A screen reader serializes the interface into a spoken sequence
Aliases: linearization · spoken sequence · screen reader
What it is
On a rebooking page, the eyes can hold the flight on the left and the notices on the right at once; the ear can hold one utterance. A screen reader flattens a two-dimensional interface into a timeline of speech. That flattening is linearization. It is not “reading the page aloud for people who cannot see”; it is rewriting a glance as a sequence.
Why it happens
Speech is a temporal medium. Vision can sample two columns, a card, and the chrome in parallel; hearing has to queue. The reader lines objects up and synthesizes them one by one, so the user’s model is a list, not a map — “the card next to this” has no counterpart in the stream.
Once the sequence exists, what comes next is decided by document structure, not by who sits nearest on screen. “Visible at the same time” becomes “arrives one after another.” Experienced users push speech rate past three hundred words a minute to buy back some of the lost parallelism; they are not restoring two dimensions.
Studying it
Sensory deprivation: cover the display and complete find-flight / rebook / confirm-amount by ear only. Then ask people to sketch the page’s regions — wherever reconstruction fails, linearization dropped a spatial relation.
Independent variables: object count, availability of jump hooks, speech rate. Dependent variables: time on task, how much of the page was heard linearly, whether reported spatial relations match the visual comp.
Run NVDA+Firefox, JAWS+Chrome, and VoiceOver+Safari; engines chunk objects differently. A satisfaction score is not a substitute for “after listening, can they say how many regions the page has.”
Where it stops holding
Low-vision users stack residual vision on top of speech, so linearization is not their only channel and the claim weakens. A refreshable braille display is also serial, but tactile rather than spoken — similar failure, different medium. Spatial audio can hint at direction; it is not the default model of mainstream readers. Testing a reader as a “read the whole article” tool measures listening, not wayfinding in a sequence.
Applying it
- Walk the primary flow in listening order: when each utterance arrives, does the user already have the context it needs.
- Do not split information that must be compared across two places that are visually adjacent but far apart in the sequence.
- How to check: turn the display off and finish one purchase or checkout by ear. Every point where someone has to ask “which region was that sentence in” is a relation linearization cut.