Fake news about MT
Thread poster: Phil Hand
Phil Hand
Phil Hand  Identity Verified
China
Local time: 05:07
Chinese to English
May 22, 2019

I hope this isn't too negative a topic, but I thought that as a bunch of human translators it might be nice to collate some of the fake news that I see pumped out continuously about the marvels of machine translation. I'm not trying to be critical of MT or its use. But I honestly think that much of the publicity surrounding MT is borderline deceitful. And I think it will help us to understand the ways in which they are attempting to fool us, so that we can better educate customers.
... See more
I hope this isn't too negative a topic, but I thought that as a bunch of human translators it might be nice to collate some of the fake news that I see pumped out continuously about the marvels of machine translation. I'm not trying to be critical of MT or its use. But I honestly think that much of the publicity surrounding MT is borderline deceitful. And I think it will help us to understand the ways in which they are attempting to fool us, so that we can better educate customers.

This is one I happened to run across:
https://www.engadget.com/2018/12/03/huawei-storysign-childrens-books-app-for-deaf/
Huawei app uses AI to help deaf children read
StorySign turns words on the page into sign language.

Sounds like automatic translation, doesn't it? Clever tech company has solved the English-to-sign translation problem!

Except...no.

Each language currently has just one book -- for English audiences, that's Eric Hill's Where's Spot. It could take a long while before you have a collection of stories to read at bedtime.

What we have here is not automatic translation, but an audiobook for the Deaf community. Each book has to be separately recorded. The AI in the headline? Is actually used in the camera, helping the phone to accurately recognise what page of the physical book the child is on. It does no linguistic processing at all.

So was a lie told? Maybe not. But the headline was incredibly misleading - and deliberately so, I think.

That's the kind of issue I wanted to collect together. Unrepresentative headlines. Misleading statistics (like "Microsoft translator is 95% accurate"). Overblown predictions. If you spot one, please add it to the thread!

[Edited at 2019-05-22 06:39 GMT]
Collapse


neilmac
Arabic & More
Dmytro Nehrii
Michele Fauble
 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 22:07
Member (2006)
English to Afrikaans
+ ...
Fake news vs misunderstanding May 22, 2019

Phil Hand wrote:
I honestly think that much of the publicity surrounding MT is borderline deceitful.


This article is not about MT, though. It uses the word "translate" but it does so in an ambiguous way.

If I were to put two translations of a children's book side by side (say, one in English, one in Chinese), and read them out loud (to a child who understands only Chinese) while pointing with my left hand at the English word and with my right hand to the Chinese "word" while I read, would you call that "translating"? Am I "translating" the English book into Chinese while I read?

Or, if I were to pre-record a spoken translation of a slide show, and then play the recording while presenting the slide show, am I using the recording to "translate" the slide show? From a users' perspective I suppose it could be said that I'm "translating" the slide show using the recording.

* ARTICLE: Point your phone at certain children's books and the app will use AI to translate individual words on the page to sign language performed by an avatar.
* Closer to REALITY: Point your phone at certain children's books and the app will use AI to highlight individual words on the page while the avatar performs sign language.

The app detects on which page you are, and then shows a pre-recorded animation of the signed translation of the words on the page.

I'm actually happy that they don't use MT for this, because if you compare the videos of this app in different languages, you'll see that it doesn't follow the word order of the written text. For example, "Good boy, Spot" first highlights "Spot" and then highlights "Good", without ever highlighting "boy", because that is the word order in the English signed language that the app uses (I'm not sure which English signed language they use, though... UK, US, what?).


[Edited at 2019-05-22 07:02 GMT]


 
Michael Wetzel
Michael Wetzel  Identity Verified
Germany
Local time: 22:07
German to English
An example shared here recently May 22, 2019

Here's a discussion of an article on MT that, in my opinion, is either based on some very broad and misleading journalistic license or, less generously, makes a claim that is simply not true:
... See more
Here's a discussion of an article on MT that, in my opinion, is either based on some very broad and misleading journalistic license or, less generously, makes a claim that is simply not true:

https://www.proz.com/forum/translation_theory_and_practice/333146-"future_translators_may_be_closer_to_editors_and_quality_control_experts".html

However, when discussing MT, I think it is also important to point out the elephant in the room: A lot of human translations are also "fake translations" in the sense that their creators have no idea what they're talking about and are basically just exchanging words in one language for more or less plausible sounding alternatives in another language. A computer can do a large portion of that task better than a human, but it will lag behind in some areas related to pattern recognition.

This passage from a German law came up in a discussion here:
"Personen, die wegen ihrer wirtschaftlichen Unselbstständigkeit als arbeitnehmerähnliche Personen anzusehen sind; zu diesen gehören auch die in Heimarbeit Beschäftigten und die ihnen Gleichgestellten."

Here is the English translation published online and in print by the German Ministry of Justice:
"persons of similar status on account of their dependent economic status, including those engaged in home work and those equal in law to home workers."

"equal in law to" is almost certainly Globish, but that's not what I'm worried about: The intended meaning is clear and it is not particularly distracting. The Gibberish is what gets me: The legal terms "arbeitnehmerähnliche Personen" and "Heimarbeit" have been translated as though they were just any old words.

"Heimarbeit" refers to a specific German law. What does "those engaged in home work and those equal in law to home workers" even mean? People who work at home? I guess so, but that's wrong.

And "dependent economic status" seems like an incredibly vague and unhelpful way to refer to the very specific concept of people incorrectly classified as independent contractors even though their work is actually based on an employer-employee relationship.

This is a fake translation: It looks more or less like it says something in English, and anyone who can't read German has good reason to assume it says the same thing as the German law (after all, it was published by the Ministry of Justice). However, anyone who actually tries to use this translation to do something, such as have an informed conversation about anti-discrimination regulations in Germany, will end up completely lost.

And we all mess up. We're human, after all. I mess up, I'm human. Still, there is a gigantic difference between the occasional, inevitable slip or misinterpretation and a systematically flawed HT or MT process.
Collapse


Christine Andersen
Michele Fauble
Kuochoe Nikoi-Kotei
Philip Lees
 
Emma Page
Emma Page
United Kingdom
Local time: 21:07
French to English
+ ...
Fake and bad are different.... May 22, 2019

Michael Wetzel wrote:
when discussing MT, I think it is also important to point out the elephant in the room: A lot of human translations are also "fake translations" in the sense that their creators have no idea what they're talking about and are basically just exchanging words in one language for more or less plausible sounding alternatives in another language. A computer can do a large portion of that task better than a human, but it will lag behind in some areas related to pattern recognition.


I think the key is that "fake" and "bad" aren't the same thing. When we talk about "fake" MT, the deception is that the machine didn't "do" the translation on its own, so someone who is uninformed about the actual abilities of AI translators comes away with a false impression of what the MT can do.

I would call your example a "bad" translation, because as you point out, it isn't fit for purpose (even if it was published and touted as such).

I have done projects, as I'm sure many others have, where I was tasked with translating large volumes of short phrases which would then be put into an "AI" translator, inevitably destined to give someone the impression that the computer could "translate" (rather than just go through a large database of pre-existing translations, which is what it was doing). In this case I provided "good", "real" translations...which then became "fake" translations in the sense that they were to be re-used without explicit acknowledgment of that fact to the end user.

I guess it all comes down to why we care. Are we annoyed that people are getting sub-par translations from machines and calling it good enough because they can't tell the difference? Or are we annoyed at the constant death-sentences being leveled at our trade by people who think the computers are a lot better than they are? Probably a bit of both...


 
Phil Hand
Phil Hand  Identity Verified
China
Local time: 05:07
Chinese to English
TOPIC STARTER
The Microsoft classic May 22, 2019

Those are salutary points, Michael and Emma. I'm going to hold the line and say: it really doesn't matter what we're annoyed by, or how many hairs can be split, a lot of bull has been talked about machine translation. The example I gave in the OP is quite clear, I think: the press release was deeply misleading, if not technically untrue. That's what I'm trying to collect here. This MS classic is another example, and it's quite subtle.

Microsoft reaches a historic milestone, using AI to match human performance in translating news from Chinese to English

https://blogs.microsoft.com/ai/chinese-to-english-translator-milestone/

It took me a while to work out what was going on here. I eventually had to download all the texts to work it out. The answer is that the vast majority of the "translations" done by the machine in this experiment were in fact back-translations. It handled back translations of Chinese translated news OK. The translations from original Chinese news were pretty much incomprehensible, as expected. (There are some other oddities in the data, but that's the one that leapt out at me when I first opened up the datasets.)

So again, I don't think MS "lied" as such. But any normal person reading that headline would assume that as of today, you can simply point the Microsoft translation engine at a Chinese newspaper, and get a perfectly readable English translation. That is not the case by a long chalk.


Michele Fauble
neilmac
Christine Andersen
Emma Page
 
Michael Wetzel
Michael Wetzel  Identity Verified
Germany
Local time: 22:07
German to English
My personal favorite (IBM was at it long before Google) May 23, 2019

"New York, January 7[, 1954] ..... Russian was translated into English by an electronic "brain" today for the first time."

https://www.ibm.com/ibm/history/exhibits/701/701_translator.html

Emma: I still think it is legitimate to refer to a translation - or, more to the point, a translation process - that consistently makes the kinds of errors that
... See more
"New York, January 7[, 1954] ..... Russian was translated into English by an electronic "brain" today for the first time."

https://www.ibm.com/ibm/history/exhibits/701/701_translator.html

Emma: I still think it is legitimate to refer to a translation - or, more to the point, a translation process - that consistently makes the kinds of errors that I talked about in my example as "fake." Like the "translations" of GoopL (or DoopLe, if you like), it looks more or less like someone or something took a bunch of words in one language and then reproduced them in another language so that people could use them to do things, but this first impression proves entirely misleading upon closer examination.

Someone paid and someone was paid to produce a text that would help people who cannot read German understand the legal situation in Germany. If the kinds of errors I pointed out appear throughout the translation, then I think it really is a "fake" translation: It looks like a translation, but it isn't. It's a car without an engine, and it seems misleading to refer to this as nothing more than "[not] fit for purpose" or "sub-par". If the errors I pointed out are exceptions, then I agree that it makes sense to distinguish between "bad" and "fake," even if those exceptions occur frequently.

What I'm saying is that if someone doesn't know or doesn't care about the most basic principles of translation, like the fact that specialized terminology exists, or if someone works on the basis of a process that prohibits them from implementing those basic principles, then they are doing the same thing as MT.

That said, I do understand your distinction between the fraudulent (fake) claim that AI is translating and the fact that it is really just regurgitating, and I don't disagree. This point is also more directly relevant to the topic of this discussion.
Collapse


Emma Page
 
Phil Hand
Phil Hand  Identity Verified
China
Local time: 05:07
Chinese to English
TOPIC STARTER
Blast from the past! May 23, 2019

That's beautiful, thank you, Michael. I can't resist giving a few more quotes. Bear in mind, this was in 1954!

A girl who didn't understand a word of the language of the Soviets punched out the Russian messages on IBM cards...A handful of men had been individually engaged in research at various institutions for almost a decade to make a machine convert the meaning of words clearly from one language to another.


...the most versatile electronic "brain" extant, the IBM 701.

This amazing instrument was interrupted in its 16-hour-a-day schedule of solving problems in nuclear physics, rocket trajectories, weather forecasting and other mathematical wizardry.

Doctor Dostert predicted that "five, perhaps three years hence, interlingual meaning conversion by electronic process in important functional areas of several languages may well be an accomplished fact."
Ah, I remember it well, that 1959 dawn of electronic interlingual meaning conversion!
Concretely, if electronic language translation makes possible, in due course, the translation into the languages of the less developed areas of the world, the basic references and scientific literature in existence in Western languages, this in itself would be significant.
Such altruism, helping the poor residents of the less developed areas.

It's all a bit terrifying. Presumably we will sound just as pompous to people 70 years from now. I wonder how?


 
Emma Page
Emma Page
United Kingdom
Local time: 21:07
French to English
+ ...
Fair point! May 23, 2019

Michael Wetzel wrote:

Someone paid and someone was paid to produce a text that would help people who cannot read German understand the legal situation in Germany. If the kinds of errors I pointed out appear throughout the translation, then I think it really is a "fake" translation: It looks like a translation, but it isn't. It's a car without an engine, and it seems misleading to refer to this as nothing more than "[not] fit for purpose" or "sub-par". If the errors I pointed out are exceptions, then I agree that it makes sense to distinguish between "bad" and "fake," even if those exceptions occur frequently.


Hmm, interesting point. I see what you mean, in the sense that the translator in this case presumably was aware that their translation wasn't equivalent to the source text (made no effort to properly translate the specialist terms), and passed it off as acceptable...much like when people stick something through google translate and go, "look, it's in Japanese now!" etc.

In both your scenario and the MT scenario, the problem (in my view) is the implicit devaluing of the actual work required to produce a good translation. Since laypeople generally have NO idea what translation involves or requires, it's somewhere between insulting and dangerous to our livelihood when folks go around giving the impression that machines can do the work instantly when they really can't, or that someone can translate a legal phrase with no specialist background and no research.

All very interesting to think about, regardless.


 


To report site rules violations or get help, contact a site moderator:


You can also contact site staff by submitting a support request »

Fake news about MT






TM-Town
Manage your TMs and Terms ... and boost your translation business

Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.

More info »
Protemos translation business management system
Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!

The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.

More info »