Full Text Search - i18n

Apologies if this question has been previously covered, but I was not
able to find something similar in any of the mailing list archives.

With full text search, if you need to support a table where the content
of individual tuples/rows may be in different languages with the
language of the content known, and the content of any individual row
does _not_ contain a mixture of languages, what is the best way to
design for this?

Would you create a different GIN/GiST index of the tsvector for each
language? Or if using columns to store the tsvector, would you have a
different column per language, or the one column with the associated
trigger using the appropriate language when building the tsvector for
that row?

Are there better alternatives than what I have listed?

Thanks,

Andy