13 Choosing strategies to recover from miscommunication
This chapter covers
- Building a robust VUI that avoids catastrophic failure
- Implementing the best recovery and backoff technique for a context
- Designing for getting back on track when recognition fails
Recognition is never perfect, not even with today’s amazing ASR. Nor is it so accurate that you can just ignore any utterances that fall off the happy path because of misrecognition or incorrect intent mapping. You’ve already learned about many of the reasons why recognition fails: not covering the words spoken by users or the way users pronounce words that are covered, interfering similar words, noisy background and side speech are just a few reasons. Or suddenly someone changes a feature or adds a branded product name, and bam! People don’t get recognized and get frustrated. This chapter is all about getting back on track.
13.1 Recovery from what?
You tried so hard to maximize recognition performance with robust grammars, good prompts and a design that fits the users’ mental models. And yet—things went off track! Why? And how do you even know? Even with the best of intentions, failure to communicate will happen.