Good Reason

It's okay to be wrong. It's not okay to stay wrong.

How to improve T9

I recently won the contest for ‘Last Person on Earth to Get a Mobile Phone’. First prize was a mobile phone. I like it. It was worth outlasting that guy from the Amazon. He got a toaster, and nowhere to plug it in. Ha.

My phone uses T9, the predictive text algorithm. It was invented in the early 90’s, and don’t you think we would have come up with some improvements in language technology since then? But no, we’re still stuck with it, and every day I text Ms Perfect to tell her that I’ll be ‘good room’ instead of ‘home soon’. ‘Good’ and ‘home’ are textonyms, you probably know, both keyed as 4663. I can change from one to the other by hitting zero, but it irritates of.

Irritates ‘me’, sorry.

I’ve seen very little out there on improving the T9 algorithm, so here are my suggestions.

  • At the very least, correct gibberish words. Even a relevant word like ‘texting’ comes out as ‘textiog’ on my Samsung mobile.
  • Auto completion. When I type a long word like ‘predictive’ or ‘abracadabra’, it should have a way to complete the word for me. If there is one, someone let me know.
  • Long-term memory on training. T9 does try to adapt to your usage. I’ve noticed that if I type the same textonym over and over, changing it to another variant each time, it’ll select the variant automatically on the fourth time. But only for that message. Next message you send, it forgets all your training. What is the point?
  • And this is the big one: Word bigram modelling. Many textonyms could be disambiguated simply by looking at one or two previous words. For example, ‘good’ and ‘home’ are both 4663, but the previous words are very often different. If the previous word is ‘coming’, choose ‘home’. If ‘is’, ‘was’, or an adverb like ‘very’, choose ‘good’. It’s very simple to check this. When I compared ‘home’ and ‘good’ in the Brown Corpus, there were no duplicates in the top 100 lists of words previous. Same with ‘am’ and ‘an’, another pair of textonyms. Which tells me that just looking at the previous word would be enough to disambiguate in the majority of cases. And that means we can stop hitting zero so many times.

10 Comments

  1. I have been lurking around your blog for awhile now and find it insightful – Thanks!

    As far as your phone dilemma – i only have one word for you –

    BLACKBERRY! It has word prediction that I love and i VERY rarely ever have texting problems!

  2. Another thing I can be last at getting!

    So any ideas on how it does it?

  3. You have certainly asked the wrong person that question! All I know is it has something to do with “intellitype” – which undoubtedly is the layperson term for non-techies like me! But hey – it does work whatever it is!

  4. Forget texting. Morse code!

  5. Telepathy!

    (BTW: Just watched ‘Messiah’ again. How does Derren Brown do it?)

  6. With considerably greater wit and skill than most of the celebrity psychics.

    As for the mechanics, I’m still working that out.

    We need a UWA magic circle.

  7. my mobile phone suggests long words so I can click on them and avoid pressing buttons. It’s a Sony Erikson. But instead of Canterbury (surely one of England’s major cities?) it suggests Canteratsy – you what??? That isn’t a word.

  8. I didn’t realise that T9 was so “old” ’cause I only just got my first mobile last year and I’m a teenager. He he. I came across your blog ’cause I use your font on Picnik.

    Thanks for everything, it’s all quite interesting.

  9. Daniel,

    How is it that you didn’t get an iPhone. They really are very fun.

  10. They haven’t been available in Australia. They are coming out here soon, but I still probably won’t get one. I already have an iPod, a Mac for browsing, and a mobile phone. I wouldn’t try to consolidate them into an iPhone when it only has 16 GB.

    Gad, I just said “only 16 GB”.

    My phone needs are fairly modest at this stage. They do look cool though.

Comments are closed.

© 2024 Good Reason

Theme by Anders NorenUp ↑