We've always used pg_trgm with GiST indexes, with the same sort of materialized view described here, containing unaccented search terms. One of the big benefits

We've always used pg_trgm with GiST indexes, with the same sort of materialized view described here, containing unaccented search terms. One of the big benefits here is we can then do really fast '10 best matches' by text similarity. Obviously this is a slightly different use case than full-text over largish documents, but it's great when you've got a database of foreign names that people only vaguely know how to spell.