How it works
The motor map stays put. Context adapts beside it. The communicator remains the author.
Three layers, three different jobs
EyeSay separates the motor map, adaptive context and permanent agency layer. Labels, numbering, borders and text identify each plane independently of colour.
Voice Canvas
Stable plane
The 8×6 motor map stays in the same coordinates. Forty-eight describes the number of stable positions — not the number of words EyeSay can express. Category positions open additional vocabulary without moving the learned map underneath.
Learned positions stay put
Context Halo
Adaptive plane
Eight context slots can surface relevant possibilities beside the motor map without replacing it. The Halo does not decide intent and does not close the choice set: SOMETHING ELSE remains an explicit escape path back to broader language.
Suggestions never become decisions
Agency Strip
Permanent layer
STOP, NO, HELP, BREAK and TOILET live outside both the stable and adaptive language planes. They remain available through context changes because access to refusal, help and basic agency is not conditional on a prediction or routine.
context cannot remove it
Step 01
EyeSay can offer
Contextual words and partner models can be shown as possibilities.
Step 02
The communicator chooses
Nothing becomes their message until they deliberately select it.
Step 03
Their voice speaks
Only their chosen words can reach the utterance ribbon and SPEAK action.
Explicit communicator-selection boundary
Suggestions can cross this boundary only through a deliberate communicator action. Context, analytics and partner modelling cannot cross it on the communicator's behalf. The communicator's selection is the final authoring action. A partner model never becomes the communicator's utterance.
Model Mode boundary
A partner can demonstrate language without requiring imitation or response. The two lanes below share the map, but not authorship state.
Partner lane
Demonstration only
A partner can point to or model language on the same stable topology without creating communicator selections.
Communicator lane
Authorship stays here
Only explicit communicator selections can enter the communicator utterance and reach SPEAK.
Boundary rule: partner model events never become communicator utterance, communicator history or communicator-device speech.
The Voice Canvas is an 8×6 map with 48 permanent motor positions. A position can hold a directly selectable word or act as a clearly marked category doorway. Opening a category expands the available vocabulary and then returns to the same learned coordinates; expansion does not turn EyeSay into a 48-word system and does not relocate the base map.
The Context Halo has eight visually distinct slots alongside the Voice Canvas. It can respond to approved context such as the current routine or communication partner, but those suggestions remain possibilities rather than asserted intent. The Halo is frozen while a selection is in progress and SOMETHING ELSE provides a visible route away from the current suggestions.
STOP, NO, HELP, BREAK and TOILET are always reachable and cannot be removed by routines, editing or context. Self-advocacy is not a reward for compliance.
EyeSay can offer contextual words and longer sentence options, but it does not decide what the communicator means. Suggestions never auto-speak. A suggestion only becomes part of the message when the communicator deliberately selects it, and only their chosen words can reach SPEAK.
Model Mode uses the same stable motor topology so a partner can demonstrate requests, refusals, comments, questions, social language and self-advocacy without creating a second map to learn. Partner model events stay technically separate: they never become the communicator's utterance, cannot insert into their ribbon and cannot trigger their voice.