Word prediction

See it for yourself

A tiny predictor living entirely in this page. Type, and take a suggestion with Tab or a click.

Try it

How it works

A guess about the next word

Underneath every predictor is a language model: something that looks at what you have written so far and ranks which words are likely to come next. Type "read the" and the model puts its money on words that usually follow those two, in roughly the order it has seen them before. Prediction is not magic — it is counting, done fast.

From word pairs to neural models

The simplest version is a table of word pairs: after this word, that word came next, this many times. That is exactly what the demo above runs — a tiny table of pairs, kept deliberately small so you can see both what the idea does and where it runs out. Modern predictors climb the same ladder with bigger steps: longer stretches of context, and large neural models instead of a table. The ranking idea underneath is the same.

Learning you

The assistive systems this idea came from added one more trick: they learned each user. Words you type often rise up the list, and words you typed recently rise faster — frequency and recency, per person. That is also where assistive prediction still parts ways with the phone keyboard: you can tune how many suggestions appear, have them spoken aloud, and keep a word bank that is yours.

Where it came from

Word prediction was born in augmentative and alternative communication — AAC — research in the 1980s. It was built for people for whom each keystroke was expensive: if pressing a key costs real effort, a system that offers the whole word after one or two letters gives you most of the word for a fraction of the work. Those early systems already learned frequency and recency per user, so your own words floated to the top.

Then the idea escaped the lab. T9 brought prediction to phone keypads in the 1990s, where nine keys had to cover a whole alphabet, and smartphone keyboards made it universal — the strip of suggested words above your keyboard is this same idea, worn smooth by a few billion thumbs.

It is a familiar arc in assistive technology: a tool built for people at the edges turns out to be useful for everyone. What the assistive version keeps, and the phone version dropped, is the tuning — suggestion counts you can set, suggestions read aloud, a word bank that belongs to you.

Who it helps — honestly

Prediction helps most when words are expensive — in one of two ways. For writers with dyslexia or dysgraphia, the cost is spelling: you know the word you want, and picking it from a list beats building it letter by letter with the spelling risk that carries. For typists with motor difficulties, the cost is the keystrokes themselves, and every accepted suggestion is keystrokes you did not have to make.

Now the honest limit: suggestions can derail a sentence. A predictor offers the most likely next word, not your next word, and if you follow it uncritically you can drift, word by plausible word, into a sentence you never meant to write. The tool is at its best when you arrive knowing what you want to say and let it handle the spelling — not when you let it decide where the sentence goes.

You can watch that failure happen in the demo above. Its model is a tiny table of word pairs, so the drift is not subtle: take every suggestion it offers and within a line you are writing its sentence instead of yours. A bigger model drifts more gracefully, but it is the same failure — which is exactly why seeing it on the small model is worth your time. Treat suggestions as offers. The sentence stays yours.

Try it in Helperbird

Helperbird brings these tools to any website, PDF, or Google Doc — no copying text anywhere, and nothing you read leaves your browser.

Word prediction Suggestions as you type, on any website and in Google Docs.

Step-by-step: How to use word prediction on any website or Google Docs · How to clear learned words

Questions people ask

Does Google Docs have word prediction?

Docs has Smart Compose for finishing phrases, but it is not built as an assistive tool — you cannot tune how many suggestions appear or have them spoken. Helperbird adds assistive-style prediction on top of Docs and any other site.

How is word prediction different from autocorrect?

Autocorrect fixes a word after you type it; prediction offers the word before you type it. For someone for whom each word is costly to spell, offering beats fixing.

Who invented word prediction?

It grew out of augmentative and alternative communication (AAC) research in the 1980s, built for people for whom every keystroke was expensive. The phone keyboards that popularized it arrived a generation later.