Beyond the Chat BoxEssay 03 / 10
Voice and turn-taking
ESSAY 03

Silence Is an Interface State

The best voice assistant knows when to yield the floor.

Voice interfaces often assume a simple rhythm: the person speaks, the system listens; the system speaks, the person listens. Real conversation is more fluid. People pause to think, hesitate, restart, laugh, breathe, and change direction. Silence can mean uncertainty, concentration, consent, refusal, or simply that the person is finished. Treating every pause as a gap to fill can make an assistant feel impatient.

The cost of premature completion

Someone practising a presentation pauses after a difficult sentence. An assistant that immediately supplies the next line may be technically helpful but socially disruptive: the speaker may be searching for their own words. A person asking for directions may pause to inspect a street sign. The right response may be to wait, not add more information.

Beyond listening and speaking

A voice interface could expose several states: listening, thinking, waiting, ready, paused, and uncertain. The user should be able to see or hear when the microphone is active, when input has been received, and when the system is holding the floor. Such cues make turn-taking more predictable and allow people to correct the system.

Timing is a design decision

The right pause length depends on task, language, speaking style, accessibility needs, and setting. An assistant that interrupts quickly may dominate; one that waits too long may feel unresponsive. Users need control over interruption sensitivity and a clear mute or pause mechanism. An open microphone must not be treated as consent to store or share everything it hears.

Where this can fail

The assistant may mistake a pause for a completed request, quiet speech for silence, or a non-native speech pattern for uncertainty. It may speak over another person or reveal private content in a public place. These are failures to respect conversational boundaries, not only speech-recognition errors.

Questions for research

  1. How do people signal that they have finished speaking across conversational styles?
  2. Which state cues make voice systems feel more predictable?
  3. When does proactive completion help, and when does it undermine confidence?
  4. How should users tune timing without being stereotyped by speech patterns?

Design hypothesis

A good voice assistant is not merely accurate at recognising words. It is skilled at turn-taking, including the decision to wait.

Research starting points

  1. Sacks, Schegloff & Jefferson (1974), A Simplest Systematics for the Organization of Turn-Taking for Conversation, Language 50(4), 696–735.
  2. Clark (1996), Using Language, Cambridge University Press.
  3. Amershi et al. (2019), CHI. https://doi.org/10.1145/3290605.3300233
← Back to all ten essays