Online machine translation software: advice needed
Thread poster: Zoe77
I have been asked to find an online machine translation software for my company. I realise that machine translation quality is often poor; I will be revising the translation myself afterwards. Can anyone recommend a one of the better free online softwares? Is Google Translator any good? Otherwise, if we were to purchase one, which would be the best one to buy?
Thanks for your help!
[Subject edited by staff or moderator 2007-07-06 21:59]
[Subject edited by staff or moderator 2007-07-06 22:00]
| | Owen Davies
Local time: 13:40
Japanese to English
I use these guys
a lot when checking localized text for European languages. Seems to be pretty reliable when it's only a word or two. But as you say, for complete text you need a translator. The site is powered by Systran and a colleague of mine has used the Professional Systran Suite to translate entire technical documents and then edit them. He seemed pretty happy!
Hope that helps.
| | Heike Behl, Ph.D.
Local time: 21:40
English to German
| Be careful what you ask for... || Jul 9, 2007 |
Post-editing machine-translated text into a translation with a quality comparable to human translation might take a good deal longer than translating it from scratch. Post-editing can also get very tricky since MT can deliver structures that look perfect but are simply not idiomatic. These types of errors, which you as human translator wouldn't even make, are much harder to spot than obvious mistranslations or grammatical errors. You might end up having to do much more work in much less time since your company is convinced that most of the work has already been done by the MT system.
For what purpose would your company be using the machine-translated texts? For what kind of texts? For what target audience?
The best use of MT (particularly with systems right out of the box or online) is for "gist translations", for instance to get an idea what a specific text is all about or to sift through huge amounts of text in order to find those 5% that are important enough to be translated by humans.
Some systems are fairly customizable; however, the better and smarter the system, the more inside knowledge (both linguistic and of the inner works of the system) you'd need to take full advantage of this option. Adding incomplete or incorrect information in user dictionaries, for instance, can cause lots of problems.
Free online TMs tend to be the low-end versions of commercially available MT systems. You should be able to find better systems if you're willing to pay for them. But most upgrade versions of systems available online won't be much better than the free version since they usually are based on the same translation engine, they just have some additional options.
I wouldn't rely on recommendations alone since the usefulness and translation quality of the individual systems depend of course on the type and quality of the input text. You'd have to do some testing on your own to find the best match.
That your company asked you to find a system when you have no previous experience and no background in this area seems to indicate that they themselves have no idea about MT, its shortcomings and advantages. Maybe the first thing would be some kind of reality check to find out what exactly their expectations are and whether these can actually be met by any MT system. Particularly when they're only considering free online MT systems...
| || || |
| | nickiy
Local time: 06:40
English to German
A couple of months ago I came across a company which provides (instant) machine translation. They've implemented the multilingual-tool "translate" into MS WordXP, MS Word2003 and upwards.
I'm not sure whether this is the machine translation tool that also Microsoft uses for their (instant-)website translations.
I haven't checked the quality of the tool yet and I'm also not quite sure whether it's for free.
| | Carol Chen
Local time: 12:40
English to Chinese
Agree with both of you.
Firstly, when I get started with translation, I wanted to rely on translation websites for automatic translation. However, then I realized it was not what I wanted. It deviated from translation itself. Translation actually covers various fields. Precise translation should be obtained through elaborated work, repetitious courses and discussion with experts, greatly out of ability of these websites.
| | Jeff Allen
Local time: 06:40
| answers to several MT related questions in this thread || Jul 14, 2007 |
Sorry to be late in replying. But this information should help. I'll cover several points in posts above in my replies below.
1. questions to ask about even considering MT
Heike gave a good start of questions to ask.
Having worked for several MT development and deployment/professional services organizations over the past 12 years, I've seen a wide variety of contexts for using MT.
I provided a complete list of questions which I ask for MT probable situations in a powerpoint presentation on this topic.
Inbound versus Outbound Translation. Presented at the Localization World conference, Bonn, Germany, June 29 - July 1, 2004. by Jeff ALLEN.
Don't try to postedit Google Tools or Babelfish output. That is not what those online systems are designed for. See posts above, especially the Inbound section of the Localization World conference presentation.
Postediting can be effective and provide very good productivity results for a variety of contexts. Statistics and case studies provided in:
Getting started with Machine Translation. by Jeff ALLEN. In the special supplement "Guide to Translation" of MultiLingual Computing & Technology, Number 69, Volume 16, Issue 1, January/February 2005. pp 8-12.
Postediting: an integrated part of a translation software program. By Jeff Allen. In Language International magazine, April 2001, Vol. 13, No. 2, pp. 26-29.
What is Post-editing? By Jeff Allen. Translation Automation Newsletter, Issue 4. February 2005. 2 pages.
Case Study: Implementing MT for the Translation of Pre-sales Marketing and Post-sales Software Deployment Documentation. By Jeff Allen. 6th Conference of the Association for Machine Translation in the Americas, AMTA 2004, Washington, DC, USA, September 28-October 2, 2004. pp 1-6.
I also have a much more complete and more recent case study that I cannot post here, but for anyone who wants a copy of it, just send me a request by email via the ProZ system.
The real key to successful postediting is upfront dictionary building. This is described in the articles above and many other articles at http://www.geocities.com/mtpostediting/ as well as posts on ProZ:
general vs domain-specific machine translation dictionaries
machine translation and noun compounds
and more examples of productivity and a link to Lorena Guerra's MA thesis on benchmarking Human Translation against MT + postediting is at:
As for the question about Babelfish, I've posted quite a number of replies on this topic on ProZ before. Here is one:
Babelfish is not a system, but instead a portal powered by an MT system
4. Google Tools
See links above.
However, also add to it that some of the Google Tools language directions now include their own in-house built Statistical MT engine.
5. WorldLingo and Microsoft.
The WorldLingo system is integrated into the MS Office menu
This is not the same MT system as the one developed by Microsoft's Natural Language Processing dept which is used for their own Help/Support knowledge base web site.
6. IM Translator
This is a chat interface built on top of the PROMT MT online system
Hope that helps.
| || || |