liveBook · Manning

Ambiguous intent

As explained in Effective Conversational AI

An ambiguous utterance is conversational input that is understandable but does not provide enough information to determine a single user goal.

A user who types “account” clearly signals a banking topic but not a single goal: opening, closing, or checking an account. Treating that input as one testable intent can distort classifier evaluation and create a poor user experience. section 4.4.2

An ambiguous utterance is understandable conversational input that does not identify one user goal. Examples include “driver license,” “that one,” and “2.” Context may clarify some phrases, but an isolated utterance can map to several intents or lose its earlier reference. section 4.4.2

Handle ambiguous utterances according to their role in the conversation:

  • Usually exclude them from classifier accuracy tests because no single golden intent can be assigned. section 4.4.2
  • Set them aside for negative testing and measurement of how often users communicate without a clear goal. section 4.4.2
  • Include responses from buttons or information-gathering flows only when they align with an intent used in that flow. section 4.4.2
  • Evaluate whether disambiguation and clarifying questions guide users toward a resolvable goal. section 4.4.2

How Effective Conversational AI covers it

How deep it goes
Defines the condition, gives concrete examples, and specifies testing treatment without detailing implementation.
Examples
Uses “account,” “driver license,” “that one,” and “2” to illustrate unclear goals, lost references, and contextual responses.
What you can do after
Diagnose ambiguous test data and evaluate whether conversational recovery features handle it appropriately.
How it connects
Connects ambiguous utterances with intents, classifier testing, negative testing, disambiguation, and clarifying questions.

Related concepts

This explainer was generated by Manning's knowledge graph from the text of Effective Conversational AI. The section links point to the parts of the book it draws on.