Automated/Machine translation – put to the test
We’re always being asked “why can’t I just use Google Translate/Babelfish/[insert name of machine translation tool here] instead? It’s free!” Where do I start?…
Well, it’s true of translation as much as of anything else: you do get what you pay for. So if you are paying nothing for your translation, you can guess how good (or not) it’s likely to be.
The free automated translation tools can sometimes be very useful for getting an understanding of the text. But if you intend on publishing the text, this is the last thing you would want to use as the automated translations are very literal.
For example, there’s nothing French about french fries, but a translation machine doesn’t know that and you could end up with a very odd text!
Machine or online translation tools are exactly that: a useful tool for trained linguists, or just to get a vague idea of what something is about, but are no substitute for human expertise and experience. A machine will not accurately translate anything more than the simplest phrases, and we’ve all seen the results when what comes out is totally wrong!
If you want to publish or use the text you are translating for any serious purpose, then steer well clear - don’t make your company or brand a laughing stock by cutting corners.
This is not a machine translation-bashing post, in fact I, like many linguists, am intrigued to see how this technology continues to develop, and use it frequently myself as one of several tools to help when I do need to use or understand something in a foreign language.
However, every tool has its correct and proper use – try using a hammer instead of a small screwdriver to repair your glasses, and you’ll soon see what I mean.
Machine translation is a really quick and handy way to decipher emails, websites you are browsing, and similar short snippets where your aim is to get a vague understanding. These tools work best at translating short standard or formulaic phrases, as these are things that can be taught to a machine in the form of rules.
Anything longer than a phrase or short sentence, and you will find that a translation machine will struggle to produce anything coherent.
We put machine translation to the test while working on improvements to our Email-translation.com technology.
Currently, the Free and Light versions of Email-translation.com we offer use automated/machine translation to translate incoming messages.
Here are the detailed results of our findings:
In the pdf comparison you can see red for wrong translated segments, yellow for mistranslations due to wrongly identifying the word order in English, and purple for things that shouldn’t have been translated (as they are company names, book titles…) or things that have not been translated and should’ve been.
June 25th, 2009 at 8:08 pm
read:
Getting started with Machine Translation. In “Guide to Translation” of MultiLingual Computing & Technology magazine, Number 69, Volume 16, Issue 1, January/February 2005. pp 8-12.
Summary: This is a step-by-step guide on how to transition from a free online translation portal like Babelfish or Google Tools to using a commercially MT software package which includes sets of productivity-related features.
https://www.multilingual.com/downloads/screenSupp69.pdf
June 25th, 2009 at 8:12 pm
The examples cited in the article and the PDF containing the comparison of MT raw output vs human translation do not take into consideration that there are MT software packages available that allow you to handle the mistranslations and the not-to-translate company names.
Why would a translator want to spend time identifying all the mistranslations, and retranslating the entire text, when less time could have been spent on creating dictionary entries to modify the translation output? See the MT case study indicated below which provides the examples and gives concrete details on the time necessary to analyze the text and create the dictionary which an MT software package or system can use.
Improved Translation Quality with Machine Translation Dictionary Building. June 2006. Published by Translatorscafe.
http://www.translatorscafe.com/cafe/Articles.asp?ArtID=59
June 28th, 2009 at 5:10 am
[...] Automated/Machine translation – put to the test | Web Translations [...]
June 29th, 2009 at 4:12 am
[...] Automated/Machine translation – put to the test | Web Translations [...]
June 30th, 2009 at 4:39 am
I found similar examples where the machine translators produce rather funny results. Please check out my blog at http://www.jollo.com/blog for more information
July 3rd, 2009 at 6:56 am
Machine translations may be a quick and cheap tool to use on the fly, but it is overly optimistic and wishful thinking to believe that these machine translations will one day come to replace human translations in the future. Like you said, it lacks style and can’t recognize context, which is the heart and soul of language and communication. However useful or easy it may be to get “gist” translations, the most reliable and accurate method of quality communication and exchange is turning to services (such as mygengo.com) for your translation needs. Don’t settle for anything less than 100% accurate translations!
July 10th, 2009 at 10:45 am
@Daniel: A machine will not accurately translate anything more than the simplest phrases… If you want to publish or use the text you are translating for any serious purpose, then steer well clear…
Alternatively, write clear and simple text. For business communication, simple text is good. Usually, if text is simple, machine translation gives satisfactory translations (http://www.international-english.co.uk/mt-evaluation.html).
August 25th, 2009 at 9:25 pm
Learning spoken Chinese language one should know the pinyin method.