Is Google Translate accurate enough for professional use?
Written by ForeignExchange Translations on Wednesday, November 25, 2009
Earlier this year, Google Translate single-handedly improved the reputation and usability of automated online translations. Until then, online machine translation solutions (especially of the free kind) were marginally useful at best.
Google Translate has changed that, largely by using a statistical approach to machine translation and by allowing users to help the system learn adaptively. 
And with a service that allows millions of amateur translators to pour their translations into an open tranlsation memory, Google hopes to change professional translation in the way that flickr and istockphoto have changed commercial photography. Online machine translation has the power to revolutionize communication by eliminating language barriers, bridging the gap between cultures, providing services and information to speakers of minority languages, and transforming global e-commerce by allowing even the smallest online vendor to serve the international market.
But will Google Translate really have that far-reaching an impact?
Translation practioners and clients/users of machine translation tools see two main challenges that must be overcome before a system like Google Translate can hit the big time:
- Quality of translated text needs to improve
- Copyright law needs to be amended
Quality
Just how good is the output from Google Translate? Well, it depends.
Sometimes, it stinks. Other times it is quite effective. And some people even claim that there are instances that Google’s language translator can provide a translation as if a human translates it.
A lot of it clearly depends on the subject matter and language pair being used. For instance, a review of the system's French abilities gave it a middling grade. The reviewer thought "that it's ok to get the gist of it, but the grammar isn't great and there are words missing here and there, also French words appear instead of English words in the translation." Not exactly a ringing endorsement.
Interestingly, we have had some of our medical device clients use Google Translate as part of their in-country review process. One medical device manufaturer uses Google Translate to determine the completeness of a translation. "Often translations come up in perfect English syntax", the client notes. "When translations come up with very questionable meanings I at least like to have them checked."
While the resulting back translation quality is rough, the client feels that the quality is good enough for the system to provide a back translation that is used as a QA tool.
Copyright
Machine translation will create massive copyright infringement on an unprecedented global scale - or that is how Erik Ketzan views the situation, as voiced in his 2006 research paper, Rebuilding Babel: Copyright and the Future of Machine Translation Online.
Why? In America, at least, copyright law considers a translation a derivative work. That means that translators must obtain permission from the copyright or derivative right holder of the source language text.
It's one thing to translate a text that has been assigned from medical device company to translation company to freelance translator. Money changes changes hand and deliverables are delivered.
But what happens when a Google Translate gets used? First off, the system's translation memory contains content from other companies/users. While they have given their permission by agreeing to Google's Terms of Service, the clients in our hypothetical example might not want these translations used in their work.
Second, when a linguist helps improve already existing texts, these translations now are part of Google Translate's corpus as well - even though the end-client did not give permission for this.
This has "ticking timebomb" written all over it...
Want to know more? Here is some additional reading on machine translation:
- Primer: Machine translation vs. translation memory
- "Employee-sourcing" the new crowdsourcing?
- Congrats to the office of the tape recorder!
- Assessing quality in machine translation
Are you on Twitter? Follow ForeignExchange Translations at @fxtrans!
In my opinion Google Translate can be a good tool to assist in translation, but it can't replace a real human. There are many nuances in languages and translation that a statistical or holistic approach of a computer program can't "understand". If you are looking for a quick rough translation, Google Translate is great. For a high level professional translation someone needs to at least edit the translation.
Post-editing of a machine-translated text for flexible languages with a free word order (e.g. Latvian, Lithuanian) may take more time than human translation when you translate it from scratch.
Sorry, but I can't see how anybody can seriously think of using MT for professional applications. This technical invention is great, when you don't speak a language and want to get a gist of a text, written in it. Editing a machine translated text would probably last just as long, as translating it from the scratch (please, google-translate the phrase "from the scratch" and you'll see what I mean).
Check out a tool I have made: Google Translate for Translators.
http://dallascao.com/gt4t
The key idea is to let the translators select which parts of sentences to be translated by Google and by hitting a shortcut, the selected source text is replaced by its translations from google. It can really save time.
If text has a simple grammatical structure, and uses words literally, Google Translate gives satisfactory translations. An article in English was translated into Spanish and into Norwegian using Google Translate. Professional translators evaluated the translations (http://www.international-english.co.uk/mt-evaluation.html).
As Mike says, (but in my borrowed words) if a rose is a rose and nothing but a rose then MT might work. If it goes beyond anything than that, and you have to interpret the text or decide between nuances then MT will have a more difficult time. Part of the thing with MT for me is that you know that something who does not have a clue about languages (a processor) made it ...
@Paulina,
If you google translate the expression "from scratch" (not from the scratch), then the result that I get is correct and helpful.
The title question alone "Is Google Translate accurate enough for professional use?" made me laugh.
Seriously, how could any professional even consider any MT "accurate enough for professional use"?
For example, Croatian is highly inflected language (7 cases, singular-plural, 3 genders). So, any MT translation would require more time to edit then translation from scratch.
Google Translate can help only in short expressions.