Google: Machine translation is web spam -- user translation possible?

[deadsea]

Matt Cutts announced today in a webmaster help video that websites that are machine translated into multiple languages are henceforth considered web spam.
[youtube.com...]

OK. I'm officially blackhat now as one of my sites is available in several languages through the miracles of machine translation. In my case the main content of the site is language agnostic (images), and its only the navigational elements and boilerplate that get machine translated. I hope that is somewhat of a mitigating circumstance and I don't get penalized by Google tomorrow.

In any case, I do get occasional emails from people upset about the poor quality of the translation, so I do see Google's point. Occasionally I'll get somebody that volunteers to help, and they correct some of the worst translations for their language.

I believe that many more of my users would be happy to help translate if this were easy for them to do. Does anybody have recommendations for web software or services that I can use to make it easy for my users to correct my bad translations?

[tedster]

There is no software to do this - software translations are, as the video points out, just plain awful.

If the site is important to you, I'd hire some help. Sometimes you can find excellent help through colleges and universities.

[chrisv1963]

Don't use machine translation! It simply doesn't work and machine translations are often completely wrong.

[deadsea]

Sadly, my image site doesn't make enough to spend money on the translations. I'll probably have to put the foreign language versions in robots.txt unless I can find a way to crowd source the translation work.

While I agree that the machine translations are far from great, the bounce rate on the foreign pages are not much worse than the English pages. I believe that most users will put up with bad translations on the navigation when it is a matter of not being able to use the site at all otherwise.

[austtr]

So we can expect Google to withdraw their language translation tools?

[tedster]

I certainly hope not - they are useful when you know you are getting a machine translation, and they are better than they were. I can't imagine Google dropping it.

[Leosghost]

Machine translations are awful..and even the most expensive software has to be fed such simplified versions of the first language in order to get a translation in the target language that still requires hours of correction for even a few hundred words on a simple subject..that it's not worth using it/them.

I've tested many software "tools" ( for others )..you have to feed them semi literate 7 year old child level "pidgin" English to get out near totally illiterate and near incomprehensible 5 year old level French..( and vice versa ) which has to be nearly totally rewritten to get to the level of a semi literate 7 year old ( with a severely restricted vocabulary ) non French born ( immigrant ) kid who's skipped school for 50% of his /her time since they got here ..

Much easier and better just to go from a literate 7 year old child in English to a literate 7 year old child in French ( or vice versa ) ..or any other "language pair" and age or "target demographic" ..via good human translation ..which isn't that expensive as an investment ..long term .

The same applies for all other languages ( according to fluent speakers of other languages that I know who have run the same tests in their own language(s)) ..don't use machine translation ..it sucks .

Google's online language translation tools get you from the 7 year old to the semi literate 7 year old step ..without going through the flaky incoherent 5 year old step in the middle ..and at no expense ..but the result is still crap compared to actual fluent human translation ..and any translation also depends on your target demographic ..one doesn't use the same vocabulary and syntax to communicate with a 14 year old as to a 40 year old and a 64 year old..even for the same subject matter ..not if you want to "connect" with each of them.

And then there is the question of regional accents in writing and the changes that makes in the vocabulary you need to use ..and of course the subject matter ..can it be handled jokingly or is it serious sales or scientific ..

I think Matt is saying ..real good human translation is best and will get you better ranking and even better still ranking if you also give us all the other "easy to read" signals regarding TLD(s) and hosting etc .( and my personal advice is also to get your incoming links in the correct language to the correct pages and if possible from the sites with the correct TLD(s)..and again if possible hosted in the correct countries )..believe me that helps a great deal ;-)

Then he hints "Google's" tools are a way behind human translation ( obvious but worth saying again ), but better than software translation ( he is so right there !) ..and that if you use Google tools you should still bear in mind to give them as many of the signals as per "above" as you can.

Link to their translation tools from your pages if you can't get human translation done for cost reasons or whatever ..I agree ( never thought I'd say that regarding translation tools ..but Google's are way waaay ahead of any commercial tool ) with him 100%.

He is quite clear that autotrans software doesn't score well with them and is considered by them to be a bad user experience ( rightly so ) and will affect your ranking negatively.

And in passing he reinforces the point that auto generated content ( as is being discussed in a "spinner "thread here ) is easy for them to spot ..they don't believe ( again rightly so ) that it gives a good user experience ..and using it will result negatively on your rankings.

And before it comes up ..ehow is not auto generated ..it's almost always shallow and frequently inaccurate or "lifted" and "respun" ..but the "spinning" plagiarising is done by humans ..and Google can spot and down rank machine spinning now ..

They may get to down ranking human spinning of plagiarised content later ..we can but hope ..but that is more complex for any combination of "algos" to identfy.

[LifeinAsia]

websites that are machine translated into multiple languages are henceforth considered web spam

Wow, Google is a few years behind what we have been saying here for years... Anyway, glad that what we've been preaching has not been deemed gospel. :)

Sadly, my image site doesn't make enough to spend money on the translations.

Then it's probably best to not translate them at all. My suggestion, if you really must have a "translated" site, is to pay the small price to have your navigation translated and replace the boilerplate content with the original language. Then translate the boiler plate as you find resources.

[Leosghost]

Wow, Google is a few years behind what we have been saying here for years

Kinda makes you think that maybe even if we dont see Matt around much here nowadays, that he maybe looks in once in a while ..apparently just takes a while for some of the more obvious things we say here to penetrate the inner sanctuaries of the plex.;-)

[wheel]

Sadly, my image site doesn't make enough to spend money on the translations. I'll probably have to put the foreign language versions in robots.txt unless I can find a way to crowd source the translation work.

If your visitors are loyal, you could maybe get them to translate bits and pieces. I'd get a javascript tool built that lets visitors highlight text and submit a correction. Get 2 or 3 corrections that look the same and you could change the text.

[Staffa]

So we can expect Google to withdraw their language translation tools?

Well they should and I hope they do ... sauce for the goose ......

I certainly hope not - they are useful when you know you are getting a machine translation, and they are better than they were.

Why differentiate between the junk translations G turns out and ditto via other software.

If you speak any other language(s) have you looked at what they churn out ?

I'm fluent in five languages and G's translations are at best risible. People who use the G translate this page buttons on their web site don't know what they are letting themselves in for.

[Leosghost]

I'm fluent in five languages and G's translations are at best risible.

If you speak any other language(s) have you looked at what they churn out ?

D'accord mais..
Agreed ..but babylon and systran et al are even worse..encore pire ..which I think is Matt's point.

<added>Btw ..for clarity ..fluent in 2 ..reasonable in 2 more ...was "quoting" staffa .. not claiming fluency in 5 languages ;-)</added>

[edited by: Leosghost at 2:33 am (utc) on Mar 24, 2011]

[tedster]

sauce for the goose

There's a difference between putting a machine translation TOOL online and serving machine translated CONTENT as a big chunk of your website. And it's a big difference.

[Leosghost]

There's a difference between putting a machine translation TOOL online and serving machine translated CONTENT as a big chunk of your website. And it's a big difference.

Entièrement d'accord Ted :) Entirely agree :)

[austtr]

Whoa there..... if I put a Google translation button on my pages I'm providing the viewer with a way to see a crappy translation.

If I create multi-language pages as part of my site I'm providing the viewer with a way to see a crappy translation.

Same outcome... the common denominator is the tool that outputs the crappy translation.

Devaluing a website that has crappy translations while at the same time providing the tool that creates the crappy translation is a bit too cute for my liking.

If the translation tool doesn't do the job properly in this new era of quality, then get rid of it.

[indigojo]

wooooah, missed point here Mr Cutts, so darned hung up on cleaning up your SERP's you are forgetting the BIG picture, take the guy in Iran with some bizarre medical condition, he has to search in english to get to translate this page option from English SERP's into his native tongue. If he has a translation available in his native tongue that just gives him at the very least a CRAPPY transation that's empowered him and helped him. Surely that's for the greater good of mankind isn't it! From feedback I've heard from native speakers they actually aren't that bad either.

[Leosghost]

Whoa there..... if I put a Google translation button on my pages I'm providing the viewer with a way to see a crappy translation.

If I create multi-language pages as part of my site using offline software I'm providing the viewer with a way to see a crappier translation.

It's not the same outcome... the common denominator doesn't exist, the offline tool that outputs the crappier translation and the online tool that puts out slightly better but still crappy translation are not the same.

Devaluing a website that has crappy translations made with offline software while at the same time providing an online tool that creates the slightly less crappy translation is better than just having really crappily translate pages done by offline tools.

If the offline translation tool doesn't do the job properly in this new era of quality, then get rid of it..And try not to use the online tool because a human translated text is worlds better.

Added my bold..IMO makes more sense now :-)

[viggen]

I found the translation tool from Google always handy, i know its far from perfect, but it gives me a very quick overview of a story that isnt available (news) in english or my mother language.

Below you see a text I just copied and pasted from the german magazine Spiegel into google translate, and i believe even though this won`t win a grammar prize, you understand what the articles says...

Google Translation German to English
U.S. politicians want to lock apps to bypass breathalyser

Apps that warn drivers of alcohol checks are to be blocked by the will of influential U.S. politicians. Four Democratic senators called for operators of mobile software platforms such as Apple, Google and Blackberry RIM in a provider on Tuesday (local time) published letter, take appropriate programs immediately from the quotation. "give drunk drivers a tool to circumvent controls, families and children in jeopardy, "wrote the majority leader of the Senate, Harry Reid, mentioned group.

...while i wouldnt put an autotranslate page on my website, i do find it handy for occasional use when browsing...

[edited by: viggen at 5:40 am (utc) on Mar 24, 2011]

[Leosghost]

CRAPPY transation that's empowered him and helped him. Surely that's for the greater good of mankind isn't it!

Unless the offline software produced a translation so crappy he does something he shouldn't and it kills him

From feedback I've heard from native speakers they .

presumably the

they

in

they actually aren't that bad either

refers to a specific piece of offline software ..Which ? ..The two most widely used are the two I quoted earlier ..and they output incomprehensible rubbish ..

But if you don't speak the target language you'd not be in a position to know that ..would you ..

I've dealt with people whose bosses hired them because they were supposedly fluent in English ..except they weren't ..but the boss thought their English was great ..so did all their friends and their co workers ..and they even had degrees in English ..still couldn't speak nor read and write in it ..same has happened with English people who are supposedly fluent, degree level qualified French speakers.

Saw a French site two days ago, the English pages were incomprehensible ..the translator had the equivalent of a degree in English awarded by a French university ..seen the same thing on English sites ..the French pages that were laughable .done by degree qualified translators from the UK ..both smacked of having been rushed through an auto trans ( both had the trademark fingerprint mistakes of systrans) and "tidied" for grammar later ( lack of vocabulary is the "give away" most of the time )..small onboard "dictionaries"..each "auto trans software" has it's own distinct dictionaries.

[iamlost]

Leosghost raises an extremely important, often missed point of translation:

And then there is the question of regional accents in writing and the changes that makes in the vocabulary you need to use ..and of course the subject matter ..can it be handled jokingly or is it serious sales or scientific ..

If one is looking to maximise visitor conversion (define as you will) then writing fluently in their own language is best.
Note: if you add podcasts or video the regional variations need to be carefully considered; many more differences in spoken than written language.

Being unilingual I have content translated by one person and the result translated back by another. Only partly because I'm paranoid... :) ...mostly because conversion optimisation can be language critical.

And, being in business, the reason I pay for translated content is NOT to be a nice guy but to make a profit. If I don't see the probability of a reasonable return within a reasonable time frame I leave the visitor to their own devices. I see machine translation as disrespectful to a visitor, not helpful.

IF Google actually follows through and removes multi-language machine translated sites that would be nice. However, I note that most such sites I see are sites with scraped content machine translated to some other single language so I don't see me putting down my DMCA sword anytime soon.

One point directly to the OP:

In my case the main content of the site is language agnostic (images), and its only the navigational elements and boilerplate that get machine translated.

In my experience the shorter the copy being translated, i.e. image captions, the greater the probability, through loss of surrounding context, to make a language faux pas. The number of words, phrases with some regional vulgarity or similar is truly astounding.

Given the minimal amount of copy to be translated you might find it economically viable to hire a local university student native in or majoring in the language in question.

[LifeinAsia]

I've dealt with people whose bosses hired them because they were supposedly fluent in English ..except they weren't ..but the boss thought their English was great ..so did all their friends and their co workers ..and they even had degrees in English ..still couldn't speak nor read and write in it

Been there too! I've gotten a lot of work going back and fixing the writings of these English "masters." OTOH, I've also seen a lot of non-natives who can speak/write far better than Native English speakers, but that's a thread for a different day. :)

We've spent a lot of money translating our sites. So I am utterly thrilled that Google is (supposedly) taking steps to ensure that people who use machine translated content don't appear above me in the SERPs.

[rlange]

austtr wrote:
Whoa there..... if I put a Google translation button on my pages I'm providing the viewer with a way to see a crappy translation.

If I create multi-language pages as part of my site I'm providing the viewer with a way to see a crappy translation.

Same outcome... the common denominator is the tool that outputs the crappy translation.

One big difference: I believe the crappy translations in Option #1 won't be indexed, thereby avoiding a potential wrist-slapping by Google. Of course, you loose potential foreign-language searches, too.

[Leosghost]

To demonstrate the point ..I found the text posted by viggen in the original German from Spiegel.

The subject in English was mentioned in Foo [webmasterworld.com] recently ..U.S. politicians want to lock apps to bypass breathalyser

Apps, die Autofahrer vor Alkoholkontrollen warnen, sollen nach dem Willen einflussreicher US-Politiker gesperrt werden. Vier demokratische Senatoren forderten Betreiber mobiler Software-Plattformen wie Apple, Google und den Blackberry-Anbieter RIM in einem am Dienstag (Ortszeit) veröffentlichen Brief auf, entsprechende Programme sofort aus dem Angebot zu nehmen. "Betrunkenen Fahrern ein Werkzeug zu geben, um Kontrollen zu umgehen, bringt Familien und Kinder in Gefahr", schrieb die von dem Mehrheitsführer im Senat, Harry Reid, angeführte Gruppe.

************
Followed below by the translation he posted using Google online translation ..free

Apps that warn drivers of alcohol checks are to be blocked by the will of influential U.S. politicians. Four Democratic senators called for operators of mobile software platforms such as Apple, Google and Blackberry RIM in a provider on Tuesday (local time) published letter, take appropriate programs immediately from the quotation. "give drunk drivers a tool to circumvent controls, families and children in jeopardy, "wrote the majority leader of the Senate, Harry Reid, mentioned group.

Not accurate but you can get the sense of it, its waay "less crappy" than what follows below.
*******************
Same original German text translated by Systran "premium" online/offline translation software .. cost just over $1000.oo latest up to date version used.

Apps, which warn drivers before alcohol controls, are to become closed after the will of influential US politicians. Four democratic senators demanded operators of mobile software platforms such as Apple, google and the Blackberry offerer RIM in one on Tuesday (local time) publish letter up to take appropriate programs immediately from the offer. To “to give, in order to go around controls, in danger”, wrote group aforementioned by the majority leader in the senate, Harry Reid, brings families and children drunk drivers a tool.

Nearly incomprehensibly "crappy" ..be honest..;-)

Btw .Babylon the other main translation software I mentioned is less expensive ..but even worse in its output...I don't have a working copy on any of my machines at the moment ( I had both of these over the last month to "test" for someone )..and can't be bothered to put it back on.

So Google is correct IMO ..any paid translation software companies out there or amongst members ..the message is clear..drastically improve your quality because "free" works better ..and good human translation beats all.

And I'm far far awaay from being a Google fanboy ..Google are right on this one ..and they are also giving warning and large hints what to think about if you want multilingual sites to rank on their search engine ..upto you if you listen and act on what they say ..if you don't , the fault will be all your own.

[Alex_TJ]

My reading of the OP's question was not about human vs software translation, but if there's software out there that will allow him to crowdsource translation corrections.
I remember LinkedIn (?) tried to do something similar a few months back, and wheel touched on it before we struck off on the usual gibberish translation reasoning.
Is there any user correction-type software or method out there?

[deadsea]

I loaded all my translation strings into [mygengo.com...] They have translators that can do the translations at 5 cents a word, or I can have my users sign up and I can give them access to my strings library.

Its not perfect, but it may end up working for me. If anybody has suggestions of other similar tools that have worked for them, I'd still like to hear other suggestions.

[gabbrunner]

I do not agree with Matt Cutts, and I think it is kind of weird that he would consider one of his own inventions (or Google's), google translate, spam. Isnt't it ironic? I prefer to go with a human translation service for my websites, which gives me a far better feeling of security. After all, you never quite know what comes out in the end with machine translation.

One last observation: My brother works at SDL, the world's leading translation agency, and they have recently switched over one of their biggest clients to machine translation, without telling them about it! There is still some kind of human input/edition work going on, but their technologies seem to be way more advanced than what the average user experiences when using online automatic translation services.

[edited by: tedster at 8:58 am (utc) on Apr 20, 2011]