12D notes — what each channel actually means

A reader asked for a plainer description of the twelve channels in Paper 1. The full mapping is in the preprint, but here's the short version.

The names are deliberately mundane: joy, anticipation, fear, trust, surprise, anxiety, calm, anger, disgust, sadness, pride, shame.

Each ranges from 0 to roughly 1, where 0 means "no measurable phonetic signal for this channel" and values above 0.6 are unusual in everyday text. The channels are correlated — joy and trust co-occur, fear and anxiety co-occur — and we don't pretend otherwise.

The phonetic drivers are mostly the obvious ones. Soft fricatives push trust up. Hard plosives correlate with anger. Long stressed vowels in the positive vowel-space contribute to joy. The unusual one is calm: it isn't really driven by phonetics at all, but by the absence of features driving the others. Calm is what you get when the other channels don't fire much.

The full ARPAbet feature table is in the appendix of the preprint. Email research@beyond-the-box.uk if you want a draft.

Related — Research

Research

Lifespan as state, not metadata

Most age-aware systems treat age as a variable. Paper 2 treats it as part of the agent's state.

Research

Why we're not training a model

The decision to make SERENA fully symbolic wasn't a research preference. It was an ethical one.