The glossary upload feature assumes the last two columns of a csv are PoS, Description - and are valid.
Even if this is not the case, the glossary will store whatever is in those last two columns into the glossary term fields.
It should be smarter, knowing at least which fields are PoS and Description.