Notice: the WebSM website has not been updated since the beginning of 2018.

Web Survey Bibliography

Title Dialogue Capability and Perceptual Realism in Survey Interviewing Agents
Year 2007
Access date 28.05.2007

The advent of new technologies for animated or conversational agents raises important questions about the future of survey data collection. Depending on how the next generation of self-administered survey interfaces are designed, we could end up merging the benefits of live interviews and self-administration, or end up with the worst features of both; not enough is yet known about how different features of human interviewing affect data quality and respondent participation. We are exploring which features of agents - talking head interviewing systems of the future will improve respondents' performance and satisfaction, and which hurt, with the goal of determining which features most warrant development for surveys. Agent interfaces for surveys can vary on at least three dimensions which range from most computer-like to most human-like: [1] dialogue capability - does the interviewing agent generate novel text and clarify its meaning? (2) perceptual cues of internal states - does the agent sound and look realistic? (3) intentionality - does the agent seems to hold intentions toward the respondent, e.g. can It disapprove of answers or determine that respondents do not understand the question? We report on our experiments that compare potential agent interfaces by simulating rather than building the new technologies, using a "Wizard-of-Oz" technique: a human interviewer (wizard) interacts with the respondent through a user interface but the respondent believes she is interacting with a computer agent. The agents face Is created by software that generates animation from video of the interviewer's face. !t is likely that interviewing agents' dialogue capability and perceptual realism have different effects for different domains of questioning sensitive vs. non-sensitive questions), because respondents are likely to bring different interactive schemas to bear in different dialogue contexts. ]t is also possible that agent interfaces may need to be designed differently for different kinds of respondents.

Access/Direct link

Conference homepage

Year of publication2007
Bibliographic typeConferences, workshops, tutorials, presentations
Full text availabilityAvailable on request

Web survey bibliography - The American Association for Public Opinion Research (AAPOR) 62th Annual Conference, 2007 (48)