Microsoft showed off a cool, forthcoming feature of its Skype software tonight at the inaugural Code conference in California: Real-time audio- and text-based language translation.
In this demo, Skype translates a conversation between two Microsoft employees—in English and German—almost live. It’s easy to see how this could be a compelling feature, especially for business, if it works well.
The feature will roll out gradually across Skype software starting sometime later this year. (Now wait for Google to announce a similar feature for Hangouts and Apple for FaceTime calls.)
Watch the demo here: http://qz.com/214106/watch-skype-translate-a-video-conversation-in-real-time/
See: Quartz
Subscribe to the translation news daily digest here. See more translation news.
Share an article »
ProZ.com blog »
Local time: 05:07
Member (2009)
Italian to German
+ ...
http://www.spiegel.de/netzwelt/web/microsoft-skype-kuenftig-mit-echtzeit-uebersetzung-a-972120.html
Spain
Local time: 05:07
English to Spanish
+ ...
http://www.pcworld.com/article/2198140/skype-demos-will-launch-real-time-translation-app.html
United States
Local time: 23:07
Russian to English
+ ...
Does real mean professional, high quality, accurate, or simply that it exists? I am very skeptical about it, as about all MT.
United Kingdom
Local time: 04:07
English to Italian
+ ...
Does real mean professional, high quality, accurate, or simply that it exists? I am very skeptical about it, as about all MT.
Dear All,
I have been given this news by my boyfriend just yesterday. I personally look at the development of machine translation and interpreting with fear. I do not want to seem too much conservative, but the progress of machines, though beneficial in many ways, has stolen work from people in the past decades, and sometimes enjoyable work. I wish that human beings could consider the drawbacks of modern technologies.
Japan
Local time: 12:07
Japanese to English
+ ...
But having watched the video, I don't buy it at all. I mean I don't speak German, so I don't know how accurate the interpretation is, but I just have to believe that they trained this algorithm specifically for this particular demo.
Plus, they chose English and German, two closely related languages with relatively rigid word order. I'm not sure but it seems like this would make it MUCH easier for an algorithm to correctly parse what... See more
But having watched the video, I don't buy it at all. I mean I don't speak German, so I don't know how accurate the interpretation is, but I just have to believe that they trained this algorithm specifically for this particular demo.
Plus, they chose English and German, two closely related languages with relatively rigid word order. I'm not sure but it seems like this would make it MUCH easier for an algorithm to correctly parse what it hears. I'd like to see it interpret something like Portuguese to Russian. ▲ Collapse
United States
Local time: 23:07
Russian to English
+ ...
interpreters can, sometimes. I am not afraid`that the machines will become any sort of competition--I am scared what they will do to human language, and how damaging the mistakes will be--they may cause serious accidents, malfunction of various instrument, poisoning, etc.
I would never ever want to read any literature translated by MT--what's the point? I'd rather read ads and obituaries.
[Edited at 2014-05-29 15:11 GMT]
Japan
Local time: 12:07
Japanese to English
+ ...
I wouldn't worry too much about machines causing serious accidents. I would worry more about them taking over the world, as that woman on stage in the video so poignantly remarks. I think that's what a lot of people don't really understand - for machine translation to operate on a really universal level, the machine has to pretty much think like a human. And what do you think is going to h... See more
I wouldn't worry too much about machines causing serious accidents. I would worry more about them taking over the world, as that woman on stage in the video so poignantly remarks. I think that's what a lot of people don't really understand - for machine translation to operate on a really universal level, the machine has to pretty much think like a human. And what do you think is going to happen when something with unfathomable calculating power and networked access to all of humanity's critical systems becomes sentient?
If (when) that day comes, we will have much bigger problems on our hands than translators losing a percentage of their earnings. You see, the problem is not that machines will enter into competition with translators, but rather that they will begin to compete with the human race as the dominant entity on this planet.
If you watch the video, take special notice when the Microsoft guy says that they have no idea why the algorithm learns the way it does. Think about it for a while. This should scare you. ▲ Collapse
United States
Local time: 23:07
English to German
+ ...
Hello, how are you?
The machine translates it to: Hallo, wie dir es geht?
Correct would be: Hallo, wie geht es dir?
It shows that the machine is translating words, not phrases.
This can be fixed by adding a TM base of frequently used phrases. That phrase must have not been stored or used before in a conversation, which seems strange to me.
According to one of the presenters, we are to believe that the machine will learn and get better - in a certain way, I believe it, yes; if two German speaking persons had used the phrase correctly first - "Hallo, wie geht es dir?" - the English side of the machine's "brain", having stored it, might have used the correct one for - "Hello, how are you? - because it would have already been used more often than the wrong one - I am sure that can be set that way.
However, none of this means that the machine is thinking like a human - it's simply "computing" according to an algorithm.
Every time it has a conversation, it will add (in this case) English and German phrases. But every time it runs into a completely new and never before used sentence, it cannot know/decide which is the correct word to use when a word has many different meanings.
When the word "umziehen" as in "move" is used here in the sense of moving from one city to another (in order to change one's temporary or permanent residence), it is translated correctly by the machine as "move" into English (but that's not very difficult 'cause "move" will work in many contexts in English); but when the English speaker uses it again in the same overall context (you would have to have a real brain though when you (the machine) listen(s) to it), the machine gets it wrong in German and translates it to "verschieben" which has nothing to do whatsoever with moving to change one's residence. "Verschieben" means foremost "slide", scroll, shunt (trains) and "switch" (but not as in "I'll switch from city A to city B). In the context of "moving to a city," "verschieben" is completely wrong.
In that same sentence, the English speaker says he'll move to London if "everything goes for track" which the machine can't figure out either and simply puts the word "verfolgen" (= track something, follow someone/something) at the end of a grammatically unacceptable sentence:
The English speaker says:
"I am currently planning to move there sometime between the middle of June to the beginning of July if everything goes for track."
The machine translates it and the machine voice says:
"Ja, ich plane derzeit dort irgendwann zwischen Mitte Juni bis Anfang Juli zu verschieben, wenn alles geht, aber verfolgen" which means in English:
"Yes, I am currently planning to do some sliding/shunting there between the middle of June until the beginning of July, if everything goes, but tracking.
The more complicated the sentence gets, the harder it becomes for the machine to translate it correctly.
Example:
English speaker:
Well, Skype is headquartered in London, so I want to be closer to my team. And you're right. The Indian food is fantastic and I've been told about some fabulous restaurants there.
Machine translation:
Auch Skype hat seinen Hauptsitz in London. Also ich will mein Team näher sein und du bist, Recht, Das indische Essen ist fantastisch, Und ich habe einige fabelhafte Restaurants gibt es erzählt - which means in English:
Also Skype has its headquarters in London. Well, I want my team closer is and you are. Right. The Indian food is fantastic. And I have a few fabulous restaurants there are told.
And then:
English speaker:
So what? ... so what brings you to the United States in addition to of course helping me with this demo?
Machine translation:
Also, was die Dinge? Also, was bringt Sie in die USA zusätzlich zu, natürlich hilft mir mit dieser Demo?
Which means:
Well, what the things? Well, what brings you to the United States in addition to, of course helps me with this demo?
Note: In the above sentence, the machine uses a different word for "you" - "Sie." That's very formal and wasn't how the machine translated it at the beginning of the conversation when it used "dir" - this is tricky for the machine; it cannot decide which form of "you" it should use, meaning it has no idea or is not programmed to recognize/decide to have a formal or informal conversation. That matters in German where when you address a person, you must decide if you use the formal or informal version of "you". In any case, it is not recommended to switch from informal (friendly) to formal (or the other way which is worse) in the middle of a conversation.
While the English speaker (obviously not from the US or UK) speaks rather fast, the German female speaker speaks very slowly, clearly and enunciates her words. She also uses relatively short sentences and easy sentence structure. (Just don't look in her eyes when you're skyping - no offense - she's probably a little nervous and means very well. Nice teeth by the way.
Transfer learning? Maybe, but I am sure it's more like recording words and phrases and then reusing the most frequently used ones. Speech synthesis? Not in the human way. It doesn't think. It computes. Deep neural net? No.
This research was started 15 years ago according to the presenter.
Speech recognition: works very well. Translation: Not so good, only easy text can be related fairly well.
Can it help people have simple conversations? To a certain degree, yes. Can it work as a serious interpreting or translation device or replace interpreters or translators? No!
[Edited at 2014-05-30 20:15 GMT] ▲ Collapse
United States
Local time: 23:07
Russian to English
+ ...
Sure they can. However the quality may be piss-poor in the vast majority of cases.
I wouldn't worry too much about machines causing serious accidents. I would worry more about them taking over the world, as that woman on stage in the video so poignantly remarks. I think that's what a lot of people don't really understand - for machine translation to operate on a really universal level, the machine has to pretty much think like a human. And what do you think is going to happen when something with unfathomable calculating power and networked access to all of humanity's critical systems becomes sentient?
If (when) that day comes, we will have much bigger problems on our hands than translators losing a percentage of their earnings. You see, the problem is not that machines will enter into competition with translators, but rather that they will begin to compete with the human race as the dominant entity on this planet.
If you watch the video, take special notice when the Microsoft guy says that they have no idea why the algorithm learns the way it does. Think about it for a while. This should scare you.
It will NEVER happen--that they will start thinking. They are just dummies which can count. (the computers, or robots) As I said in another thread before--MT may have some use for translating set phrases, or idioms--not as a part of larger pieces, though, but just as separate items.
[Edited at 2014-05-31 09:43 GMT]
Japan
Local time: 12:07
Japanese to English
+ ...
Sure they can. However the quality may be piss-poor in the vast majority of cases.
I wouldn't worry too much about machines causing serious accidents. I would worry more about them taking over the world, as that woman on stage in the video so poignantly remarks. I think that's what a lot of people don't really understand - for machine translation to operate on a really universal level, the machine has to pretty much think like a human. And what do you think is going to happen when something with unfathomable calculating power and networked access to all of humanity's critical systems becomes sentient?
If (when) that day comes, we will have much bigger problems on our hands than translators losing a percentage of their earnings. You see, the problem is not that machines will enter into competition with translators, but rather that they will begin to compete with the human race as the dominant entity on this planet.
If you watch the video, take special notice when the Microsoft guy says that they have no idea why the algorithm learns the way it does. Think about it for a while. This should scare you.
It will NEVER happen--that they will start thinking. They are just dummies which can count. (the computers, or robots) As I said in another thread before--MT may have some use for translating set phrases, or idioms--not as a part of larger pieces, though, but just as separate items.
[Edited at 2014-05-31 09:43 GMT]
All I can respond with is this:
https://www.youtube.com/watch?v=pxAqxI5R6RA
Afghanistan
Local time: 07:37
English to Persian (Farsi)
+ ...
I do not think that (real) would mean professional, it just means that the words would be translated very urgent.
United States
Local time: 23:07
English to German
+ ...
I do not think that (real) would mean professional, it just means that the words would be translated very urgent.
http://en.wikipedia.org/wiki/Real-time_computing
In computer science, real-time computing (RTC), or reactive computing, is the study of hardware and software systems that are subject to a "real-time constraint", for example operational deadlines from event to system response. Real-time programs must guarantee response within strict time constraints, often referred to as "deadlines".[1] Real-time responses are often understood to be in the order of milliseconds, and sometimes microseconds. Conversely, a system without real-time facilities, cannot guarantee a response within any timeframe (regardless of actual or expected response times).
A real-time system is one which "controls an environment by receiving data, processing them, and returning the results sufficiently quickly to affect the environment at that time."[2] The use of this word should not be confused with the two other legitimate uses of "real-time". In the domain of simulations, the term means that the simulation's clock runs as fast as a real clock. In the processing and enterprise systems domains, the term is used to mean "without perceivable delay".
In real time
http://www.merriam-webster.com/dictionary/real%20time
the actual time during which something takes place - the computer may partly analyze the data in real time (as it comes in)
In this translation machine case, the translation machine recognizes the spoken text at the time it is spoken - this input of original text is immediately processed - translated (based on the words the machine recognizes) and communicated by a voice program, relating the translation to the listener - all happening within a few seconds, allowing a conversation between two people speaking two different languages with just a few seconds delay between the time the participants actually speak and listen.
Now, if this would really be like a conversation between two people who know their language of communication well, it be a great tool. But as I have shown in my other entry above, that is not the case.
The listener has to be very attentive to getting the gist of what his/her counterpart is saying through the computer voice/voice program. It also helps to speak very slowly and clearly so that the translation machine can input the speech correctly. Longer sentences, idiomatic expressions, complex texts and even choosing the correct word or word order when there are many available meanings are some of the tasks the machine has a lot of trouble with. If someone halts in mid-thought, adds on thoughts, repeats certain words, abbreviates others or simply "implies" thoughts or leaves stuff out because it's not necessary for a human, the machine might be completely lost.
This is at best a very basic way of communicating and it reduces correct language and respectful conversations to something absolutely ridiculous (I had to think of baby talk) - but to me as a translator, such language demolition isn't that funny. Would I want to communicate like that? No.
[Edited at 2014-06-02 13:35 GMT]
edited for clarification:
[Edited at 2014-06-02 19:04 GMT]
Italy
Local time: 05:07
Italian to English
This is at best a very basic way of communicating and it reduces correct language and respectful conversations to something absolutely ridiculous (I had to think of baby talk) - but to me as a translator, such language demolition isn't that funny. Would I want to communicate like that? No.
... it should improve Skype communication between my brother, who doesn't speak Italian, and my mother-in-law, who doesn't speak English.
I'm rather looking forward to it.
Romania
German to Romanian
+ ...
Some top interpreters can, sometimes. I am not afraid`that the machines will become any sort of competition--I am scared what they will do to human language, and how damaging the mistakes will be--they may cause serious accidents, malfunction of various instrument, poisoning, etc.
I would never ever want to read any literature translated by MT--what's the point? I'd rather read ads and obituaries.
[Edited at 2014-05-29 15:11 GMT]
Theoretically, no machine could and will ever manage to compete against the creativity of gifted writers or persons.
And yet, there is a strong tendency of over-simplifying everything and reducing any nuance "hard-to-understand-by-any-mortal" to a standardised "computer" language.
And there's where machines could win.
That's actually what we should fear most.
We allow machines to take over because of our "lazyness", and they waste no time in doing so.
I think Orin has put it in the best way.
[Edited at 2014-06-06 04:59 GMT]
Brazil
Local time: 00:07
English to Portuguese
+ ...
I really want to thank you for all the details you gave us about the German language. Since I can't speak German, I was dumbfounded by the announcement and by the videos. At least for the first minutes. But then I thought: Well, we are talking about Microsoft and its Bing translator. Nahhh, that can't be right. And it really wasn't.
To report site rules violations or get help, contact a site moderator:
You can also contact site staff by submitting a support request »
This discussion can also be accessed via the ProZ.com forum pages.