[XML-DEV Mailing List Archive Home] [By Thread] [By Date] [Recent Entries] [Reply To This Message]

Re: Trigram tables

• To: <xml-dev@l...>
• Subject: Re: Trigram tables
• From: "Rick Jelliffe" <ricko@a...>
• Date: Tue, 16 Jul 2002 15:52:27 +1000
• References: <200207151841.OAA23879@m...>

From: "John Cowan" <jcowan@r...>

> Richard Tobin scripsit:
> 
> > lost is to the hishart commusted matimany he usider wit darly he the

Richard has never made more sense :-)
 
> It's fairly easy to spoof such tables if you are allowed to construct
> the text and it need not make sense.  Constructing well-formed French or
> Indonesian or something that spoofs the tables will not be so simple.

Of course, one use for trigram tables is for OCR error detection
without the need for a full dictionary.  The most common errors
of English OCR are confusing  0/O, 1/l, i/j and perhaps 5/S.  Trigram tables 
tend to be able to find those kinds of errors, at different implementation complexity 
than a spell check. 

Another use is for trigam indexing, which allows some kinds of faster jumping
within documents.  (For Chinese, bigram indexing is regarded as one of the
most efficient ways to index a document.) Knowledge of the occurrences
of trigrams lets you pick a suitable algorithms. 

There is a lot of academic work on character occurrences, but little of it
trickles down in a useable form (as code or tables) to the developer community.

Cheers
Rick Jelliffe

• References:
  • Re: Trigram tables
    • From: John Cowan <jcowan@r...>

• Prev by Date: Re: Trigram tables
• Next by Date: RE: Trigram tables
• Previous by thread: RE: Trigram tables
• Next by thread: Re: Trigram tables
• Index(es):
  • Date
  • Thread