Somewhere, a Telus rep is congratulating himself.
Four months ago, I was completely ready to ditch Telus and switch my phone service to an [MVNO](http://en.wikipedia.org/wiki/Mobile_virtual_network_operator). Indeed, if Zehrs hadn’t been repeatedly sold out of the handset I wanted, I’d probably be on the President’s Choice pay-as-you-go service. For light usage such as mine, it really is—by far—the most sensible and economical option.
In doing some soul-searching and bill-examination, however, it became apparent that my usage level really wasn’t light enough to realise a savings from switching to prepaid. So I called Telus and made a deal.
“Listen,” I said to the retention representative, “I’ll stick it out with your overpriced post-paid service. Just… give me a concession. Give me call display. There’s no reason for you to charge $7 for it, especially when it’s free to all prepaid users.”
And so he did. But he didn’t stop there. He actually gave me *two* concessions—in addition to the free call display, he also gave me 30 free outgoing text messages per month.
Texting is something I have always regarded with suspicion. The ordinary use cases for it have never made sense. In the past, when I’ve tried to interact with services like [Google SMS](http://www.google.ca/mobile/sms/index.html), I’ve found the input system slow and error-prone. Still: thirty free messages. How can that go wrong?
Well, by marvelous (or tragic) coincidence, about a week later was The Girl’s birthday, for which she received her first-ever mobile phone.
Instantly, everything made sense. Morning greetings, mid-day flirtations, pithy commentaries, the occasional goodnight. It’s everything that is too short and too transient for an email. It’s not instant enough to warrant a phone call, but amusing enough to be enjoyed in a spare moment during the day.
My usage of text shot through the roof.
A month and a half later, after two ridiculous overage charges, I finally crawled back to Telus, begging to be put on a plan, even if it meant sacrificing my 30 “free” texts.
The Girl has a handset with a [slide out qwerty keyboard](http://www.ubergizmo.com/15/archives/2008/11/telus_offers_samsung_m540_slyde.html). I, however, am slumming it with my lowly Motorola W385, and its [iTap input system](http://en.wikipedia.org/wiki/ITap)—better known colloquially as T9.
Now, it’s shocking that T9-based input is as usable as it is. For a lot of words, it correctly predicts your meaning and the total number of keystrokes is the same as using a full keyboard. For names and non-dictionary words, it’s a much more laborious process to enter them, as many as two, three, or even four keystrokes to produce a single character.
For those words where a certain sequence of numbers produces multiple options, the phone offers a simple a mechanism for selecting the one you want, which then becomes the default for that sequence.
On the whole, the system works well. The most annoying pair of overlapping words are *of* and *me*. Both are words I use with about equal frequency, and often—it [seems](http://en.wikipedia.org/wiki/Confirmation_bias)—in an alternating fashion. As clever as T9/iTap is, it is frustrating to note that even the slightest bit of context sensitivity (that is, knowledge of the preceeding word) would drastically increase the likelihood of making a correct guess in cases like the of/me split.
The most interesting thing in all this is occasionally stumbling across a pair of overlapping words that are different enough to be almost surreal. I mean, you expect confusion between *block* and *clock*, but what about *cheer* and *bides*? Or *astride* and *brushed*?
On the train yesterday, I wrote a quick script to find and display these amusing overlap pairs. The logic of it is simple enough:
1. Sort an English word list into buckets by T9 representation,
2. Cull the list to T9 representations that have at least 2 corresponding English words, and
3. Print out all pairs where at least four characters in the English are different between the two words.
The parameters are subject to some variation, but I found this combination worked well. Indeed, the most difficult and interesting part of the exercise was choosing the word list. I tried three:
1. The system file [words](http://en.wikipedia.org/wiki/Words_(Unix\)) from Mac OS X (~235k items),
2. The same file from an Ubuntu 9.04 system (98k items), and
3. A list called [corn cob](http://www.mieliestronk.com/wordlist.html), which I’d never heard of (58k items).
Initially, I assumed that the largest list would produce the most interesting results; it turned out that the opposite was the case. The Mac OS X list contains far too many obscure words—any interesting results were being drowned out by the noise of discovering pairs like *stoun* and *punto*.
There is more that could be done with this, but you’re welcome to [try it out yourself](http://sandbox.mikepurvis.com/js/t9words/). If you’d like to pass in parameters, you can do it [like so](http://sandbox.mikepurvis.com/js/t9words/?words-ubuntu.txt,4,6).
The first is the name of the word list (options are `words-ubuntu.txt`, `words-corncob.txt`, and `words-osx.txt`), the second option is the minimum number of differing letters, and the last is the minimum overall word length.
What strange ones can you find? My favourites so far are baptise/acquire, scorched/rampage, and beget/adieu.