Skip to content

I wonder, I wonder

The problem is compounded because many new computer devices are designed to recognize English but they do not understand Icelandic.
‘Not being able to speak Icelandic to voice-activated fridges, interactive robots and similar devices would be yet another lost field,’ Jonsson said.
Icelandic ranks among the weakest and least-supported language in terms of digital technology – along with Irish Gaelic, Latvian, Maltese and Lithuanian – according to a report by the Multilingual Europe Technology Alliance assessing 30 European languages.
Iceland’s Ministry of Education estimates about 1 billion Icelandic krona, or $8.8 million, is needed for seed funding for an open-access database to help tech developers adapt Icelandic as a language option.

This is all done by machine learning. Feed great vast gobs of stuff to the machines until they start to recognise it.

Languages which don’t have vast great gobs of stuff aren’t going to get recognised…..

I’ve heard that we’re seeing the same in machine translation. There needs to be a critical; mass of stuff in the language pair to be able to teach the machines.

28 thoughts on “I wonder, I wonder”

  1. It will be the next decade’s inequality – rich nations can open the fridge from the front room by shouting, but the poor will have to get up and go into the kitchen.

    A problem that will require socialism to solve.

  2. So Much For Subtlety

    Oddly enough I have about three dozen different types of English to choose for my word processor. Not that I believe it. Who claims New Zealand English even exists? But I have a choice of hundreds of other languages as well.

    It is insane to think that fridges won’t come with an Icelandic option. It is so cheap to do once you have done it once.

  3. This can be an advantage too. A closed, untranslateable language means you and your colleagues can discuss your negotiating position out loud in front of your client. There’s also less chance of your trade or military secrets being stolen, since erstwhile thieves don’t know if they’ve got their hands on your security detail or on your cleaning rota.

    The Navajo code talkers came in useful during WWII.

  4. Bloke in Wiltshire

    With all this stuff, it’s also about scale. Translation costs money. If I write an app and think of going international, I know that translating my app into French, German, Japanese Tagalog, Welsh or Icelandic costs the same per language. But French and German get me 100m native speakers and Icelandic gets me 400,000.

    Fuck’s sake. Defending languages? Going the way of Latin? And the problem with that is what? We can’t read Pliny in the original? How does that affect the price of fish?

  5. So Much For Subtlety

    Bloke in Wiltshire – “But French and German get me 100m native speakers and Icelandic gets me 400,000.”

    So it would cost about $22 per Icelander to develop such a database. Which, presumably, would work for a whole range of devices. You might not bother at first. But once you have sold to the 100 m French and German speakers, you might be a little interested in the 400,000 Icelandic speakers.

    This is a non-issue. In the future every dialect and even accent will be catered to. Software of this sort is not that expensive but even if it was, spread out over many devices the cost becomes minimal.

  6. Who claims New Zealand English even exists?

    Anyone who has watched the Super XVs and listened to the commentary. I still can’t differentiate between “position” and “possession” as a Kiwi commentator says it.

  7. Tim N, you’re on to something there. It’s all very well to celebrate the richness and variety of the language, but the problem with the claim that there is no correct way to speak it, is that a Weegie or a Kiwi or a Bostonian can understand me but quite often I cannot understand them, nor they one another.

  8. So it would cost about $22 per Icelander to develop such a database. Which, presumably, would work for a whole range of devices.

    That only works if you’re happy with machine translation, c.1990.

  9. Tim N

    I’d prefer to talk to myself rather than talk to my fridge.

    And I don’t want the government knowing how much foie gras I buy nor whether I eat foods deemed unfit for me.

    Sod them all.

  10. I wonder if an intelligent speaking Smeg fridge would feel ashamed if it knew what its name meant in English?

  11. So Much For Subtlety

    bilbaoboy – “And I don’t want the government knowing how much foie gras I buy nor whether I eat foods deemed unfit for me.”

    I would be more worried that the Guardian will get its way and your fridge will report you for using sexist language. How long can it be before someone’s microwave reports them for a domestic with the wife?

    Luckily there is little chance Corbyn will win. But I wouldn’t put it past May either.

  12. ‘There needs to be a critical; mass of stuff in the language pair to be able to teach the machines’

    Teaching machines is hard, teaching monkeys is easy.
    Have the machine only accept one type of input and train the monkey how to pronounce it.

    Icelandics just have to learn that ‘Mjólk’ is pronounced ‘Milk’.

    If you teach the 7 billion people on the planet to speak properly you only have to train the machine once (and those people too lazy/thick to learn will lose/die out).

  13. You aren’t going to be having a fucking conversation with the fridge. It needs to understand at most a few hundred words or phrases. You won’t be dictating your latest novel to it.

  14. The Icelanders do think a lot of their language. They have specifically developed Icelandic neologisms to avoid having too many loan words. With modern machine learning/AI code I would guess it shouldn’t be too hard to develop the necessary algorithm & database. It might even be a labour of love for a language geek group.

  15. Bloke in Wiltshire

    “So it would cost about $22 per Icelander to develop such a database. Which, presumably, would work for a whole range of devices. You might not bother at first. But once you have sold to the 100 m French and German speakers, you might be a little interested in the 400,000 Icelandic speakers.”

    Your numbers are wrong. It’s $220 times as expensive per Icelander.

    I can only really talk about apps and applications, but you can’t do a “database” for that. You might have error messages and prompts specific to an application. But even if you could just drop things in, you still need to test them. That could cost a couple of grand.

    The problem is so much bigger than talking fridges and apps, though. If you’re releasing a movie, is it worth you subtitling it for Iceland? That costs thousands to do. How many blu-rays are you going to sell? So, you don’t bother. So, Icelanders learn English instead. After a while, well, why bother with Icelandic any longer as we all speak English?

  16. I’ve been to Iceland a few times now (lovely place) and even got married there. I didn’t meet one person who didn’t speak English at least well enough to ask directions.

    Bearing in mind that we didn’t just stay in the capital – we went to out of the way places like Akureyri (admittedly the second largest city in Iceland) and on tours to places like Siglufjörður.

    So if they all speak English passably, the people selling the gadgets are going to develop the English, German, French languages, and then look at what they’d actually gain from developing for Icelandic. It’s not even 400,000 customers – it’s 400,000 people who may have a slightly better experience but were potential customers anyway.

    Doesn’t make economic sense.

    (p.s. I like talking to inanimate objects – my wife just got me a Google Home, and I love it!)

  17. Bloke in Wiltshire

    SMFS,

    Sorry – I misunderstood where you got your numbers. But my point stands – it’s not as simple as they think it is (it’s probably more of a job creation exercise).

  18. Probably the best argument for Iceland to join the EU. Then all those millions of pages of law and discussions have to be translated into Icelandic at the German taxpayers expense. Google can then use the EU texts for their machine learning. Simple, if you don’t worry about the fish.

  19. Odd that Microsoft consider 17 million (perhaps) Greeks merit their own version of Windows, but 60 million Brits have US spelling foisted on us. As for Icelandic Windows for less that half a million …

  20. Tim Newman:
    “Anyone who has watched the Super XVs and listened to the commentary. I still can’t differentiate between “position” and “possession” as a Kiwi commentator says it.”

    That’s niver true.

  21. Interestingly, Google Translate seems to do a fairly good job in translating Icelandic. Found this out last week when I was there. It does much better than it does translating Finnish (my mother tongue) even though there are 10-15 times as many Finnish-speakers in the world.

    I guess this is in large part due to Icelandic being a Germanic language developed from Old Norse, just like English (which is just bastardized by too much French). Finnish, on the other hand, is not even an Indo-European language.

    Icelandic is somewhat more recognizable if you can read Swedish and understand spoken modern Norwegian, but still, not really understandable. For instance, the police is “lögreglan”, and when I think of it, yes, “lag” is Swedish for law and “regel” is “rule”, so it’s about “rule of law”, but this is not so easy to see.

  22. Witchie, what? There are US and UK English versions of Windows. Also South African, Australian, and others.

  23. Bloke in Costa Rica

    Nordics are all embarrassingly multilingual. I met a Norwegian girl in Bergen years ago who worked as a translator for Microsoft doing product localisation. But she wasn’t translating English software into Norwegian. She was translating it into Spanish and Italian. She spoke something like seven languages utterly fluently. Plus she was utterly scrumptious.

  24. BiCR, yes, many Nordic people are multilingual, but the fact that you are encountering them makes it more likely that they will speak several languages.

    There are plenty who have poor language skills, but you are unlikely to ever have anything to do with them. That said, they all have 10 years of free, reasonable quality English lessons at school, and subtitled American media is easily available at home

Leave a Reply

Your email address will not be published. Required fields are marked *