Do Smart Speakers Respond to Their Errors Properly? A Study on Human-Computer Dialogue Strategy

As smart speakers with voice interaction capability permeate continuously in the world, more and more people will gradually get used to the new interaction medium–voice. Although speech recognition, natural language processing (NLP) have been greatly improved over the past few years, users still may encounter errors from time to time like “cannot understand”, “no requested audio resource (such as music)”, which can frustrate users. Therefore, when an error message is reported, it is vital that the smart speaker gives an effective and proper response. However, currently the response strategies adopted by leading smart speaker brands in China differed mainly on two dimensions: “apology or not” and “humor or neutral”. We explored user’s preference of response strategies under two error scenarios——“cannot understand” and “no requested audio resource”. A 2 (apology: yes vs. no) × 2 (error message expression tone: humor vs. neutral) within-subjects experiment was conducted. Two dependent variables (satisfaction and perceived sincerity of response) were measured. The results showed that participants were more satisfied and perceived higher sincerity when smart speaker apologized in both error scenarios. In the “no requested audio resource” scenario, humor had no significant impact on the perception of satisfaction and sincerity. But in the “cannot understand” scenario, humorous expression decreased perceived sincerity.