How many options you can speak is capped by working memory

What it is

A speaker asks “which white noise: rain, waves, campfire, fan, forest, stream, thunder.” Seven new labels enter the ear in sequence; what can be held at once and used as a set to choose from is typically around four. Anything past that capacity is not “one more alternative” — it pushes an as-yet-unchosen item out of the buffer. The working-memory cap on spoken options limits how many items can be spoken and still treated as a choice set, not how many words a synthesizer can pronounce.

Why it happens

A visual list offloads capacity onto paper or glass; working memory holds only the current focus. Spoken options have nowhere to unload, so the whole set must stay active by rehearsal in the phonological loop. Capacity is not a fixed “seven plus or minus two”: for unfamiliar, similar labels, Cowan’s focus of attention is closer to four, sometimes less; acoustic neighbours (“forest / fan”) substitute for each other during rehearsal. Each added item raises rehearsal load and leaves less resource for “which one do I actually want.” People truncate early — the third item sounds “fine” and they take it — not from preference, from a buffer that is already full.

The cap and “can I look back after hearing” are different layers: three items can still be chosen wrong because look-back is missing; seven items do not even fit the buffer, so look-back is no longer the main cause. Treating the cap as a reading-technique problem sends people off to polish how those seven names are spoken, instead of cutting seven to three.

Studying it

Run a spoken choice-set span: N from 2 to 8, each item a new unchunked label (noise names, alarm timbres, cook modes). After the readout, pick the item that matches a criterion (“quietest,” “good for sleep”) or name the one wanted. Dependents: hit rate, substitutions to a neighbour, requests to hear the whole set again. The curve usually kinks near 4, not 7.

Add load: simple mental arithmetic or a tracking secondary task during the menu. Under occupancy the kink moves earlier, so the cap shrinks with available working memory; it is not a product constant. If the lab lets people tick options on paper, the measure is no longer spoken capacity.

Where it stops holding

A closed, daily vocabulary (“play / pause / next”) has been practised into a chunk; three surface items occupy one slot. Expert dispatch phrases can run longer because the structure is theirs, not learned this turn. If world knowledge filters on contact (seven city names, only one the user ever goes to), the effective set is already smaller than N before hearing finishes. With a screen present, capacity unloads visually and spoken count is no longer under this cap — the cap applies to screenless turns whose options are new to the user.

Applying it

• Design a spoken presentation of new options with a 3–4 item ceiling. From the fifth item, filter first (“outdoor or indoor sound?”) and then present a subset.
• Do not accept “the TTS can still read it.” Reading seven names costs the synthesizer almost nothing; it does not cost working memory nothing.
• Frequent, stable small vocabularies may run a little over, but accept them as chunks (the user can name the item without a replay), and do not mix one-shot long menus into the same quota.
• How to check: the same skill at N=3 and N=7. If N=7 drops hits, raises full-set replays, or people take the last-heard item as a filler, the set is still outside the cap.

Related

• Same group: M1.02.1 Heard options vanish; there is nothing to glance back at · M1.02.3 Long spoken answers need structure, not a linear dump
• Nearby: M2.02 Open and Closed Questions · M3.08 Difficulty of Reading Long Lists Aloud · M3.11 Trimming Speech Output
• Search terms: working-memory cap on spoken options · spoken choice-set size · Cowan focus of attention