Apologetic Interfaces

Polite computers win users’ hearts and minds [New Scientist]

A study by Jeng-Yi Tzeng at the National Tsing Hua University in Taiwan suggests that computer users would be more tolerant of software programs if the programs were more apologetic, or at the very least, less likely to always blame the user. This is especially relevant to speech recognition applications.

In our speech applications at work, we build in error messages like “I’m sorry, but I didn’t understand you.” This is obviously a lot friendlier than “Stop mumbling” or “Take the marbles out of your mouth”, not that any speech vendor would build an app like that. Of course, the recognition process inherent in speech applications is more prone to misunderstandings than the control driven UI of your typical desktop or web application, so it’s necessary to be more forgiving and to openly acknowledge the inability of the system to really understand natural language.

One unfortunate downside (for the designer, anyway) of a really well designed speech application, is that users sometimes develop overly high expectations. If the recorded prompts are natural sounding and stitched together well, callers often seem to forget they aren’t speaking to a human. During one phone call that one of our apps handled, the app could’t match what the caller said to an active grammar, so the app responded with something like “I’m sorry, but I didn’t understand you.” She then replied, “That’s all right, dear.”

One aspect that really distinguishes the quality of a speech app is the ability to identify that a comment like hers, while very sweet, was irrelevant to the task of prompting her for information. Many speech apps I’ve interacted with would have confused her response with a new response to the original prompt, thus setting off another round of telling her that it didn’t understand what she was saying. That ends up making the caller feel like she is talking to someone who is rude and isn’t paying attention to her.