T9 keys are large and rarely missed, but candidates are more ambiguous
Aliases: nine-key pad · T9 · multi-tap
What it is
A T9 / 3×3 letter pad packs the twenty-six letters onto about twelve keys. Each key is larger than on a full keyboard, so fingers hit it more easily. The cost is that a press no longer names a letter: it names a set of three or four. The system must recover the intended word by multi-tap cycling or by dictionary disambiguation. Hit rate paid for by key size is billed as candidate ambiguity.
Why it happens
In Fitts’ law, wider targets cut aiming time and neighbour misses. The twelve-key pad buys that width with reuse: one press is a set, not a point. Multi-tap leaves ambiguity to timing—too fast merges into the next letter on the same key, too slow is read as the next character. Predictive T9 leaves ambiguity to a lexicon: the digit sequence 43556 matches hello and other words, and a dictionary plus language model pick one. Chinese pinyin on a nine-key pad does the same: initials and finals share keys, so one press string names many spellings and then many words. Larger keys, larger sets, longer candidate lists. “Hit the intended key” is one error class; “picked the wrong word” is another, after the presses are done.
Studying it
On the same touchscreen, compare a nine-key pad with a full keyboard on per-key hits and word-level accuracy. Independent variables include key width, prediction on or off, and whether the test words sit in the lexicon; dependent measures include neighbour misses, multi-tap timing errors, wrong-candidate selections, and completion time. Log “wrong key” separately from “right keys, wrong word,” or the nine-key pad will look universally better because keys are large, or universally worse because candidates err.
Where it stops holding
Physical twelve-key phones add travel and ridges that further cut misses; a touch nine-key pad has no ridges, so the size advantage shrinks. If prediction is off and the user multi-taps, ambiguity moves from the lexicon to timing, and fast fingers merge keys more. For proper names absent from the lexicon, disambiguation has nothing to stand on and must fall back to per-letter multi-tap: the large-key advantage remains, the ambiguity cost peaks.
Applying it
- Offer a nine-key pad for short messages where keys are easy to aim and the lexicon covers the language, with a one-tap default candidate.
- Treat “missed the key” and “picked the wrong candidate” as different error displays: key feedback for the first, an explicit choice in the candidate bar for the second.
- Test candidate hits on the target users’ real word frequencies; do not make the nine-key pad the only keyboard when the lexicon does not cover them.