That change won't do much, as collation only has to do with how data is sorted.
Yes, you definitely have a problem in your database, as seen in the XML file: except for the second question, all the others have erroneous extra binary characters: ┬á in the first and third, and ┬áÔêÆ in the last one.
It certainly seems that you have what is known as a "doubly encoded" database, which was caused by having set wrong parameters during various export/import processes.
Ok, that is the problem. Now, let's see about the solution.
To begin with, there are two ugly truths you really need to be aware regarding this kind of problems: 1) they do not have a one-step-do-this solution; and 2) they definitely need to be solved by someone with experience in them, that is, this is not the kind of issue to be left in the hands of a non-IT-technical person.
If it were me, I would start by trying to see if there are any source files (most certainly, Word files) from which the quizzes were made. This in order to see if the questions include some extra uneeded characters at the end and to determine which are they (in case there are).
I would also try to see if there are any database backups (asuming they are in text format) which could be studied and compared. This in order to see if the problem appeared at a particular point. Backups may also include information regarding the character set.
From the possible information gathered, a strategy would then need to be formulated.