>Thanks for the suggestions. I'll experiment with it although I'm afraid that the eventual SELECT that matches records from the original table with the ID's of the matching records will slow things down. I'll let you know how I make out.
Peter,
It's an assumption. However I did it in real world data and a 8086 returning results in max 30 secs on 75K records is not slow from my POV.
As I stated it's not the original code I used and I coded it here off the top of my head just to give the idea. If creating too many indexes is a concern than you have to find another way.
I also tried a hashing schema (all words kept in a dictionary with a pointer list of all records they occurred). But there was a problem, if searched thing was composed of words then I had to not only find where those words occurred but in requested sequence as well (not commenting on maintaining the pointer list).
PS:If you own phdbase than I would go with it.
Cetin