Charlie's Diary

[ Site Index] [ Feedback ]


Sat, 27 Aug 2005

Why input devices suck

I'm behind schedule on updating this blog because I've been away from my desk for two of the past four weeks, I'm behind schedule on a book, and I'm tired from too much traveling. I'm also behind schedule because I'm recovering from an attack of RSI (repetitive strain injury). I get stabbing pains in the backs of my hands every couple of years, and it's invariably -- in my case -- the result of using a lousy chair/desk combination. RSI is usually triggered by postural whoopsies of one kind or another, and I'm lucky enough to have discovered that mine can invariably be cured within a week by switching to a better chair.

In the case of the current attack, it started after my wife's office chair died of old age; I gave her my old Aeron, switched to a cheap camp chair while the order for the new one went through -- after all, this was about two days before I had to fly off to Austin -- and by the end of the first day my arms were feeling funny. Fast-forward across a trans-Atlantic flight to my arrival at a hotel, where the office chair turned out to be impossible to adjust to the correct typing height for the hotel room desk, and the provided ethernet cable was about a metre too short for typing on my lap, and crossing the road to the nearest mall involved a ten minute drive, and you've got a recipe for disaster.

(Luckily, I knew what not to do, and allowed myself to get further behind schedule rather than damaging my wrists -- a two-week hiatus I can deal with, two months with my arms in splints is another matter. The new second-hand Aeron chair is due to arrive on Tuesday and normal deadline-meeting activities will be resumed immediately afterwards. There is no cause for alarm.)

This got me thinking about input methods -- trivially, ways of shoveling data into computers -- and about why they all suck.

The oldest and most familiar input method still in use is the QWERTY keyboard layout, from 1876 or thereabouts. There are a number of explanations for its persistence and prevalence. Personally, I think the fact that you can spell the word TYPEWRITER using only keys from the QWERTY row is a bit of a giveaway, especially if you think in terms of underpaid traveling salesmen with no time to learn how to type properly. Another giveaway: despite the semi-random looking layout, English word frequency is such that the letters you type don't cluster at one side of the keyboard but alternate from one side to the other, reducing the risk of two typekey arms colliding (if you ever used a manual typewriter, where the keys are connected to type characters on the end of a hinged arm, this was a major annoyance and interruption).

QWERTY is, all in all, a fairly reasonable compromise. Despite the advocates of alternative layouts such as DVORAK, there's not a lot of evidence that we can speed up text input and accuracy by switching to a new two-handed layout. The speed record for QWERTY is around 130 words per minute which is slower than continuous speech but fast enough for most purposes -- if you could come up with the ideas fast enough and your fingers didn't catch fire you could write an entire 150-page novel in eight hours! (Take that, Lionel Fanthorpe!)

What's wrong with QWERTY is more subtle. It was designed for a desk-mounted machine that would be operated two-handed by a typist. Old manual typewriters involved lots of arm-waving and muscle contractions to run the carriage back over, insert paper, remove paper, and so on. It was all good exercise, back in the day. Modern computer keyboards don't. So we sit for long periods in a fixed position while our fingers clatter like crazy and we seize up like rusty pieces of machinery. Worse: because QWERTY takes a while to learn, it has achieved institutional inertia -- it blocks out alternative input methods. And because it's designed for two-handed typing it makes using a mouse, or a PDA, kind of a drag. What are we supposed to do, grow another arm?

Now we get into the horrible alternatives the computer industry has tried to inflict on us ...

First and worst are all the virtual QWERTY layouts. These are what you get when a programmer with no idea about ergonomics and a short deadline tries to come up with a way to let punters get data into a computing device without a physical QWERTY keyboard. You get a picture of a QWERTY keyboard on-screen to peck at with the mouse or a stylus. In extreme cases you get a little laser doohickey the size of a cigarette packet that projects a picture of a QWERTY keyboard onto whatever's in front of it -- a tray table, a sleeping cat, your neighbour's lap -- and monitors where your fingers block the light.

One-fingered QWERTY typing is no fun; because of the word-frequency element of the design (what I was talking about earlier, right?) you have to keep jumping from one side of the keyboard to the other, which really slows you up. I haven't done any precise timing tests (life's too short, especially for a rant) but I figure the slowdown is roughly 90%, if not more.

The thumb-board isn't much better. I've got one of those on my phone; itty-bitty buttons in a QWERTY layout designed so that you cup the phone in both hands and alternate between thumbs. Two-thumbed typing is a lot faster than one-fingered typing but it still tops off at around 20 words per minute. I don't know about you but as I speak at around 200 words per minute, I don't call that particularly useful. Also, you tend to sprain your thumb sooner rather than later, which is a bit of an embarrassing drawback.

Far better to look beyond QWERTY and consider alternative input methods, right?

Let's start with handwriting recognition. I'm left handed and my handwriting resembles the perambulations of a drunken spider who's just blundered through the inkwell. After signing my name fifty times in a row I can write "Cthulhu P'thagn" on the title page of a book and nobody will be any the wiser. Thanks to years of typewritten debauchery I get writers' cramp holding a pen for long enough to write an address on a mailing label: the fine motor control isn't there any more so I unconsciously grip the pen harder and harder until my arm cramps up. Even back when I was condemned to scrawl answers to exam questions every couple of months, I had trouble: after one particularly harrowing set of finals I had a crescent-shaped depression with associated bruising in the tip of my left index finger that didn't fill in for a week.

Computers are not good at recognizing my handwriting. Handwriting recognition works best on neat copperplate. Neat copperplate is incompatible with speedy text entry. The fastest I can go by hand is about twenty words per minute -- timed during a desperate mock exam many years ago -- and to be readable I'm down to maybe ten words a minute, enough to make one-finger QWERTY appealing.

About twelve years ago, Apple was busy failing to sell the Newton on the basis of its' vaunted handwriting recognition. To be blunt, the original Newt was too feeble and underpowered to do the job in real time for anyone who could write faster than me. A couple of guys with a software company called Palm figured out that if they just taught the owners a rudimentary shorthand that the computer was better at reading, they could speed things up by an order of magnitude. What they came up with (with some input from Xerox's labs -- those guys invented everything) was a piece of software called Graffiti. Graffiti is written in a letter-sized box, one character at a time (not joined-up), with simplified strokes that even a dumb Motorolla 68000 can recognize. It became the must-have piece of software for Newton owners, which just goes to show how dumb Newton owners are. They went on to sell a tiny tablet-shaped notepad called the Pilot that was basically a gadget that used graffiti for taking names, addresses, and notes, and the rest is history.

What's wrong with Graffiti is that it's still handwriting recognition, of a sort. It's improved, more accurate handwriting recognition, but it relies on the use of a pen. PDAs were mis-sold in the early 90s as a tool for thrusting MBA-wielding executives who (back then) weren't expected to know how to type. High-end ones even had -- and still have -- built-in voice recorders. These people don't need high-speed data entry. They need a cute secretary back in the outer office taking dictation, and an extra round on the golf course. They were not a large enough market to count. And they've permanently warped handheld computing in a very counterproductive direction indeed. (Spot the disgruntled former Psion user, okay?)

Speaking of dictation, I haven't whacked on speech recognition yet. Speech recognition has been the great white hope of dumb engineers who want to put computing power into the hands of the unwashed and functionally illiterate masses for, oh, about thirty years now. It doesn't work because while the English language (and others) are phoneme based, our written representations are ambiguous and often make no sense. Quick: say "Cholomondley-Featherstonehaugh" to your computer. (It's pronounced "CHUM-lee FAN-shaw.") Speech recognition software gets around this problem by using dictionaries to map phoneme sequences to words. If you want to see how brain-dead this is, fire up your copy of ViaVoice or DragonDictate and see how it recognizes "Charlie Stross". I don't know about you, but the usual results are guaranteed to stress me out.

I've experimented with speech recognition. For starters, my BBC English accent doesn't work very well with most speech rec systems, which are trained to expect English with a mid-Western drawl. Excuse me, but I do not expect to have to pretend to be Texan just so my computer can take dictation! But with a few hours of training I can reach the dizzy heights of speech recognition accuracy -- say, 90-95% accurate -- and then the fun begins.

When you make a mistake with QWERTY input, you are most likely going to end up with two transposed letters, a missing letter, or an incorrect letter. Spelling checkers are pretty good at spotting this kind of blooper and putting a wiggly red line under it.

In contrast, when you make a mistake with speech recognition the most likely error state is that the recognition software will insert a correctly spelled homophone in your text: "their" instead of "there", "bear" instead of "bare", "narcotic" instead of "narcissist", and so on. This is a complete pain in the arse to correct because at a stroke it renders spelling checkers impotent. Worse than the ludicrous errors ("narcotic"/"narcissist") are the plausible ones -- if it mis-hears "lovely" as "lonely" you may end up with one adjective replacing another, in which case a grammar checker or even a mark one sub-editor's eyeball may not be up to spotting the error even though it changes the meaning of the text drastically.

Finally, I'd like to add that I write science fiction. That's made-up stuff. Including some of the words. Long experience shows me that in a 150,000 word novel I probably coin some 500 neologisms (plus another 500 plurals and tenses based off them) that aren't in Microsoft Word's dictionary. I have a functional vocabulary that exceeds any speech recognition system's competence. This drastically increases the rate of homophone errors to the point where, when dictating to a computer, I am forced to back up and fix every error immediately before I lose track of what I'm saying, because on average one in every two short sentences comes out with an error in it that completely changes its meaning.

Overall text input speed using speech recognition: two words per minute. Backwards.

Finally, we come to an entirely different input method. It's pen based but it's not cursive real-time handwriting recognition, or shorthand. It bears about the same relationship to QWERTY that Graffiti bears to handwriting recognition: it's the alternative single-finger keyboard layout. There are several of them. The idea is, you use a virtual keyboard on a PDA or tablet PC that is optimized to minimize pen or finger movement, so that the commonest characters are grouped at the center of the keyboard where they're easy to find.

FITALY is what I use on my PDA. It's a pain in the neck to learn -- takes an hour to get minimally proficient at hunt and peck, then several hours more to acquire some speed -- but once it sinks in it's much faster than single-finger typing. The folks who sell it claim that it maxes out around 70 words per minute, which I think is an exaggeration, but I'd certainly believe 30-40wpm is achievable. Subjectively it's faster than Graffiti and much more accurate.

An alternative is MessageEase. Rather than being an optimized layout, MessageEase is a stroke-based system -- you get a grid of nine squares, the commonest characters are entered by tapping in a square, and the rest are accessed by scribing a stroke from one square to another. It's a bit more confusing than FITALY and takes a little longer to learn but is similarly fast -- and supports more devices. The drawback is, it uses the same fine motor control as handwriting; less of it, so your hand doesn't cramp up as fast, but it's still a pain. FITALY, in contrast, is amenable to hunt'n'peck; it'll eventually get you when you've been holding the pen wrong for three hours at a stretch, but it's not intrinsically the work of the devil.

Either of MessageEase or Fitaly is probably fast enough for creative writing, avoids the problems of one-fingered QWERTY, doesn't insert new and curious grammatical errors on the fly, and works in places where a full-sized keyboard won't go. They suffer from the learning curve (I estimate 10-20 hours until they begin to sink into your fingertips and become instinctive). They suffer from the size of PDA screens; even my iPaq hx4700, which is huge by PDA standards, is much smaller than a reporter's pad, so you have a choice between seeing maybe twenty words on-screen at any time and having to do the fine motor control thing again. Or you could buy a Tablet PC with Windows for Tablets on it and add the PC versions of these input methods (if you're willing to run Windows at all -- and I'm not).

Damn it, is it possible that the entire behemoth of the IT industry is unable to produce what I need -- namely a gadget the size of a reporter's pad with a battery life of at least four hours that lets me input text as conveniently as that pad, that doesn't force me to re-learn longhand, and that lets me edit on the fly? Unfortunately it appears that the answer is "yes". We're trapped in here with no way out until the industry learns to pay attention to ergonomics. And I'm still waiting for my new chair to arrive ...

PS: That's 2600 words in 90 minutes and my left hand is still mostly okay. Right. Back to work ...

[Discuss Writing (2)]



posted at: 14:26 | path: /writing | permanent link to this entry

specials:

Is SF About to Go Blind? -- Popular Science article by Greg Mone
Unwirer -- an experiment in weblog mediated collaborative fiction
Inside the MIT Media Lab -- what it's like to spend a a day wandering around the Media Lab
"Nothing like this will be built again" -- inside a nuclear reactor complex


Quick links:

RSS Feed (Moved!)

Who am I?

Contact me


Buy my books: (FAQ)

Missile Gap
Via Subterranean Press (US HC -- due Jan, 2007)

The Jennifer Morgue
Via Golden Gryphon (US HC -- due Nov, 2006)

Glasshouse
Via Amazon.com (US HC -- due June 30, 2006)

The Clan Corporate
Via Amazon.com (US HC -- out now)

Accelerando
Via Amazon.com (US HC)
Via Amazon.com (US PB -- due June 27, 2006)
Via Amazon.co.uk (UK HC)
Via Amazon.co.uk (UK PB)
Free download

The Hidden Family
Via Amazon.com (US HC)
Via Amazon.com (US PB)

The Family Trade
Via Amazon.com (US HC)
Via Amazon.com (US PB)

Iron Sunrise
Via Amazon.com (US HC)
Via Amazon.com (US PB)
Via Amazon.co.uk (UK HC)
Via Amazon.co.uk (UK PB)

The Atrocity Archives
Via Amazon.com (Trade PB)
Via Amazon.co.uk (Trade PB)
Via Golden Gryphon (HC)
Via Amazon.com (HC)
Via Amazon.co.uk (HC)

Singularity Sky
Via Amazon.com (US HC)
Via Amazon.com (US PB)
Via Amazon.com (US ebook)
Via Amazon.co.uk (UK HC)
Via Amazon.co.uk (UK PB)

Toast
Via Amazon.com
Via Amazon.co.uk


Some webby stuff I'm reading:


Engadget ]
Gizmodo ]
The Memory Hole ]
Boing!Boing! ]
Futurismic ]
Walter Jon Williams ]
Making Light (TNH) ]
Crooked Timber ]
Junius (Chris Bertram) ]
Baghdad Burning (Riverbend) ]
Bruce Sterling ]
Ian McDonald ]
Amygdala (Gary Farber) ]
Cyborg Democracy ]
Body and Soul (Jeanne d'Arc)  ]
Atrios ]
The Sideshow (Avedon Carol) ]
This Modern World (Tom Tomorrow) ]
Jesus's General ]
Mick Farren ]
Early days of a Better Nation (Ken MacLeod) ]
Respectful of Otters (Rivka) ]
Tangent Online ]
Grouse Today ]
Hacktivismo ]
Terra Nova ]
Whatever (John Scalzi) ]
GNXP ]
Justine Larbalestier ]
Yankee Fog ]
The Law west of Ealing Broadway ]
Cough the Lot ]
The Yorkshire Ranter ]
Newshog ]
Kung Fu Monkey ]
S1ngularity ]
Pagan Prattle ]
Gwyneth Jones ]
Calpundit ]
Lenin's Tomb ]
Progressive Gold ]
Kathryn Cramer ]
Halfway down the Danube ]
Fistful of Euros ]
Orcinus ]
Shrillblog ]
Steve Gilliard ]
Frankenstein Journal (Chris Lawson) ]
The Panda's Thumb ]
Martin Wisse ]
Kuro5hin ]
Advogato ]
Talking Points Memo ]
The Register ]
Cryptome ]
Juan Cole: Informed comment ]
Global Guerillas (John Robb) ]
Shadow of the Hegemon (Demosthenes) ]
Simon Bisson's Journal ]
Max Sawicky's weblog ]
Guy Kewney's mobile campaign ]
Hitherby Dragons ]
Counterspin Central ]
MetaFilter ]
NTKnow ]
Encyclopaedia Astronautica ]
Fafblog ]
BBC News (Scotland) ]
Pravda ]
Meerkat open wire service ]
Warren Ellis ]
Brad DeLong ]
Hullabaloo (Digby) ]
Jeff Vail ]
The Whiskey Bar (Billmon) ]
Groupthink Central (Yuval Rubinstein) ]
Unmedia (Aziz Poonawalla) ]
Rebecca's Pocket (Rebecca Blood) ]


Older stuff:

June 2006
May 2006
April 2006
March 2006
February 2006
January 2006
December 2005
November 2005
October 2005
September 2005
August 2005
July 2005
June 2005
May 2005
April 2005
March 2005
February 2005
January 2005
December 2004
November 2004
October 2004
September 2004
August 2004
July 2004
June 2004
May 2004
April 2004
March 2004
February 2004
January 2004
December 2003
November 2003
October 2003
September 2003
August 2003
July 2003
June 2003
May 2003
April 2003
March 2003
February 2003
January 2003
December 2002
November 2002
October 2002
September 2002
August 2002
July 2002
June 2002
May 2002
April 2002
March 2002
(I screwed the pooch in respect of the blosxom entry datestamps on March 28th, 2002, so everything before then shows up as being from the same time)



[ Site Index] [ Feedback ]


Powered by Blosxom!