Babes & sucklings
The joys of young children and infant technologies.
A few years ago, about the time I won the AIIC Most Promising Newcomer Award, one of my children took a break from her differential calculus homework to fix me with a beady stare. "What do you do?" she asked.
"I'm a linguist, I speak languages", I said.
"But what do you really do?"
So I explained about my busy and terrifically important schedule of meetings and conferences, how we kept the wheels of business turning and how only the week before I had been instrumental in getting seaweed supplies through to a stricken Welsh village.
She fixed me with the gamma rays: "Is that all?"
Well, it's a situation we've all faced. People often seem surprised that machines haven't replaced us. Surely, they argue, with powerful computers these days they should be able to deal with something as simple as translating.
SOMETHING AS SIMPLE!! Such language probably constitutes lese-majeste and heresy all rolled into one. But it gives us pause. Could a machine do what we do?
Recently I splashed out on one of those voice-recognition packages that my computer shop assured me was the absolute kippers knickers. The expert told me to do the start-up sessions properly so the machine could be trained to my voice. I spent an evening in my what we laughingly refer to as the study (junk room with father attached, might be more accurate) saying things like "the broom swept the room as the gun went boom". You get the picture, lots of phonemic minimal pairs. "The gnat hit the cat with a bat". "At the meeting I gave him a fleeting greeting". Maybe that last sentence was from an AIIC report. Whatever. I spent hours at the thing. One of the children came in to ask me if I wanted a cup of tea. "If me drink tea, want to pee", I warbled, now really into the swing of voice recognition. Said child left, muttering about the more worrying manifestations of the ageing process. And there was a time when progenitors commanded respect. Or is it progenitors who are the thief of time? Your call.
Eventually the computer told me in one of those smug messages that we had completed phase one. How many phases were there? I now had to start using the thing in real situations. That is what the on-screen prompt said. Not a pretend or bogus situation. A real one. Luckily at my house real situations are there for the asking.
The machine told me to use a normal speaking voice, not to over-articulate.
"The quick brown fox jumps over the lazy dog"
The screen said
"fhg qiek barn ox hulp er zy rag"
I scratched my head. I gave the computer one of my quizzical looks. Famous they are, those looks. I'll do it for you next time we work together.
The instructions said that the system learns as it goes, so all I had to go was correct its errors and it would get it right next time. I tried ordering the cursor around. Well, I don't know what cursors were like in the old days, but mine had a mind of its own.
"Top of page," I barked.
The cursor blinked at me lazily. It may have winked.
I think I heard it mutter, "You must be joking".
This technology involves donning a very fetching microphone arrangement as worn by pop stars. I was still wearing the microphone and giving vent to my general feeling of mild discombobulation. Suddenly on the screen I see:
"B*%$%Y H)*___ _____ ___
Perhaps I've got the army version. Good on your basic Anglo-Saxon. Finally I admitted defeat and removed the program from my computer. This was not because it was intrinsically bad but because I can already sort of touch-type and did not have the patience to train the new system. Now that "sort of touch-type" might have you wondering, because you are shrewd people who will say, well either you can or you can't. Fair play. I can manage the letters, but need to look for number and speech marks. And a French keyboard is a strangely surreal experience, but that is another story.
All this set me thinking about how hard it is to capture the infinite variation and intricacy of human speech. Even mine, and everyone knows I speak in short, pithy sentences. Were a machine to replace us, it would have to deal with the infinite complexity of just one language. If we just consider the variations of my benighted homeland then the system would need to deal with any number of accents: Yorkshire, Scottish, Welsh, Irish (N & S), Liverpool (aka Scouse) and Newcastle (Geordie). accents. And that's just one small island that some of you can perhaps point to on a map. Then you have to consider all the North American speech patterns. We move smartly on to Australia and NZ, South Africa and the Sub Continent. Next consider people who use English as a second or third language with varying degrees of accuracy and opacity. Is it not asking a lot of a machine, which is at least distantly related to my voice program that struggles with the minimal pairs in its own set up routine? Consider German, which approaches us at various heights (think about it). Or the considerable regional variations of Spanish. Would it ever work for natural, flowing speech?
I can see you are all glazing over, but I enjoyed the rant while it lasted.
Machine translation has been under discussion since the 1970s and although no expert (you may have noticed) I would imagine that the technology has progressed. However the people selling machines have done their products no favour by plugging them as complete replacements for humans. I recently attended a conference and exhibition on innovation - an area with a high translation load - and saw an advert for translation software: "It eats translators for breakfast". Now be honest, do you warm to the product?
Of course what would be useful is a machine that could translate teenagers. Grunts into limpid speech. Put me down for one.
Articles published in this section reflect the views of the author(s) and should not be taken to represent the official position of AIIC.