Speaking at a Search Engine Strategies conference last year, I said you should never translate your keywords or page titles when working on internationally targeted sites. A member of the audience publicly blogged the presentation, accusing me of having lost the plot.
Apparently I advocated "using English page titles" instead of the appropriate foreign language. Nothing could be further from the truth.
But this illustrates the point that "translation" has many different interpretations. By "translation," I mean taking an original text and working with tools (e.g., dictionaries) to find the closest possible match or equivalent meaning in the new target language (the language we're translating into).
By definition then, surely it's a simple matter of taking the original words and finding those equivalent words in the new language, no? Hell no -- that isn't even a good description of the translation process -- but for keywords that's positively unequivocally indubitably criminal.
Why doesn't translating keywords work? Because keywords are the fruit of a language, hanging on the branches of trees that grew and were nurtured in the local climate and are rooted in the local culture. As markers of someone's intent when they search -- they spring from local habits and behaviors that will vary from country to country -- or even region to region.
Compare the U.S. and U.K. use of English -- the same language. In the U.K., we're in the habit of saying "holiday" when folks in the U.S. would say "vacation." For "football boots" Americans would say "soccer cleats" (Can someone please tell me what a "cleat" is? Because, as a sailor, I think that's somewhere you fix a rope to stop it slipping). And as for baseball, well that's just not cricket.
How does this work when two different languages are involved? Let's look at the term "car insurance."
Google Translate gives an accurate translation with "l'assurance automobile." When you then ask the Google Keyword Tool to give you the volume of searches for "l'assurance automobile" in French, in France, it responds with "not enough data." You have to use "assurance automobile" to get a measly 40,000 searches.
That's strange, because Google's Keyword Tool also tells us that more than 6 million people search for "car insurance" monthly in the U.K. and the next most popular term is "cheap car insurance," with 246,000 monthly searches. Given that the population of the U.K. is very similar to France -- and both need car insurance -- there must be some searches for it somewhere?
It turns out that the most popular keyword, by far, in France is "assurance auto" with just 301,000 searches. "Assurance voiture" has an additional 145,000 monthly searches.
But then there's a wide variety of other forms of search for the equivalent of "car insurance" that you would need to use in order to be successful at reaching the market. Choosing the correct, more popular term over the translation would give you at least seven times as many searches.
This is one good example of how the translation problem works. "L'assurance automobile" was the correct translation and was given because Google uses data from its index to power its translation systems using statistical analysis.
That analysis will have shown it that Web sites in French tend to use "L'assurance automobile" when they mean car insurance. Quite right. French insurance firms want to use the most correct terms on their Web sites and not necessarily the most popularly used keywords.
English-speaking companies do the same, but "car insurance" happens to be correct in English. Thus, it appears on English Web sites, enabling the match with the French term. However, when searching the French, do what we all do and use everyday language as well as common abbreviations for popular words, such as reducing "automobile" to "auto."
So let's imagine you'd simply translated the keyword. Your U.K. PPC campaign might well have shown that just using "car insurance" generated considerable success, whereas in France, there was no interest. In other words, you'd have drawn the wrong conclusions.
Later, if you were very lucky, you would've realized that you had used the wrong keyword and needed to expand the list more widely to achieve the level of interest you'd expect. From a SEO perspective, you might well have achieved your targeted rankings, but failed to see a good business result.
So keywords can't be translated because they've evolved directly to serve the everyday needs of the people of a particular country. Translation can't predict that.
The only solution is to use a native-speaker to conduct research and use those keywords to craft your page titles, description, and keyword tags, as well as your PPC keywords and accompanying sponsored links. These should all be in the same language as the main body content. Leaving page titles or meta tags in English on non-English pages is never cool.
Tomorrow I'm speaking on the international aspects of keyword research at the SES San Jose. I've considered asking members of the audience to confess their sins and put their hands up if they've ever translated keywords, but realized it probably isn't a good idea to ask for participation and then publicly chastise the victims. So don't worry if you're there -- you can tell me your international horror stories in private.