The problem remains that until you can describe exactly what a "sentence" is, you cannot hope to instruct the computer to isolate it for you.
Say if you came across a badly written sentence with 2000 chars you wouldn't want to display it all would you?
If you came across a sentence which was 2 words, that would not transfer much information to the user either would it?
What you are trying to do is also described as creating a "[Document Surrogate" one of the things I found out in [URL="http://paulgeraghty.posterous.com/search-patterns"]search patterns](http://searchuserinterfaces.com/book/sui_ch5_retrieval_results.html).
There seems to be two ways to go on this, you either;
a) try and constrain your mysql full text search in the database first - only bring back x chars from the table
b) bring back everything from the table
if you decide on a) explode on . and choose the array item which contains the word.
If you elect for b) you potentially are able to show EACH full-text scoring term, eg if you have 2 articles mentioning virus, the first says virus once but the second says it twice, then mysql will score the second article higher than the first - so shouldn't you display BOTH words in some context? e.g.
1) I was going to buy a virus checker and thought, hell why bother? Just connect to the web when they are asleep. Take that virus."
2) "I think I caught the virus when working in the laundry, all those sleeves."
Its less about sentences, more about what confers the most, yet somehow manageable information to your users.