| Pages in topic: [1 2 3] > | Google’s New Service Translates Languages Almost as Well as Humans Can Thread poster: Jeff Whittaker
|
|---|
Jeff Whittaker 
United States
Local time: 18:33
Member (2002)
Spanish to English
+ ...
| Olga Koepping
United Kingdom
Local time: 23:33
German to English
+ ...
| Well that hasn't improved my mood! | Sep 27, 2016 |
So how do we continue to earn our crust then?
| | | | Dan Lucas
United Kingdom
Local time: 23:33
Member (2014)
Japanese to English
| It's 1983 all over again... | Sep 27, 2016 |
Oh no, AI will take over the world! Where have I heard this before? Oh, right, the 1980s. A bit before the AI bubble collapsed in 1987...
We'll see how this new Google Translate. At least in my pairs, a 70% improvement would still not give production-ready text. Might make it a more useful tool though.
Dan
| | | | | Sounds great | Sep 27, 2016 |
When can we see some output?
I am really looking forward to seeing some results for English-German.
Am I scared? Not at all.
| | |
|
|
|
Jean Lachaud
United States
Local time: 18:33
English to French
+ ...
| Almost doesn't count | Sep 27, 2016 |
Almost means: "Nope! wish it would, but it doesn't."
Let's see some of this "translation".
| | | | Jeff Whittaker
United States
Local time: 18:33
Member (2002)
Spanish to English
+ ...
TOPIC STARTER | | Approaching the cusp | Sep 27, 2016 |
Heart stopped a bit when reading the OP's article, especially since this is being rolled out in my language pair first. Much calmer now after having clicked on the link embedded in the words "today is its official debut" (https://research.googleblog.com/2016/09/a-neural-network-for-machine.html). Scroll down to the four column comparison table in this link. It's... See more Heart stopped a bit when reading the OP's article, especially since this is being rolled out in my language pair first. Much calmer now after having clicked on the link embedded in the words "today is its official debut" (https://research.googleblog.com/2016/09/a-neural-network-for-machine.html). Scroll down to the four column comparison table in this link. It's quite obvious that the entire title of the first article was a clever ploy--the human translation is pretty bad, akin to something an inexperienced translator might produce in their second language. So yes, the new Google Translate is almost as good as a human translation--if the human translation is a bad one to begin with
Did a quick test in Google Translate (which apparently is using this technology for all of its Chinese to English translations now). Used an article from Xinhua today, one of those monotonous, repetitive party releases that reuses lots of slogans developed over the past 5, 10, 20 years; this should be the perfect thing for the new technology to handle.
For the input: 会议强调，这两个文件稿经全会审议通过后，要抓好贯彻落实。中央政治局的同志要带头深入学习、深刻领会，带头按照文件规定严格要求自己。要一条一条学习，一条一条领会，牢记在心中，落实在行动上，特别是要在坚定理想信念、坚持党的基本路线、坚决维护党中央权威、严明党的政治纪律、保持党同人民群众的血肉联系、坚持民主集中制原则、保持清正廉洁政治本色、自觉接受党内监督和人民群众监督等方面作出新的努力。
We get: "The meeting stressed that the draft documents should be put into practice after being approved by the plenary session. The comrades of the Political Bureau of the Central Committee of the CPC should take the lead in in-depth study and profound comprehending and take the lead in stricter demands on themselves in accordance with the documents. We must adhere to the party's basic line, resolutely safeguard the authority of the Party Central Committee, strictly observe the party's political discipline, and maintain the party's relations with the masses of the people. Flesh-and-blood ties, upholding the principle of democratic centralism, maintaining a clean and honest political nature, and consciously accepting inner-party supervision and supervision by the masses of the people. "
Besides the inevitable grammatical issues, the first two sentences could be post-edited into a correct rendering of the source relatively easily. Fairly significant mistake in the third sentence, however, in which what should be translated as something like (very rough and literal translation) "We must study and understand these documents line by line until they have been memorized" was simply omitted (but see my doom & gloom footnote below). The entire "Flesh-and-blood ties" sentence is obviously grammatically incorrect, as the program was not able to successfully add a 'Moreover' and new verb to complete its correct breaking up of one long Chinese sentence into two English sentences.
My initial thought is that this new program is an improvement, and one that moves Google dangerously close to the cusp of producing something that really could significantly cut into our businesses. However, I'm still rather pessimistic about their chances of getting there. Google's machine translation may be getting close to producing product that is 60-70% correct, but my suspicion is that the last 30% they need to make clients switch from human translation to them will be very difficult, perhaps impossible.
Humans were able to drive (relatively) safely on highways at 70mph in 1950. It would have been reasonable to suppose that by 2016 we'd be able to drive safely at 100 mph on the highway, but here we are, still stuck with speed limits around 70 mph. Likewise, I'd say the chances that Google Translate has gotten as good as it ever will get are about the same as it making the last push it needs to put all of us out of business.
Would love to hear more thoughts/examples from other Chinese/English translators?
* My doom & gloom note is that it is also remotely possible that Google Translate understood that "in-depth study and profound comprehending" from sentence 2 is essentially synonymous with "We must study and understand these documents line by line until they have been memorized" from sentence 3, and omitted this on purpose. If this was the case, then perhaps we are in more trouble than I think.
[Edited at 2016-09-27 22:53 GMT]
[Edited at 2016-09-27 23:10 GMT] ▲ Collapse
| | | | RobinB
United States
Local time: 17:33
German to English
| As I said publicly at last year's MT Summit | Sep 27, 2016 |
"If Google's driverless cars are only 100 times better than the latest iteration of its MT, then there's going to be a lot of dead people on the streets."
If you set the bar low, then you can claim that almost anything is near-perfect.
As Dan says, we've seen all these NHQMT claims before. That doesn't mean that MT, on average, hasn't improved significantly over the intervening decades (my first experience in post-editing MT output was in 1990), but it is still a long, ... See more "If Google's driverless cars are only 100 times better than the latest iteration of its MT, then there's going to be a lot of dead people on the streets."
If you set the bar low, then you can claim that almost anything is near-perfect.
As Dan says, we've seen all these NHQMT claims before. That doesn't mean that MT, on average, hasn't improved significantly over the intervening decades (my first experience in post-editing MT output was in 1990), but it is still a long, long way off from reproducing what expert human translators can produce. Tech journalists will, I'm sure, be replaced by machines long before professional translators. ▲ Collapse
| | |
|
|
|
Jeff Whittaker
United States
Local time: 18:33
Member (2002)
Spanish to English
+ ...
TOPIC STARTER | The issue is not when MT will reach human quality... | Sep 27, 2016 |
... but at what point are customers willing to overlook awkward phrasing and the occasional mistranslation when they are saving thousands of dollars and can get instant translations without the need to wait. A lot of translation work only survives today because it is in a non machine-readable format (pdf, etc.).
Sample translations in... See more ... but at what point are customers willing to overlook awkward phrasing and the occasional mistranslation when they are saving thousands of dollars and can get instant translations without the need to wait. A lot of translation work only survives today because it is in a non machine-readable format (pdf, etc.).
Sample translations in Spanish, French and Chinese to English: https://drive.google.com/file/d/0B4-Ig7UAZe3BSUYweVo3eVhNY3c/view
[Edited at 2016-09-28 00:01 GMT] ▲ Collapse
| | | |
LegalTransform wrote: ... but at what point are customers willing to overlook awkward phrasing and the occasional mistranslation when they are saving thousands of dollars and can get instant translations without the need to wait. A lot of translation work only survives today because it is in a non machine-readable format (pdf, etc.). Sample translations in Spanish, French and Chinese to English: https://drive.google.com/file/d/0B4-Ig7UAZe3BSUYweVo3eVhNY3c/view[Edited at 2016-09-28 00:01 GMT]
Those translations are not particularly impressive. Plenty of ambiguity and grammatical error, especially when you consider that they were almost certainly cherry-picked to show how 'good' Google Translate is.
You are right in noting that machine translation doesn't have to be perfect to induce clients to use it. But I do think the CN>EN market is fairly instructive here: you can get a China-based, non-English native translator to translate documents from Chinese into English for a third of the price and three times as quickly as yours truly, but plenty of work continues to come my way. This is because the occasional mistranslation matters a great deal to many clients (and their insurers). There's no doubt that I already miss out on "easy" translations (like email chain translations ) because of my rates. But then again, email chains didn't exist before the Internet era. So technology gives and it takes away, but the volume of translation available for human translation is undoubtedly higher than before the Internet. Until that changes, we're still OK.
| | | |
I'm not sure at what point customers will be willing to overlook awkward phrasing and the occasional mistranslation. The good thing is that, judging from these samples, we're definitely not there yet.
At least Google is being honest about it, "they sometimes couldn’t see much difference between them" seems to be a nice way to say "they could still often see a significant difference".
| | | | Dan Lucas
United Kingdom
Local time: 23:33
Member (2014)
Japanese to English
| Those customers already exist | Sep 28, 2016 |
LegalTransform wrote:
... but at what point are customers willing to overlook awkward phrasing and the occasional mistranslation when they are saving thousands of dollars and can get instant translations without the need to wait.
Those unfussy customers are out there.
They already exist.
They always have existed.
Just think about the implications of what you're saying. In essence, your argument is that all products of a certain type are undifferentiated and that if a certain product becomes cheap enough, all potential customers will buy the cheapest product.
By your argument, if some technological miracle were to come along whereby it became possible to slash the cost of building and selling a basic dwelling from $200,00 to $100,000, all home buyers would buy the $100,000 dwelling.
Really? You really think the people who were previously prepared to spend $300,000, $500,000 or indeed $1,000,000 on a home would suddenly decide that they prefer to buy the $100,000 house just because it now costs $100,000 instead of $200,000?
Consider another example. Why do we, as humans, bother to wear anything other than a purely functional one-piece item of clothing made out of a hard-wearing material? Given that deviation from the lowest common denominator inevitably pushes up costs and prices, why does anybody accept any variation from that basic one-piece garment? Think about it.
Repeat after me: Markets Are Segmented.
If you need proof, look around you. Whether it's homes, cars, clothing, consumer goods, food or indeed services, you will see a huge range of features and price points. You will see differentiation, in other words. Purchasing decisions are driven by a complex mix of factors, and price is only one issue, often a minor issue.
If you don't differentiate your product for that of other suppliers or if you aim for the wrong segment of the market, you will suffer.
If you don't understand that markets are internally diverse, you won't even understand why you're suffering; instead you will blame it on things like MT.
Dan
[Edited at 2016-09-28 07:56 GMT]
| | |
|
|
|
| All this algorithmic stuff doesn't cut it | Sep 28, 2016 |
The point where translators really will get in trouble will be the point when software some day should be able to understand human language. This cannot be achieved by algorithms, not even the most sophisticated ones. This could only be achieved by systems that are able to learn entirely new things. To this day, there is no such thing as artificial intelligence, the only things we have are sort of "intelligence simulators", "learning simulators", and complex systems that can look very sma... See more The point where translators really will get in trouble will be the point when software some day should be able to understand human language. This cannot be achieved by algorithms, not even the most sophisticated ones. This could only be achieved by systems that are able to learn entirely new things. To this day, there is no such thing as artificial intelligence, the only things we have are sort of "intelligence simulators", "learning simulators", and complex systems that can look very smart in very narrowly defined environments. MT today does not even attempt to understand language, MT just analyses language in order to increase the chance to hit the right translation, and while it might succeed in increasing this chance, the difference to 100% will stay very significant, therefore any text that requires accuracy cannot be machine-translated. With this approach, MT results will always clearly show that the person who presents us with such a text does not care for quality. Of course there is a market for this, as well as there is a market for even the worst product if it is just cheap enough.
To this day, no software was able to master the Turing test. As long as this does not happen, MT will be exactly what it is today: one particular reason why the world needs competent human translators. ▲ Collapse
| | | | RobinB
United States
Local time: 17:33
German to English
| They already do | Sep 28, 2016 |
LegalTransform wrote:
... but at what point are customers willing to overlook awkward phrasing and the occasional mistranslation when they are saving thousands of dollars and can get instant translations without the need to wait. A lot of translation work only survives today because it is in a non machine-readable format (pdf, etc.).
A number of my clients already use MT, some in a really big way. They use it for "information only" translations within their global groups of companies. And if they're using MT, many other multinationals are most likely doing the same.
But frequently awkward phrasing and a high level of mistranslations (both of which are common in German-English/English-German MT) are material barriers when it comes to "decision-useful" translations, which is why humans will continue to be involved to a significant extent for this category of translations. I don't know about you, but I guess (at least) 95% of my translation work involves "decision-useful" texts, and I'm sure this is the case for the majority of expert translators. Of course I'm also looking forward to being able to integrate high-quality MT into my own translation environment, but only if I'm in a position to control input and output across the entire process.
[Edited at 2016-09-28 08:29 GMT]
[Edited at 2016-09-28 08:30 GMT]
| | | | | Potential clients do use MT | Sep 28, 2016 |
RobinB wrote:
A number of my clients already use MT, some in a really big way.
And if they're using MT, many other multinationals are most likely doing the same.
And not just for information only.
A few weeks ago I saw some turkey legs in a Romanian Lidl supermarket that were described as "Turkish legs" (i.e. using the country name Turkey) in Hungarian. The word for turkey should have been 'pulyka', so it's not even close phonetically to what they printed. I had to stop to laugh and take a photo. The interesting thing here was that a Romanian person obviously used EN-HU machine translation, probably because they didn't trust Google Translate for the RO-HU pair.
Some fields of translation are probably more threatened by MT, but I don't think any client would risk using MT for legal or financial texts.
| | | | | Pages in topic: [1 2 3] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Google’s New Service Translates Languages Almost as Well as Humans Can | Wordfast Pro | Translation Memory Software for Any Platform
Exclusive discount for ProZ.com users!
Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value
Buy now! » |
| | 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 » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
Translation news
