Machine translation: integrating automation into an effective localization strategy

By Jennifer Beaupre,
Senior Marketing Manager,
SDL

Become a member of TranslationDirectory.com at just $12 per month (paid per year)

Market Trends

As the pace of globalization increases, companies are faced with new challenges including staying ahead of the competition by reaching global markets faster, communicating with global customers in multiple languages, and the increasing cost and complexity of doing business on a global scale. Yet at the same time, selling products and services to a global marketplace introduces new costs and challenges with varying language and cultural requirements for each target market.

Since communicating with global customers in their own language is a key differentiator for global enterprises, how can companies communicate in multiple languages and keep their costs down?

One solution is machine translation (MT). MT in its most basic form is the use of computer software to translate text or speech from one language into another. While this seems to be a relatively simple concept, language translation can be a complex, cognitive process requiring in-depth knowledge of the grammar, context, and culture of the speaker. Therein lies the challenge of machine translation. How can computer software understand the source language as a person would, and how can it translate into the target language so that it sounds like it had been translated by a person?

Approaches

There are several types of machine translation solutions that can produce translated content with varying degrees of quality. The first form of machine translation is a Rule-based MT, a system that processes translations based on linguistic rules and dictionary databases. Other forms include Statistical MT, Example-based MT and Hybrid MT (Statistical and Rule-based MT) solutions. These solutions include a probability based approach that detects patterns in large quantities of human translated content.

These approaches on their own translate files that contain text in a computer-readable format, such as plain text files, HTML, MS Word, XML, etc. This basic level of MT is extremely fast and can provide a “gist” of the meaning of the original language. However, these approaches do not on their own produce high quality translations that can be used for every type of content across a global enterprise. MT is not a replacement for human translation; yet, it can be a replacement for no translation at all.

However, MT, when integrated with translation memory, terminology management, and human post-editors, can produce publishable quality output. A key component in improving the quality of MT is source language consistency. If the source content is inconsistent in its use of terminology, the output of translated content will not be high quality. Real quality improvements can be achieved by integrating computer aided translation database memories, or TM’s, with MT and then post-edited by a human translator. When translators are trained to check the source content, work with TMs and to post-edit the machine translation, the result can be high quality, publishable translations faster and cheaper than traditional human translation.

When to use MT?

One of the challenges in deciding when to use machine translation is that different content types have different quality requirements. Marketing and advertising materials are highly nuanced and emotive to support the overall corporate brand. Product, legal and HR documentation and user manuals need to be factual and accurate. Large volume content with a small potential audience or occasional readership lends itself to MT, in that it will provide for some degree of understanding and allow for human translation of the specific content if required, while keeping costs low. Additionally, MT can be very useful when a person only needs to get a “gist” translation, such as communication via instant messaging or email.

How can companies decide when to use MT and what kind of MT solution should they be using?

Baseline MT

Baseline MT or Raw MT, is the most cost effective use of machine translation. The end user sees un-edited MT translations. This is best used when errors can be tolerated or when the alternative is no translation at all. For example, a foreign language email may need to be understood, yet the reader does not have the time or resources to get a high quality translation. Running the email through a quick translation engine, such as BabelFish or FreeTranslation.com, can provide a fast translation that provides a “gist” meaning that is sufficient to understand the nature of the email.

Trained MT

Trained MT solutions combine linguistic resources such as TM’s, dictionaries and managed terminology to provide higher quality translations for less cost. Over time the system can be customized and “trained” to produce more reliable results. The user sees un-edited translations that have been processed through a “trained” system. This approach can provide reliable results, but does not avoid errors and variances without human post-editing. This approach is good for translating large volumes of content that would normally be slow and costly to translate using a traditional human method.

Post-edited MT

Post-edited MT output can match the quality of human translation with significant time and cost savings over traditional methods. SDL’s Knowledge-based Translation System (KbTS), for example, combines translation memory, automated translation, terminology extraction and dictionaries with trained, post-editors to check the source content and edit the MT output. The solution tightly integrates the TM technology, the MT technology, and human translation skills. The result is faster, high quality translations at a significantly reduced cost. When accuracy is critical, it is essential that MT output is edited by trained post-editors to maintain superior quality.

Why use MT?

Delivering multi-lingual content can present a number of challenges to global enterprises. Most companies have large volumes of content such as product documentation, technical guides, web content, support materials, sales collateral, and much more.

Additionally, they have pressure from their stakeholders to release new products and information quickly. Global markets demand that this content is available in their native language and companies that meet that demand succeed above the competition. However, for such large volumes of content, traditional translation methods can take too long and be very costly.

Another challenge for global companies is a distributed workforce. When employees are living and working in different countries and speaking different languages, internal and external communication can become challenging. MT can address many of these challenges, especially when it comes to translation speed, cost and consistency.

The main reasons to use MT are speed, cost savings, and consistency:

Accelerate time-to-market- Oftentimes, traditional human methods cannot translate content fast enough. Global companies are experiencing more pressure than ever to deliver new products to market quickly. However, localizing the content for software, user documentation, and marketing and support materials can delay the release. Using MT can shorten the translation cycle and speed time-to-market for localized content.

Decrease costs- By reducing the need for human involvement in the translation cycle and accelerating time to market, MT can significantly reduce costs. For raw MT, the human translator can be eliminated and translations can be provided instantly. A tightly integrated MT solution with human post editors combined with language technologies, such as SDL’s KbTS, can save up to 40% over human translation alone. The system is trained over time, allowing for continued efficiency and quality improvements.

Improve translation consistency- When MT is integrated with TM’s, dictionaries, and human post-editing, the consistency of the translations can improve over time. Since the MT software improves with high quality source content and integrates corporate terminology, the output is superior to traditional methods. Additionally, with some models, post-edited content can be put back into the MT system and it can “learn” the correct terminology.

Summmmary

Global enterprises are experiencing tremendous pressure from stakeholders to introduce new products on an international scale quicker and more effectively. At the same time, they must always look to improve cost and time deficiencies while maintaining the highest levels of customer satisfaction. These pressures and concerns present unique challenges for companies who communicate with global customers across multiple languages and cultures. To effectively meet these challenges, companies are continually combining MT solutions with their existing localization processes to help them localize their content more efficiently and for less cost.

Those companies that are able to implement business processes that are faster and lower in cost, deliver better customer service and support, and improve productivity among their employees, will succeed in the marketplace. The use of MT alone or in combination with other technologies, such as translation memory, terminology management systems, and human post-editing, can provide significant cost savings, accelerate time-to-market and provide high quality results. In sum, global enterprises can greatly benefit from including MT into their overall localization strategy.

Published - February 2010

Submit your article!

Read more articles - free!

Read sense of life articles!

E-mail this article to your colleague!

Need more translation jobs? Click here!

Translation agencies are welcome to register here - Free!

Freelance translators are welcome to register here - Free!