Translution logo

www.translution.com


Translation Quality

Translution delivers the best possible Machine Translation available today.

Follow these links to learn more:

How and for what purpose should Machine Translation be used?

The vast majority of communications in business these days is by email, often with attachments. Translution provides a completely automated translation system which is perfectly suitable for day to day email collaboration and communication without language barriers.

Users can now communicate in their own language and have it automatically translated in to the language of the recipient, without changing in any way the way that they work. For the first time you will be able to sell in the language of your customers and buy in your own language regardless of the language of the seller.

  • Translution's technologies are suitable for communication between businesses and individuals who want to collaborate at a much deeper level because both parties can express themselves in their own language. Examples would be colleagues in offices overseas, suppliers and distributors.
  • Translution's technologies are suitable for communication between consumers. Now you can communicate instantly with friends and acquaintances throughout the World regardless of your language skills. Examples would be interest groups and classes of school children wanting to send emails to penpals.
  • Translution's technologies are suitable for communication from consumers to businesses. Now you can communicate with foreign suppliers, book hotel reservations, and even deal with your plumber and estate agent for your overseas holiday home.

Where you have to be careful is in using Translution for communications from your business to consumers. Whilst the technology can be used with consumers that you know well e.g. a property owner that has contacted you using Translution and you are the local plumber, it shouldn't be used for translating marketing literature or even your website. You need human translation for this.

Having said the above, where it can be effectively used is for tightly defined and technical areas such as customer support, where you have invested in building the terminology to improve translation quality. If you have these sort of requirements, contact Translution and we can help you deliver this.

Back to top

So How Good is Machine Translation Today?

Machine Translation ("MT") is perfectly suited to casual communication like email and for browsing the web. Its fast and its inexpensive, in comparison to human translation.

We have run extensive tests on MT suitability for most day to day communication with Professor Tony Hartley of the Centre for Translation Studies at Leeds University and hundreds of human evaluators.

His conclusion was that MT, in the vast majority off cases, gets over the meaning perfectly adequately. Where it often falls down is in the fluency of the email. Don't expect a translated email to be as easy to read as an email in your language. However it will usually get over the meaning accurately with a little study and perhaps further clarification. And because we always include the original text as well as the translated text, you can always get your dictionary out if you are not entirely sure!

It's also very useful for browsing the web. There are already 3bn translations every month using MT technology, many to translate web pages in foreign languages. That's 4 for every person that uses the internet! It can't be that awful if so many people use it - unless of course you are one of those people who expect a machine to provide a perfect translation every time! You will be disappointed.

In actual fact the translation quality can also be improved dramatically.

Back to top

How Can Machine Translation Quality be Improved?

In three ways:

  • Firstly write grammatically correct, well spelt, short sentences (5 - 20 words) without using idioms or catch phases.

    The first rule of computers is Garbage In, Garbage Out (GIGO). And this is the same for translation.

    We all communicate in different ways. For instance a teenager will communicate using SMS in ways that an adult would never understand. And we all use words that have particular meanings to us.

    This is a big problem for machine translation systems. They are simply not intelligent enough to understand all the nuances of communication between people.

    However if you try and follow the translation quality tips we provide, you wont go far wrong.

    And don't try and re-translate already translated text - it's bound not to work under the GIGO rule.

  • Secondly use the tools provided to improve translation quality. These are as simple to use as a spelling checker, and with a small amount of thought and effort over time, you will both personalise and improve your translations and your communications with colleagues that don't speak your language as well as you do.
  • Thirdly use dictionaries to collect your specialised terms and terms that you don't want translated.

Back to top

What Tools do Translution Provide to Improve Machine Translation quality?

Included in all Translution Products are the following tools:

  • User Specific Dictionaries
    We each use words and phrases that, to us, have particular meanings. Examples of this are parts lists, where a specific translation is required for an individual part description. You can set up your own User Specific Dictionary which controls how Translution Central handles the translation of those specific words.
  • Departmental Specific and Corporate Specific Dictionaries
    With Translution Business and Translution Corporate your organisation will be able to develop its own Dictionaries which determine how specific words are translated and which can be shared with the whole organisation.
  • Handling Words that should not be translated
    There are some words that you won't want translated; for example your address or a proper name. Translution allows you to mark up text within a specific document that you do not want translated. It also supports Do Not Translate lists i.e. words which you or your organisation would never want translated.
    Note: Recent research undertaken by Professor Tony Hartley of the University of Leeds shows that implementing Do Not Translate lists improves local translation accuracy by, on average, 23%. (Ref: Improving Machine Translation Quality with Automatic Named Entity Recognition, EAMT Workshop at EACL 2003 Budapest 13 April 2003).
  • Handling "Not Found" Words
    Machine Translation (MT) engines are usually supplied with dictionaries of the most common words used in a language. However no MT engine could possibly include all the words in a language, particularly "rich" languages such as English, French and German. Until these words are translated and added to the dictionaries of MT engines held centrally, they will not be recognised by the translation engines and will simply pass through un-translated.
    As part of its ongoing commitment to improve translation quality, Translution collects and analyses all Not Found Words generated by all users and, after researching them, continuously adds selected new words to its dictionaries. In this way overall quality of translation will be improved for all users of Translution products.
  • Managing Already Translated Text
    It is essential that already translated text is not re-translated as this reduces overall translation quality. This is important, particularly when replying to or forwarding emails across language barriers. Translution ensures this does not happen by always including the original text with the translated text, and by identifying already translated text in email replies and forwarded messages.

Back to top

Translution's Quality Mission

  • License the best available machine translation (MT) technologies for specific language combinations rather than relying on a single MT engine.
    We used the services of the Centre for Translation Studies at Leeds University under Professor Tony Hartley to evaluate five of the leading MT engines, including a statistical MT system, and have chosen the best of these to support our initial five European languages.
  • Integrate and use multiple MT engines to provide (eventually) a wider range of languages and combinations than has been available previously.
    Currently we only use one single MT. However we are planning to introduce additional languages and language combinations which will require investment in new MT engines.
  • Keep pace with Machine Translation technology change as it improves.
    MT technology will improve dramatically over the next few years and Translution will be at the forefront of these developments.
    We have already evaluated statistical Machine Translation as it holds the promise that dramatic improvements in machine translation can be achieved. Unfortunately, today, for general purposes, this technology does not deliver any improvement in translation over the rules based systems we use. We are however continuing to work closely with Statistical MT suppliers, and when the benefits can be demonstrated, we will integrate their technology.
  • Provide additional tools to improve overall quality for all our users.
    Our products include a number of tools that have been demonstrated to significantly improve translation quality, and more are planned. They have been designed with the non-linguist in mind but are suitable for use by professional linguists too.
  • Provide additional dictionaries and other content to improve translation quality.
    We have already started to build two additional resources for the benefit of all our users. The first resource will hold words and phrases that have not been correctly translated or are not in the dictionaries used by our MT engines. The second resource includes a number of common idiomatic expressions. We are also currently working on other dictionaries and linguistic resource to improve overall quality for all our users.

Back to top