Translation by Machine or Human?


Let's assume you have an English-language, web-based business, including downloadable documents. To expand, you have to reach out to non-English-speaking customers. You must translate and transform your content for non-English speakers. When you get into the details, you realize this is complicated: Numbers may need different formatting, print and web layouts will likely need to be adjusted, cultural differences can be a mine field, specialized technical terms require deep expertise, and myriad other issues. You quickly realize that localization is a lot of work, and you may have to repeat it for each target language or culture.

Where do you start? You must choose between automated and human translation-or some combination of the two. Automated translation is fast but generally low quality. Using Google Translate, I round-tripped a former General Electric tag line: "We bring good things to life." I translated that from English to French, and back to English to see what, if anything, changed. The round-tripped English was, "The good things in life." Clearly, something was lost in translation.

Human translators provide top-notch results, but only work 8-hour days and can be quite expensive, especially if knowledge of specialized terms is required. So can you create a middle ground by improving the automated translation process and combining it with human review?

I discussed these issues with Lee Densmer, senior manager of new products at Moravia, a firm specializing in translation and localization services. Before going any deeper, Densmer pointed out that these three terms mean different things: translation, localization, and internationalization. Translation means simply replacing words in one language with those in another, as in my Google Translate example. Localization has a much broader scope, translating all the elements in a document, product, or website: text, graphics, idioms, and cultural references. You must localize for each language. This can be quite expensive, especially if the source text and elements are not designed to facilitate reuse. That facilitation is internationalization, and done right and early in the process, it can reduce the repetitive localization costs and resources applied to each language and culture.

Internationalization is holistic and identifies issues and weak points in everything from source code to formatting and design, striving to make these as easily localized as possible. You must consider your product's content. Is text embedded in graphics? Is the content highly technical? How much text is involved? Even the structure of calendars and how numbers and time are formatted is important. And you must review the look and feel: How does the text flow from column to column, given the use of different fonts and languages?

Internationalization requires a host of language, editorial, design, and testing skills. Since you don't want to repeat the same steps unnecessarily, the best approach is to create your content in as much of a context-neutral fashion as possible. That could mean tagging dates, so they are easy to spot and localize. And it could mean using a neutral fabricated language, if one is available. For example, there is a neutral version of Spanish that avoids idioms and local terminology. Since roughly 26 countries speak some version of Spanish, localization would require adapting that neutral Spanish to a specific country's dialect. Unfortunately, this neutral language is very specific for certain kinds of technical text, not for marketing or other content. Still, working from a neutral version of a language to culture-specific versions beats re-creating text from scratch. Moravia's approach to localization involves both human and translation tools, including a fee-based version of Google Translate as well as similar tools from Microsoft. Also, if you expect to localize frequently, you should build a full life cycle business process starting with analysis and going through quality assurance test, review, and deployment. Life cycle processes are neither trivial nor cheap.

Given both the potential benefits and costs, how do you proceed, especially if you are just getting started? As with any complex project, you must carefully analyze the effort, the size and type of your content, how much quality you need, and how much you can afford to spend. Do you need to internationalize 200 words just once and can tolerate a result that will be recognized as imperfect? Try this yourself and, perhaps, have the result reviewed by a knowledgeable editor. Is your scope larger, and will it need to be repeated for many languages? Better develop a full life cycle with all participants on board. The good news is that over time, you can reuse your translations and experience to reduce future costs.