Information Age: News, analysis & insight for IT & business leaders

One of the early visions for the Internet was that it would connect organisations and people across the globe, irrespective of their location. But it did not take long for businesses to understand that globalisation meant more than just connecting. If the customer cannot understand the content on a website - or is not confident enough in the language to commit to a transaction - then, even if they are an ideal target for the business, the whole opportunity is lost.

Localisation - the adapting of information for language and cultural differences - has begun to spread across the web, with companies maintaining multi-lingual versions of their web site or multiple versions aimed at different audiences.

Either way, they need to minimise the cost of translation and adaptation of web sites and still ensure they are not missing out on major opportunities by not addressing the customer in a language they understand.

"For me this is just one of the fundamentals of doing business: If I can't understand what you're selling me, I'm not buying," says Don DePalma, the president and chief research officer at localisation analyst Common Sense Advisory.

It is a message that has been well understood at Philips Electronics. It has 50 separate websites across the globe, with content provided in multiple languages. That global presence meant that, historically, any changes made to content would involve up to four months of laborious, manual translation to keep the sites consistent.

The same was true of Best Western. Based on an affiliation of 4,000 independent 'member' hotels, it needed to ensure that all these company's websites were available in seven further languages, including French, Spanish, Italian, German, Chinese, Japanese and Korean. That involved the translation of more than four million words from English into the languages and a continual flow of updates.

Both companies systematised that process and implemented effective translation management software and related translation services, dramatically reducing the translation overheads and simultaneously reducing turnaround times. At Philips, a major change to all 50 sites now rolls out in one month; Best Western reckons it can save $2 million over the cost of traditional translation within the first year of operation.

What is underscored by these two examples (both use a translation package from SDL International), is that the Internet and translation have made uncomfortable bedfellows, and that to create and maintain multi-lingual web sites, companies have to understand the core issues and technologies.

Speaking in tongues

One of the most common associations between the Internet and translation are translation engines such as Systran available for free in a limited form at the company's own site and at AltaVista's Babelfish.

However, most users are well aware of the limitations of this type of translation. The raw, literal result may be useful in giving a sense of the text, but it is still not sufficiently capable of understanding the nuances of language which is required when translating an enterprise-class site.

"Language is a complex beast," says Common Sense's DePalma. "At best, machine translation is a reasonable alternative to zero translation."

Historically, organisations have turned to third-party translation specialists - or if they are large enough, to their own in-house teams - to create the multi-lingual text for content such as finely-tuned corporate messages, tightly-worded legal documents or product manuals where errors might lead to legal issues. But such approaches are time-consuming and expensive, even if the changes being made in documents are relatively minor.

And that assumes they are actually tracking the cost. All too often businesses have little idea of the amount spent on localising websites and corporate documents around the globe, says Ron Rogowski of IT advisory group Forrester Research. "Beyond per-word translation costs, firms incur hidden costs in the form of process inefficiencies, marketing inconsistencies and go-to-market delays."

While 'go-to-market' delays may seem an extreme example, it is far from being a superficial issue, particularly in industries such as pharmaceuticals. Here, as part of the European Union drug licensing regulations, manufacturers are required to submit documentation in multiple languages simultaneously; with millions riding on first mover advantage, any delay in translating the material can be extremely costly. Rogowski proposes that savings can be made through centralising the translation management process and eliminating manual processes where possible.

And it is this centralisation of the translation process that demands a large and sustained input from the IT function.

Translation management needs to be linked into the content management process, thereby reducing the likelihood of duplication and ensuring that workflows incorporate translation in the content publishing process. What organisations do not want to do is re-translate a whole web page each time a small change is made.

Such demands have given rise to a group of fast-growing suppliers promising to ease the translation burden, including SDL International, Lionbridge Technologies, Trados (soon to be acquired by SDL), Idiom Technologies and Merrill Brink International.

These companies typically provide a mix of tools to automate elements of the translation - notably translation memory and terminology management - but also for managing the processes associated with human translation - all with the aim of taking cost out of the equation and minimising error.

Translation memory is fundamentally a file, spreadsheet or database which contains source text and accepted translations of that text. Translation programs can then use the source text to identify either perfect matches or close matches, potentially reducing the volume of text that requires translation.

While direct machine translation has proved too unsophisticated for corporate use, translation memory provides a useful bridge between the worlds of human and machine translation that is enterprise-ready, says Gary Muddyman, CEO of translation service provider Conversis. "If you're being charged per word for translations, this is one way to cut your costs dramatically."

US automotive giant DaimlerChrysler has used translation memory software to make dramatic cuts to translation overheads for its vehicle service manuals. It provides online versions of the 8,000 to 10,000 page documents in Spanish, French, Italian and German - as well as English.

As many models share the same components, translations can be re-used. "Initially we thought we could save up to 30% of translation costs this way," says Paul Mansfield, manager for service information distribution at DaimlerChrysler. "Our initial results indicate the savings we can make will be even higher."

In many cases, it is the service provider that controls the translation memory, explains DePalma, where it is used to strip out as many words as possible before the rest of the text is passed to the (more expensive) human translators. If that is the case, "insist on owning the memory files," he says. "And manage these memory files as a corporate asset."

Alternatively, organisations can control the translation memory database themselves, says Forrester's Rogowski. This can then "centrally track content and filter previously processed words out of pages before they are sent for translation."

According to Forrester, organisations are creating progressively-larger volumes of content; by dint of that, as the translation memory continues to expand the number of words it recognises, cost savings grow (see graph).

To further reduce the volumes of text sent to human translators, companies can also introduce terminology management systems. Much like translation memory, these are a repository of known words - but the terminology management tools are used in the content creation phase.

For example, if a manual for a computer monitor is being translated, the terminology management system can help impose a consistent style, ensuring that the equipment is always referred to as a 'monitor' and not VDU, that the power button does not become the 'on-off switch'. This reduces duplication of the translation effort.

"There will always be some parts of documents that need human translators. But the more we can translate automatically, the more we can reduce the costs of translation," says Terry Lawlor, worldwide marketing VP at SDL International.

Translation management

Automating translation is not sufficient for most organisations: what is needed is some form of automated workflow, says Lawlor. By implanting a translation management platform to work alongside a content management system, businesses can truly reap the benefits, he explains.

Such a platform will apply the terminology management and translation management, within the scope of a well-defined workflow, from file movement, document conversion through to review and publishing. At DaimlerChrysler, once its service manuals are written, the company's Paul Mansfield does not have to worry about them until they arrive back at the other end of the workflow. The documents are then reviewed and approved. "A nice thing for us is that we can get them back tagged in XML, which makes the process of uploading them and searching for them on the website a whole lot easier," he says.

But even here, the most carefully managed process can get complicated. Most web page designs use templates as the cornerstone for ensuring that something like a polished corporate marketing messages has the maximum impact. That work is soon undermined however, when language is translated - for example, the German language contains very long compound words, so on a simple character count, a faithful translation is unlikely to fit a web page design originally written in English.

Again, organisations need to build rules into their translation management systems to resolve the issue, says Lawlor. For example, Philips Electronics puts character limits on translations so that a chunk of 100-character source text is limited to a maximum of 135 characters in translation, capping the amount of space used for translated texts in templates.

As that underscores, there is no 'silver bullet' when it comes to translating web content. But as web commerce becomes an increasingly global phenomenon, it will be an insular business that leaves the polyglot opportunities hanging.