Picking letters with a D-pad is extremely costly
Aliases: on-screen keyboard · KSPC · D-pad typing
What it is
Letters on television sit in a grid. A direction key walks one cell; OK commits the current letter. Cost per character is “walk to it + confirm”; cost of a string is the sum, plus space, shift, backspace, and script switching. D-pad character selection is extremely costly names this character-picking channel, not how far general navigation is. Navigation can place frequent items near; the alphabet’s couple of dozen cells cannot be deleted.
Why it happens
A QWERTY on-screen keyboard is about six to ten columns. From a centre default key to a corner letter is already three or four steps, plus confirm. Letters that are adjacent on a physical keyboard may still sit on different rows on TV, so the row-crossing tax still applies. Backspace often lives on a function key outside the letter grid; fixing one typo means walking out and back. Shift, digits, and symbols are another grid; switching is a mode tax, then each symbol is another walk.
Prediction and T9 can cut steps, but the first press still has to reach a key, and a wrong prediction has to be cancelled. Pinyin turns a character into a longer Latin string and then a candidate walk—two layers of cost. People slow down, get angry, and shorten queries to two or three letters on luck. Demand did not shrink; the channel made a full string an irrational choice.
Studying it
TV text entry keeps the desktop keyboard’s dependent variables, on a D-pad grid: keystrokes per character (KSPC), words per minute, backspace rate. Tasks are titles, email addresses, short search strings, not prose.
Independent variables: layout (QWERTY / ABC / T9), prediction on or off, which key holds default focus, key-repeat on or off. Dependent variables: KSPC, completion time, errors, abandonment.
Do not use physical-keyboard speed as a ceiling—that is another channel. The right controls are spoken query and phone typing of the same strings. Lab “please type this,” printed on paper, is copying, without the real hesitation of “I want to search but I am unsure of the spelling.” Title tasks are closer to the living room than random letter strings. Continuous typing leaves default focus near the previous letter, so KSPC drops; the first word after opening the keyboard is dearer. Measure both.
Where it stops holding
A four-digit PIN, yes/no, picking an existing address from a list, is not free character selection; cost is not this layer. Trackpad tracing, handwriting, a physical keyboard dock replace the grid walk. Gamers are fluent on a D-pad; KSPC remains, only time per step shrinks, and long strings still collapse. Number keys on the remote are a direct path for digits—if the task really is digits.
Applying it
- Prefer picking to typing: make recent queries, popular titles, and address-book entries a focusable list. The on-screen keyboard is the last path.
- If typing is required, land default focus on the prediction strip or a central letter, offer word-level prediction on QWERTY, and put backspace where a direction key can reach it without leaving the letter region.
- For Pinyin and similar, make candidates the first focus landing after a syllable; do not walk back to the letter grid after every chunk.
- Verify with ten real titles and one email, D-pad only, logging KSPC and abandonment. If KSPC sits stably above a couple of confirms plus a few walks, change the task (to selection) before restyling the keyboard.