When it comes to multilingual sites, I am not that knowledgable but it is important as more and more companies and web sites need to go multilingual.
A Google Webmaster Help thread has an SEO claiming there is a bug in Google but the reality, Google is confused by how the content is written and the URLs are designed.
The specific example is this URL, which has the eng in the URL path:
Now, the snippet in Google shows mixed content, both French and English.
John Mueller explained the issue is related to having primary content untranslated in many cases, making Google think they should not necessarily trust the href lang implementation. John wrote:
One of the main issues we noticed here was that the primary content is often untranslated. Taking a random URL of a property, and swapping out the language code, results in a translated template, with some of the attributes translated, but the main description remains in English. This results in us assuming that these URLs are to some extent equivalent. Having the full content translated would be the optimal solution, since it would allow us to really treat these pages separately. If that's not possible, and if the choice of canonical that our algorithms pick isn't your preference, using the rel=canonical (in addition to the hreflang) would help push for your preferred canonical version. The hreflang in this case would still show the best-fitting URL to the users, but the snippet would at least be in your preferred language. Personally, that feels a bit suboptimal to me, so if at all possible, I'd recommend working to make sure that the main description is also translated.
Forum discussion at Google Webmaster Help.