The aim of this research was to we explore the role of culture in systems that recognize facial expressions to respond/react accordingly.
We studied Greek and Dutch people that were selected based on their differences in Uncertainty Avoidance Index (UAI), one of the five cultural dimensions defined by Hofstede (1994) to interact with a speech interface that imposed controlled errors. Their facial expressions were videotaped and analyzed with a perception study. The results display that when encountering errors, cultures that rate high in the UAI are more expressive than ones that rate low. Also, when encountering errors in a low UA interface, high UA cultures are more expressive than low UA ones.
Experiment Design: The study was set up as a 2X2 mixed factorial design given that we had two independent variables: culture and type of interface. Both variables had two levels. Culture was divided into Greek and Dutch and type of interface was divided into Hi UA and Low UA. The dependent variables were frustration experienced by participants and the amount of expressiveness displayed. UAI for each culture group was measured with the VSM94 questionnaire, frustration was measured with the NASA-TLX questionnaire and expressiveness was measured through a perception test.
System Design: The system used for the purpose of this study was a speech interface. The speech messages were recorded with Iwona (http://www.ivona.com/) online text to speech software and edited with Wavosaur audio editor. Microsoft Powerpoint was used to present the speech messages in a controlled manner. The dialog of a Train Booking system was designed, where the participants were asked to make a booking from London to Newcastle for a specified date and time. The study interface used a Wizard of Oz approach where one of the experimenters controlled the system as a response to the users’ utterances. Two versions of the system were developed, one for High UA and one for Low UA culture. The interfaces were developed using the guidelines available (Marcus and Gould, 2000)
Quantitative and Qualitative Results:
A 2X3 repeated measurements ANOVA with 2 independent factors (nationality, 2 levels: Greek and Dutch; videotype: 3 levels: Low UA, Neutral UA and High UA) was conducted on the data. The results display a significant main effects for nationality F(1,16)=27.109, p=.00 and for VideoType F(2,15)=125.466, p=.00 as well as a significant interaction F(2,15)=31.804, p=.00.
According to the interviews conducted after each experiment, five out of eight Greeks preferred the high UA interface, commenting that they liked the step by step structure of the dialogue, the confirmations it provided and that they found it less complicated. Five out of eight Greek participants disliked the low UA interface, describing it as stressful and frustrating and claiming that it presented them with too many details. Moreover, three out of eight Dutch participants liked the high UA interface, commenting positively about the confirmations it provided and describing it as more structured. Three out of eight Dutch preferred the low UA interface, stating that it was challenging, flexible and quicker. Finally, also three out of eight Dutch participants disliked the low UA interface, stating that it provides them with more options and that it is more demanding.
Team: Beant Kaur, Rafal Kocielnik, Ioannis Politis
Role: Research, Experiment design, Moderation and Analysis.
Final paper available upon request.