LOTR Meets Monty Python: The hilarious saga continues in this thread from lotricons. shiny_emma is quite inspired.
The Twelve Days of Christmas: Here are the twelve songs (carols, hymns, and assorted other goodies) I presented over the last 12 days.
- 1. 25 Dec 2003 - How Great Thou Art
- 2. 26 Dec 2003 - The First Noel
- 3. 27 Dec 2003 - The Last Unicorn
- 4. 28 Dec 2003 - Hark! The Herald Angels Sing
- 5. 29 Dec 2003 - Away In A Manger
- 6. 30 Dec 2003 - Good King Wenceslas
- 7. 31 Dec 2003 - Un Flambeau, Jeanette Isabelle
- 8. 01 Jan 2004 - I'm Dreaming of A Black Downfall
- 9. 02 Jan 2004 - The Out of Tune Song
- 10. 03 Jan 2004 - Do You Hear What I Hear?
- 11. 04 Jan 2004 - O Come All Ye Faithful (Adeste Fideles)
- 12. 05 Jan 2004 - It Came Upon A Midnight Clear
Banazir, the Movie: First, the adult action film...
Banazir in Bad Blood Loss
|In this thrill-per-minute rollercoaster, banazir (Tom Arnold) is a naïve cop with a pet squirrel. He elects to get to gondhir (Robin Williams) before his archenemy, jereeza (Jennifer Aniston), outwits him. Following a conversation that seems to justify all violence, he attempts a break-in at a hidden facility. From there on in it gets messy.|
Banazir in Seen Girls
|banazir (Jake Lloyd) has one last chance at saving his seat on the school council. With the hopes of the school pinned on one basketball game, his lack of schoolyard weaponry is endearing to watch, and deire, his best friend (Alexa Vega), is struck with a bout of love-sickness at a most inconvenient time. phawkwood (Daryl Sabara) is allayed by emergency council powers, which brings out the best in everyone. From there on it, conversation is replaced by song.|
"Are you an angel?"
Banazir in Wither With You
|Abounding with what's right about love, this painfully beautiful tale, set in a sixty-four cubic-metre underground box, takes a look at lovers, banazir (Bronson Pinchot) and scionofgrace (Michelle Pfeiffer), who discover themselves troubled indeed due to the fact that a plot to keep them apart is set in motion by her menacing father, barahirion (David Spade). Michelle Pfeiffer was paid twenty-five million dollars for her performance, reportedly due to all of the 'yukky kissing' involved.|
(Cousin Balki! Wot's yucky about kissing a Meeposian, nazwaz?)
Last but not least: A thousand thanks to angelfate for introducing me to Vienna Teng! "Enough to Go By" is quite the addictive song.
Edit, 13:00 CST: Scott (yahvah) has posted a brief summary of some ideas on knowledge representation (especially abstract data types and relational KR), domain knowledge, and development of common sense and multi-domain knowledge bases for interactive applications. He cites psychoanalysis (using weblog data) and text summarization (presumably of similar natural language documents) as an example.
yahvah poses the questions:
- Why are not the lexicons and the knowledge bases more closely-knit?
- If we're to give semantics to a computing system, why not begin the long and arduous task of ontologically defining the words of the English vocabulary?
My first thoughts are as follows:
- The cold, hard truth about reuse: Although AI researchers often talk about commonsense reasoning and general-purpose ontologies that are reusable across domains, lexicons and knowledge bases tend to fragment quite.
- Ontological dictionaries: Though we often compare ontologies to dictionaries as a first approximation, they aren't exactly the same. A dictionary is a document in a natural language that defines words and phrases, usually using more basic terms, but not necessarily with formal structure. A good ontology, we might argue, should support the task of generating dictionaries in different lingos. This may or may not be easier than full translation or even summarization. (I suspect that compiling a good thesaurus is akin to summarization and yet easier than freeform translation.) The problem is that there is a gap between formalizing the ontology and knowledge bases and formalizing the lingo.
Apropos of which, have you seen the following?
- Decidability: As zaimoni remarked: depending on the application, natural language is likely to be Type 1 (context-sensitive) or Type 0 (Turing complete), pitting the developer of the reasoning and learning system agains decidability issues. Planning systems have run up against this problem in practice, and I wouldn't expect learning to reason and converse to be any exception. The jury is still out on whether summarization (to get the gist of a document) or machine translation (MT) can be cast into easier subproblems. In limited MT applications, of course, this is done willy-nilly (with hilariously bad reeults).
Lest my original reply to yahvah look like a complete non-sequitur: I should mention that though I am definitely interested in this topic, my bias is a machine learning-oriented slant. Moreover, I tend to think in terms of information extraction (IE) and collaboration, from groupware to collaborative filtering and recommender systems. I (perforce) approach these applications from available data sources and measurable objectives - e.g., one can always pay people to elicit facts and enter them into a KB, but if one wants to see whether it scales up, the proof is in the pudding. (This is also where I was coming from with my Google in a Box remark.1 Of course, it doesn't have to be the "whole web", all of LJ, etc., but you have to start with a document database or other information repository.)
1 Google in a Box is just a search appliance for your own intranet. Google keeps an archive of the pages that are cached by its spider, of course, but does not sell these to just anyone. Alexa, an Amazon company, does (as a 60Tb archive).