Week 4 Lab: AAC scanning simulation

A lab spent composing sentences through a scanning-and-selection interface with a deliberately mediocre word predictor, to feel the cost of a bad list

18 March 2027

Before the lab#

Nothing to bring. The simulation runs in the browser: a small on-screen keyboard that scans through rows and columns on a timer, selected with a single key, with a five-word prediction bar you can also select from.

In the lab#

You’ll compose short, assigned sentences twice — once using only the letter grid, once allowed to use the prediction bar — and log the selection count for each. Then you’ll do it again with a prediction list quietly swapped for one trained on a mismatched vocabulary, so the words it predicts are confidently wrong for what you’re being asked to say.

The debrief is the point of the lab: most people find the mismatched predictor more frustrating than no predictor at all, because a wrong-but- confident suggestion costs a selection to dismiss, on top of the selections it was supposed to save.

Afterwards#

Nothing here is separately assessed, but if your Assignment 1 build touches accessible input at all, the selection counts you logged today are the kind of evidence a marker will expect to see cited, not asserted.