Finish My Sentence

Why it works

Chatbots and the word suggestions on a phone keyboard work by predicting what is likely to come next, based on enormous amounts of text they have learned from. Children are excellent predictors too. Playing with familiar and unfamiliar sentences shows them, from the inside, what prediction is good at and where it goes wrong: it is great at what usually happens, and can be confidently wrong about what actually happened.

How to play

1. Start with something everyone knows: “Once upon a…” Everyone shouts the next word.
2. Try a few more familiar starts: “Happy birthday to…”, “Brush your…”
3. Now start a sentence about something only you know: “Today at work, my friend told me she…” Let everyone guess.
4. Compare. Why were the first ones easy and the last one hard?
5. Take turns being the one who starts.

Make it easier, make it harder

• Easier: stick to nursery rhymes, songs, and family sayings.
• Harder: after each guess, ask “how sure are you, from 1 to 10?” Then check whether the sure guesses were the right ones.

Talk about it

• Why were some sentences easy to finish?
• Was anyone very sure and still wrong?
• If a computer finishes sentences by guessing what usually comes next, when should we double-check what it says?

If your child asks

“So does a chatbot actually know things?” It has learned patterns from a huge amount of writing, so it is often right about common things. But it is built to produce a likely-sounding answer, not to check whether the answer is true. That is why it can sound completely confident and still make things up. Checking with a book, a teacher, or another source is a good habit.