A caller asks for a reminder by text or phone, then answers “Yes.” Repeating the same question as an error makes the caller feel stuck, even though they tried to cooperate. Disambiguation keeps the exchange moving: “Great! Which would you like: a text message or a phone call?” section 8.2.4
Disambiguation asks the user to choose among plausible interpretations instead of treating an ambiguous response as a failure. For an open-ended request, a chatbot can present the classifier’s leading intents to clarify the user’s primary goal. Within a dialogue flow, a short follow-up can clarify which offered option the user meant. section 4.4.2 section 8.2.4
When one message contains several goals
“Do you have the COVID booster? How can I make an appointment?” expresses both an availability question and an appointment request. Without disambiguation, it has no single definitive intent for a classifier accuracy test. If the chatbot offers disambiguation choices, test whether both relevant intents appear among them. section 4.4.2
Use disambiguation without making the user work harder than necessary:
- Reserve it for genuine uncertainty; assess whether a clarifying question or another dialogue strategy better resolves the confusion. section 5.5
- Accept a good-faith answer when it provides enough information. On voice channels, where buttons may be unavailable, keep questions direct and concise rather than cycling through retries. section 8.2.4 section 11.3.3
- Check ambiguous and multi-intent utterances separately from ordinary single-intent accuracy tests; verify that the offered choices reflect the goals users actually expressed. section 4.4.2
