I was reading the insightful qwghlm (because all the other domain names were taken) recently. I'm a private sudoko fan, I do a puzzle here and there under the covers when R isn't looking. Seriously though, I liked the idea that something like Sudoko is a much more tractable problem to a computerised solution than crossword puzzles. However, I must take issue with Chris Booth's comments.
I love The Independent's "Get the picture" for that reason. But the best example of this class of puzzle (often done for charity) are those where the clues are initial letters and lengths of the key words in a phrase (e.g. "The T----- D--- of C--------"). What fascinates me about them is that there doesn't seem to be any analytical process one can go through to solve them, but the answers are almost always universally accepted as right. And they can pop into my head at any time.
There is a clear analytical process to solve these based on a combination of understanding sentential structure and knowledge of common phrases. The sequence abolve follows the structure determinant adjective noun of (possessive) noun (potentially a proper noun!). Therefore solving it could be achieved using a search program (another opportunity to use the Google API?)
which builds up [Letters|Rest_Of_Word] signatures of common phrases based on verbal categorisation based on accepted English grammar. I'm not discounting the complexity of parsing a vernacular or highly idiomatic phraseology here! I recommend anyone interested in computational linguistics should get the following book: Using Computers in Linguistics: A Practical Guide (Paperback)