Sunday, 1 June 2008

So near, and yet so far...

I have always been a fan of predictive text systems. On a mobile phone - to use just the least surprising possibility - something as simple as T9 - which can't even claim to be properly predictive, being instead interpretive - can slash necessary keystrokes and make even a simple numeric keypad competitive with a full qwerty keyboard. And on any computer equipped with a decent mouse equivalent - hell, even an actual mouse - dasher is not only the best implementation of a predictive system that I have yet seen, and not only a fantastic solution to the problem of text entry for the digitally disabled, but insanely cool as well.

My association with predictive systems has been a time of searching for the next better thing, at least partly because I know that we're not already getting the best that's available.

I was eager to try out T9 before I had a phone equipped with it and was scratching my head wondering if anybody had tried a similar thing with PDAs, to save having to tap at a 10-letter keyboard compressed into the width of the touch screen, before I got my hands on an early Sharp Zaurus and found out that someone had (Qtopia/Opie's pickboard).

At some point I stumbled across dasher and never stopped swearing that it never worked on the Zaurus.

And now I have a mobile that is a sight more powerful than my computer was when I first got that grandfather's-axe perpetual upgrade project, and yet something as fundamental as text entry has advanced not one iota, not one tiny fraction of an atom. We're still stuck with an interpretive system when I have enough processing power on hand to replace a hand-held games console, view several binary file formats, browse the web at full broadband speeds while playing either a radio station or any one of a number of music file formats on-board... You get the general idea.

Has anybody out there in the industry stopped to think "Hang on... Are we really even trying, here? Aren't we perhaps being a little, well, complacent?"

Which is, of course, were third-party developers come in.

I am currently (yes, this has had a point to it) using a trial version of a third-party true predictive text system called Adaptxt to write this entire blog post. If I can stand to.

You see, it doesn't entirely work.

Let me explain. The way that Adaptxt works is by plugging itself into the phone and hijacking - well, extending - the native input systems. It then attempts to guess what word you're in the process of writing and gives you a tiny little pop-up window with options in it that you can scroll through. Potentially, you can hit one letter and then get a really long word with as little as one extra press on the D-pad. If you want a word it doesn't know, adding new ones is extremely simple.

But wait, there's more! Just like dasher itself, Adaptxt attempts to also predict what you are about to type, giving you an option on the next few words as well.

Ultimately, I can imagine that after getting used to this system, and having the bugs worked out, it would be a handy improvement to your speed using unassisted T9, particularly given it's final trick - it learns as you use it, making future predictions theoretically more accurate.

But there are nonetheless bugs in the bouillabaisse, and I'm not talking Moreton Bay.

For a start, it doesn't work consistently across all applications. In most places, you can switch smoothly between lower case, capitals and the same with T9, but for some reason the pure number mode is unattainable - the command that's supposed to work, doesn't. All the time time. It does sometimes, you see. In the text editor that I am currently using, JBak's rather excellent DEdit (can anyone tell me why so many ebook readers and text editors for handheld devices are created by Russians?), I can't get T9 at all, which slows me right back down again.

There's also an odd little encoding problem: Adaptxt replaces the existing punctuation-and-symbols dialogue, and not only is the new one less useful but the newline character it enters isn't recognised as such by all programs, or is recognised only when Adaptxt is enabled.

Then there's the human habit factor. On no practice I was excruciatingly slow because I was entering one letter, checking the options, entering the next letter... This has improved with practice, but I still feel slower than with vanilla T9 and the intrusive way it works is still more annoying than helpful.

If it was consistent across all applications, was slightly smoother and all the advertised functions worked correctly and all the time, it may become a valued part of my armoury. But at the moment, no. It still feels like an application that is about saving keystrokes but not time, and which interrupts a mental flow rather than aiding it.

Which brings me to my final point: I'd make bug reports, and offer my humble (stop laughing) suggestions, but it's a commercial application, and there doesn't seem to be any way for me to do so. The best I can do is send an email to customer support and hope that it ends up somewhere useful, and I'm sorry, but I'm just too cynical for that.

So I'm back to trying to be as smooth and fast as possible on the N95's odd looking but surprisingly effective (bit like most Nokia phones, really) keypad.

Did I bear to write the entire post in Adaptxt? No. Well, I wrote it in Adaptxt, but I edited it in standard T9, and wrote this last paragraph on a full size keyboard. Which is still the fastest system available.

No comments:

Search This Blog