Tuesday 21 July 2009

"Predictive text" that actually is.

(This has been updated, because I was an idiot. For some reason, I attributed the software to the wrong company. Ah, the power of Twitter and a company that knows how to use it.)

Some time ago, I got unduly optimistic and went looking for a proper predictive text solution for S60, assuming that somebody had implemented one.

What I found was a couple of commercial betas and vague promises.

But there was one company, Keypoint Technologies, who are Scottish and therefore clever (seriously, have you any idea how many inventions pivotal to the creation of the modern world that Scotland can claim credit for?), that were making a trial of their solution available.

It was called Adaptxt, and I promptly went and tried it, and wrote of my experiences here, just barely over a year ago. It was ... Beta. Very, very beta.

Times move on, however, and programmers keep on working, and Adaptxt became a more mature beta, and it's time to give it another go.

So I am.

How it works, in a nutshell, is that instead of just working out possible options based upon what you've already entered and displaying that much of the first-choice possible word, it shows the rest of that word as well and lets you select it before actually finishing entering it. Which can save you considerable keypresses and, theoretically, time.

In addition, it displays the other possibilities in a pop-up menu, which allows the user to scroll down for a different option. Very nice.

The final piece of prediction is that it can attempt to predict several words in advance, theoretically (there's that word again) allowing you to finish an entire sentence by hitting Select > Select > Select ...

Does it work?

Straight away, my main complaints with the earlier model - not working well with dedit, in which text editor I spend rather a lot of my time - has been resolved. It works just as well in dedit as it does anywhere else. Notice I didn't say "perfectly", but I'll get to that.

There are two major problems left, both of which are hugely annoying.

Problem number one, the easier one to explain, is that it is a truly enormous RAM hog. On my N95, it can use up to about 10MB, which is not only startling for a text entry system, but also highly unpleasant on a phone which has about 20MB free if you're lucky. I opened a support ticket about that, and they assure me that they're working on it.

Problem number two is a little more subtle. You can't chose between the normal case options of Abc (first letter capital, then lower case), ABC (all capitals) and of course abc (all lower case). No, it decides. This is great with "I", and of course it knows about sentence starts, but otherwise ...

Names are generally okay, because they're usually in the dictionary with a capital first letter. But suppose you want to put a capital inside quotation marks, or a regular word is being used as a proper noun, or indeed anything else? How do you force it to enter what you want it to?

There are two ways: enter a new word into the dictionary, or: start a new line, enter the desired word, then go back and join the lines up again.

This is, not to put too fine a point on it, highly annoying.

They claim, however, that they're working on that, as well. I have my fingers crossed.

What else doesn't quite work? Well, for me, the look-ahead prediction, which is claimed to learn from what you enter. This is probably, to be perfectly honest, because I will write anything and everything - twitter updates, this blog post, random fiction, emails - all on this phone, which makes predicting what I will type next difficult for me, let alone a tiny little piece of software.

What's brilliant? The dictionary is, get this, user-editable! Learn from this, Nokia!

Is it faster? Well ... Maybe. I haven't actually done any speed tests, but the time taken to examine the provided options, enter more text, scroll, select... Makes it feel no faster than using full T9. This, however, needs to be considered in light of my rather impressive T9 speed. The real advantage of Adaptxt is the key presses it saves, which may not save time but certainly saves energy, and is a lot more relaxing.

Unfortunately, they have, for a while, removed the menu entry which shows you how many key presses it's saved you, how many first-choice options you use, how many second-choice... etc. Which is not useful, but is very cool to know.

Will I continue to use it? Yes. It feels very relaxing, which is an odd reason to have, but it works for me. I will, however, be extremely disappointed if they don't release a new beta, and soon, with some significant updates in memory management and grammar options.

Would I pay for it when it comes out of beta?

That, I'm afraid, will remain to seen.

No comments:

Search This Blog