Archives

2009
July
May
March
January

2008
November
September
July
May
March
January

2007
November
September
June
March
January

2006
November
June
March
January

2005
November
September
July
May
March
January

2004
November
September
July

Three out of four online translators that we tried did, in fact, translate the Spanish text in this picture as "Two young donkeys, please."

Machine translation (MT), also known as automated translation, has come a long way in only the past few years, as anyone who has been using Google Translate continually can testify. Is the professional translator, an endangered species? Absolutely not. Although MT continues to improve, there are definite limits to how far this technology can take the process anytime soon.

Take, for example, the experience of La Tribune, the French financial newspaper, that introduced a multilingual version of its website using an MT system by Systran to render the site in English, German, Spanish and Italian. The results can make for entertaining – if sometimes confusing – reading.

"[T]he salaried employees raise their threat to make jump the factory," read one recent headline in the English version.

A Little Background
MT systems are either rules-based (RBMT), statistical (SMT) or a hybrid (HMT) of the two. The first attempts at machine translation were all RBMT, with linguists and software developers attempting to encode grammar and syntax rules for different languages and map them to each other. If you think about how many rules there are in English, how many exceptions there are, and how often people break the rules, you'll see why this approach had definite limitations.

SMT involves analyzing large amounts of data consisting of parallel texts in different languages to come up with the "best" – that is, the most likely – translation for any text. It is only recently that the necessary computing capacity has been widely affordable and that massive bodies of data in different languages, like UN documents, have been digitized and made available.

Many recent systems are HMT, but it is the statistical approach that primarily accounts for the improvement in MT technology. Training data can only take it so far, though.

The Obstacles
One big hurdle to MT is the importance of context to meaning. Think of how many words can mean a variety of things depending upon where and how they are used. Some words can even serve as different parts of speech - verb, noun, adjective, etc.

For example, English readers of a La Tribune article in August might have been surprised to learn that, in response to a court decision against the company, Microsoft announced that "it would make phone call." In the French version, Microsoft "ferait appel" (would appeal or, literally, would make appeal) the decision, but the MT opted for the primary meaning of appel, "call."

But the biggest obstacle to fluent MT is the vast differences in grammar, word order and word usage in different languages (and even within languages). Translation is never simply the word for word replacement of one language with another.

Here's just one small example: in a September La Tribune article, the English version referred to Texas Instruments as "he." In French, all nouns are either feminine or masculine and the pronouns that replace them are always masculine or feminine, as well – there is no "it" in French. The MT system has to figure out whether to replace "he" or "she" with it. In the article, Texas Instruments had been referred to as a "giant" of the semiconductor industry; the MT apparently saw giant as masculine. Many more examples can be given.

Remember that French and English are fairly close in structure. The problems multiply when the languages involved in the translation are very different from each other, such as English and any Middle Eastern or Asian language. The fact that all living languages are constantly evolving - meanings, rules and structures constantly in flux – adds one more layer of difficulty for designing MT systems.

Human vs. Machine
Despite its limitations, machine translation is good for many everyday purposes. Unless a text is very complex, MT may very well give you the gist of a document, email or website so that you can determine if it is of interest or extract basic information. MT is used for a lot of material that would never otherwise be translated.

But, if accuracy and fluency of language are important, a professional translation service is the only way to go. This applies to all marketing documents or other materials, such as websites, that help create an impression of your company and brand. A fluent translation, tailored to your audience, is simply more accurate and persuasive.

Product instructions, user manuals, training applications – anything that needs to be completely accurate and comprehensible to be useful – should also be professionally translated. In these cases, MT can sometimes be used to pre-translate a document. Even for pre-translation, however, MT only works very well when it is combined with the use of a controlled vocabulary in a limited subject domain, such as the Simplified English that was originally developed for the aerospace industry. The pretranslated documents still need to be reviewed by professional translators to ensure they are accurate; and, even with a limited subject matter, the MT software must be "trained" using the controlled vocabulary in order to get results of high enough quality to help the translators.

If a mistranslation can lead to legal difficulties (a contract, for example) definitely use a human every time.

This is the situation now and there is no indication that MT will be able to take over from humans for any kind of quality translation in the foreseeable future.

You've laughed with us at the many examples of bad translations that we've highlighted in past issues of this newsletter. Be careful using machine translation or your company may be the one being laughed at.

	Myriam meets a new friend in Japan

In founding MTM LinguaSoft in 2003, Myriam drew from her own cross cultural experiences as an information technology consultant with an international background. Born in Paris, France, she came to the U.S. to get her MBA at Drexel University and ended up staying in Philadelphia. For 12 years she was a successful project manager who implemented several ERP systems and e-commerce systems for both Fortune 500 and startup companies within the US, Europe and Latin America. As a result of this experience, she has developed an awareness of and a methodology for addressing cultural issues that arise in global business settings.

Before getting her MBA, Myriam received a B.S. in Computer Science at a leading French engineering school and a Masters from the European School of Management in Paris (formerly, Ecole Superieure de Commerce de Paris).

Myriam and her husband, Tim, both love traveling and earlier this year went on their first trip to Japan. They also make an annual trip to Paris to visit Myriam's family there. When she is not traveling or taking care of business (usually both at the same time), she likes sharing and trying out new recipes. The staff particularly appreciates her crepes.