Warmtype Join the Android test

The thinking behind the suggestions.

What research says about connection, how we test our own autocorrect, and what none of it shows.

Here is what research says about why being understood matters, and how we test our own autocorrect before it reaches your phone.

Why being understood matters

A check-in, a shared thought or a reply is one small way people stay in touch. Baumeister and Leary's 1995 review argues that belonging, the need to form and keep lasting relationships, is a fundamental human motivation [1]. Ryan and Deci's 2000 review of self-determination theory names relatedness, alongside autonomy and competence, as a basic psychological need [2].

These are reviews about human motivation, not studies of writing tools or Warmtype. They explain why relationships matter; they don't show that a keyboard creates connection.

How we test our own autocorrect

Today's app corrects in a series of steps, each able to overrule the last. We're building one engine that weighs your taps, the spelling and the word before it all at once. Below is a recording of today's autocorrect on example phrases.

Before it switches on, it has to fix more words than it breaks on our recorded typing and on real typing from a phone, and every word it breaks is listed and worked on. It hasn't passed every check yet, so it is still off in the app.

Watch today's autocorrect

RecordingA recording of the app's current autocorrect on example phrases.

Pick an example below to watch it work.

A recording of our autocorrect at work on example phrases. Nothing here runs a live engine, and nothing leaves this page.

How it decided

    Pick an example to watch it work

    Nothing typed yet.

    Recording: these answers were saved from the autocorrect in the app today, running on each example phrase. Nothing is computed in your browser.

    We also ran an early experiment on spelling by sound, using English-learner spelling pairs from ETS's TOEFL-Spell set, licensed CC BY-SA 4.0 [3]. On 1,187 held-out misspellings:

    981 of 1,187
    had the intended word in the top three, with and without the experiment.
    751 of 1,187
    ranked it first with the experiment, against 755 without it.

    So the experiment didn't help yet, and it stays off. These are English learners' spellings, not typing from phones.

    What this doesn't show

    None of these sources tested Warmtype. They don't show that it helps people connect, or that it helps with learning or any medical condition. If we ever claim more, it will be because we measured it.

    Sources

    1. Baumeister and Leary (1995). The need to belong: desire for interpersonal attachments as a fundamental human motivation. Psychological Bulletin. DOI
    2. Ryan and Deci (2000). Self-determination theory and the facilitation of intrinsic motivation, social development, and well-being. American Psychologist. DOI
    3. Flor, Fried and Rozovskaya (2019). A benchmark corpus of English misspellings (TOEFL-Spell). Paper, licence