Speak only one or two distinctive fields per item

What it is

Three flights to Beijing, each read as “flight number, times, aircraft, meal, Wi-Fi, gate.” After three, the dimension that would actually decide — nonstop versus connection, morning versus noon — is drowned in six or seven shared fields. Distinctive-field spoken items require that only the one or two fields that actually differ inside the current set earn a place in the ear. The rest of the record is not gone; it does not belong in this round of spoken comparison.

Why it happens

Auditory comparison runs on differences still in the buffer, not on record completeness. More fields per item means the shared parts (all go to Beijing, all are jets) spend time twice or three times, and the forking dimension is pushed to the middle or the tail, where comparison dies. Distinctiveness is relative to this set: if all three are Starbucks, brand is not a distinctive field and distance or the street is; if the three categories differ, category outranks rating. Treating “more information” as “read more per item” spends completeness against discriminability.

The one or two fields also have to be the right ones. Two fields that do not vary (all three serve a meal) are as good as silence. A field the current decision does not use (aircraft type) occupies the slot that time-of-day or connection should have had. Distinctiveness moves when filters move: after “nonstop only,” nonstop stops distinguishing and departure time comes forward.

Studying it

Same candidate set, two readings: full record versus the one or two fields with highest mutual information on this contrast set. Then a forced choice or ranking (“which flight fits”) and a probe for the reason. Dependent measures: agreement with an objective best, whether the stated reason equals the distinctive field, and whether barge-in lands on repeated fields. Independent variables: within-set field variance (some fields identical, some all different), fields per item.

Do not use a subjective “was it informative enough.” That feeling can come from length. Printing the fields as a table while people listen measures visual comparison, not auditory distinctiveness. In the wild, mark at which item and which field people commit — if commitment always precedes the distinctive field, everything after is noise.

Where it stops holding

A singleton set has no contrast object; distinctiveness is undefined. Speak the decision fields for this item (when it lands, which stop to get off), not “how it differs from other candidates.” Legal or safety items that must be said now (redeye, allergen) cannot be trimmed even if they do not vary in the set. Experts with a fixed template (a dispatcher listening for a tail number) distinguish on that slot, not on algorithmically chosen mutual information. When a screen already shows the full record, speech should still speak only distinctive fields — that is a split of labour, not a second reading of the table.

Applying it

• Before readout, compute which fields actually vary on this page. Put only those that vary and serve the current decision into the one-or-two-field spoken template.
• Recompute when a filter changes. Do not keep the previous template (“nonstop only” should not say nonstop three times).
• Leave trimmed fields for after the item is named, or for the screen. Do not smuggle them back into the comparison round for completeness.
• How to check: the same three flights, full-record versus “time + nonstop/connection.” Ask “what did you choose on.” If the reason is not the distinctive dimension, or people pick whichever was heard last, the fields are still wrong.

Related

• Same group: M3.08.1 The ordinal is the only handle you can name · M3.08.3 Paged readout needs a user-gated continue
• Nearby: M3.03 Reading Long Lists Aloud · M3.11 Trimming Speech Output · M1.02 Memory Load of Screenless Interaction
• Search terms: distinctive-field spoken items · contrastive description · auditory comparison