Do not blame the user for the miss

What it is

A recovery sentence puts the locus of responsibility on one side of the channel, and the user can hear which side. “I didn’t catch that name” is system-side. “Please speak more clearly” and “your accent is quite strong” are user-side. On a benefits hotline the second kind makes people raise their voice, apologise, or hang up, and the next turn is not necessarily more recognisable. A miss is a channel event. If the wording recasts it as a defect in the speaker, face is already broken before repair starts.

Why it happens

Spoken repair is already a face-threatening act: a public admission that the last utterance is void. Attribution then decides who wears the threat. System-side wording (“I missed that”) leaves the defect with the machine; the user remains a competent speaker and will usually try once more. User-side wording (“please speak clearly,” “please use standard wording”) assigns the defect to the speaker and triggers compensation: louder, slower, over-articulated, or exit. Those compensations are not reliably better for recognition — too loud clips, too slow trips the endpoint — and they always cost more.

When computers are treated as social actors, a blaming sentence lands with force close to a person-to-person rebuke; it does not automatically soften because “it’s only a machine.” Even without a “you,” “please speak more clearly” still measures the last utterance against a standard of adequacy. The locus sits in the verb, not in the politeness marker: “could you try again” can still be blaming the user.

Studying it

After the same miss, randomise two wordings: system-side (“I missed that name”) versus user-side (“please speak clearly” / “please use Mandarin”). Dependent measures: sound-pressure and rate on the next turn, presence of apology or self-blame, abandonment, whether that turn succeeds. A further independent variable is whether bystanders are present (a public counter versus a phone).

Conversation analysis can code the responsibility component inside repair initiations. In the field, compare hang-up rates around “please speak clearly.” Do not substitute a post-task rating of “was that polite” for behaviour: some people will call the blame professional while already talking louder. A lab with only a machine and no audience will understate how much user-side attribution costs in public.

Where it stops holding

Safety-critical instructions (“keep your face away from the device”) must name the person’s action; that is not attribution of a recognition miss, it is an unfolding risk. A teaching system (spoken assessment) is in the business of evaluating wording; avoiding attribution would empty the feedback. Even then, “this attempt didn’t score” and “the device didn’t hear” must stay separate. When the user already takes the miss (“sorry, I wasn’t clear”), doubling down on “no, I missed it” can turn into an argument; a short uptake is enough. Calling every miss “the network” when the intent was never covered is a different mis-attribution, not face-saving.

Applying it

• Subject of the recovery sentence is the system or the channel: “I didn’t catch that,” “it’s noisy here — say the name once more.” Ban “please speak clearly,” “accent,” “you were too fast.”
• When the user does need to change behaviour, give the next executable action (“come closer to the mic,” “please spell it”), not a quality judgement (“please speak more standard”).
• In copy review, tag the locus of every recovery line: system, channel, user. Anything that lands on the user is rewritten before it ships.
• How to check: sample the turn after a miss for jumps in sound pressure, apologies, hang-ups. If those rise together with user-side wording, rewrite to system-side and compare a second round.

Related

• Same group: M2.04.1 Say what was not heard, not just “repeat” · M2.04.2 Offer alternative phrasings
• Nearby: M1.06 Repair strategies · M4.01 Social cost of using voice in public · M4.09 Recognition differences for children and older adults
• Search terms: non-blaming recovery wording · locus of attribution · face-threatening repair