The ordinal is the only handle you can name

What it is

Nearby coffee is read as “Starbucks, Luckin, Manner.” The user wants the third; the name was heard once, they will not risk the English, and they cannot place the stress. What they can say, and what the system can hear, is often only “the third one.” Ordinal as speakable handle means: in an auditory list, the position number is the short label that remains nameable after the list has passed. Shop names, track titles, and flight numbers may be too long, too alike, or already gone from the ear.

Why it happens

Referring needs a symbol that is still in common ground and short enough to repeat accurately. On a screen, a finger or “that one” will do; a spoken list has no spatial point, and the name itself is a novel word heard once — repeating it is dictation. The ordinal is spoken with the item, fixed in length (“first,” “second”), unique inside the current set, and tiny as a recognizer vocabulary. It turns deixis from “repeat that proper name” into “report an index I just gave you.”

Without ordinals, people invent handles: “the one that starts with M,” “the middle one.” Those phrases mean something to the speaker and often nothing to ASR or entity linking. When names are acoustic neighbours, even a willing repeat misses. Ordinals do not decide whether the list should have been this long; they decide how to name an item after it has been spoken.

Studying it

After a named spoken list, elicit a choice (“which one”) and code the referring expression: ordinal, full name, partial name, attribute (“the green one”). Independent variables: whether an ordinal is spoken before each item, name length, whether the name is non-native. Dependent measures: distribution of referring types, selection success, and whether “say it again” follows a failed name.

A control prints the same names on cards and allows looking. Named referring rises with a screen, which shows the ordinal as compensation for the auditory channel, not a universal preference. Do not substitute “was the menu nice” for a coding of referring forms — people can find the reading clear and still never dare the shop name.

Where it stops holding

Two items with short, punched names (“on / off,” “red / blue”) are themselves the handle; the ordinal is a second vocabulary nobody needs. Users who already know the set (three regular cafés) will name the proper noun, and the ordinal gets in the way — “the third” requires recalling order. Ordinals are unique only on the current page; after a page turn “the second” binds to someone else, so page must travel with the index or the count must reset. Treating ordinals as “therefore announce how many there are” is a different job: the total is a preview; the ordinal is the handle for pointing afterwards.

Applying it

• Open every spoken candidate with a stable ordinal (“first… second…”) and map “the Nth,” “number N,” “the last one” onto the current page index in recognition.
• Do not expect users to repeat a proper name heard once. Still speak the name; the pass condition is being able to select without saying it.
• After a page turn, either restart ordinals on the page and say so (“next are four through six”) or keep a global index — do not mix the two.
• How to check: listen to the list with no screen, speech-only selection. Count ordinal successes. On hard-to-pronounce names, if almost nobody uses the name and the system does not take ordinals, the handle is missing.

Related

• Same group: M3.08.2 Speak only one or two distinctive fields per item · M3.08.3 Paged readout needs a user-gated continue
• Nearby: M3.03 Reading Long Lists Aloud · M1.02 Memory Load of Screenless Interaction · M1.09 Context Retention and Reference Resolution
• Search terms: ordinal as speakable handle · deictic · spoken list index