ACC-02 · AAC and language models

Evaluate an AI-assisted AAC workflow for rate vs. authorship; propose safeguards for voice and agency.

One AI-assisted AAC or predictive-text workflow (school-approved tool or published case). Authorship/rate worksheet. Consent required for any partner who uses AAC; otherwise use published evaluations only. WAI principles as framing (https://www.w3.org/WAI/) when discussing operable, understandable interfaces.

Celebrating speed while erasing the user’s voice. Testing only on people who do not use AAC. Proposing voice cloning of a partner without specific consent. Declaring “agency preserved” without a measure.

Evaluation memo: rate metrics you can actually collect; authorship criteria declared before testing; safeguard list (what the system must not do); partner or published-evidence appendix; refusal of any voice-synthesis path without consent.

CT: rate vs authorship. CR: pre-declared authorship criteria. CO: peer red-teams safeguards. CM: safeguard list for a therapist. CZ: the AAC user whose words are at stake.

SLP, ATC, or special-ed teacher reviews authorship criteria. Standing instruction: reject speed-only success. School-supervised.

R1: one workflow. R2: criteria and measures written. R3: non-AI or current AAC baseline. R4: agency risks named. R5: memo usable. R6: voice/consent refusals explicit.

(a) Words-per-minute as the only score. (b) A demo that speaks for someone who never consented.

Lane A immediately. No verified credit claim.