Machine translation

I played around with Google’s translator a little after adding some notranslate directives as discussed in my previous post. Google did honor my requests to mark some sections as literal text to not be translated. Google’s translator was also able to recognize my name as a name without special markup. Yahoo, on the other hand, translated my name, turning “Cook” into “Cuisinier” in French.

Google treated text inside <code> tags as literals that should not be translated. That is, Google would leave my source code snippets alone and only translate the English prose surrounding the code. Yahoo, on the other hand, would translate everything, including source code. For example, I had some PowerShell code on my page with the keyword matches that Google left alone but Yahoo translated into “allumettes,” presumably good French prose but not a legal PowerShell keyword.

One puzzling thing about the Google translation engine was that it would change which text was hyperlinked. For example, the text “My résumé” was changed to “Mon CV,” linking on the translation for “my.” Yahoo produced what I expected, “Mon résumé.” There were several other instances in which Google produced odd links, such as hyperlinking the | marker between words that were linked before. For example, the footer of my website has these links:

Home | Sitemap | My blog | Search

Yahoo turned this into

Maison | Sitemap | Mon blog | Recherche

while Google produced

Accueil | Plan du site | Mon blog | Recherche

So Google incorporated the separator bars as part of words, and moved the last link from “Recherche” to the bar separating “blog” and “Rescherche.”

One advantage of Google’s translation is that it lets you hover your mouse over a line of translated text and see the original text.

One problem with machine translation is that machines don’t know when to stop translating. For example Yahoo’s Babel Fish translator translates my last name “Cook” literally to “Cocinero” in Spanish and “Cuisinier” in French.

Today Google announced a way to tell its translator that text should not be translated. Place such text inside a <span> tag with the attribute class="notranslate". I tried this on a web page that explained that a certain piece of code printed out “Hello world.” Since “Hello world” is literal output, it should be left untranslated, not turned into, for example, “Bonjour le monde.” The solution was to modify the HTML to say

The code above prints “<span class="notranslate">Hello world</span>.”

To prevent an entire page from being translated, add the following tag in the <head> section of the page.

<meta name="google" value="notranslate">

I suppose other machine translation efforts, such as those from Microsoft and Yahoo, will follow Google’s lead and support the class=notranslate directive.

Comparing Google and Yahoo automatic translation

Giving hints to automatic translators