After @joel’s presentation at the conference last week I came across this article (which may have informed his presentation?).
That is right! I have it linked in the slides along with the link to the paper it was referring to. It is a pretty interesting piece of work.
The headline and strapline of the article don’t reflect the content very well at all. So here’s a brief summary in case it’s helpful. It’s best to test with between 3-5 users then fix the issues before testing again. If there are different kinds of users, e.g. translators, consultants etc, then each of those groups need 3-5 test users.