Week 4: Prediction as access — AAC and assistive input

What predictive text looks like when it's the only way someone composes language at conversational speed, not a convenience layered on top of typing

15 March 2027

Everything so far has treated prediction as a shortcut for someone who could, in principle, type the whole word. Augmentative and alternative communication devices — the tablets and dedicated hardware used by people who can’t reliably produce speech or handwriting at conversational speed — remove that assumption. For a user selecting words one scan-step at a time, prediction isn’t shaving a few keystrokes off something they could do anyway; it’s the difference between finishing a sentence in the length of the conversation or finishing it after everyone has moved on.

That changes what the design problem is. A word-prediction error in a messaging app costs a laugh and a correction. A word-prediction error in an AAC device, or a prediction list that buries the word someone needs behind six more common ones, costs turns in a conversation that don’t come back. The lecture treats accessible input as the setting where every property this course has discussed so casually — how ranking works, how a vocabulary gets built, whose language counts as “common” — turns out to matter enough that getting it wrong is a rights issue, not a UX complaint.

Outline#

• scanning input and why prediction quality is a speed multiplier, not a nicety, in this setting
• vocabulary design: whose words end up ranked first, and who decided
• what “the user can always type the long way” fails to account for
• the lab’s device-free simulation, and why it’s more frustrating than you expect