essential pa tomatic evaluations of speakers. In addition, perceptions based on this informeffect interactants conversational behavior, leading to successful, smooth communication. Service Delivery is a Special Case of Goal-Directed Interpersonal Interactpointed out that a service encounter is dyadic and both the customer and pappropriate, coordinating roles for it to proceed smoothly. The interactants fscript to increase the probability of attaining their goals. Both providers and cusscripts, which form the expectations for each service encounter. The authors a 19


satisfaction, providing support for the idea that provider behavior is an integral part of the service ; Humphreys, 1996; ted a theoretical istics of the service provider and customer (e.g., background similarity, interaction frequency, script strength, number of subscripts, experience with complementary role, goal compatibility) affect their role taking degree to which expections and actual role behaviors are similar. Role taking acturn influences Scripts associated with service delivery would not be altered simply because a speech technology system replaces a human service provider. Users would continue to assume that the oothly and ter. Because ch means of interaction, it seems reasonable that the expectations of human service providers would also apply to speech user interfaces. However, user perception of e-service encounters and other types of hual literature to date. I of human interaction ogies The notion that human users might perceive social characteristics in a conversational computer is not new: Turing (1950) proposed a test in which human and computer interactants converse as an evaluative method for the humanness of computer partners. The Turing Test has remained a gauge by which technological progress is still measured (Saygin & Cicekli, 2002). A role (Baydoun, Rose, & Emperado, 2001; Cran, 1994; Holland & Baird, 1968Mount, Barrick, & Stewart, 1998; Yagil, 2002). Mohr and Bitner (1991) presenmodel of mutual understanding in service encounters, in which several characteraccuracy, or the curacy and cognitive similarity influence mutual understanding, which in customer satisfaction. system is in place to support their goals and that the service encounter should proceed smpleasantly, in a behavioral sequence consistent with a human service encounspeech-based e-service makes use of the most common, natural, and socially-ri man-computer social interaction has had limited treatment in the empiricnow turn to a review of the few studies that have examined the social aspectswith speech technology. Human Users Have Social Cognitions about Interactive Technol 20


growing body of current work suggests that social psychological findings about interpersonal spch technology. like to be perceived rch in voice and human emotion has used synthetic speech, because it allows complete control over every acoustic parameter (Kappas, Hess, & Scherer, 1991, p. 220) and thus provides better internal validity for roblems relative to vocal affective e text-to-speech (TTS) is the most commonly used form of synthetic speech (Henton, 2002). This type of speech is generated from units of recorded human speech; thus, even synthetic speech in the majority of loyed applications is human speech. Alternatively, speech interface designers may use rections to a dialogue. In speaker (system itself). Consistent with a social-cognitive perspective, listeners negative perceptions of a erface, even when their etic speech only erformance, perception, ormance as compared with the mixed-voice condition. However, TTS also resulted in more repetition of prompts and more negative perceptions of performance, clarity, liking, and effort than did mixed voice prompts. Thus, synthetic speech can differentially affect users actual performance, perceived performance, and attitude, all of which may affect users preference and future use of a eech and language in humans may generalize to human interactions with speeThere is general recognition that synthetic speech should sound humanmore positively by users (Henton, 1999, 2002). Interestingly, much of the reseaexperimental work. Although this choice may present some generalization pspontaneously produced human speech, it strongly supports the applicability ofresearch to synthetic speech. In commercial speech applications, concatenativdep orded responses of a professional voice talent as the systems contribuboth cases, the use of either type of speech will elicit trait associations about thesystems speech may have implications for users overall judgment of an intperformance is unaffected. Gong and Lai (2001) measured the effect of synth(TTS) and mixed human and synthetic speech (mixed) prompts on user pand attitude. They found that the TTS condition resulted in improved user perf 21


system. ly influenced ean theory to gain insight ) developed 14 guidelines for man-machine interaction based on analysis of dialogues from a prototype flight reservation system. The authors later observed the similarity between their principles and Grices ed their model irically validated for oal dialogue, people, just like machines, should communicate their communication deficiencies, take background knowledge into account, and initiate repair and clarification metacommunication when needed. ed-goal dialogue (pp. ound that use of Grices maxims differentially affected perceptions of a computer conversational partner. They provided participants with 14 conversation excerpts from annual Turing test competitions, in which er. Participants read cean maxims (e.g., formation than uter. Results indicated that violation of the Manner maxim caused participants to perceive the computer partner as more humanlike, while violations of Relation and Quantity revealed that the partner was a computer. Saygin and Cicekli (2002) suggest that empirical work in pragmatics and conversational interfaces is in its early stages and observe that pragmatics is a crucial component of linguistic Although the pragmatics of conversation does not appear to have broadapplied practitioners in interface design, a few researchers have used Gricinto human-computer interactions. Bernsen, Dybkjaer, and Dybkjaer (1996theory, although Bernsen, Dybkjaer, and Dybkjaer (1996) claim to have developindependently. They concluded that the Gricean maxims have been emptask-oriented, spoken human-machine dialogue. When performing shared-g Failure to do so detracts, sometimes seriously, from the rationality of the shar234-235). In further support of applied pragmatics, Saygin and Cicekli (2002) fhuman interlocutors interact with a human or machine conversational partndialogue transcripts, then rated their agreement with statements based on GriAs contribution to the conversation is irrelevant, A provides more or less inrequired), and also rated whether the speaker was a person or comp 22


communication. Although this is a widely acknowledged fact among pragmatics researchers, other mate how hard the d Dybkjaer (1996), modification to account fully for human-computer interaction, but this research indicates the general applicability of pragmatic theory to human-computer interaction. r the interpersonal viduals would hold ervice provider should be friendly, helpful, and polite. Individuals probably evaluate their interaction based on this internalized script by comparing the behaviors of the system to the prototypical behaviors expected iently similar to the e heuristic processing duncounter will proceed smoothly and result in positive affect. However, if the system does not take its expected role, negative affect and controlled processing to understand the discrepancy may result. have identified a now review these ice Various literatures have implicated a number of social-cognitive variables as important to an understanding of speech-based e-service outcomes for users. To identify the relevant variables for speech-based e-service adequately, a broad review of several literatures (e.g., interpersonal communication, social psychology, service marketing, and human factors) is needed. Although the scientists and writers, notably computer scientists, sometimes tend to underestiproblem of pragmatic modeling can be (p. 255). Like Bernsen, Dybkjaer, anSaygin and Cicekli (2002) acknowledge that Grices maxims may require some Social-cognition research provides a broad explanatory foundation foimpact of speech-based e-service on human users. It seems reasonable that india script or event schema for customer service encounters that specifies that the sof a human customer service provider. System characteristics, if they are sufficinternal script, will activate the service knowledge structure and engag ring the encounter. If the speech system responds as expected, the service e The literatures concerned with various aspects of this type of encounternumber of affective variables that may be applied to speech-based e-service. I variables. Toward a Social-Cognitive Psychology for Speech-Based e-Serv 23


literatures are distinct and largely parallel, they converge on the notion of satisfaction as a vitally n outcome of summarize the constructs, its antecedents, and consequences as a basis for a social psychology of speech technology e-service. Service Quality d measures in the sed interchangeably with stomers normative expectations of service and their perception of the actual service received (Gilmore, 2003; Parasuraman, Zeithamal, & Berry, 1985). Parasuraman and colleagues developed a SERVQUAL been a centerpiece of n, Zeithamal, & unds, including its measurement of customer expectations (Cronin & Taylor, 1992, 1994), use of difference scores, factor structure, and psychometric properties (Gilmore, 2003). Parasuraman and colleagues have y but acknowledging there may be considerable interdimensional overlap, especially among responsiveness, as. More recently, sure to create a conceptual framework of e-service quality and identified 11 dimensions of perceived e-service quality. Although service quality is an important variable in services marketing, it appears to measure specific aspects of service provision by an enterprise, rather than an affective customer important outcome of any customer service interaction (Gilmore, 2003) and atechnology use (Gilmore, 2003; Wickens, Gordon, & Liu, 1998). In this light, Iliterature on customer satisfaction and its related Perhaps one of the most important, controversial, and widely-researcheservices marketing literature is service quality, a construct that has been ucustomer satisfaction. Service quality is defined as the congruence between cuscale, which has been widely adopted as a service measure in industry and has the service quality literature until the present time (Gilmore, 2003; ParasuramaBerry, 1994). However, SERVQUAL has been challenged on a number of grocontinued to refine and evaluate their scale, asserting its psychometric qualitthat surance, and empathy (Parasuraman, Zeithamal, & Berry, 1994, p. 221)Zeithaml, Parasuraman, and Malhotra (2000) modified the SERVQUAL mea 24


response to service. Therefore, customer satisfaction may be a better operationalization of the le of interest in the present study. Cureceived extensive attention in the services literature. Wirtz and Lee (2003) indicated that satisfaction, like service quality, is customers evaluation of a product or service relative to their expectations. However, the reler satisfaction and elopment. Service quality is defined in the literature as the ability of an organization to meet or exceed customer expectations. Customer expectations have been defined as the desires or wants definition of ality management and iterature have been operationalized as predictions of service performance while expectations in the service quality literature are interpreted as what the service provider should offer. Different etween and within f different theoretical olls, Gilbert, & Roslow, 1998). Wirtz and Lee (2003) reviewed nine commonly used satisfaction measures and found that the distinction between affective and cognitive measures was not supported empirically. Instead, they showed that the highest quality scales (as demonstrated by their high satisfaction loadings, high reliability, and low error variance) comparably measured satisfaction for services customer variab stomer Satisfaction Customer satisfaction, a construct similar to service quality, has also ationship between satisfaction and service quality is ambiguous: There has been considerable debate about the concepts of customservice quality. Each has its own research background and theory devof consumers or what they feel a service provider should offer. Thisexpectations differs from the way the term is used in the service quthe consumer satisfaction literature. Expectations in the satisfaction l authors use different meanings for the concept of expectations both bsatisfaction and service quality studies (Gilmore, 2003, p. 23). The nature of satisfaction also has come into question. A number oand operational definitions have been proposed (Babin & Griffin, 1998; Nich 25


considered primarily hedonic (e.g., ice cream restaurant) and utilitarian (e.g., ATM machine). e increased empirical technologies, al. (2000) examined operational antecedents of customer satisfaction with 514 callers to call centers representing 15 industries. The researchers found that center responsiveness to caller goals (service level) was eed of answer, goal mount of time to resolve h customer satisfaction. However, stepwise regression analysis indicated that only percentage of goal achievement with first contact and average call abandonment rate predicted customer satisfaction, 2n, Cummane, and tisfaction. In their d satisfaction with the service center were significant predictors of satisfaction with the call center. In general, callers indicated they were less satisfied with call center service than face-to-face service delivery, and 43able. Although the not provide data to A similar construct, user satisfaction, has been a measure of affective response to computer technologies in the human factors literature. Nielson (1993) proposed that user satisfaction is one of five factors that define usability: learnability or how rapidly a user learns to use a system, efficiency or the extent to which a system supports user performance, memorability or the extent to Recently, customer satisfaction in call centers has begun to receivattention. This research is of particular relevance to speech-based self-service because such systems often replace human call center operators. Feinberg etpositively correlated with customer satisfaction, and several variables (spachievement with first contact, average call abandonment, work time or acaller problem, and percentage of blocked calls) were negatively correlated witaccounting for minimal variance in customer satisfaction (R=0.05). BenningtoConn (2000) used stepwise regression to determine predictors for customer sastudy of an Australian call center, callers quality expectations, rating of staff, an % of the sample indicated they would use a different service if it became availauthors suggest that customer loyalty was associated with satisfaction, they did support this assertion. User Satisfaction 26


which appropriate system use can be recalled, errors or incorrect actions performed during system satisfaction is verall assessment of the extent to wh. The operational definitions of user satisfaction and customer satisfaction appear to have much in common, although they seem to be conceptualized somewhat differently. In services evant construct faction as only one (like the remaining usability categories) may range on a continuum from greater to lower priority, depending on the task and user (Wickens, Gordon, & Liu, 1999). In keeping with these theoretical orientations, ttention in services fined and measured in the y & Salzman, 1998a, 1998b; Olson & Moran, 1998). In general, satisfaction has cross-disciplinary support as a central variable in various types of interaction, making it clearly relevant to the affective outcomes of speech user interfaces in e-service. ntified that impact is largely based on ll as their newest incarnation, e-service (Rust & Kannan, 2002). From a marketing perspective, understanding of the user is vital to the effective positioning of new goods and services as they enter the marketplace. A less developed stream of research is concerned with characteristics of an e-service interface that influences users subsequent attitudes and behaviors. Although most of this research has been use, and the subjective factor of satisfaction. Like customer satisfaction, usertypically measured using rating scales that ask users for their o ich a system pleased them (Nielsen, 1993; Wickens, Gordon, & Liu, 1999)marketing, researchers have asserted that perceived satisfaction is the only rel(Gilmore, 2003). Conversely, Nielson (1993) conceptualized subjective satisfactor in a broader construct of usability, and others suggest that satisfactionsatisfaction has received extensive theoretical and empirical measurement amarketing (Gilmore, 2003), but the construct has not been consistently de human factors literature (Gra Antecedents of Customer Satisfaction In the scant but emerging literature, a number of variables have been idecustomer satisfaction with self-service technology. The bulk of this literatureresearch efforts in the services marketing and management literatures, as we 27


conducted with visual interfaces (e.g., Internet webpages), previous empirical efforts illuminate a pahat impact speech owever, previous research efforts point to a number of personality variables that may impact use of speech interfaces. One of the most informative empirical sources of personality variables is e-service model of self-service tors, including t demographic variables (e.g., gender, age, education, socioeconomic status) are the least relevant consumer traits because most individuals are familiar with basic technologies regardless of group membership. In king, need for interaction emographic or n is at the heart of consumer attitude formation and behavior intentions (p. 187). Results of their structural equation modeling indicated that self-efficacy, inherent novelty seeking, and self-consciousness moderated relationships between riables. Colby to adopt and embrace ld be defined as anything that is (1) cutting-edge and (2) removes a significant part of the human element from a product or service it replaces (p. 27). Parasuraman (2000) developed the Technology Readiness Index (TRI), a 36-item scale that measures two drivers (Optimism, Innovativeness) and two inhibitors (Insecurity, Discomfort) of technology adoption. Subsequent research with the scale th for research in speech or auditory interface design as well. Individual Differences of Users. Individual difference characteristics ttechnology use have not been systematically explored by researchers to date. H research. Dabholkar and Bagozzi (2002) present and evaluate an attitudinaltechnology in which consumer traits (individual differences) and situational facperceived waiting time and social anxiety, serve as moderators. They argue thacontrast, personality variables such as self-efficacy, inherent novelty seewith a service employee, and self-consciousness are of greater interest than dpsychographic factors because such variatio ease of use, fun, performance, and participant attitude. Other researchers have also examined technology-based personality va(2002) describes technology readiness as the propensity of an individual new technology for personal use or at work. The technology in question wou 28


demonstrated that its scores effectively differentiated consumers groups (Colby, 2002) and their f technology-based technologies (e.g., cable television, answering machine, caller ID, cellular phone, desktop computer) and did not involve speech-based e-service interfaces. to technology use, ity to take the ess of a conversation. Because speech interface use mimics human-human conversation, a users skill in picking up subtle social cues provided by the systems intonation, emphasis, or lexical choices may social skills may also ell-studied variable social environment and plan their own behavior accordingly (Snyder & Cantor, 1980). High self-monitors respond to social norms and have been found to adapt quickly to social situations, ractive partners, and monitors (Fiske & he role of self-monitoring in responses to product brands, advertising, and other forms of consumptive behavior (Allen, 2002; Auty & Elliott, 1999; Czellar, 2003; DeBono, Leavitt, & Backus, 2003; OCass, 2000; Snyder & Gangestad, 1986). Lehman and Winer (1997) argue that a general personality approach has relevance for market analysis and planning: Given the relatively limited rates of ownership of technology-based products and services, usage rates oservices, and perceived desirability of future technology-based services (Parasuraman, 2000). However, this research was mostly concerned with Internet and home-based Beyond these user characteristics thought to relate specificallyHoltgraves (2002) argues that personality may also influence individuals abilperspective of a communicative partner into account, which influences the succinfluence how he or she responds to the system. In this sense, users differentially impact how an individual perceives an interactive system. A wknown as self-monitoring refers to the tendency of individuals to respond to their communicate a wider range of emotions, initiate conversation, seek out attpoint to situational factors to explain their own behavior more than low self-Taylor, 1993; Snyder, 1974, 1979). More recently, the consumer marketing field has started to investigate t 29


predictive power of demographic and socioeconomic variables, the fact that many people in basis for segmenting urprising that an attempt has been maAs Snyder and Gangestad (1986) point out, high self monitors are highly responsive to social and interpersonal cues of situationally appropriate performances (p. 125). Applying this ore sensitive than ehavior. In addition, e more successful with a speech interface because they quickly adapt their interpersonal behavior to the interfaces interaction style. Thus, self-monitoring may impact perceptions of a speech interface and the er behaviors with taped samples of a speech interface before it is designed, to give stakeholders and other observers a better understanding of how the final system will sound to the intended user group (e.g., an audio in implementing may not perceive a y users do. This issue has not been addressed in applied work. Hoteractants and listeners observers differ in their perceptions of, cognitions about, and memory for conversational interactions, as shown by a number of studies that use interactants and observers as a between-subjects independent variable. Conversational interactants (speakers) and observers may perceive incoming information marketing are trained in psychology, and the natural desire to find a general consumers that will be useful over many situations, it is not s de to use personality traits as a basis for [market] segmentation (p. 101). characteristic to interaction with a speech interface, high self monitors may be mlow self monitors to an interface designed to imitate prototypical service bbecause self-monitoring predicts social behavior, high self monitors might b enterprise for which it provides service, and even be associated with specific usthe interface. Point of View. A common current practice in speech interface design is to provide audio clip of a simulated interaction may be played for an executive who is interestedspeech technology). This practice may be problematic because observers speech interface in the same wa wever, the previous psychological literature indicates that conversational in 30


about target persons differently. Burgoon and Newton (1991) showed only limited agreement Geers, and Apple e information in four sing observed behavior into units of action. Variations in rate, pattern or both indicate perceptual effects. The researchers demonstrated that individuals anticipating a speaker role parsed observed behavior into o showed decreased sed affect as compared that potential speakers adopt a processing strategy that allows a clear, concise summary of information for transmission to others. Indeed, a variety of studies have suggested that conversational interactants 9; Inman, Reichle & enoit & Wilke, ich, 1989; Miller & deWinstanley, 2002) are different than observers. However, observers may have a more accurate perception of a speech-based e-service system because they have more cognitive resources with which to make the required social judgments than thodological ose of actual rvice attitudes (Dabholkar, 1994, 1996; Dabholkar & Bagozzi, 2002). Similarly, in previous speech interface research, audiotaped user-system interactions have been used as stimuli (Polkosky, 2003). The use of an audiotaped, simulated interaction would appear to be more similar to an actual interaction with a speech system than a written narrative because listeners hear the actual voice of the system, between participants and observers in judgments of nonverbal cues. Lassiter(2002) examined perception by individuals expecting to transmit or receivstudies using a behavior unitization technique. Unitization is a method of parfewer units than individuals without this expectation. Further, speakers alsconfidence in their trait impressions of targets, poorer memory, and decreawith control and observer groups. Lassiter, Geers, and Apple (2002) explainedcognitions (Burgoon & Newton, 1991; Gilovich, 1987; Guerin & Innes, 198Baron, 1993; Lerner & Tetlock, 1999; Monahan, 1995) and memory (Benoit, B1996; Frank & Gilov do users (Patterson, 1996). In service research, this issue has not been recognized as a potential meproblem. Researchers have assumed that observer judgments are similar to thconsumers and have used written narrative scenarios to evaluate models of e-se 31


its prompts, and a users response to those prompts. In interpersonal communication studies, irect, non-interacting Ryan, & Bradac, 1968; Schlegoff, Jefferson, & Sacks, 1977; Schlegoff & Sacks, 1973; Trafimow & Wyer, 1983; White & Carlston, 1983). Although an observer method has been used in the disciplines that inform the present study, sually replace human nsumers have expectations about the service provider role and these expectations have implications for the success of the interaction (Solomon, Surprenant, Czepiel, & Gutman, 1985) and customer satisfaction ersonality and eech user interfaces. e of the most important design issues in the adoption of speech technology by enterprises (Cohen, Giangola & Balogh, 2004; Kotelly, 2002). Indeed, the literature suggests that discrepancies between users havior) and their actual y and behavior that n service providers, employee selection is a critical factor in customer satisfaction with service delivery. In their meta-analysis of performance in jobs requiring interpersonal interaction, Mount, Barrick, and Stewart (1998) found that Conscientiousness was most strongly related to performance for dyadic service providers (r=.29), followed by Openness to Experience (r=.17), Agreeableness (r=.13), researchers have also used audiotaped telephone conversations and other indobservational methods of evaluating partners in conversation (Cargile, Giles, 1994; Kreuz & Roberts, 1993; Sacks, Schlegoff, & Jefferson, 1974; Schlegoff, the impact of point of view warrants empirical attention. Personality and Behavior of Service Providers. Speech interfaces uservice providers for constrained or repetitive interactions. Because co(Parasuraman, Zeithamal, & Berry, 1985, 1988, 1994), findings on desirable pbehavior characteristics of service providers provide a model for designing spIn the speech industry, finding the correct persona of an interface has been on internalized script for customer service (including the service providers beinteraction may influence cognitive processing and affective responses. Past research has analyzed aspects of human service provider personalitsubsequently influence customers attitudes and behaviors. Thus, with huma 32


Emotional Stability (r=.12), and Extraversion (r=.07) when these personality traits were measured rsonal Characteristics Inventory, a measure of the dimensions of the Five-Factor Model of peave more narrowly defined variables thought to impact customer satisfaction. Cran (1994) defined customer service orientation as a set of basic individual predispositions and an inclination to provide service, to be ersonal skills, or a aird, 1968, p. 503) have prenant, Czepiel, & Gutman, 1985). To investigate this variable further, Alge et al. (2002) measured personality characteristics and performance of 115 bus operators in a metropolitan transit authority. on (r=.65), general iated with supervisor pport for the relationship between interpersonal competence and performance, interpersonal skills may be more strongly associated with customer attitude and performance in jobs that require interaction as a prole and expected ervice orientation s, and helpfulness) Yagil (2001) examined service providers use of ingratiation and assertiveness (influence tactics) and found that ingratiation increased customer satisfaction but assertiveness decreased satisfaction. The study was concerned with repeated service encounters, or service relationships, in 115 customer-provider dyads in several industries (e.g., banking, insurance, government). Yagil by the Pe rsonality. In contrast to broad measures of personality, other researchers hcourteous and helpful in dealing with customers and associates (p. 36). Interpservice providers acquired ability for effective interaction (Holland & Bbeen shown to impact customer satisfaction positively (Humphreys, 1996; Solomon, SurThey found that interpersonal skills were significantly correlated with extraversidisposition (r=.39), and self-reported performance (r=.31), but were not assocratings or other performance indicators. Despite relatively weak empirical su imary activity. Baydoun, Rose, and Emperado (2001) note that the general rbehavior is relatively clear: In general, the existing measures of customer sinclude similar subscales (e.g., interpersonal skills such as friendliness, politenesand most tend to emphasize an individuals overall energy level (p. 618). 33


(2002) speculated that the relationship between service providers influence tactics and customer ertive behavior vey a degradation of the service. On the other hand, the ingratiatory behavior of the service provider, which is deliberately designed to please the customer, grants the customer a respectable status, enhances his or her self-esteem, and results in satisfaction (p. 350). This interpretation is consistent with a broad litrown & Levinson, 1987; Price, Arnould, and Tierney (1995) proposed a framework that describes service encounters in terms of their duration, affective content, and spatial proximity. The authors found customers e encounter Price, d be appropriate for speech technology or self-service in general (e.g., extended, affectively charged tour guide-traveler relationships). Despite this, the variables they identify show some similarity to constructs in other el and the use of mer interaction. ework to evaluate nters at hotel front desks. They found that ratings of displayed emotions were positively correlated with customers evaluation of the service encounter and their overall assessment of the hotel. In addition, whereas demographic (ethnicity, gender) and duration of the interaction failed to add incremental variance in a hierarchical regression analysis, customer mood and rating of displayed emotions did account satisfaction was mediated by customers self-esteem: A service providers assmight be interpreted by the customer as reflecting a lack of respect; it may conthe customers status and thus lead to a general sense of dissatisfaction withconsequently erature on power and politeness, which are conveyed linguistically (BHoltgraves, 2002). that these three predictors influenced the service providers behavior, as well asaffective response and service satisfaction. However, the specific type of servicArnould, and Tierney (1995) studied was not the type of encounter that woul domains, most notably the emphasis on affective variables throughout the modinterpersonal distance (proxemics) as a moderator of the service provider-custo Mattila and Enz (2002) used Price, Arnould, and Tierneys (1995) framaffective response to and customer evaluation of service delivery in brief encou 34


for unique variance in the service encounter evaluation and overall hotel evaluation. This work in service encounters, and fuctions. enbeck (2003) define trust in an online environment as an attitude of confident expectation in an online situation of risk that ones vulnerabilities will not be exploited (p. 740). Trust functions as a cognitive ter & Kaluscha, 2003). This ns (e.g., risk, rmational or transactional websites. The online environment is similar to more typical human social situations in which individuals exhibit trust, including interpersonal interaction (Corritore, Kracher, & W construct that has been 2003; Corritore, Krasher, and Wiedenbecks (2003) online trust model does not specify the measurement of the various constructs and is not empirically validated. Their definition of trust Grices (1969) maxims, particularly the Quality and Rede only true and tion through speech a speech systems voice that may impact a users trust in the system. Comfort. Spake, Beatty, Brockman, and Crutchfield (2003) argued that the psychological construct of comfort, or a psychological state wherein a customers anxiety concerning a service has been eased, and he or she enjoys peace of mind and is calm and worry free concerning service draws from a research base suggesting the influential role of mood rther indicates the relevance of mood to even brief, everyday service interaTrust. In a recent review of the literature, Corritore, Kracher, and Weidheuristic to simplify complex, uncertain environments (Grabner-Kraumulti-dimensional construct occurs in the presence of several common conditiovulnerability, expectation, confidence, and exploitation), especially in info eidenbeck, 2003). Researchers identify trust as a complex, multifacetedoperationally defined in numerous ways (Corritore, Kracher, & Weidenbeck, Grabner-Krauter & Kaluscha, 2003). appears to overlap at least partially with lation maxims. These two maxims require that conversational partners provirelevant information. Trustworthiness is also conveyed in interpersonal interacand vocal cues (Polkosky & Lewis, 2003), which points to at least one aspect of 35


encounters with this provider (p. 317), aids in the selection, development, and maintenance of tial scale and ommitment, and ease of els of affective variables and demonstrated that a model with comfort provided the best fit and explained more variance than alternate models, adding incremental understanding to the more common constructs to be a specific n, a speech act consists of meaning, speaker intention, and a consequence on the hearer (perlocutionary dimension). In a service encounter, comfort may be viewed as the desired psychological effect of service on a er. When service is provided by a speech interface, it seems reasonable that comfort may alsman and Crutchfields ased customer satisfaction. Ease of Use and Usability. Several researchers, especially in the service marketing nology and iedenbeck, 2003; of ease of use and how complicated, confusing, effortful, time-consuming, and reliable a self-service technology was. The scales were again used in Dabholkar and Bagozzi (2002). Zeithaml, Parasuraman, and Malhotra (2002) identified this construct using two dimensions of e-service quality, which they referred to as Ease of Navigation and Efficiency. In contrast, Nielsen (1993) argued that usability is composed of service relationships. They developed an eight-item 10-point semantic differendemonstrated that comfort is positively correlated with satisfaction, trust, ccomplaining when problems occur. They also evaluated three structural mod studied in the marketing literature. In addition to the empirical support for comfort, the construct appears example of Austins (1962) perlocutionary dimension. According to Austicustom o be the desired effect on the user. In both cases, Spake, Beatty, Brock(2003) study suggests that increased comfort will be associated with increliterature, have defined constructs relating to the usability of a self-service techincorporated them into models of service delivery (Carritore, Krasher, & WDabholkar & Bagozzi, 2002). Dabholkar (1994) developed a six-item measurefour-item measure of performance, which tapped consumers perceptions of 36


several components, including ease of learning, efficiency, ease of recalling the interface, and few of the greatest en in the human factors atic and controversial (Bevan, 1995; Gray & Salzman, 1992, 1998a, 1998b; Hartson, Andre, & Williges, 2001; Hassenzahl, 2001; Hertzum & Jacobsen, 2001; Olson & Moran, 1998) and a universally e Gricean (1975) maxim ilar to Bernsen, Dybkjaer, and Dybkjaers (1999) cooperative principles to avoid superfluous or redundant interactions (p. 122) and be clear and brief. A general ease or cooperative construct appears to y have been investigated in the previous literature. They are positively associated with behaviors such as repeat use of self-service technologies and intent to purchase (Gilmore, 2003). or the behavioral of similar brands, over a oyalty as a and finally to behavioral intention and action-based loyalty (Gilmore, 2003). Rust and Kannan (2002) point out that, although the e-service environment may hinder customer loyalty due to the ease of comparison shopping and lack of human interaction, it is these features that may also permit greater customer loyalty when they are part of an appropriate customer service strategy. They assert that user errors. Wickens, Gordon, and Liu (1998) assert that usability is oneconcerns for those working on software interface design (p. 453). Evliterature, definitions and measures of usability have historically been problem acceptable definition and measure of usability continue to be elusive. The concepts of ease of use and usability bear some similarity to ththat contributions should be brief, orderly, and to the point. They are also sim have broad theoretical support across disciplines. Consequences of Customer Satisfaction The consequences of customer satisfaction and perceived service qualit An important consequence of customer satisfaction is customer loyaltyoutcome of a customers preference for a particular brand or a selectionperiod of time (Gilmore, 2003, p. 24). Most researchers have conceptualized ldevelopmental process that evolves from cognitive loyalty to affective loyalty 37


customer equity manifests itself in e-loyalty outcomes, both attitudinal (repeat purchase intention, al (repeat-purchase ending) (pp. distinct variable that influences customer expectations, attitudes, and behaviors. A positive association between service quality and word of mouth has been demonstrated empirically (Bof the major affective outcomes of interaction with a speech-based e-service. There are several important but uninvestigated antecedents that may influence a users satisfaction when customer service is om script-based int of view. vice experience may influence longer-term attitudes and behaviors toward businesses, as well as have implications for the behavior of customers and other potential customers with whom they interact. Although ma extremely limited, e relevant The purpose of the current research was to begin to examine the relationships among user characteristics and perceptual and affective outcomes of speech-based e-service. A review of the literature indicates that several disciplines have been concerned with aspects of speech user interfaces when they are used to provide customer service. However, these literatures have rarely willingness to share information, and positive word of mouth) and behaviorbehavior, frequency of visits, cumulative sales, and e-tailer share of customer sp14-15). Harrison-Walker (2000) has also recently highlighted word of mouth communication as a oulding, Kalra, Staelin, & Zeithaml, 1993). Summary of Potential Speech-Based e-Service Constructs Taken together, several literatures suggest that customer satisfaction is one provided by a speech interface, including perceptions of the interface derived frknowledge of human service and conversation, individual differences, and poCustomer satisfaction and other affective responses to this modern customer ser ny variables are implicated in this emerging field, the empirical work has beencreating a need for research to further elucidate the interrelationships among thconstructs. Overview of the Current Study 38


interacted, have generated a broad range of theoretical variables, and disparate findings have not definitions and with similar constructs, which maology. Therefore, the current research uses two studies to address three main categories of antecedents to customer satisfaction in speech-based e-service. In the first study, expert judgments r the second study. variables (need for elf-monitoring). These variables were selected for further investigation because they each have been implicated by several studies and even across disciplines that inform the present research. In addition, a broad set behavior and aluated as potential ites a multi-factor construct (Nielsen, 1993). Finally, comfort and customer satisfaction also are measured in the second study to determine their relationships to the individual and interface perceptual variables. interface usability, cknowledged that ons for customers overall perceptions of service from an organization (Ostrom, Bitner, & Meuter, 2002; Rust & Kannan, 2002). However, most of this research has been concerned with service delivery via the Internet. Therefore, the current study identified a subset of perceptual and affective interface variables that are related to and predict customer satisfaction for speech technology interactions. been systematically applied to speech technology. Despite discipline-specificempirical treatment, many of the implicated fields are concerned kes them potential candidates for investigation in applied speech technare used to assess the quality of several speech interfaces and select stimuli foThe second study uses student observers to collect data on, individual differenceinteraction with a service provider, inherent novelty seeking, age, gender, and sof interface antecedents are measured, based on the literature concerned withpersonality of a service provider, trust, and ease of use. These variables were ev ms for a usability scale for speech interfaces, with usability broadly defined aLittle is known about the relationships between individual differences, comfort, and customer satisfaction with automated systems. It is generally ae-service interfaces may have broad implicati 39


Rationale ons for conducting the ech and linguistic ing an e-service encounter helps designers predict the success of their interfaces before they are commercially deployed. Such knowledge is likely to promote more usable and satisfactory designs in the by predicting customer tion from interface characteristics, enterprises that use a speech user interface for their cutechnology on their customers. On a theoretical level, this research begins to integrate separate literatures that are tly, the interpersonal arallel, with s do acknowledge the importance of speech technology as a mechanism for future e-service delivery (Rust & Kannan, 2002). Similarly, the findings from interpersonal communication and social-cognitive psychology haservice, and hue of variables in a melevant application for social and cognitive psychologists who study psycholinguistics for two reasons. First, experimental research has illuminated general principles about processing, representation, and interaction that can be applied directly to explaining, predicting, and improving human-computer interaction. Without such underlying principles, progress in interface There were several practical, theoretical, and methodological reascurrent studies. From a practical perspective, an understanding of the speperceptual variables that impact customers social cognitions and affect duremerging and highly competitive field of speech technology. In addition, satisfac stomer service may have clearer expectations about the impact of this simultaneously addressing similar issues from different perspectives. Currencommunication, e-service, and speech technology literatures are advancing in prelatively little interaction among them. However, e-service researcher ve had limited penetration into the applied fields of speech technology, e-man-computer interaction. Thus, the current studies begin to integrate a ranganingful way. Brennen (1998) asserts that [the] domain of human-computer interaction is a particularly re 40


design will be ad hoc at best, especially for multimodal, intelligent systems that use rovides an ideal rting models from dying such models in software has the potential to bring additional clarity and pragmatism to these fields (p. 222). minary cross-fertilization of these disparate literatures will help promote a more unified peisciplinary applied Finally, this research also presents several methodological advances in using correlation and regression for analyzing human-computer interaction via speech. Although such analyses have y, and other applied latively limited use d range of methods applied to communication research has left relevant questions unanswered by the empirical literature. Thus, the current research uses variables and statistical analyses borrowed from other ilding in speech this research offers the potential for another methodological advance: the development of a scale for assessing the usability of speech user interfaces. No empirically-validated scales currently exist for speech interface evaluation, which presents a significant measurement problem that limits both theory development and practical understanding of these technologies. speech and language. At the same time, human-computer interaction ptestbed for demonstrating and testing models and principles Transposocial and cognitive psychology to electronic communication and embo Preli rspective, which may sharpen theoretical perspectives and encourage transdresearch. been effectively utilized in education, industrial-organizational psychologfields to address social-cognitive questions (Pedhazur, 1997), they have had rein human factors or interpersonal communication research. This restricte applied fields to promote a stronger methodological approach to theory-butechnology. Finally, 41


Research Questions rrent studies is the mer satisfaction. The itoring, need for interaction with a service employee, and inherent novelty seeking. These variables have been identified in the services marketing and psychological literatures as influencing individuals erface perceptual agmatics, nterfaces. Finally, customer satisfaction and comfort, a psychological variable that is defined similar to perlocutionary force in the pragmatics literature, were also considered. Extending analyses targeting associations ence and perceived n of affective outcomes such as customer satisfaction and comfort from perceived usability variables. Because customer satisfaction has been a central variable in the customer service literature, it seems that this variable will continue to bellow hypotheses ge developed: usability of speech interfaces? Question 2: How are individual difference variables (e.g., demographics, self-monitoring, need for interaction, inherent novelty seeking) related to comfort and customer satisfaction? Question 3: What speech user interface characteristics best predict customer satisfaction? Question 4: What speech user interface characteristics best predict comfort? Based on the previous literature, the primary issue of interest in the cuinterrelationships among individual difference, usability, comfort and custopresent research focuses on several individual difference variables: self-mondecisions about product and service consumption. In addition, a number of intvariables were considered, including aspects of voice and speech production, prorientation toward users goals, customer service behaviors, and trust of speech i among these variables, the potential interaction between the individual differusability variables was also analyzed using analysis of covariance. The second major issue of interest is the predictio a practical focus as service increasingly is provided by technology. The previous literature does not provide adequate background to a neration for the current research. Therefore, several research questions were Question 1: What items should be measured in assessing the 42


Question 5: Do individual difference variables interact with speech user interface characteristics to crlity of several speech user interfaces and prepare stimuli for Study 2. The use of expert judges with background in psycholinguistics is due to several issues in the speech technology field. First, there is some ave a necessary set have extensive h interface, due to their own personality characteristics (Capretz, 2003), a lack of specific knowledge about human communication, or potential biases related to the constraints imposed by the languages used to code d or empirically validated ot study, two scales scales demonstrated several weaknesses, specifically, limitations in scope and small effect sizes (Polkosky, 2003). Finally, a variety of guidelines exist for interface development (Balentine & promote variability ross speech applications. The first study uses expert judgments of several speech interface-user interactions to determine their relative quality. Based on their ratings of speech and language characteristics, six interfaces representing the best, average, and worst quality interfaces will be selected for use as stimuli in Study 2. eate customer satisfaction? Study 1: Speech Interface Expert Ratings The first study used experts in speech and language to evaluate the quaacknowledgment in industrial settings that experts in communication behavior hof skills for the design of effective speech interfaces. Second, engineers who knowledge of the underlying technology may hold biased perceptions of a speecthese interfaces (e.g., VoiceXML). Third, there is no universally acceptemeasure of usability that specifically applies to speech technologies. In a pilwere developed and used to evaluate the usability of speech interfaces, but both Morgan, 1999; Cohen, Giangola, & Balogh, 2004; Kotelly, 2003), which mayin quality ac 43


Method Pa speech and d Disorders programs at the University of South Florida. The participants included two Ph.D. level psycholinguistics researchers, three Ph.D. level speech science or speech-language pathology researchers, and one e had specific expertise ns were emailed to a known t sample, data were collected by a web survey tool that does not allow identification of participants. Therefore, the data could not be specifically matched to their provider. Measures on the review of literature faces. The items ing, use of emphasis, voice naturalness, use of wide intonation variation, effective response to communicative breakdowns, contingent, specific responses to user, pragmatic appropriateness, consistency with conversational expectations, deference to user, liking of interface, ease of conversation with this interface, overall quality of interface, and similarity to human-human conversation. Each item was presented in random order for each stimulus and for each participant. Participants rated each item on a five-point scale, ranging from 1 (poor) to 5 (excellent). rticipants The participant sample consisted of six subject-matter experts in humanlanguage, recruited from the Psychology and Communication Sciences anPh.D. psychology student who is also a speech-language pathologist. Nonin speech technology or speech interface design. Although invitatio exper Fifteen items were used to evaluate each speech interface basedand speech and language characteristics that appeared to vary among the interincluded: politeness, appropriate use of indirect vs. direct requests, use of paus 44


Table 2. Speech Interface Stimuli f System-User Interaction ength(min:System/Caller Voice Type o LInteraction of sec) Retail Or 9 Retail Purch 09 Credit D 2 Phone Bi 41 Car 36 Directory Dialer 4 Tennis Scoreboard 0:38 Male/male Available Flights 2:27 Female/female der 1:5Male/male ase 2:Male/female ispute 1:4Male/male Package Delivery 2:26 Female/male Flight Status 2:36 Male/male ll 3:Female/male Rental 2:Male/female 0:3Male/female Patient Claim 1:14 Female/male Movies 3:22 Female/male Report Electrical Outage 0:44 Male/male scription Refill 0:38 Female/male Fi Pre nancial Services 3:48 Female/male Trade Stock 0:51 Male/male Stimuli The stimuli consisted of 16 digitally-recorded audio files of speech interface-user interactions created at IBM and its competitors (Scansoft, Nuance). The interfaces were selected to 45


represent a diverse range of development stages (e.g., prototype and commercially deployed marizes the general -system interaction appear in Apcts. Audio files of simulated telephone calls by a hypothesized user to a speech system (known as vision clips because they provide a vision of what the completed system will sound like) to y solutions or to velopment. Vision resenting prompts with the simulated responses of a recorded caller. This technique allows stakeholders to hear a sample system-user dialogue before a speech system is actually developed. To the observer, vision a real user and a cause both depict user-system dialogue. Because vision clips have a very pr empirical attention, the current studies used vision clips as a source of stimuli, in addition to actual recordings of user calls to working systems. ings, each depicting a erators prompts and a e. Participants were recruited from the Psychology and Communication Sciences and Disorders Departments at the University of South Florida. Each participant received an email invitation, which briefly explained the study and provided a link to a webpage. The participants were randomly divided into two groups to receive one of two webpages (each webpage presented systems), durations of interactions, system voices, and industries. Table 2 sumcharacteristics of the audio clips; complete transcripts of each user pendix A. The stimuli were presented in two orders to control for order effedemonstrate a speech user interface are frequently used to sell speech technologvalidate an interface design with its financial sponsors in the early stages of declips are typically created by editing together recordings of a professional voice pclips cannot be differentiated from recordings of an actual interaction between deployed system be ominent role in the speech technology industry and have received minimal The stimuli for each participant consisted of a series of 16 recordsingle telephone call to a synthetic speech system. Participants heard the opcallers responses as if they were listening to the call on another telephone linProcedure 46


the stimuli in different orders). Participants clicked on the link in the email and then used an pleting the rating matically uploaded to an IBM-based server using a web survey and data collection tool (WebSurv1). Results A statistical software package (SPSS 11.5.1) was used for all analyses. The purpose of the 16 potential stimuli for Study 2. Analyses scriptive statistics (e.g., central tendency, dispersion, type of distribution) and a relDescriptive statistics Visual inspection of the boxplots for each rating item indicated normal distributions for ributions for 12 items ntonation, Humanlike servation occurred for the Communication Breakdown item, which was characterized by minimal variability and neutral ratings (M = 2.99, SD = .79, 95% CI = 2.82-3.16). Participant comments consistently quate variability in were used to calculate the descriptive analyses for each ce. own in Table 3. Overall means for each interface suggest that the Financial Services interface was most positively rated and Directory Dialer was most negatively rated. Reliability Reliability of the expert ratings was analyzed using all items and composite means with a Internet browser to access the audio stimuli and questionnaire items. After comitems, participants data were auto eyor 4. Study 1 was to obtain expert quality ratings of consisted of de iability analysis on the composite mean of the 15 items. two items (Quality, Liking) and approximately normal, slightly skewed dist(Directness, Naturalness, Deference, Politeness, Contingency, Pragmatics, IQuality, Consistency, Emphasis, Pausing, Ease of Use). The exception to this ob referred to the lack of breakdowns shown in the stimuli, further suggesting inadethe stimuli for this item. The items ratings interfa The composite means and standard deviations for each interface are sh 47


two-way random effects model intraclass correlation coefficient (Shrout & Fleiss, 1979). This ct and belong to a larger population of Overall Ratings of Quality for 16 Speech Interfaces f System-User Interaction M SD Low High model is used when judges are considered a random effe Table 3. ExpertsType o Prescrip tion Refill 1 5 pute 1 5 rder 2 5 Retail Purchase 3.73 0.81 2 5 Trade Stock 3.70 0.99 1 5 ient Claim 2 5 trical Outag 1 5 eboard 1 5 Movies 2.99 1.17 1 5 Package Delivery 2.94 0.69 2 5 lights 1 4 Phone Bill 2.78 0.86 1 5 Directory Dialer 2.40 1.00 1 5 3.86 0.83 Credit Dis 3.79 0.92 Retail O 3.78 0.92 Pat 3.55 0.69 Report Elec e 3.39 1.11 Tennis Scor 3.37 0.88 Available F 2.79 0.90 Financial Services 4.22 0.83 2 5 Car Rental 2.72 0.97 1 5 Flight Status 2.66 0.86 1 5 judges (Shrout & Fleiss, 1979). The results indicated that the single rater intraclass correlation coefficient was poor, ICC (2,1)=.37, suggesting that individual raters had minimal reliability 48


across items. However, the average rater statistic was acceptable, ICC (2,6)=.78, suggesting that s. As a result, only group mean ratings (as able 3) were considered in selecting Study 2 stimuli. StiThe goal of Study 1 was to evaluate the quality of a sample of speech user interfaces according to expert judges and identify stimuli for Study 2. Although the data showed poor e. In addition, terface quality refore, six interfaces were selected for inclusion in Study 2 to represent the best (Financial Services, Prescription Refill), worst (Flight Status, Directory Dialer), and average (Tennis Scoreboard, Movies) interfaces. s servers perceptual ttings, observers (such as enterprise decision-makers and other stakeholders) often judge an interface and determine the merits of the design based on their perceptions of a recorded system-user dialogue. Therefore, this study generated data using a common observer approach to interface evaluation in the speech an factors work in this The study employed a correlational design, using observers who listened to audiotaped recordings of user-system dialogues and complete rating scales for a range of affective and perceptual variables. Students were recruited to listen to one of six user-speech system interactions, then provided individual difference data and rated their group means were more reliable than individual rating shown in T muli Selection reliability within individuals, the mean group ratings were notably more reliabloverall mean ratings for each interface showed that the experts perception of incould be quantified along a positive to negative continuum. The Study 2: Observers Affective Responses to Speech InterfaceAs the main focus of the current research, Study 2 is concerned with oband affective responses to interactive speech interfaces. In applied industrial se technology industry, increasing its applicability and importance for humfield. 49


perceptions of the speech system, as well as their comfort and customer satisfaction with thility analyses were ession analyses were used to evaluate associations among individual difference variables (inherent novelty seeking, self-monitoring), usability, comfort, and customer satisfaction. Finally, quality categories derived from Study 1 were used to evaluate potential interactions between observer variables and interface ions. Participants Participants in this study consisted of 862 students recruited using the University of South the study was voluntary, of 20.61 years (SD= 3.78). The sample sizes for the six interface stimuli were approximately equivalent (Tennis Scoreboard, N=133; Directory Dialer, N=137; Flights, N=120; Movie, N=150; Financial Services, participants %, N=818). They most were full-time students (72%, N=619) and psychology majors (56%, N=479). Participants indicated that they used speech interface systems occasionally (N=308), seldom (N=285), or never (N=156), and only 13% of the sample indicated they were expert users (N=114) of such systems. Several analyses were conducted on the various demographic variables to look at the distribution of participants across the six interface (between subjects) groups. These analysis suggested that representatives of e speech system. This study was used to address three main issues. First, factor and reliabused to evaluate items for a usability scale. Next, correlation and regr percept Method Florida (USF) Psychology Departments participant pool. Participation in and the student participants received extra course credit. The sample included 688 females and 161 males, with a mean ageN=115) but the Prescription Refill group was largest (N=206). The majority ofreported normal hearing (97%, N=840) and English as their native language (95represented a very broad range of job roles and educational backgrounds but 50


different genders, ages, educational backgrounds, native languages, job types, and frequency of prative language mary variables of interest in this study. The sample included 816 native English speakers and 44 native speakers of other languages. Independent t-tests demonstrated that the two native language groups were not ovelty Seeking, cales of the Usability agmatics). However, statistically significant difference between the groups occurred for Need for Interaction with a Service Provider (p=0.03) and the Customer Service Behavior ratings on the Usability scale had less need for ving more negative nteraction M=5.30, SD=1.26; Customer Service Behavior M=5.43, SD=1.08). Because the non-native English speakers are more likely to choose technology-based self-service and are part of the general user technology, their data were retained in the sample. A similar independent nalysis was conducted to determine differences between participants with normal hearing (Ns between and their data was also retained in the sample. Procedure The participants were recruited using an automated research tracking tool (ExperimenTrak), which displayed a link to the first page of the online study to eligible students. evious speech technology usage. Several analyses were conducted to determine whether participants n(English vs. other languages) and hearing status effected their ratings of the prisignificantly different (p>0.11) on ratings of customer satisfaction, Inherent NSelf-Monitoring Sensitivity, Self-Monitoring Ability, Comfort, or four subsscale (Speech Characteristics, User Goal Orientation, Hedonic Quality, Pr(p=0.03). For these variables, non-native English speakers reported theyinteracting with a human (M=4.86, SD=1.47) but rated speech interfaces as haservice behavior (M=5.08, SD=1.24) than native English speakers (Need for Ipopulation for speech t-test a =840) and a hearing loss or disability (N=22). No statistically significant differencethe groups were observed for any of the primary variables of interest (p>0.31) 51


The ExperimenTrak interface was hosted on a USF-based server. When students clicked the e website r all participants, only tudy start page, participants clicked an audio file link, which accessed a third-party internet service provider server, and listened to the recorded user-system interaction via .mp3 audio files. The participant then 20 minutes. When the to the IBM server and credit for participation. To prevent individuals from participating multiple times, the ExperimenTrak software was configured to allow participation in only one of the six groups and WebSurveyor was against history the data for all six interfaces were collected ghout the period of August to November 2004. The ExperimenTrak software automatically assigned participants based on internal criteria, which included number of completed surveys and students personal criteria established by the Psychology department. raphic Questions. Information regarding participants gender, age, native language, anC). In addition, a oss. Participants also identified their job type and educational background to allow the assessment of the generalizability of the data obtained. Inherent Novelty Seeking. This variable was measured using the Dabholkar and Bagozzi (2002) four item scale (see Appendix D). Individuals high in inherent novelty seeking tend to survey link, they were directed to a website hosted on an IBM-based server. Thinstructions (Appendix B), questionnaire, and visual display were identical fodiffering in the audio file presented to each of six groups. After accessing the scompleted the questionnaire items, which required approximately 15 toparticipant clicked a Submit button, his or her responses were uploadedthe participant was directed back to the USF ExperimenTrak system to receive configured to eliminate multiple responses from the same IP address. To guardeffects and maintain approximately equal sample sizes, throu Measures Demog d previous experience with speech technology was collected (see Appendix question regarding hearing acuity was used to screen individuals for hearing l 52


look favorably on technology and the use of technology based products, have stronger intrinsic to approach old icated that the scales (performance, intention, fun, ease of use, and attitude), providing some empirical support for the scales validity. Dabholkar and Bagozzi (2002) reported a Chronbachs of 0.72 for their scale. The current results veloped a four-item Bagozzi (2002) in their attitudinal model. Confirmatory factor analysis indicated that the scales formed a distinct factor and moderated relationships between ease of use and attitude and between fun and attitude, e up the scale (see 3 for this scale. abholkar and Bagozzi (2002) scale was borderline (coefficient = 0.68) due to a lower item-to-total correlation for item 3 (r=0.40) than the other two items (item 1 r=0.55, item 2 r=0.60). However, because this was a previously published scale, the fers to the extent to ol their own expression (Snyder, 1974; Snyder & Cantor, 1980). Snyder (1974) introduced the Self-Monitoring Scale but this measure has drawn criticism based on its construct validity, factor structure, and psychometric qualities, leading to controversy and revised versions of the scale (Briggs & Cheek, 1986; Briggs, Cheek, & Buss, 1980; Lennox & Wolfe, 1984; OCass, 2003; Snyder & Gangestad, 1986). motivation to use such products, and enjoy the stimulation of trying new waysproblems (Dabholkar & Bagozzi, p. 188). Confirmatory factor analysis indformed a distinct factor and moderated relationships among other user variables indicated a Chronbachs of 0.85 for this scale. Need for Interaction with a Service Employee. Dabholkar (1996) descale to measure this construct, which was further examined by Dabholkar and which provided some empirical support for the scales validity. Three items makAppendix E). Dabholkar and Bagozzi (2002) report a Chronbachs alpha of 0.8In the present study, reliability for the D item was retained in the scale score for subsequent analyses. Self-Monitoring. An aspect of personality known as self-monitoring rewhich individuals attend to the expressive behavior of others and thereby contr 53


To respond to difficulties in the factor structure of the original scale and the ) presented a 13-item, 1974) scale, the ainly always false to certainly always true. Lennox and Wolfe (1984) argue that their narrow operationalization of self-monitoring provides better construct validity and is more consistent with the original tric qualities of the itoring Sensitivity) for factors and full scale score). OCass (2003) also modified the scale anchors to read strongly agree and strongly disagree due to participant confusion and misinterpretation of the original anchors during pilot ugh Snyder and r both RSMS factors may have been more strongly associated with experienced individuals perception of user responses to speech interfaces than an overall scale score. Therefore, the two factor RSMS appears to operationalize better the variable of interest than the original Snyder (1974) scale. The 12 items -Monitoring Speech Interface Usability. Currently, no measures exist that adequately operationalize usability of speech interfaces. A previous pragmatic scale demonstrated very small effect sizes (Polkosky, 2003) and was restricted in the range of items included, requiring the development of a new measure. I generated 76 items based on a broad review of the literature, including pragmatic operationalization of the self-monitoring construct, Lennox and Wolfe (1984two-factor scale. Unlike the original true/false response format of the Snyder (Revised Self-Monitoring Scale (RSMS) uses six point scales anchored by certtheoretical definition (Snyder, 1974). OCass (2003) evaluated the psychomeRSMS and confirmed a two-factor structure (Self-Monitoring Ability, Self-Monand high internal consistency reliabilities (Cronbachs alphas exceeding 0.85 testing. The OCass (2003) RSMS was used to measure self-monitoring. AlthoGangestad (1986) argue in favor of a single factor construct, in this study, one o of the OCass (2003) RSMS appear in Appendix F. In the current study, reliability for the Self-Monitoring Ability and SelfSensitivity factors was acceptable (Chronbachs 0.85 and 0.83, respectively) 54


theory (Bernsen, Dybkjaer, & Dybkjaer, 1996; Grice, 1975) and measures of customer service and aml, & Barry, 1994; ased on results of er development at IBM (Polkosky, 2002a, 2002b, 2003, 2004; Polkosky & Lewis, 2003). Ten factors assumed to be associated with speech interface usability were used to comprehensively measure the broad range y the previous literature (see Appendix G). To complete the scale, pam 1 (strongly Comfort. Comfort was measured using a modified version of Spake, Beatty, Brockman, and Crutchfields (2003) eight-item scale. These researchers define comfort as a psychological e or she enjoys peace ovider (p. 317). Alble identifies a specific case of a perlocutionary outcome of conversational interaction (Austin, 1962) during a service encounter and may be an important affective response to speech technology interactions. ty using s statistically mitment, active addition of comfort to the more traditional satisfaction-trust-commitment paradigm (Spake, Beatty, Brockman, & Crutchfield, 2003, p. 327). The anchors for seven of the 10-point semantic differential items are: uncomfortable-comfortable, very uneasy-very much at ease, very tense-very relaxed, insecure-secure, worried-worry free, distressed-calm, turbulent-serene. I modified e-service (Dabholkar & Bagozzi, 2002; Liu & Arnett, 2000; Parasuraman, ZeithZeithaml, Parasuraman, & Malhotra, 2000). In addition, items were included bprevious usability ratings and participant comments about speech interfaces undof variables suggested b rticipants indicated their agreement with each item by rating it on a scale frodisagree) to 7 (strongly agree). state wherein a customers anxiety concerning a service has been eased, and hof mind and is calm and worry free concerning service encounters with this pr though it was defined specifically as related to service encounters, this variaSpake, Beatty, Brockman, and Crutchfield (2003) established item content validiexpert judgments on the items included in the final scale. Construct validity waconfirmed by comparing comfort to related constructs (satisfaction, trust, comvoice) and by using structural equation modeling, which indicated the useful 55


the anchors for the final item troublesome-peace of mind to read troubled-peaceful so it woof 0.99, with item Brockman, & Crutchfield, 2003). Current reliability of the scale was also acceptable (Crohnbachs = 0.96). Customer Satisfaction. Satisfaction has been defined as both an affective and cognitive al importance in the t (1990) includes the unfavorable, 3) pleasant to unpleasant, and 4) I like it very much to I didnt like it at all. Participants respond by indicating the strength of their response along a seven-point scale (Appendix I). irtz and Lee (2003) of satisfaction iabilities (0.88, 0.69), lowest error variances (0.16, 0.29), and positive correlations with an affective (0.52) and cognitive (0.52) factor across two different service contexts. They summarized their evaluation by saying it seems these multi-item scales achieve finely grained measurement by tapping into satisfaction froan, and Crutchfield parable to its reliability Stimuli Stimuli for this study were selected based on expert ratings in Study 1. The two best (Financial Services, Prescription Refill), two worst (Flight Status, Directory Dialer), and two average (Tennis Scoreboard, Movies) interfaces were included in Study 2 to represent the range of uld be grammatically parallel to the rest of the scale (Appendix H). The scale was reported to have a composite reliability (coefficient ) reliabilities ranging from 0.79 to 0.91 (Spake, Beatty, construct (Crosby, Evans, & Cowles, 1990; Wirtz & Lee, 2003). It has centrservices marketing literature. The scale provided by Eroglu and Machleifollowing items: 1) satisfied to dissatisfied, 2) favorable to In their comparison of nine frequently-used measures of satisfaction, Wreported that Eroglu and Machleits (1990) four-item seven-point bipolar scale showed the second-highest loadings on satisfaction (0.94, 0.83) and item rel m different angles (Wirtz & Lee, 2002, p. 353). Spake, Beatty, Brockm(2003) reported a composite reliability of 0.97 for this scale, which was comin the present study (Chronbachs =0.93). 56


quality in current implementations in the speech technology industry. a web survey and 5) for descriptive and inferential analyses. The manipulation checks and results of data analyses for each of the five research questions are presented in turn. nterface they heard to d on their responses to three factual questions about the user-system interaction. Correct answers were scored and summed to create a number correct; if the manipulation check score was two or greater, the ses from 41 participants er the system voice was male or female. The majority of participants correctly identified a female voice for the Directory Dialer (N=126, 98%), Flight (N=115, 100%), Prescription Refill (N=196, 98%), and Fidentified for Australian voice was less in a female voice, ings submenu (primacy effect). For each interface, participants responded to an additional open-ended question and multiple choice question (see Table 4). As expected, participants provided correct answers more often on the multiple choice questions. In general, the results of the participant listening questions indicated that participants were generally accurate in their recall of the interactions. Results In Study 2, data was automatically collected using WebSurveyor 4.1, data collection tool, then transferred to a statistical software package (SPSS 11. Manipulation Checks Participants in each group responded to four questions about the idetermine whether they listened to the recording. Participants were score participants data was retained in the sample. Based on this result, responwere omitted from all subsequent analyses, resulting in a sample size of 821In the first manipulation check question, participants were asked wheth nancial Services (N=107, 97%) interfaces. Male voices were correctly iOpen (N=124, 95%) and Movies (N=118, 86%) interfaces. The Movies systemfrequently identified as male, possibly due to an extensive main menu selectionfollowed by the male voice once the user proceeded to the movie list 57


Table 4. Open Ended and Multiple Choice Manipulation Check Responses erfaceN% Correct Int Question The system provided scoring for what tenni s m? 111 (87%) atch rd Which of the for? Pete Sampras Andre Agassi What locati Which of the following commands did r giv Connect What ai JFK LAX The syste mation about? Dra Comedy What was the name of the pharm aceutical com? 159 (80%) pany ion How did D Tennis Scoreboa following players did the user request scores Roger Federer 127 (100%) on did the user request? 119 (92%) Directory Dialer the usee? Fax number Cell phone 106 (92%) rline did the user request information about? 108 (92%) Flight Status What was the arrival airport for the flight? OHare 106 (92%) m provided movie listings for what location? 115 (84%) Movies What type of movie did the user request infor ma Suspense 137 (94%) PrescriptRefill the user obtain his prescription? elivery Pickup 193 (97%) What stock did the user buy? 101 (94%) Financial SePhone Electric Alimony Mortgage 100 (93%) rvices What kind of bill was the user alerted about? In addition to the manipulation checks for participants, I conducted an analysis to assess whether the quality categories influenced customer satisfaction ratings. This step was necessary 58


due to reliability problems with expert ratings of interface quality in Study 1 and provided another of stimulus (good, 51, MSe=1.97, d by a post-hoc LSD test (average-poor, p=.002; average-good and poor-good, p<0.0001), as illustrated in Figure 1. Figure 1. Speech User Interface Quality and Customer Satisfaction indicator of stimuli variability. A univariate ANOVA indicated that qualityaverage, poor) had a significant influence on satisfaction ratings, F(2,814)=82.p<0.0001. Differences among all three quality categories were also supporte Interface Quality Data were collected on 76 items thought to measure aspects of usabilityinterfaces. Univariate analyses and visual inspection of the boxplots for the 76 ite 1234567HighAverageLowean Customer Satisfaction Rating 0M Question 1: What items should be measured in assessing the usability of speech interfaces? in speech ms revealed that most items were characterized by a normal or near-normal, slightly skewed distribution. The only exceptions to this observation were five items that showed a positive skew (items 13, 45, 58, 70, and 75) and three items that had a negative skew (items 12, 34, and 61). The sample size, means, standard deviations, and correlations for the ten assumed factors appear in Tables 5 through 13. 59


Table 5. Means, Standard Deviations, and Correlations for Hypothesized Customer Service pat Far Ite10 19 346 74 Ex ect ions cto m# M SD 1 0 64 66 4.25 1.63 1.00 4.48 1.57 .56** 1.0 4.28 1.96 .38** .49* *1 0 5.60 1.32 .35** .44* *.3 ** .00 4.51 1.61 .55** .72* *.4 **. 2** 1.00 1 100 19.0 3421 6054 32**.67**.42**1.00 66 5.14 1.42 .48** .61**.36**.44**.65**.47**1.00 74 5.35 1.43 .38** .47**.37**.53**.52**.56**.55** 1.00 64 5.45 1.33 .35** .45**. **correlation is significant at 0.01 level (2-tailed) Ta.nens, nd Coelatios for Hypothesized Pragmatics Factor Ite26 41 42 50 65 76 ble 6 Mea n s, Sta dard D viatio a rr n m# M SD 5 8 15 3.10 1.61 1.00 3.62 1.61 .14** 1.00 5.01 1.52 -* .09 -.13 1 0 5 8 15 *.0 1.00 5** -.03 1.00 42 3.10 1.61 .24** .35** -.25** .35**-.17**1.00 50 3.37 1.71 .20** .20** -.24** .20**-.23**.31**1.00 65 3.73 1.79 .13** .35** -.26** .35**-.22**.47**.34** 1.00 76 4.17 1.52 .12** .04 .30** .05 .29** .003 -.09** -.03 1.00 26 3.51 1.62 .25** .30** -.05 41 5.04 1.55 -.09** -.05 .5 **correlation is significant at the 0.01 level (2-tailed) correlation is significant at the 0.05 level (2-tailed) 60


Table 7. Means, Standard Deviations, and Correlations for Hypothesized Recognition Factor Ite7 29 30 33 35 71 m# M SD 13 2 2.61 1.54 1.00 4.19 1.41 -.27** 1.0 3.82 1.72 .26** -.08 1 0 3.93 1.70 -.13** .41** -. 8* 5 .00 6** 1.00 13 27 0 29 *.0 30 01 .0.2 4** -.04 -.08* 1.00 71 4.54 1.66 -.41** .35** -.28** .24**.37**-.26**1.00 33 4.95 1.48 -.24** .38** 35 2.97 1.59 .27** -.11**.2 **correlation is significant at the 0.01 level (2-tailed) correlation is significant at the 0.05 level (2-tailed) 8.ansandDeviations, and Colatioforpothesized Affective Response FaIte7 46 48 55 73 Ta ble Me St ard rre ns Hy ctor m# M SD 17 25 3 4.56 1.76 1.00 4.06 1.85 .63** 1.00 17 25 .00 46 4.85 1.61 .59** .45** .62** 1.00 48 2.65 1.56 -.27** -.19**-.20** -.31**1.00 73 4.44 1.72 .67** .58** .46** .53** -.22**.60**1.00 37 4.42 1.58 .50** .39** 1 55 3.85 1.72 .67** .68** .45** .48** -.19**1.00 ** correlation is significant at the 0.01 level (2-tailed) Ta.Deviations, and Correlations for Hypothesized User Goal Orientation Item# M SD 14 23 28 32 51 ble 9 Mea ns, Sta ndard Factor 14 4.86 1.45 1.00 23 4.46 1.57 .53** 1.00 28 4.64 1.74 .49** .48**1.00 32 4.40 1.54 .53** .59**.48**1.00 51 4.39 1.62 .56** .53**.53**.49**1.00 **correlation is significant at the 0.01 level (2-tailed) 61


Table 10. Means, Standard Deviations, and Correlations for Hypothesized Accuracy Factor Ite57 63 43 m# M SD 47 5.42 1.37 1.00 3.09 1.52 -.32** 1.0 47 57 0 .00 43 4.94 1.47 .50** -.40**.42**1.00 63 4.48 1.60 .32** -.30**1 **correlation is significant at the 0.01 level (2-tailed) Taans, Standard Deviations, and Correlations for Hypothesized Prompt Wording Factor Ite 59 70 ble 11 Me m# M SD 12 20 58 5.59 1.49 1.00 12 20 5.53 1.33 .53* 1.00 59 5.00 1.62 .45** .47** .08* 1.00 70 2.45 1.52 -.38** -.38**.10**-.41**1.00 58 2.87 1.69 .05 .09* 1.00 **correlation is significant at the 0.01 level (2-tailed) correlation is significant at the 0.05 level (2-tailed) Ta2aiod Correlations for Hypothesized Usefulness Factor Ite 6 21 38 ble 1 Me ans, St ndard D eviat ns, an m# M SD 3 4 9 16 3 4.54 1.70 1.00 4 5.14 1.39 .58** 1.00 6 4.46 1.62 .30** .23** 1.00 9 3.60 1.84 -.53** -.36**.24**1.00 16 3.44 2.03 .48** .40** .16**-.39**1.00 21 4.55 1.74 .13** .19** .10**.01 .06 1.00 38 4.72 1.81 .62** .63** .20**-.38**.45**.16**1.00 **correlation is significant at the 0.01 level (2-tailed) 62


Table 13. Means, Standard Deviations, and Correlations for Hypothesized Speech Impression Factor Item# M SD 2 11 24 31 39 40 44 45 52 54 6 2 68 6 9 7 2 1.048* 0* .00 1 43 ** 39 ** 1 2 3.04 1.68 1.00 11 5.37 1.41 -.24** 1.00 24 3.55 1.93 .41** -.35** 1.00 31 5.29 1.55 -.20** .47** -.38** 1.00 39 4.96 1.72 -.27** .61** -.52** .49** 1.00 40 4.65 1.49 -.19** .41** -.34** .45** .46** 1.00 44 3.86 1.88 -.23** .38** -.50** .33** .50** .34** 1.00 45 2.78 1.86 .24** -.29** .46** -.44**-.40**-.33**-.32**1.00 52 4.29 2.01 .36** -.30** .60** -.30**-.48**-.28**-.53**.36** 1.00 54 3.85 1.95 -.28** .43** -.59** .43** .61** .41** .61** -.39** -.69** 1.00 62 3.83 1.92 -.32** .41** -.63** .41** .61** .37** .58** -.39** -.70** .78**1.00 68 5.30 1.47 -.21** .61** -.36** .48** .60** .44** .37** -.28** -.32** .42**.43** 69 4.96 1.88 -.24** .50** -.51** .63** .60** .47** .47** -.64** -.45** .56**.55**. 72 3.97 1.76 -.21** .44** -.40** .25** .52** .34** .46** -.26** -.42** .51**.52**..00 **correlation is significant at the 0.01 level (2-tailed) 63


The overall goal of analyses for this research question was to create a usability scale for ble scale with 25 or reliability analysis and ntify items that clustered in subscales. First, a reliability analysis was conducted on items for each of the ten assumed subscales. Cronbachs was calculated for each factor and each items corrected item-to-total correlation and item deleted were reviewed. When an item demonstrated a ted from the subscale ntil Cronbachs reached or exceeded the criterion value of 0.70 (Nunnally & Bernstein, 1994). As shown in Table 14, this procedure resulted in elimination of 29 items (including two entire subscales, cted to determine ontrast to the a priori assumption, visual inspection of the scree plot indicated a six-factor solution was most likely. This solution accounted for 60.86% of the variance in the data. However, only a single item loaded Therefore, the four d with the reduced no items loaded on Factor 5. This solution accounted for 61.26% of the variance in the data. Loadings for the rotated five-factor solution appear in Table 15. As shown, the majority of items loaded on Factor 1. The originally hypothesized Pragmatic and Speech factors were supported with four items (Factor 4 = items 8, 26, 42, and 65) and six items (Factor 3 = items 39, 44, 54, 62, 69, 72). Two items (5 and 36) did not load on any factor and were removed for the subsequent analysis. practical usage in applied settings; therefore, the desired end result was a reliafewer items. To decrease the number of items from the original 76 items,factor analysis were used to eliminate problematic items in the scale and iderelatively low item-to-total correlation and an improvement in it was delefactor and Cronbachs was recalculated. This procedure was repeated u Recognition and Accuracy) due to reliability problems. Next, a Principal Component analysis with varimax rotation was conduwhether the subscale structure of the scale could be empirically supported. In con one factor in this initial solution and four items did not load on any factor. items (40, 50, 59, 61) were removed and a second factor analysis was conducteitem pool. This analysis again suggested a five-factor solution was most likely (see Figure 2) but 64


Table 14. Summary of Reliability Analysis for Usability Scale Assumed Factor Items Items RemoFinal ved 64 Pragma tics 6, 41 50, 65, 76 15,0.70 9, 3035, 71 .29 13, 290.61 (factor deleted) ive Response 17, 25, 37, 46, 48, 3 .0.76 48 0.86 ation 8, 32 None 0.84 Accuracy 47, 57, 63, 43 0.11 57 .67 (factor deleted) Prompt Wordi8, 59, 70 0.21 70, 58 0.74 Usefulness 21, 0.49 6, 9, 21 .80 Impres 11, 24, 31, 39, 40, 44, 45, 52, 54, 2, 24, 45, 52 0.91 0.85 5, 8, 15, 2 42, 0.51 41, 76 Recognition 13, 27, 2 33, 0 35 Affect 55, 7 User Goal Orient 14, 23, 2 51 0.84 ng 12, 20, 53, 4, 6, 9, 16, 38 sion 2, Speech Customer Service Expectations 1, 10, 19, 34, 60, 66, 74 0.87 19 0.88 62, 68, 69, 72 Mental Model/SUI Organization 7, 18, 22, 49, 53 61, 67, 75 0.69 49, 53, 56, 75 A third Principal Component analysis (varimax rotation) was conducted with the reduced item pool and constrained to a four-factor solution, which accounted for 58.67% of the variance.. Factor loadings appear in Table 16. Similar to the previous solution, the majority of items loaded on the first factor, the proposed Pragmatics construct was supported by four items (Factor 5), and the Speech Characteristics factor was supported by five items (Factor 3). However, the Pragmatics 65


Figure 2. Scree Plot for Usability Scale Factor Analysis Scree PlotComponent Number4037343128252219161310741Eigenvalue2010 0 factor included items that specifically r elated to the talkativeness of the interface, making the orion. Therefore, a oad variety of items from several of the hypothesized factors. Additional analyses were used to reduce the number of items in the scale to make it suitable owest loadings were ocess resulted in ctor 1 and one item from Factor 2. Additional analyses indicated that three of the factors had acceptable reliability: Factor 1, User Goal Orientation (with eight items; Factor 2, Customer Service Behavior (.89) with eight items; and Factor 3, Speech Characteristics (.87) was acceptable with three items. Factor 4 (Verbosity) was marginally acceptable with four items (.69). iginally hypothesized factor name too general to provide an adequate descriptmore specific label, Verbosity, was adopted for this factor. Factors 1 and 2 included a brfor use in applied settings. To reduce the size of the scale, items with the leliminated from Factors 1 and 2 until they each included eight items. This prremoval of 13 items from Fa 66


Table 15. iosmactoFactor 2 Factor 3r 4Factor 5 In tial Fact r Loadin gs for U ability S cale Ite # F r 1 Facto 1 .67.12 -.067 9 91 .154 .01 0 3 54 .317 .09 80 .114 .09 0 9 50 .328 .13 56 .207 .06 6 9 390 .130 .05 608 .252 .13 5 8 6 022 -.118 -.63 -.025 -.70 .035 -.70 8 032 -.141 -.67 7 1 86 .379 .10 5 3 41 .430 .13 7 2 32 .097 .04 6 2 48 .169 .06 5 5 38 .373 .21 3 1 45 .407 .05 4 0 82 .099 .02 3 0 72 .188 .07 8 2 09 .233 -.07 2 0 81 .223 .18 1 0 01 .252 .02 90 .071 -.04 47 .086 -.00 3 6 91 .185 .29 4 2 42 .144 .13 6 5 040 .211 .26 8 4 38 .211 .18 47 .346 .03 2 69 .294 .08 8 87 .482 .08 .543 .16 .771 -.05 1.70.29 -.005 34 .228 .78 -.025 6.72.28 .122 64 .234 .76 -.041 6.67.7 .223 74 .343 .6 .146 .030 -.089 .138 -.414 -.421 -.07-. 0 .010 26 .069 -.050 8 -.162 42 -.304 -.061 2 .008 65 -.27-. 8 .070 1.67.26 -.062 2.55.15 -.258 3.67.26 .176 4.71.21 .257 5.62.14 -.173 7.61.20 -.162 1.64.31 -.023 2.67.29 -.025 2.56.38 -.129 3.69.14 .010 5.57.30 -.013 12 .238 .52 .244 20 .277 .64 .220 .63.14 .074 .64.40 .105 1.55-.0 -.024 3.72.24 .182 11 .266 .66 -.045 31 .31.41 .446 36 .44.27 .421 39 .383 .472 6 .003 44 .206 .161 1 .097 54 .267 .177 .801 .095 .084 62 .287 .184 .768 .148 .098 69 .429 .291 .533 .086 .448 72 .233 .285 .602 .001 -.223 68 .360 .666 .278 .046 -.091 7 .620 .341 .110 .037 .247 18 .682 .271 .108 .046 .281 22 .497 .544 .176 .101 .187 67 .628 .365 .132 .073 .294 67


yUsability Scale mactoFactor 2 Factor 3Factor 4 Table 16. cond Fac tor Anal sis Load ings for Ite # F r 1 Se 1 .67.168 .216 -.015 2 0 7 28 .184 0 2 699 .276 6 4 639 .254 .022 .023 1.68.258 .350 .096 34 .198 .7.057 6.71.303 .323 .161 64 .206 ..023 6.66.464 .116 .080 74 .315 ..146 -.059 -.030 -.106 -.647 26 .082 -.106 -.731 42 -.295 -.059 -.692 65 -.259 -.025 -.150 -.687 1.64.276 .410 .114 2.53.072 .494 .126 3.66.290 .101 .047 4.70.345 .135 .097 5.61.089 .433 .195 7.58.204 .462 .049 1.62.363 .157 -.005 2.65.271 .224 .081 28 .543 .270 .284 -.082 3.67.198 .245 .192 5.55.301 .283 .024 12 .215 .6-.017 20 .248 .7.027 .61.233 .184 .319 .62.470 .163 .131 1.54-.028 .222 .267 3.71.308 .200 .205 11 .236 .6.013 31 .291 .6.141 39 .352 .466 .167 44 .191 .192 -.043 54 .239 .222 .775 .134 62 .253 .226 .748 .175 69 .411 .451 .415 .167 72 .208 .205 .671 -.023 68 .331 .606 .353 .009 7 .605 .428 .086 .069 18 .672 .372 .079 .077 22 .472 .597 .168 .121 67 .613 .469 .089 .115 8 7 9 5 2 7 7 6 1 5 0 3 8 4 7 3 3 2 5 1 1 54 .049 06 .065 3 8 4 4 6 8 8 1 04 .399 12 .205 .569 .708 68


The final usability scale consisted of four factors, named User Goal Orientation (items 1, 4, 68, 74), Speech his result indicated ncluded items that related to the systems efficiency, user trust and confidence in the system, and clarity of a speech interface based on user needs. Customer Service Behavior included items that were related to the iar terms. The made up of items that measured the talkativeness and repetitiveness of the sy enthusiasm of the system voice. To determine whether the final factors discriminated among the sets of interfaces, a series of four univariate ANOVAs was conducted, with interface as the independent variable and each of ignificantly different =0.196), Customer h Characteristics (F(5,814)=96.15, MSe=1.41, p<0.0001,=0.373), and Verbosity (F(5,814)=36.42, MSe=1.18, 2n reverse-scored). n in Table 17 ill interface in terms of Customer Service Behavior and User Goal Orientation but the two differed on perceptions of Speech Characteristics and Verbosity. Similarly, the Movie interface was perceived as similar to the Tennis Scoreboard and Flight interfaces on User Goal Orientation but was rated more positively on Speech Characteristics and Customer Service Behavior but more negatively on Verbosity. In general, the pattern of results indicated that the four factors of the usability scale did differentiate 10, 18, 32, 37, 38, 46, 60), Customer Service Behavior (items 12, 20, 22, 31, 34, 6Characteristics (items 39, 44, 54, 62, 72), and Verbosity (items 8, 26, 42, 65). Tthat 25 items assess the usability of speech interfaces. User Goal Orientation ifriendliness and politeness of the system, its speaking pace, and its use of familVerbosity factor was stem and Speech Characteristics included items related to the naturalness andthe usability factors as the dependent variables. The sets of interfaces were son ratings of User Goal Orientation (F(5,814)=53.54, MSe=1.36, p<0.0001, 2Service Behavior (F(5,814)=38.27, MSe=0.92, p<0.0001,2=0.191), Speec2p<0.0001,=0.184). The mean ratings appear in Figure 3 (Verbosity is showPost-hoc LSD tests also supported differences among the factor means, as show(p<0.05). For example, the Financial Services interface was similar to the Ref 69


among the interfaces. urface QuQualstomer Svior ser GorientatpeecCharVerbosity (reverse-scored) Table 17. Post-Hoc Resity Cu lts for Inteervice U ality al S Beha O ion h acteristics (5.02) Te Best Refill (5.92) Financial (6.05) Refill (5.12) Financial (5.36) Financial (5.52) Refill (5.22) RefillFinancial (4.69) Flights (4.63) Movie (5.46)nnis (4.20) Flights (4.27) Movie (4.36) Movie (4.01) Dialer (4.30) Tennis (4.28) Worst Dialer (4.82) Dialer (3.66) Tennis (3.33) Tennis (4.95) Flights (5.14) Dialer (3.23) Flights (3.07) Movie (3.67) (non-significantly different means are shown in each cell, significantly different means in different cells) 70


Figure 3. Usability Factor Scores for the Different Interfaces 1234567FlightsDirectoryTennisMoviesPrescriptionFinancialInterfaceMean Rating ristics Verbosity Customer Service Behavior User Goal Orientation Speech Characte Question 2: How are individual difference variables (e.g., self-monitoring, need for interaction, e statistics and Pearson product moment correlations were calculated for each of theoratory and comparative purposes, the four factor scores from the speech interface usability scale were also included in the analyses. ormal distributions for all ooled within-group ) to adjust for the effect of different interfaces on the relationships among the correlated variables. Results showed that Inherent Novelty Seeking, Self-Monitoring Ability, and Self-Monitoring Sensitivity were weakly positively associated with both Comfort and Customer Satisfaction. Need for Interaction with a Service Provider (NISP) was weakly negatively inherent novelty seeking) related to comfort and customer satisfaction? Descriptiv individual difference variables, comfort, and customer satisfaction. For expl Visual inspection of the boxplots indicated approximately nvariables. Means, standard deviations, and correlations appear in Table 18. Pcorrelations were calculated (interface defined as an independent variable 71


Table 18. Descriptive Statistics and Bivariate Correlations for Individual Differences, Usability Factooa Cio Item M SD 1 2 3 4 5 7 8 9 10 rs,6 C m fort nd us tom er S atis fa ct n .49* 1. Customer Satisfaction 5.08 1.54 1.00 2. Comfort 7.24 1.85 .75** 1.00 3. Need for Interaction with a Service Provider 5.27 1.28 -.17** -.11** 1.00 4. Inherent Novelty Seeking 5.07 1.19 .11** .16** .17** 1.00 5. Self-Monitoring Sensitivity 4.55 0.88 .07* .08** .18** .26** 1.00 6. Self-Monitoring Ability 4.61 0.91 .12** .13** .10** .25** .42** 1.00 7. User Goal Orientation 4.53 1.30 .71** .65** -.13** .16** .14** .12** 1.00 8. Speech Characteristics 4.09 1.50 .43** .40** -.05* .02 .09** .08* .53** 1.00 9. Customer Service Behavior 5.42 1.06 .40** .40** -.002 .11** .20** .21** .64** 1.00 10. Verbosity 3.49 1.20 -.26** -.25** .14** .03 .06 .004 -.28** -.15** -.17** 1.00 *Correlation is significant at the 0.05 level (2-tailed) **Correlation is significant at the 0.01 level (2-tailed) 72


associated with these two variables. In contrast, the interface variables (User Goal Orientation, ted with Comfort and association. Quatisfaction? The analyses for this question consisted of simultaneous multiple regressions to determine whether the usability scale factors and items predicted unique variance in customer satisfaction. best unique rovide a shorter, more s.The multiple regression using factors as the predictors indicated a significant model (F(4, 813)=302.79, MSe=0.95, p<0.0001) with an R2 of 0.60. Diagnostic statistics indicated no evidence of collinearity. Sir Service Behavior 01), Verbosity oal Orientation (b=0.86, t=21.10, p<0.0001). The stepwise analysis indicated a model with one predictor was significant (F(1, 813)=1121.39, MSe=0.99, p<0.0001) with an R2 of 0.58 and a resulting equation of Customer a significant model ) with an R2 of 0.63. Significant b weights were observed for nine items: item 1 (b=0.07, t=2.51, p=0.01), item 10 (b=0.15, t=4.14, p<0.0001), item 18 (b=0.06, t=2.04, p=0.04), item 20 (b=-0.08, t=-2.33, p=0.02), item 34 (b=-0.08, t=-198, p=0.05), item 38 (b=0.19, t=5.96, p<0.0001), item 46 (b=0.12, t=3.62, p<0.001), item 60 (b=0.18, t=4.78, p<0.0001), item 65 (b=-0.05, t=-2.18, p=0.03). Customer Service Behavior, Speech) were strongly positively associa Customer Satisfaction. However, Verbosity showed a moderate negative estion 3: What speech user interface characteristics best predict customer sThese analyses were followed by stepwise multiple regressions to identify thepredictors of customer satisfaction. The item analysis was conducted to pefficient set of items that predict customer satisfaction in applied setting gnificant b weights were observed for all four factor scores: Custome(b=-0.15, t=-3.22, p=0.001), Speech Characteristics (b=0.13, t=4.36, p<0.00(b=-0.12, t=-3.83, p<0.0001), and User G Satisfaction = 0.99 + 0.90(User Goal Orientation). The analysis was repeated using the usability scale items, and indicated(F(25, 800)=52.28, MSe=0.91, p<0.0001 73


The stepwise analysis showed that a model with four items was significant (F(4, 2andardized regression stomer Satisfaction = 0.119 + 0.036(item 60) + 0.030(item 38) + 0.0In general, this analysis suggested that 58% to 63% of the variance in customer satisfaction can be accounted for by the factors or items from the Usability scale. A more efficient set of four The resulting ws that customer satisfaction ratings may be predicted from the 60 and 10), expected repeat usage (item 38), and user confidence (item 46). Question 4: What speech user interface characteristics best predict comfort? taneous multiple rs for the criterion tem ratings. For the factors, the multiple regression indicated a significant model (F(4, 813)=216.98, MSe=1.65, p<0.0001) with an R2 of 0.52. Diagnostic statistics indicated no evidence of collinearity. n (b=0.90, t=16.80, 001), and Verbosity (b=-0.12, t=-3.07, p=gnificant (p=0.37) ant (F(1, 812)=824.06, MSe=1.69, p<0.0001) with an R2 of 0.50. The resulting standardized regression equation (p<0.0001) is Comfort = 0.17 + 0.035(User Goal Orientation). The analyses were again repeated with the items to determine an efficient set that may be used in lieu of a complete scale or subscale in applied settings. The stepwise analysis showed that 800)=292.25, MSe=0.96, p<0.0001) with an R of 0.60. The resulting stequation (p<.0001) is Cu 34(item 10) + 0.033(item 46). usability items accounts for slightly less customer satisfaction variance (60%).stepwise regression equation sho User Goal Orientation score or ratings of general quality (item As with question three, analysis for question four consisted of a simulregression and a stepwise multiple regression to identify the best unique predictocomfort. These analyses were conducted with both the factor scores and the i Significant b weights were observed for three factors : User Goal Orientatiop<0.0001), Speech Characteristics (b=0.15, t=3.57, p<0.0 0.002). The b weight for Customer Service Behavior was not statistically siThe stepwise analysis showed that a model with one factor was signific 74


the model was significant (F(25, 800)=37.25, MSe=1.59, p<0.0001) with an R2 of 0.54. Significant em 18 (b=0.14, p=0.001), item 39 1), item 60 (b=0.10, t=1.93, p=0.05), item 62 (b=0.09, t=2.29, p=0.02), and item 64 (b=-0.10, t=-2.03, p=0.04). The stepwise analysis showed that a model with four items was significant (F(4, MSe=1.72, p<0.0001) with an R2 of 0.50. The resulting standardized regression eq 0.039(item 18) + As with the previous research question, this analysis suggested that 50 to 54% of the variance in comfort can be accounted for by one factor or up to nine items from the Usability scale. f the variance in the criterion. The resulting stepwise Goal Orientation r ratings of general quality (item 60), user confidence (item 46), navigation ease (item 18), and naturalness of the system voice (item 62). Question 5: Do individual difference variables interact with speech user interface quality to create pendent (predictor) edictor) variables (age, Neonitoring Ability, Self-Monitoring Sensitivity). The categorical variables were both effect coded for the analysis and vectors were constructed to represent the two-way interactions among all seven variables to complete the analysis. Customer satisfaction served as the dependent (criterion) variable. Results of the first analysis with only main effect predictors indicated a significant model, b weights were observed for nine items: item 10 (b=0.16, t=3.37, p=0.001), itt=3.45, p=0.001), item 31 (b=0.11, t=2.81, p=0.005), item 37 (b=0.14, t=3.35, (b=0.12, t=2.68, p=0.007), item 46 (b=0.20, t=4.45, p<0.000 800)=198.47 uation (p<.0001) is Comfort = 0.163 + 0.043(item 60) + 0.041(item 46) +0.028(item 62). However, only four items account for 50% o regression equation shows that comfort ratings may be predicted from the User subscale score o customer satisfaction? This analysis consisted of an ANCOVA with two categorical indevariables (interface quality, gender) and five continuous independent (pr ed for Interaction with a Service Provider, Inherent Novelty Seeking, Self-M 75


F(7,794)=32.27, MSe=1.86, p<0.0001, with an R2 of 0.22. Six predictors were significant: need ovelty seeking 12), quality (b=0.705, .96, p=0.05). Self-Monitoring Sensitivity failed to be significant (p=0.54). The resulting regression equation is: Customer Satisfaction = 5.48 + 0.135(Inherent Novelty Seeking) + 0.151(Self-Monitoring Ability) ider) ividual difference eking, Self-Monitoring Ability, Age, Gender) plus interface quality ratings will yield the customer satisfaction score. Furthermore, these five predictors account for 22% of the variance in customer satisfaction. fects and six interaction vectors, was significant, F(13,794)=18.22, MSe=1.83, p0.10). Discussion n, ginnings of a theoretical Although speech ery, it is thought to have huge market potential: ABI Research predicts that the speech technology market will grow to over $5 billion by 2008 as it is used to replace or supplement call centers (Allied Business Intelligence Research, 2003) and analysts acknowledge that the quality of speech technology has improved to acceptable levels for broad user acceptance (Harris Interactive, 2003; Popova, 2004). for interaction with a service provider (b=-0.235, t=-5.92, p<0.0001), inherent n(b=0.135, t=3.13, p=0.002), Self-Monitoring Ability (b=0.151, t=2.53, p=0.0t=11.80, p<0.0001), age (b=-0.042, t=-3.33, p=0.001), and gender (b=-.121, t=-1+ 0.705(Quality) 0.042(Age) 0.232(Need for Interaction with a Service Prov0.121(Gender). This equation indicates that the additive combination of indvariables (Need for Interaction with a Service Provider, Inherent Novelty Se The second regression, including main ef 0.0001. Th edictors observed in the first model continued to be significant The present research integrates research in interpersonal communicatiopsycholinguistics, human factors, and services marketing to create the beand methodological foundation for future applied work in speech technology. technology has only recently emerged as a mainstream means of e-service deliv 76


However, despite the substantial business interest in this technology and broad acceptance of the basic esponses to these from a broad literature to develop a scale of usability for speech interfaces and examine interrelationships among this construct, individual differences, comfort and customer satisfaction. These results have implications for both the methodology and theory of speech technology, as well as the disciplines that inform the current s Metholodological Implications Although the development of a measurement scale formed a substantial part of the present h. The Usability ity in industrial ts, recorded clips of interfaces, and third-party observers provides an efficient means of understanding affective responses to speech technology. Nonetheless, critical examination of the current methodology suggests a number of issues that The dein speech technology, because no empirically-validated scales for measurement of this construct currently exist. In addition, although usability is widely regarded as an important consideration in designing interfaces, little has been known about it in the context of speech technology. Seventy-six items for the usability scale were generated based on a broad review of the critical need for well-designed interfaces, there is limited research that appliessocial-communicative findings to speech interface design, and users affective rinteractions have been only minimally explored. The present research drew tudy. research, its general methodology has implications for applied e-service researcScale for Speech Interfaces advances practitioners ability to measure usabilsettings. The use of speech and language exper should be empirically addressed in future applied research. The Usability Scale for Speech Interfaces The first research goal was to develop a scale of usability for speech interfaces. velopment of a usability scale marks a significant methodological advance 77


literature in social psychology, communication and services marketing. The resulting usability Customer Service ale for Speech predecessor and related scales, and is consistent with the theoretical research that underlies its development. Each of these issues will be strengths will be discussed in turn. cale for Speech iability for four of because it indicates that usability for speech interfaces is a multifaceted, complex variable that incorporates technological features such as ease of use, communicative features such as voice naturalness, and human with greater face or pragmatics ount of variance in customer satisfaction and comfort, suggesting its predictive validity for e-service outcomes. The predictive capability of the scale is especially important to its usefulness as speech system usability asure because it may be used to predict user outcomes proactively during development. Finally, theere previously found such as voice Rose & Emperado, 2001). Despite the apparent strengths of the Usability Scale, the Verbosity factor demonstrated somewhat weaker reliability than the other four factors. Nonetheless, there are several reasons for retaining this factor and its associated items in the Usability Scale. First, the Verbosity score did scale contained 25 items and four factors, known as User Goal OrientationBehavior, Verbosity, and Speech Characteristics. In general, the Usability ScInterfaces appeared to be psychometrically sound, compared favorably with its Psychometric Quality of the Usability Scale. In general, the Usability SInterfaces appeared to be psychometrically sound with acceptable levels of relthe five factors. The new measure also appears to have construct validitycustomer service behaviors. In this sense, it measures speech system usabilityvalidity than more narrowly operationalized scales concerned with only speech (Polkosky, 2003). In addition, the Usability Scale accounted for a substantial amme new scale appears to have content validity because it includes items that wto be important to observers liking and expected future use of a speech interfacenaturalness (Polkosky, 2003), and friendliness of service providers (Baydoun, 78


discriminate among the six Study 2 interfaces in a different pattern than the other factors, so it is mprehensive nable that items d be included in a usability scale, because participant comments during informal usability testing often include these perceptions of speech systems. Third, a more general version of this scale (Pragmatics) had atic perceptions are h & Ferguson, 2000), of pragmatics to speech interface design is consistent with previous approaches (Bernsen, Dybkjaer & Dybkjaer, 1996; Brennen, 1998; Saygin and Cicekli, 2002), although the current scale measures a specific as service contexts. ned in the Usability Another potential weakness of the scale was the small effect size for each Usability factor, based on the Cohens (1969) interpretation guidelines. A possible interpretation of this finding is xt of current trends usiness-related a speech system of that same call. If 10% of the callers opt out of the speech system (i.e., hang up or transfer to an operator) due to a negative perception of the system itself, the additional financial cost to the enterprise is approximately $702,000 yearly. There is also the potential loss of the customers who are dissatisfied to the extent that they take their business elsewhere. A seemingly small change to the interface, such as likely that the combination of the five factors will provide designers with a counderstanding of a speech systems usability. Second, it seems intuitively reasolike perceived talkativeness, amount of detail, and repetitiveness shoulextremely small effect sizes in previous research (Polkosky, 2003) and pragmlikely to be automatic and beyond the conscious awareness of observers (Bargwhich may result in reduced measurement reliability. Finally, the application pect of pragmatic behavior that appears to be intuitively relevant to customerTherefore, despite the psychometric weaknesses of this factor, it should be retaiScale. that the effects are too small for continued investigation. However, in the contein speech technology, even small effects may result in large differences in boutcomes. For example, consider an enterprise that receives 100,000 calls in toeach month, at a savings of $5.85 per call as compared to human handling 79


substituting a more pleasant or friendly-sounding system voice, may effect large numbers of users, ystem use, thout transfer to a e, understanding even small effects may have a substantial impact on the use of speech technology usage in applied settings. Technology and that incorporates a the Usability Scale for Speech Interfaces is significantly broader in its scope. The present results of scale development indicate that speech system usability is a related but unique variant of visual and human forms of lop a usability scale nion Scale Expanded (MOS-X) to measure speech characteristics of synthetic speech. The MOS-X scale, which measured Intelligibility, Naturalness, Prosody and Social Impressions, subsequently was used by gh the MOS-X its singular focus on e conversational was developed to discriminate among different synthetic voices speaking the same text, not recorded human voices nor interactive dialogue usability. The Usability Scale includes speech items measuring the naturalness of a system voices; similarity between a system voice and radio, television or regular human voices; and the enthusiasm or energy of a system voice. This result is consistent with resulting in not only improved customer satisfaction, but increased accuracy of sincreased preference for e-service, retention of the user in a speech system wihuman operator, operational savings, and faster return on investment. Therefor Comparison between the Usability Scale and Other Measures of Speeche-Service. The current scale suggests that usability is a multi-faceted construct broad variety of items. In contrast to previous measures for speech technology customer service. Results of this study reveal the limitations of previous attempts to devefor speech interfaces. Polkosky and Lewis (2003) developed the Mean OpiPolkosky (2003) to measure observer perceptions of speech interfaces. Althouexpanded the scope of its predecessor (the Mean Opinion Scale or MOS), speech characteristics appeared to be too limited for measurement of interactivsystems (Polkosky, 2003). This outcome was expected, because this scale 80


previous MOS-X research, which showed that Naturalness and Social Impression of system speech ression ratings to usion of items that voices suggests observers expectations for interactive systems may be defined by vocal standards in other forms of popular media or technology. In contrast to the Mean Opinion Scale-Expanded or MOS-X (Pe appear to suggest that than the previous Pragmatics Scale for Dialogues, which was focused on the interactive and social-communicative aspects of conversational interaction (Polkosky, 2003). Results with the Pragmatics Scale for ent, it indicated it had very lkosky, 2003). In cerned with repetitiveness and talkativeness of a system, a more targeted set of items that address frequently-heard comments about speech systems. The items included in the Usability scale appear d conversational ogues are formed by of other forms of e-service is also important, because they contributed to the current scale development effort. The usability scale revealed a number of similarities and differences with the Zeithaml, Parasuraman and Malhotra (2000) measure of e-service quality (e-SQ), which provides 11 dimensions of service quality, specifically targeting self-service via Internet websites. Similar to their scale, the User were most strongly correlated with Liking of a system, but refines social impinclude those items most relevant to customer service. Interestingly, the inclmeasure the similarity between a speech systems voice and radio or television olkosky & Lewis, 2003), the speech items included in the Usability Scalspeech perceptions play a prominent role in e-service judgments. The current Usability Scale more broadly operationalizes usability Dialogues indicated its potential benefit for applied e-service measuremlow effect sizes and correlations between its factors and affective outcomes (Pocontrast to the previous work, the current Verbosity factor included items con to be those that are more specifically related to appropriate customer service anbehavior, providing a potential indication that expectations for interactive dialhuman service scripts (Solomon, Surprenant, Czepiel & Gutman, 1985). Comparison of the Usability Scale for Speech Interfaces to measures 81


Goal Orientation factor included items that measure several dimensions of e-SQ: Ease of e Customer Service e e-SQ, which ludes items related to auditory presentation (Speech Characteristics) and expectations of appropriate conversational behavior (Verbosity), which is more similar to expectations of human conversation (Holtgraves, of a Price Knowledge dimension, whrity of A final comparison may be made between measures of human service provider personality and the Usability Scale for Speech Interfaces. The Usability Scale includes items that map to items easuring the confidence n extraversion and ount, Barrick & Stewart, 1998). Mount, Barrick and Stewart (1998) found that Conscientiousness was most strongly correlated with job performance. Following from this finding, it is intuitively reasonable conscientious service provider would provide efficient service, behave with the customers bems included in the d by Baydoun, Rose, friendly, polite, and helpful). In summary, comparison of the Usability Scale for Speech Interfaces with its predecessor and related e-service scales shows a number of similarities with other measures, as well as a number of differences. The scale refines specific factor scores (i.e., Speech Characteristics, Navigation, Reliability, Efficiency, and Customization/Personalization. ThBehavior scale is similar to their Assurance/Trust dimension. In contrast to thincluded items related to the visual design of a website, the present scale inc2002). Another contrast with the e-SQ measure was its inclusion ich was not included in the speech usability scale, and may reflect the populacomparison-shopping on commercial Internet websites (Rust & Kannan, 2002).measured in human service delivery. For example, Usability scale items mand enthusiasm of a speech system are related to findings of associations betweeinterpersonal skills (Alge et al., 2002) and performance in service providers (Mthat a st interests in mind, do their work correctly, and infer customer needs, all iteUsability Scale. The scale also includes several items specifically identifieand Emperado (2001) as being part of a customer service orientation (i.e., 82


Verbosity) for the customer service context more than do previous scales, which defined these om both human and e has much in common forms of service, the usability of this new form of customer service is uniquely operationalized. Use of Experts and Students to Measure Speech System Usability ge behavior and college to measure usability. Generally, both groups demonstrated relatively consistent imps in applied measurement. Both experts and students rated the Financial Services interfaces as highest in quality. The Services followed son of the overall dicates that experts rated all interfaces more negatively than did observers on the seven-point scale. This difference was potentially due to their greater or more conscious understanding of conversational norms, which caused them to rate the d and Movies interfaces: nd observers rated veral potential reasons for this discrepancy. First, the most obvious difference between these interfaces was that the Tennis interface used a British English dialect as the system voice. Experts may have recognized the dialect without judging it but this speech pattern may have reduced the intelligibility of prompts for students, leading to their more negative impressions of the interface. Second, the factors in a more open-ended manner. In addition, the scale includes items frInternet service quality measures. Thus, although speech-based e-servicwith other The current research used both experts in human speech and languastudents pressions of six speech interfaces, suggesting the potential of using both grourelative ranking of the six interfaces was similar for both groups (e.g., Financialby Prescription Refill, Flight Status and Directory Dialer poorest). A compariUsability Scale score with mean expert ratings in interfaces more negatively when these norms were violated. A group difference in ratings occurred for the Tennis Scoreboarexperts rated Tennis (expert M=3.37, observer M=4.01) as superior to Movies aMovies (expert M=2.99, observer M=4.33) as superior to Tennis. There are se 83


experts, who by definition have deep expertise in speech and language, may perceive interface groups is likely d extensive knowledge tion (i.e., metalinguistic skills), instead of heuristic processing. The third reason that the student and expert ratings differed may be due to differences in the interface characteristics measured by each group. is and intonation, ngs were limited to ents rated a more extensive set of questions that included speech and language, but included a broad array of items concerned with technology usage and customer service as well. Finally, the students may have had nerally interested in her perceived efulness to the students. In general, considering the differences in expertise and measurement of the two samples in these studies, both expert ratings and students demonstrated agreement about the relative quality of the six example interfaces, suggesting the concurrent validity of the Usability vice users and are ample in this study was n for speech-based e-service. This result contradicts Dabholkar and Bagozzis (2001) assertion that basic demographic variables do not have an influential role in affective outcomes to e-service. However, future research required to more fully elucidate the potential impact of a broad age range on e-service outcomes. Much of the human factors literature has suggested that elderly individuals characteristics differently than students. A general difference between thesebecause experts perceptions may be subject to controlled processing anabout underlying cognitive and physiological factors involved in speech producFor example, experts rated contingency of system responses and use of emphasitems that were not specifically measured with the Usability Scale. Expert ratiaspects of speech and language, consistent with their expertise. In contrast, studa more positive affective response to the Movies interface because they are geand frequent this form of entertainment. Thus, the interface itself may have higus Scale. Although students do comprise a large proportion of speech-based e-serlikely to be the majority of future users of this technology, the participant srestricted. Nonetheless, age was a significant predictor of customer satisfactio 84


require special considerations based on changing cognitive, motor, and other skills related to aging ith speech systems as that the present study should be replicated with a broader range of potential e-s The current results indicate that both speech and language experts and naive observers have a number of advantages for applied measurement. Because experts appeared to be highly be an efficient velopment, expert user population is given access to an e-service system. Given the significant business implications of poor interface design, expert ratings may prevent significant financial loss if they are used as part of l weakness of expert evaluate the same also has an advantage, in that they are more numerous and possibly easier to recruit than speech and language experts. Ideally, a representative sample may be selected from the user population and their ratings may be uswever, an adequate limit their One of the most important potential limitations of the present research was the use of observers instead of actual interface users. Findings from social cognition highlight this issue for not only applied speech technology research, but also marketing and interpersonal communication studies, which frequently use observers to generate data on conversational and service interactions. (Rogers, 1997). These skills are likely to impact their ability to interact wwell, suggesting ervice participants. critical of and sensitive to speech and language behavior, their impressions mayindication of a speech systems use of conversational norms. Used early in deratings may be used to identify and refine problematic aspects of an interface design before the an iterative process of evaluation and design during development. A potentiaratings is that they have poor reliability, suggesting that they will change if theysystem multiple times. Use of students (and other non-expert potential users) ed to better assess the post-deployment outcomes of a speech system. Hosample size will be more expensive and time-consuming to obtain, which will practicality of sampling users in industrial settings. Use of Third-Party Observers to Measure Affective Outcomes 85


In contrast, findings from the social-cognitive literature warn that interactants and observers may vers of speech ethodological problem has cal means of conducting applied research. In addition, vision clips have a central role in the proliferation of speech technology because they are the primary means of demonstrating speech technology fu of future research lored. not be underestimated. Because vision clips and an observer point of view are involved in the success of speech technology, the current research is a critical first step to understanding affective responses to this form of ast point of view as an e outcomes of speech to compare their perceptions provide an excellent methodology that may be utilized in future studies (Gilovich, 1987). Increased variability in stimuli provides an empirical challenge even with yoked y and quickly or have to resolve h system at all. To avioral differences (e.g., number of turns, average time for turn exchange) and statistically control behavioral variability across user-technology interactions. Thus, although the user-technology interactions will differ across the user and yoked observer pairs, methods of statistical control will remove variance due to interaction differences, allowing the researcher to measure the effects of point of view more have different affective outcomes. Thus, the present results are limited to obserinterface usage and do not necessarily apply to users themselves. This mimportant implications because the use of observers is an efficient and practi nctions to stakeholders before a system is designed. It should be a central goalefforts that potential differences in user and observer affective responses be expIn spite of this weakness, the validity of the present research shoulde-service. Continued advancement in this field requires that future studies cindependent variable to better elucidate potential differences in affectivtechnology. Studies that have yoked conversational interactants and observers user-observer pairs, in that some users may proceed through an interface easilsignificant problems, requiring not only more time but more system messagescommunicative difficulties. In some cases, users may not be able to use a speecresolve this issue, researchers may rate specific aspects of user-system beh 86


effectively. o have several theoretical r pragmatic and social-cognitive theories, extending their relevance to a new applied discipline. The current findings highlight the relevance of conversational and service expectations for affective responses to for applied work. nd social-cognitive theory provides useful explanations of observations and empirical findings in speech technology. As pragmatic theory suggests, social goals are realized and inferred through the use of language. s is the Usability for speech e-service seemed efficient, The system allowed me to do things that are important to me, The system seemed to know my needs). In addition, the Usability Scale also included items that refer to the power relationship like I was in control) e repetitive, The system logue, to be perceived positively, a sp conversational behavior. Thus, the content of the Usability Scale itself appears to add to the existing empirical support for the applicability of pragmatic theory to the design of high-quality conversational technologies (Bernsen, Dybkjaer, & Dybkjaer, 1996; Saygin & Cicekli, 2002). Also consistent with the existing research, the Usability Scale included a number of items Theoretical Implications In addition to their methodological implications, these studies alsimplications. In particular, the present results provide empirical support fo e-service but underscore the weaknesses of a number of construct definitionsThe Explanatory Role of Pragmatic and Social-Cognitive Theory The results of the present research reinforce the notion that pragmatic aPerhaps the strongest evidence that this view holds for speaking technologieScales inclusion of a number of items that refer to user goals and priorities usage (e.g., I could complete my business quickly using this system, The systemimplied by a speech systems use of language (e.g., The system made me feeland appropriate adherence to conversational norms (e.g., The messages wergave me more details than I needed). As with human-human dia eech systems language usage should be in line with the context and expected 87


concerned with impressions of speech and language usage. For example, items that measured the reinforce the 2; Cargile, Giles, strated with human speakers, including those with various types of disordered, accented, dialectal, and typical speech (Clopper & Pisoni, 2004; Collins & Missing, 2003; Feinberg, Jones, Little, Burt & Perrett, 2005; Myers, Faragasso, non-human speakers. r satisfaction of the four Usability Scale factors. This finding provides some empirical support for the potential importance of speech characteristics over other aspects of interface behavior and is consistent with e automatic trait theory-based explanation for the non-volitional perception of a speaking technologys humanlike traits than other alternative explanations. Nass and Lee (2001) acknowledged that individuals recognize that avior may be an speech interface is ike perceptions about n specific speech and language behaviors and their resulting perceptions is an important priority for not only for basic research, but also interface design. Ideally, interface designers should have an empirical basis for designing specific speech and language variables and should be able to articulate the desired affective response in the user. Conversely, it would seem that speech interface design offers an naturalness, confidence, professionalism, and enthusiasm of the system voice importance of speech behavior in social impression formation (Holtgraves, 200Ryan & Bradac, 1994; Patterson, 1996). Although this finding has been demonFlege, 1998; Flege & Fletcher, 1992; Munro and Derwing, 1995, 1998; St. LouisTownsend, & Gallaher, 2004), the current results extend these findings toInterestingly, the Speech Characteristics factor had the largest effect on custome previous results (Nass & Lee, 2001; Polkosky, 2003). Although the current research did not directly examine automaticity, thinference literature would seem to provide a more intuitively satisfying andtheir trait inferences about speaking technologies are inappropriate. This behexample of postconscious automaticity, because individuals are aware that the the basis of their perceptions, although they cannot explain having humanltechnology (Bargh & Ferguson, 2000). Understanding the causal links betwee 88


excellent method for modeling human conversational behavior and controlling speech and logy modeling may magnetic resonance e impact of auditory, speech, and linguistic variables on brain activation and automatic trait inference. Items excluded from the Usability Scale may also be explained by the social-cognitive and accuracy items (e.g.., I d more than I usability scale development analyses. Anecdotal evidence in the speech industry has suggested that recognition accuracy is a frequent concern of engineers. It is asserted that users often reject poorly-designed I say), even when is attribution of poor nsion by speech systems, the current results suggest that the present measurement approach did not yield reliable ratings, possibly due to inaccurate human perception of others comprehension (Keysar, Barr & Horton, 1998; Keysar & Henley, 2002) or other misattributions er of inaccurate enley (2002) rsation. Keysar, Barr and Horton (1998) argue that both speakers and listeners process conversation egocentrically, assuming the clarity of communication and listener comprehension, until overt errors require adjustment to the stream of conversation. In a broader consideration of communicative impacts on social perception, Hilton (1995) argued that conversational norms are an independent variable that language variables to experimentally examine social perception. Speech technoalso be combined with new methods of cognitive research, such as functionalimaging, to more fully examine th pragmatic literature. For example, the elimination of perceived recognitionfelt frustrated because the system didnt understand me; The system understoothought it would) due to low reliability was a somewhat surprising outcome ofspeech interfaces due to poor recognition (e.g., It doesnt understand what recognition accuracy rivals that of human conversational partners. Despite thcomprehe (Hilton, 1995). Several research streams suggest that individuals may have a numbperceptions about and explanations for communication behavior. Keysar and Hshowed that speakers tend to overestimate listener comprehension during conve 89

PAGE 100

affects attributional processes. Hiltons attributional model of conversational implicature draws on bout the interaction. e in mental ttributions form the foundation of social perception and judgment in interpersonal settings. Accordingly, because the social cues provided by speech systems may be designed ambiguously or inappropriately, users attstems poor he users utterance. uman communication. In non-human animal research, despite empirical demonstration of animal comprehension of language comparable to human children (Kaminski, Call & Fischer, 2004) and 1; Sims & Chin, and frequently rejected teraction research, Nass and Moon (2000) also reject anthropomorphism as an explanation of their demonstrations of human social responses to computers. Applying social-cognitive and pragmatic theory to expect technology to be cooperative and engage in t would be negative affective Expectations and Social Perception of Speech-Based e-Service Expectations seem to underlie many of the results observed in this research. As previously suggested, conversational norms such as the Gricean (1975) maxims are one important source of expectations about speech system behavior. Glovers (1995) discussion of various types of Grice (1975) to assert that partners enter conversations with expectations aWhen a speaker violates Grices (1975) conversational maxims, partners engagcalculus to arrive at a causal explanation of the speakers intention, and these a ributions about the cause of communicative difficulties are likely to involve the sycomprehension because its behavior does not appear to be contingent on tInterestingly, misattribution about comprehension also occurs in non-hhuman perception of animal comprehension (Pongracz, Miklosi, & Csanyi, 2002002), the attribution of mental states to animals remains a controversialexplanation of findings (Schilhab, 2002; Wynne, 2004). In human-computer in human-computer interaction means that users will appropriate conversational behavior. When these expectations are violated, ireasonable to assume that attributions about the cause of atypical behavior andresponses would result. 90

PAGE 101

conversations and their associated rules suggests that all conversations are subject to a set of f the self and behaviors such as an, 1976; Sacks, Schegloff, & Jefferson, 1974; Schegloff, 1968; Schegloff, Jefferson, & Sacks, 1977; Schegloff & Sacks, 1973). Behaviors within a conversational script influence both participants social cognitive the literature ank & Abelson, 1977). The current findings lend empirical suend these findings to the realm of human-computer conversation. In addition to expectations about the structure and content of conversation, items included ese technologies ervice providers 94; Holland & Baird, 1968; Yagil, 2001). Items that were included in the Customer Service Behavior factor included the friendliness, politeness, enthusiasm, and courteousness of the system, as well as its use of familiar o require speech of expectations, e the usability scale included items that measure the naturalness of the system voice and the exrofessional voice talent recordings for system voices in the speech industry seems to be in line with the Usability Scales inclusion of an item that measures the similarity between a system voice and those on radio and television. In the services marketing literature, the role of expectations has been identified as an expectations that include role schema about the relative power relationship oconversational partners. In addition, conversation includes a set of expected greeting, turntaking, closing, and repair of communicative breakdowns (Goffmprocessing and subsequent behavior, a perspective that is broadly supported by(Abelson, 1981; Holtgraves, 2002; Sh pport to the association between expectations and affective outcomes and extin the Usability Scale for Speech Interfaces also suggest that expectations for thmay originate from a variety of sources. For example, the behavior of human sseems to figure prominently in the Usability Scale (Alge et al., 2002; Cran, 19terminology. However, in addition to these general expectations, users seem tsystems to be efficient. The general behavior of humans may also be a sourcebecaus tent to which it sounds like a regular person. Finally, the dominance of p 91

PAGE 102

important determinant of affective responses such as customer satisfaction. Researchers in this d actual service e suggested that nating roles in a ritualized script (Mohr & Bitner, 1991; Solomon, Surprenant, Czepiel & Gutman, 1985). As with conversational expectations, service expectations seem to also be important to the usability oners, because they rk in applied speech technology has focused on designing a unique personality into a speech system (Cohen, Giangola & Bargh, 2004; Kotelly, 2003), using an upbeat, energetic prompt presentation by a professional n that speech interfaces troversion) is approach, the current research indicates that systems that incorporate a constrained set of behaviors (those associated with the service provider role, such as helpfulness, friendliness, awareness of user needs) Gutman, 1985) are e current state of the k for users individual However, this behavior underlies the relatively simplistic but common design of many current systems that ask users to identify their preferred language for interaction (e.g., To continue in English, say English), a behavior that is often mentioned by individuals as a disliked aspect of e-service. Continued research in sources of expectations and direct examination of expectations will reveal field argue that service quality is determined by the gap between expected an(Parasuraman, Zeithamal & Berry, 1985, 1988, 1994). Several researchers havfavorable affective outcomes of service occur when the provider and customer adopt coordi and affective outcomes of speech-based e-service. These findings are particularly important for speech technology practitiprovide parameters for interface design. One of the predominant strains of wovoice talent. This design strategy has been adopted based on Nass assertioshould be designed to have personality characteristics (such as introversion or exconsistent with the personality of the user (Nass & Lee, 2001). In contrast to thand are designed to adhere to a service script (Solomon, Surprenant, Czepiel &more effective than unconstrained personality (persona) design. Given thtechnology, it is impractical for a speech interface to infer or explicitly asdifferences and then adapt itself to each unique individual in a user population 92

PAGE 103

design strategies that appropriately accommodate users and result in desirable affective outcomes. Ret tool was an important goal ay have implications for speech technology research. A broad review of relevant literatures indicated that customer satisfaction was an important potential outcome of speech technology usage. It also ide findings provide a theoretically defined in very similar ways, including usability, service quality, user satisfaction, customer satisfaction, and perlocutionary force. The current study did little to resolve this issue but it did indicate that tion in speech fort was included ary force, but it was predicted by virtually the same set of usability items as customer satisfaction. Unfortunately, comfort did not appear to contribute a unique or additional perspective on affective responses to speech technology, beck, 2003; tisfaction in industrial her than comfort for future onstructs, a useful extension of the current research would be to use advanced statistical techniques such as path analysis or structural equation modeling to compare several hypothesized models of antecedents and affective outcomes to e-service. An empirical approach to these construct interrelationships would be useful in addressing the lack of clarity in their theoretical boundaries. fining Constructs to Accommodate New Applications of Theory Although the development of an applied usability measuremenof this research, equally important was evaluating the potential variables that m ntified a number of potential antecedents and related constructs. The presentnew perspective on these variables and their applicability to speech technology. The literature review also suggested that a number of variables areusability is strongly associated with customer satisfaction, an important assumptechnology that was not previously based on empirical findings. In addition, comin the present study because of its apparent similarity to perlocutionexcept to highlight the importance of trust (Corritore, Kracher & WeidenGrabner-Krauter & Kaluscha, 2003). Due to the emphasis on customer sasettings, research efforts should probably use this dependent variable ratwork. To resolve unclear relationships among c 93

PAGE 104

One issue in defining the relevant constructs was clarified in these studies. The current ale for Speech imilarity to other rry, 1994) and e-SQ (Zeithaml, Parasuraman & Malhotra, 2000). These findings suggest that service quality and usability may be operationalized similarly for speech-based e-service. This perspective suggests d contextually-defined constructs tha the type of service Another question that guided this research was to determine whether customer satisfaction was more closely associated with individual differences or interface perceptions. The results y seeking, and In contrast, perceived on. Interestingly, the related disciplines seem to handle these antecedents in different ways: many marketing researchers seem to examine individual differences and human factors researchers appear to be perceptions. The observed discrepancy in the relative associations amecause it suggests that consideration for g some design recommendations (Nass & Lee, 2001). The present research was also concerned with individual difference and interface perception as predictors of customer satisfaction. Wickens, Gordon and Liu (1998) identify prediction as a primary characteristic of human factors as a scientific discipline. They state that it work did not attempt to measure service quality. However, the Usability ScInterfaces includes items that measure service expectations and has notable sservice quality scales such as SERVQUAL (Parasuraman, Zeithamal & Bethat both usability and service quality are flexible, abstract, an t depend largely on expectations of the user or receiver of service, as well asprovided (e.g., Internet e-service, speech-based e-service, human provider)indicated that need for interaction with a service provider, inherent noveltself-monitoring are weakly positively correlated with customer satisfaction. usability variables showed strong positive correlations with customer satisfactimore concerned with interface ong these variables is important for development of speech systems bperceptions of speech interfaces should be a more important and practical interface designers than catering to user personality characteristics, contradictin 94

PAGE 105

is important to be able to predict that solutions that are envisioned to create good human factors will of largely sfaction than a similar reinforces the relative importance of interface perceptions to speech technology affective outcomes. A practical outcome of this work is that the regression equations will provide expected customer satisfaction he value of conducting usest practice task for speech otelly, 2002). Nonetheless, the weak correlations between individual differences and customer satisfaction and lack of interactive effects for quality and individual differences were somewhat have been partially 997), this type of methodology, known as Ateractions due to a wide variety of design problems, including weak effects (p. 585). It is likely that the interactive effects are the result of weak effects in the current research. teristics and interface nt methodology. ean the differences ately cost savings for enterprises that implement speech technology. In accordance with their doctrines of human factors, Sanders and McCormick (1995) suggest that individual differences should be considered in interface design, thus implying that user and system characteristics interact. Although the current results provide one approach to examining potential user-system variable interactions, research on actually succeed when put into practice (p. 7). A regression equation made upindividual difference predictors accounted for less variance in customer satiequation with interface perceptual (usability) interface items. This finding alsoscores when only usability scores are known. This information will add to t ability evaluations during development, already considered a critical bapplications (Balentine & Morgan, 1999; Cohen, Giangola & Balogh, 2004; Kunexpected. The methodology used to respond to research question five mayresponsible, however. According to Pedhazur (1 tribute-Treatment Interaction (ATI) research, has a paucity of findings of intIn spite of this problem, the potential interaction between user characdesign characteristics should not be dismissed due to the limitations of the curreInteractive effects do appear to have practical significance because they may msuch as customer satisfaction, repeat usage of speech-based e-service, and ultim 95

PAGE 106

this topic (like ATI research in general) is still in its early stages (Pedhazur, 1997). Therefore, tion for what to methodologies to explore d customer satisfaction. Although this method has been attempted previously (Polkosky, 2003), its primary limitation was that there were no scales for measuring usability, nor a comprehensive understanding of important afrrent work is an important step in this line of Although the present research began to establish a social-cognitive psychology by developing a basic measurement tool and illustrating the explanatory value of theory, it also al should be to sychology, es marketing and human factors; and applied research in speech technology and conversational design. As Brennan (1998) observed, applied work with speech technologies will promote questions that challenge and n for applied design. Past resis imperative that edge base for both s that could facilitate more sophisticated research in all of these fields and exploit cross-disciplinary findings. In addition, the current research provided an empirical and theoretical basis for positing several important affective outcomes of speech technology, but its design did not allow for interpretation of causal relationships. Because these studies were the first to integrate variables because the current research has developed a theoretical and empirical foundameasure, future research efforts should begin to focus on experimental these possible user-system interactions, as well as their impact on usability an fective outcomes. The cu applied research. Future Research highlights the significant amount of work that remains. An important future goexpand the symbiotic relationships among basic research and theory in social ppsycholinguistics, and interpersonal communication; applied research in servicstrengthen theoretical models and basic theory will provide a foundatio earch has seen each of the relevant fields progress largely in parallel but it more cross-disciplinary research be undertaken to provide a robust knowlresearchers and practitioners. The current studies have suggested several topic 96

PAGE 107

from different disciplines, it was a necessary step to identify which variables should be measured dvance the field by d measuring their impact rchers may manipulate syntactic forms, lexical choices, gender of system voice, and messages provided by the system when a miscommunication occurs (Berger, 2001; Holtgraves, 2002) to determine their will promote better an-computer al psychology and interpersonal communication researchers are important because designers should have control over linguistic and speech variables in creating an interactive e-service and should be able to manipulate y did not provide adequate re research. A nipulate types of errors depicted in auditory clips and evaluate their impact on observers, similar to Kreuz and Roberts (1993) methodology in which they manipulated articulation, syntactic, and pragmatic errors and measured attributions about which conversational lso an important ng ssible, difficult, or highly unusual with human interactants. Such modeling will also provide a means for testing theoretical models of interpersonal communication and cognition (Bradac, Cargile & Hallett, 2001; Patterson, 1996), a task that should be undertaken to better explain social impressions about speech systems. A paucity of research that empirically demonstrates causal relationships among specific and to examine associations among them. Future researchers will continue to amanipulating speech, language, and social aspects of speech interfaces anon affective outcomes (e.g., customer satisfaction). For example, applied reseaimpact on customer satisfaction. Research designs that allow causal inferences user-system conversations and facilitate the next generation of naturalistic humdialogues. Experimental studies following the examples provided by basic socithem to achieve desired user responses. As an example, the present studvariability in dialogue error, but this variable has considerable potential for futuresearcher may ma partner caused the error. Applying findings from the present research back to basic research is afuture task. For example, speech interfaces provide an ideal medium for modelisocial-communicative interactions and controlling them in ways that are not po 97

PAGE 108

speech and language behaviors, observer (or interactant) perceptions, and affective outcomes exists ge variables are are very few rface designers understand specifically how to combine interface variables to facilitate desired outcomes. Another area challenged by the present research concerns the selection of interactants for oncerned with & Redmond, 2002), al, goal-oriented interactions. By contrast, service delivery appears to provide a constrained yet highly social and goal-oriented domain for investigating affective and cognitive impacts of both typical and atypical nversational ally, applied work cores the importance and relevance of basic research in a number of areas that have had limited exploration to date, including communication failure (Berger, 2001) and automatic trait inferences (Bassili, 1976; Brown & Bassili, 2002; Ham & Vonk, 2003; Skowronski, man factors is a case with other, more efit from application of basic knowledge about human social, cognitive and communicative behaviors to develop increasingly usable, intuitive, and pleasurable conversational interactions with technologies. Perhaps the best argument for continued research at the intersection of social cognition, communication, psycholinguistics, services marketing, and human factors was advanced by in basic literature. Although it is generally accepted that speech and languaimportant to social perception and affective responses (Holtgraves, 2002), therefindings that help inte the majority of interpersonal communication research. Much of this work is clong-acquainted pairs or couples who are romantically linked (Beebe, Beebe, making a substantial portion of this research difficult to generalize to more casuspeech, language, and social behavior. Increased focus on a broad variety of cointeractions will provide more robust findings for applied work and theory. Finin speech technology unders Carlston, Mae, & Crawford, 1998). As Sanders and McCormick (1993) observed over a decade ago, hudynamic, ever-expanding area of research and applied work. As is thetraditional areas of human factors, speech technology design will also ben 98

PAGE 109

researchers who may not have even foreseen this field: The brave new world of the future n factors discipline pments will, in reality, general (Sanders & McCormick, 1993, p. 754). It is to this larger goal that speech technology practitioners and researchers should aspire. should indeed be developed with people -all of us -in mind. Thus, the humamust be at the cutting edge of future developments to ensure that such develocontribute to the improvement of the quality of working life and of life in 99

PAGE 110

Abelson, R. (1981). Psychological status of the script concept. American Psychologist, 36(7), 715-729. Masters, R. (2002). Measuring cuinary test in a public Allen, M. (2002). Human values and product symbolism: Do consumers form product preference by comparing the human values symbolized by a product to the human values they enn market to exceed $5 bilApple, W., & Hecht, K. (1982). Speaking emotionally: The relation between verbal and vocal communication of affect. Journal of Personality and Social Psychology, 42, 864-875. 976). The voice of personality: Stereotyped judgments and their relation to -220. with words. Oxford: Clarendon Press. Auty, S., & Elliott, R. (1999). Reading advertising and writing identity. Advances in Consumer Research, 26, 439-444. Babin, B., & Griffin, M. (1998). The nature of satisfaction: An updated examination and analysis. Journal of Business Research, 41(2), 127-136. References Alge, B., Gresham, M., Heneman, R., Fox, J., & Mc stomer service orientation using a measure of interpersonal skills: A prelimservice organization. Journal of Business and Psychology, 16(3), 467-476. dorse? Journal of Applied Social Psychology, 32(12), 2475-2501. Allied Business Intelligence Research. (2003). Speech recognitio lion by 2008, according to new ABI study. Oyster Bay, NJ: Author. Aronovitch, C. (1 voice quality and sex of speaker. The Journal of Social Psychology, 99, 207Austin, J. (1962). How to do things 100

PAGE 111

Balentine, B., & Morgan, D.P. (1999). How to build a speech recognition application: A styGroup. behaviorism: On the automaticity of higher meBaron, R., Byrne, D., & Johnson, B. (1998). Exploring social psychology, 4th ed. Boston: Allyn and Bacon. eption of social events. logy, 33(6), 680-685. Journal of Personality and Social Psychology, 79(6), 941-952. Baydoun, R., Rose, D., & Emperado, T. (2001). Measuring customer service orientation: l of Business and Psychology, les influencing perception of the communicative competence of an adult augmentative and alternative communication system user. Journal of Speech and Hearing Research, 35(5), 1105-1113. on and call centers: (2), 162-173. t, W., Benoit, P., & Wilkie, J. (1996). Participants and observers memory for conversational behavior. Southern Communication Journal, 61, 139-155. Berger, C. (2001). Miscommunication and communication failure. In W.P. Robinson & H. Giles (Eds.), The new handbook of language and social psychology ( pp. 177-192). New York: John Wiley & Sons. le guide for telephony dialogues. San Ramon, CA: Enterprise Integration Bargh, J., & Ferguson, M. (2000). Beyond ntal processes. Psychological Bulletin, 126(6), 925-945. Bassili, J. (1976). Temporal and spatial contingencies in the percJournal of Personality and Social Psycho Bavelas, J., Coates, L., & Johnson, T. (2000). Listeners as co-narrators An examination of the validity of the customer service profile. Journa15(4), 605-620. Bedrosian, J., Hogg, L., Calculator, S., & Molineux, B. (1992). Variab Bennington, L., Cummane, J., & Conn, P. (2000). Customer satisfactian Australian study. International Journal of Service Industry Management, 11Benoi 101

PAGE 112

Bernsen, N., Dybkjaer, H., & Dybkjaer, H. (1996). Cooperativity in human-machine and hu W. (1997). Linguistic bases of social peBevan, N. (1995). Measuring usability as quality of use. Software Quality Journal, 4, 115-130. Boulding, W., Kalra, A., Staelin, R., & Zeithamal, V. (1993). A dynamic process model of seing Research, 30, Bradac, J., Cargile, A., & Hallett, J. (2001). Language attitudes: retrospect, conspect, and prospect. In W. Robinson & H. Giles (Eds.), The new handbook of language and social psychology y & Sons. h and through computers. proaches to interpersonal communication. (pp. 201-225). Mahwah, NJ: Lawrence Erlbaum. Briggs, S., & Cheek, J. (1986). The role of factor analysis in the development and 06-148. nitoring scale. l Psychology, 38, 679-686. Brown, R. & Bassili, J. (2002). Spontaneous trait associations and the case of the superstitious banana. Journal of Experimental Social Psychology, 38, 87-92. Brown, T., Churchill, G., & Peter, J. P. (1993). Improving the measurement of service quality. Journal of Retailing, 69, 127-139. man-human spoken dialogue. Discourse Processes, 21, 213-236. Berry, D., Pennebaker, J., Mueller, J., & Hiller, rception. Personality and Social Psychology Bulletin, 23, 526-537. rvice quality: From expectations to behavioral intentions. Journal of Market7-27. (pp. 601-623). New York: John Wile Brennen, S. (1998). The grounding problem in conversations witIn S. Fussell & R. Kreuz (Eds.), Social and cognitive ap evaluation of personality scales. Journal of Personality, 54, 1 Briggs, S., Cheek, J., & Buss, A. (1980). An analysis of the self-moJournal of Personality and Socia 102

PAGE 113

Brown, P., & Levinson, S. (1987). Politeness: Some universals in language usage. Ca W.P. Robinson (Eds.), Haohn Wiley & Sons. Burgoon, M., & Klingle, R. (1998). Gender differences in being influential and/or influenced: A challenge to prior explanations. In D. Canary & K. Dindia (Eds.), Sex differences anations of sex and gender plying a social meaning model to relational message interpretations of conversational involvement: Comparing observer and participant perspectives. Southern Communication Journal, 56, 96-113. ro-ed. ational Journal of HuCargile, A., Giles, H., Ryan, E., & Bradac, J. (1994). Language attitudes as a social process: A conceptual model and new directions. Language and Communication, 14, 211-236. ersity Press. rk, H. H. (1989). Contributing to discourse. Cognitive Science, 13(2), 259-294. Clopper, C. & Pisoni, D. (2004). Some acoustic cues for the perceptual categorization of American English regional dialects. Journal of Phonetics, 32, 111-140. Cohen, J. (1969). Statistical power analysis for the behavioral sciences. New York: Academic Press. mbridge: Cambridge UP. Burgoon, M. (1990). Language and social influence. In H. Giles & ndbook of language and social psychology, (pp. 51-72). New York: J d similarities in communication: Critical essays and empirical investigin interaction, (pp. 257-285). Mahway, NJ: Lawrence Erlbaum. Burgoon, J., & Newton, D. (1991). Ap Cairns, H. (1999). Psycholinguistics: An introduction. Austin: PCapretz, L. (2003). Personality types in software engineering. Intern man-Computer Studies, 58(2), 207-214. Clark, H. H. (1996). Using language. Cambridge: Cambridge UnivCla 103

PAGE 114

Colby, C. (2002). Techno-ready marketing of e-services: Customer beliefs about Rust & P. Kannan (Eds.), e-service: Nen of deindividuation in synchronous computer-mediate communication. Computers in Human Behavior, 15, 51-65. Collins, S. & Missing, C. (2003). Vocal and visual attractiveness are related in women. ncepts, evolving uter Studies, 58, 737-758. Cosmides, L. (1983). Invariances in the acoustic expression of emotion in speech. Journal of Experimental Psychology: Perception and performance, 9, 864-881. idation of the service orientation construct. The Services In & Lamias, M. (2001). Web surveys: Perceptions of burden. Social Science Computer Review, 19(2), 146-162. Cronin, J., & Taylor, S. (1992). Measuring service quality: A reexamination and exus SERVQUAL: Reconciling pe quality. Journal Crosby, L., Evans., K., & Cowles, D. (1990). Relationship quality in services selling An interpersonal influence perspective. Journal of Marketing, 54(3), 68-81. Czellar, S. (2003). Consumer attitude toward brand extensions: An integrative model and research propositions. International Journal of Research in Marketing, 20(1), 97-115. technology and the implications for marketing e-services. In R w directions in theory and practice, (pp. 25-44). New York: M.E. Sharpe. Coleman, L., Paternite, C., & Sherman, R. (1999). A reexaminatio Animal Behavior, 65, 997-1004. Corritore, C., Kracher, B., & Wiedenbeck, S. (2003). On-line trust: cothemes, a model. International Journal of Human-Comp Cran, D. (1994). Towards val dustries Journal, 14, 34-44. Crawford, S., Couper, M., tension. Journal of Marketing, 56, 55-68. Cronin, J., & Taylor, S. (1994). SERVPERF vers rformance-based and perceptions-minus-expectations measurement of serviceof Marketing, 58, 125-131. 104

PAGE 115

Dabholkar, P. (1994). Incorporating choice into an attitudinal framework: Analyzing mo21, 100-118. -based self-service alternative models of service quality. International Journal of Research in Marketing, 13(1), 29-51. Dabholkar, P., & Bagozzi, R. (2002). An attitudinal model of technology-based se. Journal of the Academy 2001). Spoken interaction with computers in native or non-native language Same or different? In Proc. Of 8th International Conference on Human-Computer Interaction, (pp.294-301). Washington: IOS Pres arising in the social intDeBono, K., Leavitt, A., & Backus, J. (2003). Product packaging and product evaluation: An individual difference approach. Journal of Applied Social Psychology, 33(3), 513-521. DePaulo, B. (1992). Nonverbal behavior and self-presentation. Psychological Bulletin, 11, 203-243. Dixon, J., Mahoney, B., & Cocks, R. (2002). Accents of guilt? Effects of regional accent, race, and crime type on attributions of guilt. Journal of Language and Social Psychology, 21(2), 162-168. dels of mental comparison processes. Journal of Consumer Research, Dabholkar, P. (1996). Consumer evaluations of new technologyoptions: An investigation of lf-service: Moderating effects of consumer traits and situational factorsof Marketing Science, 30(3), 184-201. Dahlback, N., Swamy, S., Nass, C., Arvidsson, F., & Skageby, J. ( ess. Darley, J. & Fazio, R. (1980). Expectancy confirmation process eraction sequence. American Psychologist, 35, 867-881. 1 105

PAGE 116

Eide, E., Aaron, A., Bakis, R., Cohen, P., Donovan, W., Hamza, W., Mathes, T., Picheny, ents to the IBM trainable ference on Acoustics, Speech, and Signal PrEroglu, S., & Machleit, K. (1990). An empirical study of retail crowding Antecedents and consequences. Journal of Retailing, 66(2), 201-221. Feinberg, D., Jones, B., Little, A., Burt, D. & Perrett, D. (2005). Manipulations of fumale voices. Animal Feinberg, R., Kim, I., Hokama, L., de Ruyter, K., & Keen, C. (2000). Operational determinants of caller satisfaction in the call center. International Journal of Service Industry Mew of Psychology, Fiske, S. T., & Taylor, S. E. (1991). Social Cognition. New York: McGraw-Hill, Inc. Flege, J. (1988). Factors affecting degree of perceived foreign accent in English sentences. ee of perceived foreign ank, M. G., & Gilovich, T. (1989). Effect of memory perspective on retrospective causal attributions. Journal of Personality and Social Psychology, 57, 399-403. Fussell, S. R., & Benimoff, I. (1995). Social and cognitive processes in interpersonal communication: implications for advanced telecommunication technologies. Human Factors, 37(2), 228-250. M., Polkosky, M., Smith, M., Viswanathan, M. (2002). Recent improvemspeech synthesis system. IEEE International Con ocessing Proceedings, (pp. 708-711). New Jersey: IEEE. ndamental and formant frequencies influence the attractiveness of human Behavior. anagement, 11(2), 131-141. Fiske, S. T. (1993). Social cognition and social perception. Annual Revi44,155-194. Journal of the Acoustical Society of America, 84(1), 70-79 Flege, J. & Fletcher, K. (1992). Talker and listener effects on degraccent. Journal of the Acoustical Society of America, 91(1), 370-389. Fr 106

PAGE 117

Fussell, S., & Krauss, R. (1992). Coordination of knowledge in communication: Effects ers assumptions about what others know. Journal of Personality and Social Psychology, 62ondon: Sage. Gilovich, T. (1987). Secondhand information and social judgment. Journal of Experimental Social Psychology, 23, 59-74. r: Context ynthetic speech and human speech? Impact on users performance, perception, and attitude. Conference on Human Factors in Computing Systems (CHI 2001) Proceedings (pp.158-165). Seattle: CHI. ch in on-line trust: A review ans, 58, 783-812. 998a). Damaged merchandise? A review of experiments that compare usability evaluation methods. Human-Computer Interaction, 13(3), 203-261. Gray, W. & Salzman, M. (1998b). Repairing damaged merchandise: A rejoinder. ds.), Syntax and erin, B. & Innes, J. M. (1989). Cognitive tuning sets: Anticipating the consequences of communication. Current Psychology: Research and Reviews, 8, 234-249. Ham, J. & Vonk, R. (2003). Smart and easy: Co-occurring activation of spontaneous trait inferences and spontaneous situational inferences. Journal of Experimental Social Psychology, 39(5), 434-447. of speak 378-391. Gilmore, A. (2003). Services marketing and management. L Glover, K. (1995). A prototype view of context and linguistic behavioprototypes and talk. Journal of Pragmatics, 23, 137-156. Gong, L., & Lai, J. (2001). Shall we mix s Grabner-Krauter, S., & Kaluscha, E. (2003). Empirical resear d critical assessment. International Journal of Human-Computer StudieGray, W. & Salzman, M. (1 Human-Computer Interaction, 13(3), 325-335. Grice, H. (1975). Logic and conversation. In P. Cole and J. Morgan (Esemantics 3: Speech acts, (p. 41-58). New York: Academic. Gu 107

PAGE 118

Harris Interactive. (2003). Harris Interactive Speech Satisfaction Study. Menlo Park, CA: Nuommunication and an quality and customer commitment as potential antecedents. Journal of Service Research, 4(1), 60-75. Hartson, H.R., Andre, T., & Williges, R. (2001). Criteria for evaluating usability mputer Interaction, 13(4), 373-410. duct appealingness. 4), 481-499. Henton, C. (2002). Challenges and rewards in using parametric or concatenative speech synthesis. International Journal of Speech Technology, 5, 117-131. the International Ph, N. (2001). The evaluator effect: A chilling fact about usability evaluation methods. International Journal of Human-Computer Interaction, 13(4), 421-444. Holland, J., & Baird, L. (1968). An interpersonal competency scale. Educational and 02). Language as social action: A social psychology of language use. Hillsdale, NJ: Erlbaum. Humphreys, M. (1996). Exploring the relative effects of salesperson interpersonal process attributes and technical product attributes on customer satisfaction. Journal of Personal Selling and Sales Management, 16, 47-57. ance. Harrison-Walker, L.J. (2001). The measurement of word-of-mouth cinvestigation of service evaluation methods. International Journal of Human-Co Hassenzahl, M. (2001). The effect of perceived hedonic quality on proInternational Journal of Human-Computer Interaction, 13( Henton, C. (1999). Where is female synthetic speech? Journal of onetic Association, 29(1), 51-61. Hertzum, M. & Jacobsen Psychological Measurement, 28, 503-510. Holtgraves, T. (20 108

PAGE 119

Inman, M. L., Reichl, A. J., & Baron, R. S. (1993). Do we tell less than we know or hear d? Exploring the teller-listener extremity effect. Journal of Experimental Social Psischer, J. (2004). Word learning in a domestic dog: Evidence for fast mapping. Science, 304, 1682-1683. Kappas, A., Hess, U., & Scherer, K. (1991). Voice and emotion. In R. Feldman & B. Rial interaction, (pp. ork: Cambridge UP. ed. Upper Saddle River, NJ: Prentice Hall. Keysar, B., Barr, D. & Horton, W. (1998). The egocentric basis of language use: Insights fro, 46-50. & Henly, A. (2002). Speakers overestimation of their effectiveness. PsKotelly, B. (2003). The art and business of speech recognition: Creating the noble voice. Boston: Addison Wesley. 96). Nonverbal behavior and nonverbal coa (Ed.), Advances in 91). Perspective-taking in communication: Representations of others knowledge in reference. Social Cognition, 9(1), 2-24. Krauss, R. M., & Fussell, S. R. (1996). Social psychological models of interpersonal communication. In E. T. Higgins & A. Kruglanski (Eds.), Social psychology: A handbook of basic principles (pp. 655-701). New York: Guilford. less than we are tol ychology, 29, 528-550. Kaminski, J., Call, J. & F me (Eds.), Fundamentals of nonverbal behavior: Studies in emotion and soci200-238). New Y Keppel, G. (1991). Design and analysis: A researchers handbook, 3rd m a processing approach. Current Directions in Psychological Science, 7(2)Keysar, B. ychological Science, 13(3), 207-212. Krauss, R. M., Chen, Y., & Chawla, P. (19 mmunication: What do conversational hand gestures tell us? In M. Zannexperimental social psychology (pp. 389-450). San Diego, CA: Academic PressKrauss, R., & Fussell, S. (19 109

PAGE 120

Kreuz, R., & Roberts, R. (1993). When collaboration fails: Consequences of pragmatic errraction. In M. He5-928). New York: Elsevier. Langer, E. (1989). Mindfulness. Cambridge: Harvard University Press. Lassiter, G. D., Geers, A., & Apple, K. (2002). Communication set and the perception of going behavior. Personality and Social Psychology Bulletin, 28(2), 158-171. ng (4th ed.). Chicago: Lennox, R., & Wolfe, R. (1984). Revision of the self-monitoring scale. Journal of Personality and Social Psychology, 46, 255-275. ffects of accountability. Psfactors associated with web site success in the context of electronic commerce. Information and Management, 38(1), 23-33. Lombard, M. & Ditton, T. (1997). At the heart of it all: The concept of presence. Journal d Communication [Online], 3(2). Available: htt, D., & Skowronski, J. (1999). Spontaneous trait transference to familiar comunications: Is a little knowledge a dangerous thing? Journal of Personality and Social Psychology, 77(2), 233-246. Mattila, A., & Enz, C. (2002). The role of emotions in service encounters. Journal of Service Research, 4(4), 268-277. ors in conversation. Journal of Pragmatics, 19, 239-252. Landauer, T. K. (1988). Research methods in human-computer inte lander (Ed.), Handbook of human-computer interaction (pp. 90 on Lehmann, D., & Winer, R. (1997). Analysis for marketing planniIrwin. Lerner, J., & Tetlock, P. (1999). Accounting for the e ychological Bulletin, 125(2), 255-275. Liu, C., & Arnett, K. (2000). Exploring the of Computer Mediate p:// Mae, L., Carlston m 110

PAGE 121

McCann, C. D., Higgins, E. T., & Fondacaro, R. (1991). Primary and recency in sive audiences and multiple encodings influence sulf-service technologies: Understanding customer satisfaction with technology-based service encounters. Journal of Marketing, 64, 50-64. petence in memory lletin, 28, 78-89. ers and employees in service encounters. Advances in Consumer Research, 18, 611-617. Monahan, J. L. (1995). Information processing differences of conversational participants ommunication Me-factor model of personality and performance in jobs involving interpersonal interactions. Human Performance, 11(2/3), 145-165. Munro, M. & Derwing, T. (1998). The effects of speaking rate on listener evaluations of nain synthetic speech: A motion. Journal of the Acoustical Society of America, 93, 1097-1108. Nass, C., & Lee, K.M. (2001). Does computer-synthesized speech manifest personality? Experimental tests of recognition, similarity-attraction, and consistency-attraction. Journal of Experimental Psychology: Applied, 7(3), 171-181. communication and self-persuasion: How succes bsequent evaluative judgments. Social Cognition, 9, 47-66. Meuter, M., Ostrom, A., Roundtree, R., & Bitner, M. (2000). Se Miller, J., & deWinstanley, P. A. (2002). The role of interpersonal comfor conversation. Personality and Social Psychology Bu Mohr, L., & Bitner, M. (1991). Mutual understanding between custom and observers: The effects of self-presentational concerns and cognitive load. C onographs, 62, 265-281. Mount, M., Barrick, M., & Stewart, G. (1998). Fiv tive and foreign-accented speech. Language Learning, 48(2), 159-182. Murray, I., & Arnott, J. (1993). Toward the simulation of emotion review of the literature on human vocal e 111

PAGE 122

Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Joocial interactions: uating the impact of negative expectancies. Journal of Personality and Social Psychology, 56, 374-386. Neumann, R., & Strack, F. (2000). Mood contagion: The automatic transfer of mood 3. Nicholls, J., Gilbert, G.R., & Roslow, S. (1998). Parsimonious measurement of customer sal of Consumer Marketing, 15(3), 239-253. Nielsen, J. (1993). Usability engineering. San Diego: Academic Press. an, D. (2004). Emotional design: Why we love (or hate) everyday things. Cay, 3rd ed. New York: McGraw-Hill, Inc. OCass, A. (2000). A psychometric evaluation of a revised version of the Lennox and Psychology and Marketing, 17(5), 397-419. equity and 53, 21-35. Olson, G., & Moran, T. (1998). Commentary on damaged merchandise? Human-Computer Interaction, 13(3), 263-323. Ostrom, A., Bitner, M., & Meuter, M. (2002). Self-service technologies. In R. Rust & P.K. Kannan, (Eds.), e-service: New directions in theory and practice. (pp. 45-64). Armonk, NY: M.E. Sharpe. urnal of Social Issues, 56(1), 81-103. Neuberg, S. (1989). The goal of forming accurate impressions during sAtten between persons. Journal of Personality and Social Psychology, 79(2), 211-22 tisfaction with personal service and the service setting. Journa Norm mbridge, MA: Basic Books. Nunnally, J., & Bernstein, I. (1994). Psychometric theor Wolfe revised self-monitoring scale. Oliver, R., & Swan, J. (1989). Consumer perceptions of interpersonal satisfaction in transactions: A field survey approach. Journal of Marketing, 112

PAGE 123

Parasuraman, A. (2000). Technology readiness index (TRI): A multiple-item scale to mech, 2(4), 307-320. conceptual model of service qu-50. Parasuraman, A., Zeithamal, V., & Berry, L. (1994). Alternative scales for measuring service quality: A comparative assessment based on psychometric and diagnostic criteria. Journal multiple-item scale y. Journal of Retailing, 64, 12-40. Paris, C., Thomas, M., Gilson, R., & Kincaid, J. P. (2000). Linguistic cues and memory for synthetic and natural speech. Human Factors, 42(5), 421-431. allel process approach. ds.), What's social about social cognition? Research on socially shPedhazur, E.J. (1997). Multiple regression in behavioral research: Explanation and prediction. Fort Worth: Harcourt Brace. ns: Reflections on ita Hill, and the like. Applied Cognitive Psychology, 7, 299-310. R. Sproat, J. Olive, & g (Eds.), Progress in speech synthesis (pp. 541-560). New York: Springer. Polkosky, M. (2002a). Usability evaluation of an alphanumeric speech user interface prototype. Unpublished data. Polkosky, M. (2002b). Usability evaluation of a food ordering speech user interface. Unpublished data. asure readiness to embrace new technologies. Journal of Service ResearParasuraman, A., Zeithamal, V., & Berry, L. (1985). A ality and its implications for future research. Journal of Marketing, 49, 41 of Retailing, 70, 201-230. Parasuraman, A., Zeithamal, V., & Berry, L. (1988). SERVQUAL: Afor measuring consumer perceptions of service qualit Patterson, M. L. (1996). Social behavior and social cognition: A parIn J. L. Nye & A. M. Brower (E ared cognition in small groups (pp. 87-105). Thousand Oaks, CA: Sage. Pezdek, K. & Prull, M. (1993). Fallacies in memory for conversatioClarence Thomas, An Pisoni, D. (1997). Perception of synthetic speech. In J. van Santen, J. Hirschber 113

PAGE 124

Polkosky, M. (2003). Measuring the interpersonal consequences of interactive speech tecosky, M. (2004). Usability evaluation of a directory dialer speech user interface. UnPolkosky, M. & Lewis, J. (2002). Effect of auditory waiting cues on time estimation in speech recognition telephony applications. International Journal of Human Computer Interaction, t and psychometric echnology, 6, 161-182. Pongracz, P., Miklosi, A. & Csanyi, V. (2001). Owners beliefs on the ability of their pet dogs to understand human verbal communication: A case of social understanding. Current nd reality. Retrieved Price, L., Arnould, E., & Tierney, P. (1995). Going to extremes: Managing service eng, 59, 83-97. ure and quantify usability attributes of man-machine int67). Piscataway, NJ: Rogers, W. (1997). Designing for an aging population: Ten years of human factors/ergonomics research. Santa Monica, CA: HFES. Russell, S. & Norvig, P. (1995). Artificial intelligence: A modern approach. Upper Saddle River, NJ: Prentice Hall. hnology. Unpublished manuscript, University of South Florida. Polk published data. 14(3&4), 463-488. Polkosky, M. & Lewis, J. (2003). Expanding the MOS: Developmenevaluation of the MOS-R and MOS-X. International Journal of Speech T Psychology of Cognition, 20(1-2), 87-107. Popova, E. (2004, Nov.). The speech technology market: Hype aDec. 18, 2004 from the Frost & Sullivan website: counters and assessing provider performance. Journal of MarketinRauterberg, M. (1996). How to meas erfaces. In Proceedings of the IEEE International Workshop, (pp. 262-2IEEE. 114

PAGE 125

Rust, R., & Kannan, P.K. (2002). The era of e-service. In R. Rust & P. Kannan (Eds.), e-sNY: M.E. Sharpe. 74). A simplist systematics for the orSalza, P. L., Foti, E., Nebbia, L., & Oreglia, M. (1996). MOS and pair comparison combined methods for quality evaluation of text to speech systems. Acta Acustica, 82, 650-656. ing and design (7th ed.). in computer mediated communication. Computers in Human Behavior, 15, 185-194. Saygin, A., & Cicekli, I. (2002). Pragmatics in human-computer conversations. Journal of ersational openings. American AnSchegloff, E. A., Jefferson, G., & Sacks, H. (1977). The preference for self-correction in the organization of repair in conversation. Language, 53, 361-382. 8, 289-327. hab, T. (2002). Anthropomorphism and mental state attribution. Animal Behavior, 63Schmidt-Nielsen, A. (1995). Intelligibility and acceptability testing for speech technology. In A. Syrdal, R. Bennett, & S. Greenspan (Eds.), Applied speech technology (pp. 153-179). Boca Raton: CRC Press. Searle, J. (1969). Speech acts. Cambridge: Cambridge UP. ervice: New directions in theory and practice (pp. 3-21). Armonk, Sacks, H., Schegloff, E., & Jefferson, G. (19 ganization of turntaking in conversation. Language, 4, 696-735. Sanders, M. & McCormick, E. (1993). Human factors in engineerNew York: McGraw Hill. Savicki, V., Kelley, M., & Oesterreich, E. (1999). Judgments of gender Pragmatics, 34, 227-258. Schegloff, E. A. (1968). Sequencing in conv thropologist, 70, 1075-1095. Schegloff, E. A. & Sacks, H. (1973). Opening up closings. SemioticaSchil 1021-1026. 115

PAGE 126

Shank, R., & Abelson, R. (1977). Scripts, plans, goals, and understanding. Hillsdale, NJ: Er (1979). Intraclass correlations: Uses in assessing rater reliability. PsSims, V. & Chin, M. (2002). Responsiveness and perceived intelligence as predictors of speech addressed to cats. Anthrozoos, 15(2), 166-177. on, D., Mae, L., & Crawford, M. (1998). Spontaneous trait traurnal of Personality Slowiaczek, L., & Nusbaum, H. (1985). Effects of speech rate and pitch contour on the perception of synthetic speech. Human Factors, 27(6), 701-712. ehavior. Journal of Personality and Soowitz (Ed.), Advances in experimental social psychology (pp. 85-128). New York: Academic Press. Snyder, M. & Cantor, N. (1980). Thinking about ourselves and others: Self-monitoring 34. Matters of Psychology, 51(1), 125-139. Snyder, M. & Haugen, J. (1994). Why does behavioral confirmation occur? A functional perspective on the role of the perceiver. Journal of Experimental Social Psychology, 30, 218-246. Solomon, M., Surprenant, C., Czepeil, J., & Gutman, G. (1985). A role theory perspective on dyadic interactions : The service encounter. Journal of Marketing, 49, 99-111. lbaum. Shrout, P., & Fleiss, J. ychological Bulletin, 86(2), 420-428. Skowronski, J., Carlst nsference: Communicators take on the qualities they describe in others. Joand Social Psychology, 74(4), 837-848. Snyder, M. (1974). Self-monitoring of expressive b cial Psychology, 30(4), 526-537. Snyder, M. (1979). Self-monitoring processes. In L. Berk and social knowledge. Journal of Personality and Social Psychology, 39, 222-2 Snyder, M. & Gangestad, S. (1986). On the nature of self-monitoring:assessment, matters of validity. Journal of Personality and Social 116

PAGE 127

Spake, D., Beatty, S., Brockman, B., & Crutchfield, T. (2003). Consumer comfort in sevice Research, 5(4), 316-332. ies of & H. Giles (Eds.), The new handbook of language and social psychology (pp. 601-623). New York: John Wiley & Sons. 2004). Perceptual 998). Similarities and differences between the impact of traits and expectancies: What matters is whether the target stimulus is ambiguous or mixed. Journal of Experimental Social Psychology, 34, 227-245. : What do people learn in vidndane social events. Journal of Personality and Social Psychology, 64(3), 365-376. Turing, A. (1950). Computing machinery and intelligence. Mind, 59, 433-460. Sloan, M., & Nass, C. (2001). Effects of prational Conference on ington: IOS Press. Walther, J. (1996). Computer-mediated communication: Impersonal, interpersonal, and hyperpersonal interaction. Communication Research, 23(1), 3-43. Walther, J. (1997). Group and interpersonal effect in international computer-mediated collaboration. Human Communication Research, 23(2), 342-369. rvice relationships: Measurement and importance. Journal of SerSpears, R., Lea., M. & Postmes, T. (2001). Social psychological theorcomputer-mediated communication: Social pain or social gain. In W. Robinson St. Louis, D., Myers, F., Faragasso, K., Townsend, P. & Gallaher, A. (aspects of cluttered speech. Journal of Fluency Disorders, 29(3), 213-235. Stapel, D., & Schwarz, N. (1 Storck, J., & Sproull, J. (1995). Through a glass darkly eoconferences? Human Communication Research, 22(2), 179-219. Trafimow, D., & Wyer, R. (1993). Cognitive representation of mu Vanhoucke, V., Neeley, W.L., Mortati, M., ompt style when navigating through structured data. In Proc. of 8th InternHuman-Computer Interaction, (pp.530-536). Wash 117

PAGE 128

Wellens, R. A. (1993). Group situation awareness and distributed decision making: From ual and group decision making: Cuntion, impressions, and recall in complex social interactions. Journal of Personality and Social Psychology, 45(3), 538-549. 98). The psychology of language. Boston: Houghton Mifflin. an factors engineering. Wirtz, J., & Lee, M. C. (2003). An examination of the quality and context-specific applicability of commonly used customer satisfaction measures. Journal of Service Research, 5(4), 34ation of self-fulfilling prsychology, 10, 109-120. Wyer, R., Budesheim, T. & Lambert, A. (1990). Cognitive representation of conversations about persons. Journal of Personality and Social Psychology, 58, 218-238. informal be ascribed to Nature, 428, 606. Xue, M., & Harker, P. (2002). Customer efficiency: Concept and its impact on e-business management. Journal of Service Research, 4(4), 253-267. Yagil, D. (2001). Ingratiation and assertiveness in the service provider-customer dyad. Journal of Service Research, 3(4), 345-353. military to civilian applications. In N.J. Castellan, Jr. (Ed.), Individ rrent issues (pp. 267-291). Hillsdale, NJ: Lawrence Erlbaum Associates. White, J., & Carlston, D. (1983). Consequences of schemata for atte Whitney, P. (19 Wickens, C., Gordon, S., & Liu, Y. (1998). An introduction to humNY: Addison Wesley Longman. 5-355. Word, C., Zanna, M., & Cooper, J. (1974). The nonverbal medi ophesies in interracial interaction. Journal of Experimental Social P Wyer, R., Swan, S., & Gruenfeld, D. (1995). Impression formation in conversations. Social Cognition, 13, 243-272. Wynne, C. (2004). The perils of anthropomorphism: Consciousness should animals only with extreme caution 118

PAGE 129

Zajonc, R. (1960). The process of cognitive tuning in communication. Journal of Abframework for re research and managerial practice (MSI Report No. 00-115). Cambridge, MA: Marketing Science Institute. Zhu, F., Wymer, W., & Chen, I. (2002). IT-based services and service quality in consumer banking. International Journal of Service Industry Management, 13(1), 69-90. normal and Social Psychology, 61, 159-167. Zeithaml, V., Parasuraman, A., & Malhotra, A. (2000). A conceptual understanding e-service quality : Implications for futu 119

PAGE 130

Appendices 120

PAGE 131

Appendix A. Audio Recording Transcripts Operator: From this is the Australian Open On-Demand Scoreboard. Which ou like to hear about? Operator: The match is currently in progress. Andre Agassi is losing to Marad Saffin. Marad s 7-6, 7-6, 5-6, with Marad Saffin serving. Andre Agassi leads love-15 in the current User: Goodbye. or calling the on-demand Australian Open Scoreboard. B) he automated travel service system. You can peat or help at any time. Select flight status, available flights Available flights. you have a preferred airline? User: No. System: Departure city? w York. nt Kennedy, LaGuardia, or Newark? LaGuardia. System: Approximate departure time? User: 6:45pm System: Was that 6:45pm? User: Yes. Tennis Scoreboard (Interface A) player would y User: Andre Agassi. Saffin lead game. Would you like to hear about another player? Operator: Thank you f Goodbye. Available Flights (Interface System: Thank you for calling t say re User: (interrupts) System: Do User: Ne System: Do you wa User: 121

PAGE 132

Appendix A (Continued) city? : Was that Fort Lauderdale Hollywood International Airport? User: Yes. System: Which day? November 21st. System: Please wait. [pause] There are 16 flights that match your criteria. The closest match is lywood International User: (interrupts) List previous flights. System: The list has 8 flights. To select a listed flight, say its option number. Option 1: Delta n 2: US Airways flight 3802, departing 6:15pm. Option 3: est flight 529, departing 6:10pm. Option 4: US Airways flight 3920, departing 6:10pm. Option 5: US Airways flight 191, departing 6:01 pm. User: (interrupting) Goodbye. ure you want to end this call? User: Yes. System: Thank you for calling the automated travel service system. Goodbye. Retail Order (Interface C) System: Thank you for calling . Whats your team number? User: 447108. System: Thanks! Welcome back. I see you have 3 items in your wish list and 2 items in your shopping cart. What would you like to do today? System: Arrival User: Fort Lauderdale. System User: System: Was that November 21st? User: Yes. American flight 2431, departing at 6:45pm. It arrives in Fort Lauderdale HolAirport at 9:52pm. Say repeat, list previous flights, list following flights flight 1631, departing 6:30pm. Optio Northw System: Are you s 122

PAGE 133

Appendix A (Continued) System: Great! Your shopping cart contains 2 items: a Florida Bobcats #26 jersey in extra large air of Nike soccer shoes in size 11. Would you like to place an order? System: Alright. We have 10056 Ridgway Avenue as your shipping address and a Visa card with numbers 5714 and expiration date May 2006 on file. Would you like to use this shipping User: Visa card. System: What type of credit card would you like to use? the number? : Expiration date? : Ok, so your credit card is Mastercard with last 4 numbers 0950 and expiration date ber 2007. Is this correct? User: Yes. System: Placing your order. Your order is $284.85 with standard shipping. Your package d you like a receipt emailed to you? User: Yes. System: Is your email address still User: Go to shopping cart. and a p User: Yes. last 4 address and Visa card? User: No System: Ok, what would you like to change? User: Mastercard. System: Whats User: 555501010022089640950 System User: September 2007 SystemSeptem should arrive within 10 business days. Woul ? User: Yes. System: Great. Anything else we can do for you today? 123

PAGE 134

Appendix A (Continued) System: Thanks for calling . Well look forward to serving you again. Bye! Retail System: Welcome to by phone. If you ever need help on any question I ask elp and Ill provide additional information. Lets get started. Please say atus, request a catalogue or search for an item. ser: User: Bravo Espresso Machine. System: The Krups Bravo Espresso Machine, the price is $49.95. Say details to hear... r: other item? r: e product code or the product name. r: 18 oz. Commuter Cup. The price is $14.95. Say details to hear more about this User: Buy it. System: Would you like to order another item? User: No. System: We have several popular items that compliment your purchase. Would you like to hear them? User: Yes. User: No, thanks. Purchase (Interface D) you just say h place an order, check order st UPlace an order. System: Say the product code or the product name. UseBuy it. System: Would you like to order an UseYes. System: Say th UseCommuter Cup. System: The product. 124

PAGE 135

Appendix A (Continued) Systemur purchase. If you cart. You can also product, next to go to the next item or out to complete your purchase. Here we go. I have five items for you. The first s stainless steel cappuccino espresso cups and saucers. User: System: Enjoy coffee bar chic at home with these gourmet coffee cups and saucers, Euro styled inless steel handled cups keep cappuccino and espresso hot. User: Buy it. saucers to your shopping User: Check out. this with a ediface E) System: Thank you for calling . I can help you receive a copy of your credit dispute items on your credit report, or check the status of items you have disputed. Which of these are you calling for today? ute. l security number? User: 154-678-4932 and last name. User: Kent Montgomery. System: Finally, I need to collect your address, starting with your zip code. Please tell me your zip code now. User: 75008 System: And now the street address? : Great, I have a list of the most popular items that compliment yohear an item that interests you say buy it to add it to your shopping say details to get more information about the checkitem i Detail double walled sta Dishwasher safe, price is $16.95 per set. Each set System: Ive added the stainless steel cappuccino espresso cups and cart. The second item is oval System: The total cost of your purchases is $41.90. Will you be paying for credit card, Visa, MasterCard Crt Dispute (Inter report, User: Place a disp System: Next, I need to get your social security number. What is your socia System: Ok, now tell me your first 125

PAGE 136

Appendix A (Continued) stemOk, now well take a look at your report. Using the item number on your report, tell me the first item number youd like to dispute. User: Four. o be a listing for Dayton-Hutson. Is this correct? Its not mine. you believe this account is not yours. Are there any more items youd like to dispute? stemWhats the next item number youd like to dispute? System: We show that item number to be a listing for Sears. Is this correct? User: stemOk, whats the problem with this item? User: System: Ok, Ive sent notice to that creditor that you believe this account is fraudulent. Are there any more items that you would like to dispute? User: No. System: Ok, the creditors receiving disputes have up to 30 days to respond. Once we have all the responses, we will mail the results to you. Thank you for calling . User: 1601 Bethany Road. Sy: System: We show that item number t User: Yes. System: Ok, whats the problem with this item? User: System: Ok, Ive sent notice to that creditor that User: Yes. Sy: User: Seven. Yes. Sy: It looks fraudulent. 126

PAGE 137

Appendix A (Continued) edule a pickup using ackage using your tracking ss two, for international shipping say or press three, to find package drop off locations say or press four, for express US rates say or press five interrupt the rating system by speaking at any time. Please say or enter your account number. If you dont know your account number, say I dont know and know. System: Ok, Ill give you a quote based on standard rates, but you can always talk to a customer ending? You can say envelope, pack, box hat is the approximate weight of your package? Please round up to the next pound. System: What 5 digit zip code are you sending the package from? ping your package without any special handling to a business address. The basic cost for priority overnight scheduled to arrive tomorrow by 10:30 am is $26.14. You can say read that again or rate another package, to find out about other delivery times or special handling say lets go on, for other FedEx shipping services say menu and if you are done say good bye. User: Good bye System: Thanks for calling rate finder, good bye. Package Delivery (Interface F) System: For faster service, please choose from the following options: to schyour account number say or press one, to track a p number say or pre User: (interrupting) Five System: You can we will move along. User: I dont representative who can look into this for you. Which package type are you s User: (interrupting) box. System: W User: Two pounds. User: 33435 System: And what 5 digit zip code are you sending the package to? User: 37411 System: Please hold while I calculate the rate for ship 127

PAGE 138

Appendix A (Continued) System: Thank you for calling the automated travel service system. You can help at any time. Select flight status, available flights, My Flights, My Hotel, or exit. User: Flight status. System: Which airline? American. umber? 153 System: Was that 153? User: Yes. User: Tomorrow. System: Please wait while Pat gets the information. For Thursday, November 7, 2002, American l from gate 2c at 7:52am. It arrives in OHare International gate A19 at 10:03am. American flight 153 departs OHare International airport gate A16 at 11am. It arrives in gate [unintelligible] at 3:30pm. Say repeat User: Goodbye. sure you want to end this call? System: Thank you for calling the automated travel service system. Goodbye. Phone Bill (Interface H) System: Thank you for calling , also on the web at . To ensure quality service, your call may be monitored or recorded. To continue in English, press or say 1 now. [Spanish language prompt] User: One Flight Status (Interface G) say repeat or User: System: Flight n User: System: Which day? flight 153 departs Miami Internationa airport System: Are you User: Yes. 128

PAGE 139

Appendix A (Continued) stem: This center is using a system that can understand you when you speak. If you are currently a customer, please say or key in the number you are calling about, beginning with the area 9-2443 account. Thank you. Please tell me briefly the reason for your call today. User: I need to make arrangements for some payment. System: Thank you. To verify your account, please say or key in your 6-digit customer ID, which on telephone keypad) e agreed to pay $102.31 by October 2, 2003. Would you like to hear your agreement again? : Was that a yes? : Do you want to change this agreement? .31 is due on your account. The last date I can give you to pay is October 2, 2003. Can you make payment by this date? would you like to do? Pay each week, pay every other week, or pay the full amount in one payment? User: (no response) System: Im sorry, I missed your response. If you can pay weekly, say weekly. If you can pay every other week, say biweekly. If you can pay full amount in one payment, say one payment. User: One payment. Sy code. User: 972-53 System: One moment while I access your follows your telephone number at the top of your bill. User: (presses 6 digits System: You hav User: No, I need to change it. System User: No. System User: Yes. System: $102 User: Yes. System: What 129

PAGE 140

Appendix A (Continued) System: Youve agreed to pay full amount of $102.31 by October 2, 2003. Is this arrangement ? confirmation number. nd a security deposit on number is ontinue to receive w enough time for it to reach our office by the ate. Any new bills you receive while this agreement is in effect must be paid by the due date on the bill. If you exceed your credit limit at any time, your long distance service could be hanks. t it is so I may direct heck my balance or billing questions or Id like to order voice mail. What is the main reason for your call today? (call end) pany name> automated reservations. I can help you make, change or cancel reservations. What would you like to do? hange reservations ) User: Lewis. System: Changing reservation. What is your reservation number? User: B as in boy, 56212, t as in tack, 4567. System: Im getting that reservation information, please wait. Is that the reservation you made to pick up at the Dallas-Fort Worth airport May 30 3:40pm? acceptable User: Yes, it is. System: Please hold while I update your account. Ill be right back with your Please note the following: Information service disconnection, reconnect fee, amay apply if your payment is not received by October 2, 2003. You confirmati090410. Keep in mind that although an agreement has been made, you may cnotices. When mailing your payment, please allo agreement d disconnected. Do you want to hear this information again? User: No t System: If there is anything else I can help you with today, please state whayour call. Or you may hang up to end this call. User: Goodbye. System: You can say something like c Car Rental (Interface I) System: Welcome to
PAGE 141

Appendix A (Continued) u like to change first? The drop off date, June 8. For example, you could say a different date like May 3 or a day like Saturday, today, or g else? User: The drop off time, 1pm. System: Drop off at what time? ng else? No. lanning to pick up the car at the Dallas-Fort Worth airport May 30 3:40pm and returning to the Dallas Fort-Worth airport June 8 ur current profile? System: After applying all passes, fees, and discounts, the total charge of this reservation will be s? User: No System: Would you like me to make this reservation? User: Yes. System: The reservation number is B as in boy, 56212, C as in cat, 4567. Would you like to hear it again? User: No. User: Yes. System: What would yo User: System: tomorrow. User: June 8. System: Change anythin User: 1pm System: Change anythi User: System: Lets review to be sure I heard you correctly. Are you p 1pm using yo User: Yes. $390 with unlimited milage. Would you like to hear a breakdown of the charge 131

PAGE 142

Appendix A (Continued) e anything else I can do for you today? for using the automated reservation system. Goodbye. dbye. name? User: West Palm Beach, Brent Davis. on? User: No. Choose outside line, tieline, fax number, pager number, pin code, department number, location, or connect. You can say new name, new location, or main menu to select another listing. Other information? for calling the Provider Support System. Please say the patients nine digit I.D. Number. 119001234. System: Coverage is in effect for this patient only. Would you like to check claim status, hear plan details or hear claim history? User: Check claim status. System: This patient has twelve claims to review. Would you like to check a specific claim or hear all claims? System: Is ther User: No, thank you. System: Thank you User: Thank you, goo Directory Dialer (Interface J) System: directory dialing. What location and System: West Palm Beach, Brent Davis. Tieline 975-2177. Other informati System: User: Connect. (phone rings) Patient Claim (Interface K) System: Thank you User: 132

PAGE 143

Appendix A (Continued) he date of service? : This claim for $249.80 has been processed. It was paid on April 30th, 2003. Would you like to hear additional details for this claim? User: No. d you like to check another claim status, hear plan details or hear claim history? You can hang up to end the call. Movies (Interface L) System: Good morning, welcome to 1-800-555-tell. Get movie tickets before you find out its sold oose from: stock quote, sports, entertainment Okay, entertainment. Here are the choices in entertainment: movies, horoscopes ny name>. Do you ton, Florida? User: Yes em: Okay Boca Raton, Florida. Now looking for movies in and around that area. Remember you can always say go back to back up a step or start over to start your search again. If you know you want, just say it; to hear about new releases or other movie categories, say movie listings, for show times at your local theaters User: Movie listings System: What kind of movie do you want to see? For a list of movie categories, say tell me my choices. User: Tell me my choices User: Specific claim. System: Whats t User: January 23, 2003. System System: Woul out! Say tell me more. Main menu. Here are all the categories you can ch news center User: Entertainment System: User: Movies System: Movies. Now you can buy tickets over the phone, powered by
PAGE 144

Appendix A (Continued) System: Here is a list of movie categories. When you hear the one you want, just say it: new ovies, drama, comedy, romance System: Here are the up-coming movies. When you hear the one you want, just say it: Signs el Gibson and Jauquin Phoenix, rated pg13. To hear the next movie, say next, to end this list say stop. ack Austin Powers in my Lee Jones are show times for Men in Black II: its playing at Regal Shadow Wood 16, 9889 Glades Road in Boca Raton. You can see it today at 12:30, 2:45, 5 oclock, 7:10 and 9:20. o hear show times say next. It is also playing at Regal Delray Beach 18, 1660 South Federal Highway, Plaza of Delray, in Delray Beach. You can see it today at 12:35, 2:50, 5 oclock, 7:50 re available at this theater. System: I heard you say good bye. If you dont mean to hang up, say go back; otherwise, thanks od bye. Report Electrical Outage (Interface M) Operator: Welcome to . Before I can help you I need to find out who you are. Are you calling from home today? Customer: No. Operator: Tell me your ten digit home phone number. releases, up-coming m User: Up-coming movies starring M User: Next System: Martin Lawrence Live rated R, full frontal User: Go b System: What kind of movie do you want to see? For example, say comedy. User: Comedy System: Here are the comedies. When you hear the one you want, just say it: Goldmember, starring Mike Meyers, rated PG13; Men in Black II starring Tom User: Men in Black II System: Here Advanced tickets are available at this theater. To buy tickets, say buy tickets; tfor the next theater and 10:10. Advanced tickets a User: Good bye for connecting, go 134

PAGE 145

Appendix A (Continued) me Operatoreat, we found your information. Lets go to the main menu. You can report a ur balance or enter your meter reading. Which one would you like? Custometility outage at my house. rt a complete outage or flickering light. Which type of outage would Customer: My powers out. ou for reporting the outage and we will have the problem resolved within ace N) Operatoription refill service. Please say your name to access our service. Customer: Tom Smith. you. We have accessed your record. You are due for a refill on Imitrex. Would you like to refill your Imitrex prescription? Customer: Yes. Customer: Delivery. Operator: Are you still at 5671 Round Tree Avenue? Customer: Yes. Operator: Thank you. Your Imitrex prescription will be delivered to your home address within three to five business days. Would you like another refill? Customer: No. Operator: Thank you for using , good bye. Custor: 2018186600. : Gr power outage, check yo r: I have a u Operator: You can repo you like to report? Operator: Thank y two hours. Thank you for calling , good bye. Prescription Refill (Interf : Welcome to prescr Operator: Thank Operator: Is this for pick up or delivery? 135

PAGE 146

Appendix A (Continued) Operator name>. Ill help you get solutions for all your financial s. All you have to do is talk to me. Lets get started. So I can tell whos calling, say your account number. Customer: 94837273 Custome Operator: Your voice print has been verified. Hi, John. menu, tell me service you would like to use? You can say banking, investments or mortgages. You can also say my to access the custom meOperatorld you like? You can say account balances, credit card information or say make a payment. me Operator information say last five transactions, last five checks meOperator: Your last five transactions are all ATM transactions. Ill list them; on January first 00, on January third a withdrawal in the amount of $500, on January sixth a withdrawal in the amount of $100 s the balance on my credit card? Operator: As of January sixth your credit card balance is $3455.32. What next? You can say Customer: Make a payment to my credit card. Operator: Pay how much? Customer: All of it. Financial Services (Interface O) : Welcome to
PAGE 147

Appendix A (Continued) ould I take the money from you checking account? me Operator: Okay, $3455.32 has been paid from checking to your credit card. Your credit card now zero dollars. These changes will show up immediately on your on-line account. What next? ts. balance, stock quotes or place a trade. otes. Operator: Say the name of the security or its ticker symbol. Customer: SunTrust. Operator: SunTrust International, symbol STI is trading at 62 and 7/16 up 1 and Customer: Buy two hundred shares of SunTrust from my checking account. Operatorernational at a share price of 62 and 7/16 with funds from your checking account. Is that correct? Customer: Yes. Operatorurchase. Your confirmation number is 14432. Ill send a ry of this transaction to your home address and your on-line account will reflect the change immediately. What next? You can say Custome got your personal page; you can say news, traffic, weather, sports, stocks, vacations or alerts. Customer: Weather. Operator: Todays weather in Atlanta is sunny with a high of 62 degrees and a low of 54. For more information say get details. Customer: Sports. Operator: Sh Custor: Yes. balance is Customer: Investmen Operator: Investments. You can say account Customer: Stock qu : I think you wanted to buy two hundred shares of SunTrust Int : Okay, Ive made the p summa r: My . Operator: Ive 137

PAGE 148

Appendix A (Continued) Operatorraves just signed Chipper Jones to a five year one hundred twenty million dollar contract. You can say get details or say another ory from your page. Custome u have one bill alert. Here it is; your alimony payment in the amount of $2117.34 is due today. Would you like to pay this bill now? me Custome, my savings account. Operator: Okay, $2117.34 was paid from your savings account to Mandy Smith. Your news, traffic, Customer: I need a vacation. Operator: Vacation. Delta airlines has several special exclusive getaways for subscribers. Each week theres one unique vacation in and one a Operator: Welcome to . To get started, just tell me you Custor: Johnson. 138

PAGE 149

Appendix A (Continued) Operator: Confirming the purchase of five hundred shares of America Online at $12.53. ase say yes to confirm the purchase. meOperator: Is there anything else I can help you with today? Customer: Nope. Operator: Thank you for calling , good bye. Ple Custor: Yes. 139

PAGE 150

Appendix B. Participant Instructions psychology on how ide is strictly anonymous. Therefore, you will not need to provide your name or other identifying information. In this study, you will listen to an interaction between a speech interface and a user, similar ill be asked to rate ou to provide information about yod social interaction style, so that I can better understand how different people view speech interfaces. To hear the users interaction with the speech system, please double click on the following lick here to hear user-speech system interaction I invite you to participate in a study for my doctoral dissertation inpeople perceive and evaluate speech user interfaces. The information you prov to listening to a telephone conversation on another line. After you listen, you wvarious aspects of the speech system. The questionnaire also asks y urself, such as your age, gender, job type, preferences for customer service, an audio file: C Participation in this study is strictly voluntary. Again, your answers will be completely y questions regarding this study, please do not hesitate to let me know at tact below. Melanie D. Polkosky Advisory Human Factors Engineer, IBM Pervasive Computing 8051 Congress Ave, Suite 2207 Boca Raton, FL 33487 internet: phone: 561-862-2037 (TL: 975-2037) fax: 561-862-2988 (TL: 975-2988) anonymous. If you have an any of the points of con Thank you, 140

PAGE 151

Appendix C. Participant Demographic Questions ______ 2.ale 3.Spanish describe); used speech technology in your daily life? Never dom n 5.following best describes you as a speech technology user? t user best describes your hearing ability? rmal ring loss or disability 7.unction? Administrative Staff Researcher ofessional al professional g Other (please describe): 8. Highest educational level achieved: High school diploma Technical degree Bachelor degree Masters degree Ph.D. Other (please describe): 1. Age: ___ Gender: Male Fem Native language: English Other (please How frequently have Sel Occasionally Ofte Frequently Which of the Novice user Exper Which of the follow No Hea What is your current job f nagement Ma Engineer Other technical pr Non-tech nic Marketin Sales 141

PAGE 152

Appendix D. Dabholkar and Bagozzi (2002) Inherent Novelty Seeking Scale 1 trngige2 5 6 7 Strongly Agree S o ly D sa re 3 4 1. I am always seeking new ide as and experiences. 2. When things get boring, I like to find sand oe new unfamiliar experience. 3. I like to continually change activities. 4. I like to experience novelty and change in my daily routine. m 142

PAGE 153

Appendix E. Dabholkar and Bagozzi (2002) Need for Interaction with a Service Employee Scale 1 trngDisagree2 5 6 7 Strongly Agree S o ly 3 4 1. Human contact in providing services m ak es the process enjoyable for the customer. 2. I like interacting with th e person who provides the service. 3. It bothers me to use a machine when I could talk to a person instead. 143

PAGE 154

Appendix F. OCass (2003) Revision of the Lennox and Wolfe (1984) Self-Monitoring Scale 1 y Disagree 2 5 6 Strongly Agree Strongl 3 4 1. In social situations, I have the abmy behavior if I feel that something ecalled for. ility to alter lse is 2. I have the ability to control the wacross to people, depending on ay I cme the impression I o wish to give them. 3. When I feel that the image I am portraying to isnt working, I can readily change it something else that does. 4. I hav e trouble changing my behavior to suit ns. different people and different situatio 5. I have found that I can adjust my bmeet the requirements ehavor to of any situation in which i I find myself. 6. Once I know what a situation caleasy for me to regulate my actions accor ls for, its dingly. 7. I am often able to read peoples trucorrectly (throu e ems gh their eyes). otion 8. In conversations, I am sensitiveslightest change in the facial expres to eventhe sion of the person with whom I am conversing. 9. My powers of intuition are quite g ood when s an it comes to understanding the emotionmotivations of others. d 10. I c an usually tell when others consider a joke to be in bad taste, even though they may laugh convincingly. 11. I can usually tell when Ive said something inappropriate by reading it in the listeners eyes. 12. If someone is lying to me, I usually know it at once from that persons manner of expression. 144

PAGE 155

Appendix G. Items for Usability Scale Item 1 ree2 3 4 5 6 7 Strongly Agree Strongly Dis ag 1. The system made me feel like I was in control. 2. When speaking to me, the system paused or hesitated unnaturally. 3. I could complete my business quicklysystem. ing thi us s 4. The system seemed efficient. 5. Some of the systems responses were unexpected or surprised me. 6. The system said things in a short, brie f anner. m 7. It was easy to figure out how to do whwith this system. at I needed 8. The messages were repetitive 9. It would be too time consuming to use ts system. hi 10. The system gave me a good fee ling about beinga customer of this business. 11. The systems voice seemed confident. 12. The system used terms I am familiar with. 13. I felt frustrated because the system didnt understand me. 14. The system was designed in a way thahave my best interests in t s eemed to mind. 15. The system was easy to in teract with. 16. I would rather use this system than a ebpage w 17. This system impressed me. 18. I could find what I needed withou t anyifficul. d ty 19. The sy stem reminded me of a human operator or customer service representative. 20. This system used everyday wo rds. 21. This system reminded me of other te cnologie, bpges. h s such as touch tone phone systems or we a 22. The sy stem was organized and logical. 23. The system allowed me to do thiimporta ngs that are nt to me. 24. The systems voice sounded unusual. 25. I thought this system was interesting to listen to. 26. The system gave me more details than I needed. 27. If the system misunderstood me, it was easy to fix the problem. 28. The system provided personal, customized information. 145

PAGE 156

Appendix G (Continued) od say. 29. I had to carefully consider what I sh ul 30. The messages I heard gave me hemis lp if made atake. I 31. The system spoke at a pace that was esy to a follow. 32. The system would help me be productive. 33. The system understood more thanwould. I thught it o 34. The system seemed polite. 35. I fe lt like I couldnt say anything this system might understand. 36. I was able to easily unde rstand everyting this h system said. 37. I could trust this system to wor k correctly. 38. I would be likely to use this syste m again. 39. The systems voice w as pleasant. 40. The system emphasi zed words in a way that helped me know what I was supposed to say 41. The system provided all the information I needed. 42. The system was too talk ative. 43. This system seemed reliable. 44. The systems voice sounded like peopthe rad le I hear io or television. on 45. I felt frustrated because I couldnt unwha drstand t the system said. e 46. I felt confident using this sys tem. 47. The system gave me accurate inf ormation. 48. U sing this system required a lot of work or effort on my part. 49. I could predict what th e system would say before it said it. 50. I felt rushed when I was thinking of what to say. 51. The system seemed to kno w my needs. 52. The systems voice sounded mechartificial. anil or ca a 53. I knew what this system could and couldnt dos soon as I started interacting with it. 54. The systems voice sounded like a regular person. 55. Using this system was fun. 56. I felt like I was talking to a person. 57. Using this system would result in errors that would prevent me from getting what I want. 146

PAGE 157

Appendix G (Continued) guage I hear in my everyday conversations. 58. The system used slang, common sayings, or other lan 59. The system provided clear instrucneeded to do. tionon whaI s t 60. The quality of this system made m e want to ess. remain a customer of this busin 61. It would be easy to remembe r how tose this several days. u system if I didnt use it for 62. The systems voice sounded natural. 63. The system seemed priva te and secure. 64. The system seemed courteous. 65. I felt like I had to wa it too long for thstem e sy to stop talking so I could say something. 66. The system was helpful. 67. The system was easy to use. 68. The system seemed friendly. 69. The systems speech was easy to un dstand. er 70. T he system said things in a vague, ambiguous, or unclear way. 71. I felt confident that the things I said were understood correctly. 72. The systems voice sounded enthof energy. usiastic or full 73. The system pleasantly surprised me. 74. The system seemed professional in its speaking style. 75. This system seemed complicated to use. 76. The messages I heard helped me when I felt unsure about what to do. 147

PAGE 158

Appendix H. Modified Spake, Beatty, Brockman, & Crutchfield (2003) Comfort Scale 1 ortable 4 5 6 7 8 9 10 Comfortable Uncomf 2 3 Very u y 1 Very tense 2 3 4 5 6 7 8 9 10 Very relaxed 1 Insecu 1 neas2 3 4 5 6 7 8 9 10 Very much at ease re 2 3 4 5 6 7 8 9 10 Secure 1 ied 4 5 6 7 8 9 10 Worry free Worr 2 3 Distr 2 3 1 essed 4 5 6 7 8 9 10 Calm 1 Turbulent 2 3 4 5 6 7 8 9 10 Serene 1 Troubled 2 3 4 5 6 7 8 9 10 Peaceful 148

PAGE 159

Appendix I. Eroglu and Machleit (1990) Customer Satisfaction Scale 1 fied 2 3 4 5 6 7 Satisfied Dissatis Unfavo le 1 U 1 rab2 3 4 5 6 7 Favorable npleasant 2 3 4 5 6 7 Pleasant 1 I like it very much 2 3 4 5 6 7 I didnt like it at all 149

PAGE 160

150 About the Author in 1992 from tion Disorders from The Pennsylvania State University in 1995. At Penn State, she specialized in augmentative and alternative communication for individuals with severe communication disorders. She completed gy at All Childrens Hospital (St. Petersburg, Fl n served on a and attended art school until entering the USF Psychology Ph.D. program in 1999. In 2001, she began a year-long internship at IBM Voice Systems, which lead to full time employment in 2002. At IBM, she consults on human factors issues and designs speech technology systems for clients in a variety of industries. Melanie lives in Collierville, Tennessee with her husband, dog, and four cats. Melanie Diane Polkosky received a Bachelor of Science in Education Millersville University of Pennsylvania and a M aster of Science in Communica her Clinical Fellowship in Speech-Language Patholo orida) and received the Certificate of Clinical Competence in 1996. She the multidisciplinary pre-kindergarten diagnostic te am for Pinellas County Sc hools

xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam Ka
controlfield tag 001 001680963
003 fts
005 20060215071034.0
006 m||||e|||d||||||||
007 cr mnu|||uuuuu
008 051205s2005 flu sbm s000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0001158
BF121 (Online)
1 100
Polkosky, Melanie Diane
0 245
Toward a social-cognitive psychology of speech technology
h [electronic resource] :
b affective responses to speech-based e-service /
by Melanie Diane Polkosky.
[Tampa, Fla.] :
University of South Florida,
Thesis (Ph.D.)--University of South Florida, 2005.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
System requirements: World Wide Web browser and PDF reader.
Mode of access: World Wide Web.
Title from PDF of title page.
Document formatted into pages; contains 160 pages.
Includes vita.
3 520
ABSTRACT: Speech technologies, or technologies that recognize and respond to human speech, have recently emerged as a ubiquitous and cost-effective form of customer self-service (e-service). Although customer satisfaction is regarded as an important outcome of e-service interactions, little is known about users affective responses to conversational interactions with technology. Using a theoretical foundation derived from research in social cognition, interpersonal communication, psycholinguistics, human factors, and services marketing, two studies develop items for a speech interface usability scale, which is then used to examine interrelationships among individual differences (e.g., self-monitoring, need for interaction with a service provider, inherent novelty seeking), usability, comfort, and customer satisfaction.
Adviser: Judith Becker-Bryant.
Human computer interaction.
Dialogue design.
Customer satisfaction.
Speech user interface design.
Dissertations, Academic
x Psychology
t USF Electronic Theses and Dissertations.
4 856