I totally agree. An approach with a "learning" heuristic engine is great. Unfortunately, most projects and/or companies don't have the patience or budget.

>>I just don't think it's humanly possible to achieve perfection in something like this because of the nature of the beast. I think, over time, you can develop a rules engine to cover most weirdnesses in addresses but I don't think you can get all, especially in Tracy's situation where she may have to parse "100 Main No W Ste 109" (note the missing street type and malformed suffix).
>
>He beast is highly irregular, and I'm actually amazed that the humans somehow manage to live with it :).
>
>Even in countries where there's more of a centralised regulation on these matters, there are various oddities. When we came here, the street numbers seemed quite odd, until we got the gist of it - and the street where I worked didn't help any, because it had only two numbers: 1 on the right, and 2000 on the left :).
>
>We were used to simple sequential numbering, where #1 would be the first house on the left, #2 on the right and so on. But then if a lot was split and more houses inserted, house 26 would keep its number, and the added houses would be 26a, 26b etc - and sometimes 26c would come before 26. My street was another special case - the first block was supposed to be the warehousing area, so numbers on the odd side started on the second block, leaving first block unnumbered. So my number was 35, and the neighbor's across the street 76.
>
>The specially weird thing were what you'd call projects. The apartment number would look something like 26 L1/U2/S12 (building #26, segment 1, entrance 2, apartment 12). Then they saw that was too complicated to remember and handle, so they started assigning street numbers per entrance, where all the previously assigned numbering schemes had to be renumbered... and a few hundred thousand people had to oPaperwork.refresh().
>
>So, back to Tracy's problem - in her place, I'd write a lookup table with all the forms of an address, which would point to the exact address. The record for "100 Main No W Ste 109" would lead to "100 NW Main St Apt 109", so next time it appears, wouldn't need to be parsed again. Gradually, that'd lead to close to 99% of hits.