Google Search Translation Service Tested
05/24/2007, 1 year 3 months ago
I have had rather vast experience as freelance translator so I am particularly interested in testing all translation-related Web tools. There are some good translation and language-learning communities online (and we have already mentioned some of them). But I have never seen any good example of machine translation. I have seen funny examples, ridiculous examples, even hilarious examples. But never in my life have I seen a good (or even so-so) example of machine translation.
I always thought that using machine translation to localize your web service into various languages to attract more users usually ensures very poor results: new users only come to give it a try in their native tongue and leave for good to spread the word of new anecdotic examples of machine translations. I would not name any products not to offend anyone but those who actually have their content machine-translated into other languages will know.
So I thought no company would risk giving machine translation a real serious try. I was wrong. It looks like giants like Google can afford such an approach and they really believe it can be good. So Google has launched search translation service in beta promising they will translate the whole web to facilitate multi-language search. "We have been working on translating all of the Web to all languages," Google VP of engineering Udi Manber said. "The results are probably not perfect, but the information you want will be there." So I could not resist and I rushed to test this new product. Not to sound too critical I think I should mention here that I use at least 5 of Google Web-based services myself and in most of the cases I am satisfied with the results and occasionally I actually admire them and I suspect that when Google does something it usually does it well. But definitely this latest addition to Google's product range is not the case for me.
Basically this new search translation services can be described in 3 steps:
- You put the search query in one language (your mother tongue).
- You specify the language for the web pages to be searched for.
- Google translates the results from their original language into the language you want to view them in.
So I chose my language as Russian (my native one) and asked Google to perform a search for the term "Web 2.0 blog" (in Russian, of course) for me. The translation of the search query was fine but when it comes to more complicated queries the results may be far from true and you can offer a better translation by clicking the respective link. I should mention that I know perfectly well what blogs appear on Google's first page for my query so I was not afraid of being lost in the results. Mercifully, they have given us the option of viewing the results in both your language (left column) and in their original one (right column that can be hidden). But anyway if I did not have the results in English, I could hardly even guess what the Russian results mean anyway. It is a classic example of how machine translation (Google names it "automatic translation") works and it's really a poor work which is almost disappointing when it comes to such a prominent company.
For example, if we translate the Russian results back into English Techcrunch is described as "a group of editorial offices blog on technologies of launch" ("Group-edited blog about technology start-ups"). Profy is a "multi-purpose author of a blog" (originally "multi-author blog") but in our case a startup was translated into Russian as any person starting anything while for Techcrunch it was more related to launch of a rocket or a missile. I will not offer you more examples here because instead of being funny they are disappointing and even aggravating.
Why? The answer is simple: while I personally could compare the two versions of results, this new service is actually aimed at a general Internet user only proficient in his or her mother tongue to provide this user with additional opportunities, to allow "users across the world to find and view search results on foreign language web pages in their own native language". And without knowledge of the original language the results (the links, I mean) will be shown Ok but their descriptions will make no sense at all.
I know it is beta but I can hardly imagine they will hire all the translators on Earth to translate all the web into all the languages so even when out of beta it will remain machine translation ("translation produced by state-of-the-art technology, without the intervention of human translators" according to their About page) and I'm afraid the machine translation technology is still too immature to enable Google to provide the results that actually make sense to users.
Currently the following languages are included: French, Arabic, English, German, Italian, Japanese, Korean, Portuguese, Russian, Spanish, and traditional and simplified Chinese. So I invite you all to give the service a try (and contribute with better translations - finally, in web 2.0 users generate value, not only the technology) and if you find something interesting, I will be happy to see your opinions in the comments section below - what if the service is actually useful in some other languages?
If you enjoyed this post, make sure you subscribe to profy RSS feed!
It seems reasonable to assume that this uses the same underlying translator as Google’s language_tools page. This works differently to most, searching for matches in a database of documents already translated by humans. (”If all you have is a hammer, everything looks like a nail.”)
This beat other services for translating “The Taming of the Shrew” into Russian, but seems no better than others for normal translation work, and messed up some Tolstoy I gave it. It _may_ be good for translating commercial documents, which use a lot of pro-forma phrases, but I can’t see that their strategy is ever going to beat standard rule-based techniques. Especially in a language as heavy on grammar as Russian.
Oh, and if you try to translate rude words with it, they seem to get bowdlerised. Presumably this is an attempt not to “do evil”
Yes, Google is known to omit rude words in their search. And surely, why would they need to develop a new translation algorithm when they already have one. The thing is that there is simply no reliable machine translation algorithm (too true for grammar-heavy languages) now and I really doubt there will ever be one.
Possibly it will work better for some e-shops when you are looking for a certain product (or product range) but when it comes to coherent descriptions there is no tool to work with it with any quality. Google is no exception here, to tell you the truth.
What I like about them is that they invite users to contribute their translations (if they have any) to the existing language pairs to improve the translation results. But the problem is only professional translators have large volumes of translated materials and originals but they simply can not disclose this materials in the majority of cases. So I can hardly imagine how these contributions will work.
After reading your post, I tried Google Translate Beta.
In my opinion, even if the translator has some well known limitation, it will be useful for people that otherwise would search native language documents only. Possibly still a lot of people.
And in some case it is really funny
Pasquale,
What language have you tried it with? I’m still trying to find opinions about other languages. Honestly, I could hardly understand what the documents were about when I looked through their translations into Russian. Only looking at their originals in English could help that is why I am rather doubtful about possible use of the service. Though you are very right - in some cases it is really funny
Svetlana,
I tested my native language, Italian. Here is some example.
1. “mobili ufficio”, correct translation: “office furniture”, gives correct results. But when traslating “Home Office Furniture” to Italian, “Home Office” became the governmental body for national affairs.
2. “spaghetti al pomodoro”, correct translation: “spaghetti with tomato sauce”. You got a plenty of reasonable recipes. When translating in Italian, “Early Girls” referred to tomatoes, became a funny girls that are early. Or a “large pot” is translated with “grande POT”, (”big POT”), leaving POT in english. It sounds like a divinity
Pasquale,
Thanks a lot for your funny examples for Italian. I guess you helped me find at least one suitable purpose for the service - looking for new recipes in other languages. But taking into account my own rather limited cooking ability I really doubt I will ever need more examples than those listed in my own Russian language
:))
I’m constantly amazed at the criticism of any new endeavor by people who can’t do any better. Until you post the link to your vastly superior machine translation site, how about commenting on how the algorithm could be improved instead of saying “I am rather doubtful about possible use of the service”? Google didn’t even exist 10 years ago, and today most denizens of the web can’t imagine life without it. Was it as good when it first appeared as it is today? Uhhh, no…
They should, of course, have given up when they first created the search engine, because people were rather doubtful about the possible use of the service.
I’ll see you here again in ten years. Let’s see what the state of automated translation is then.
Chaz, it is not actually criticism of Google at all, it is more of the fact that being a professional translator myself (with a number of full-time experience) I do not believe in machine translation. I do not believe in it at all and I will never start to believe in it only because Google can do it. Google is a giant and it can do almost anything, I know, but machine translation has never worked and I really doubt it will ever work. Correct translation takes a human translator. Google declares it has developed a superior translation engine but here is my proof that it does not work either. I do hope someone will find this service useful, maybe I just don’t see how - that is why I asked for comments. If you give me an example of how well you use the service yourself I will really be thankful. But I don’t believe in machine translation, no matter how prominent a company tries to implement it.
Yes, it’s true that today machine translation doesnt work. Before when i just began to learn russian, i used babelfish service for couple of months, and the result was of course not good. But when i compared google with babelfish, I saw that babelfish is a little bit better.
Google says that they developed their own system, and they name it statistical machine translation. The interesting difference is that they allow people to submit their own suggested translations, when they see something wrong or missing. That’s some fresh hope, but requires good reviewing.
Another smilar solution is Prompt translator. From their site, I read that they wrote some technical dictionary, and with the help of that they translated MSDN information to russian with machine translation successfully. But of course need to test it
I downloaded and tested their desktop translation software. They have some nice features like integrated OCR, for recognizing the text from TIFF or other image documents, and translating it. And also have an icq add-on, which can be useful for people who only know english and want to communicate with other countries. And as I know, mail.ru also use their translation service.
Of course, today a machine translation service can not compete with human translations
It’s quite hard to do this. Taxonomy, synonyms, idioms… it’s really hard.
But think that there is an english company making business with russia, when they receive a fax in russian from they customer, they can OCR it translate with machine translation and at least they can understand what the document is about, before they receive the translated document from translation office.
As i am working on the document management area, that example is quite interesting for me. At least a good marketing point
Regards
Arda, thank you very much for your input, it is really great to see someone as interested in the subject matter. The problem with Prompt and other examples (believe me, I tested them all when I worked as a translator is that there is no technology that could recognize the meaning of the source text as well as a human translator can. Believe me, I would love to believe that one day we will see a powerful machine translation technology that will enable us to understand each other instantly without the delays caused by translators working on the letters, documents, contracts, etc. It’s just that even as an early adapter of various new technologies I don’t believe that machine translation will reach any reasonable level of accuracy to be applied at all. And the reason for this post was to demonstrate that even such a giant as Google is can not overcome the difficulties with their technology. You have mentioned a great point here about users being able to edit Google’s translations but the problem is it is only possible to edit the translation of the search query itself while the most stupid mistakes occur in the website description - those descriptions are incredibly poorly translated and I can not even help Google by translating this or that description for websites for them.