Too much personality raises information-density cost

What it is

Personality is performed in words and syllables. Those syllables compete with the payload on the same auditory channel. A morning-briefing voice that spends thirty seconds on manner and mood before reading exchange rates and stalled lines often leaves the manner in memory and lets the rates slip. Information-density cost means: character colour in prompts and answers occupies time that propositions needed. Personality is not “bad.” It is billed by the second.

Why it happens

Audition passes one timeline. Jokes, catchphrases, anthropomorphic asides are extra propositions that must play before, inside, or after the payload. Up front, they delay speech of the payload and park it in a less stable stretch of working memory. In the middle, they split the payload, with no glance-back to rescan. At the tail, the listener may already be planning the next line, so the personality never enters the ear — and still occupies device talk time.

This is the other end of the same channel constraint as “keep the prompt short”: there, shortness is the instruction; here, the surplus is character. A sentence that both performs a role and delivers a number also pits character prosody against the grouping pauses the number needs. Denser personality means more seconds per useful fact, and a higher chance of barge-in and forgetting. What is spent is density, not “not being fun enough.”

Studying it

Give the same payload two shells: minimal persona (read the numbers) versus loaded persona (a fixed-length joke or catchphrase before or inside). Dependent measures: immediate recall of payload items, time from onset to the target fact, whether barge-in lands in the personality stretch or the payload. Independent variables: length of the pad, position of the personality stretch, difficulty of the payload (a string of rates versus a yes/no).

Do not make “liked this voice” the primary measure. Liking can coexist with high forgetting. In the field, align the first syllable of useful information against turn onset; if that offset grows across versions while recall falls, density is charging. A lab that lets people read a transcript will wash the auditory cost of the personality stretch into reading.

Where it stops holding

Seconds that must be filled anyway (a lookup, a transfer) are an almost free window for personality, because the channel has no payload then. A brand that has to be recognised on first meeting can spend one sentence on character and should taper after. Entertainment products (a bedtime story, match companionship) have personality as the payload; this cost account does not apply. Hard-of-hearing or non-native listeners may drop decoration and keep the numbers — or may have the numbers wrecked by the decoration. Either way the personality stretch must be skippable, not thicker.

Applying it

• Split every system utterance into two columns: payload syllables, persona syllables. If the persona column has no job of its own (fill a wait, establish brand, entertainment as such), cut it.
• Park personality in waits, one opening sentence, or stretches the user has asked to chat. Keep number, address, and deadline sentences at minimum persona.
• Allow barge-in through the personality stretch to the payload. Do not lock the payload behind an uninterruptible overture of character.
• How to check: on critical skills, measure time-to-target-fact and later recall. If adding personality lengthens time and hurts recall, personality yields, regardless of whether a listen-through felt more entertaining.

Related

• Same group: M2.05.1 Persona must stay consistent across every reply · M2.05.2 Persona must not mask capability limits
• Nearby: M2.01 Information density of prompts · M3.11 Trimming speech output · M3.07 Rate, pauses and prosody
• Search terms: personality information-density cost · verbal overhead · prompt density