STORYTRON - Interactive Storytelling

Wouldn't you like to be a character in one of your favorite novels and participate in directing the course of the story events? Chris Crawford, an ex computer game designer, has set out to give us the technology to do just that. Writers, instead of depicting a fixed course of events, will spend their time defining characters and circumstances. The reader, instead of just following a fixed course of events, will participate as one of the characters and watch as others in the story act and react to his actions. I am looking forward to their first demos, which should be available soon. I am also fascinated with the technology from an AI perspective: in particular the question of how much depth do the characters need to have in order for the story that unfolds from their interaction to be interesting.
Full post... Related link

İnsanlar taş devrinden beri ne kadar değişti?

Teknolojinin ilerlediği ve buna uyum sağlayamayan kurum ve kültürlerin
değiştiği aşikar, fakat benim sorum insanların benlik ve
motivasyonlarının gerçekten değişip değişmediği üzerine.

Yeni popülerleşen evrimsel psikoloji (Pinker vs.) değişenden çok
değişmeyen olduğunu iddia ediyor gibi: temel istekler ve bunların
şekillendirdiği hiyerarşik toplum yapısı taş devrini bırakalım belki
maymunlardan beri değişmedi. İki örnek vereceğim:

Sürü avlandığında avın en lezzetli kısmını alpha male yer, geri
kalanlar hiyerarşik sıralarına göre kalanları paylaşırlar.
Teknolojinin ilerlemesiyle yiyeceğin üretimi ucuzladı, dolayısıyla
belki yeme isteği toplumu şekillendirici yapısını kaybetti diye
düşünebiliriz (unlike sex :). Ama insanlar hiyerarşik yapılarını
korumak ve farklı statüdekilerin birlikte yemek yememeleri için bu
sefer lüks lokantalar, pahalı şaraplar vs icat etmediler mi?

İkinci örnek Paul Graham'ın "Hackers and Painters" kitabından.
Amerikan liselerindeki nerd'lerin (inek?) gördükleri işkenceyi
incelerken vardığı sonuç, kesin bir amaca sahip olmayan ve dolayısıyla
o amaç etrafında kendini organize edemeyen insan toplumlarının yine
de sırf hiyerarşi güdülerinin etkisiyle birbirlerini didiklemeye dayalı
anlamsız yapılar kurdukları yönünde. Bu olaya lise öğrencilerinin
yanı sıra verdiği diğer örnekler ise hapishane mahkumları, ve yüksek
sosyete hanımları.

Bana öyle geliyor ki aynı tiyatroyu oynuyoruz. Sadece sahne
değişiyor. Teknolojinin değişmesiyle bireylerin yaşam deneyimleri
derinden değişiyor mu emin degilim.

Full post... Related link

Secondary structure prediction using decision lists

Deniz Yuret and Volkan Kurt. Presentation at the Protein-Bioinformatics Workshop at Koç University Center for Computational Biology and Bioinformatics.



Full post...

Why you should not use the Penn Treebank to train a parser

Wrote a script to find inconsistently labeled constitutents in penn treebank. Script is inconsistent.scm, uses escape.pl as a prefilter for the mrg format, and its output is inconsistent.out.gz. Treebank has 49208 sentences, 1173766 tokens. There are 735722 constituents (only counting ones that have more than one non-empty child). 36306 constituent strings appear more than once, of these 5646 (15%) have multiple parses. It is a reasonable guess to say that the annotation for the whole treebank has at least this much inconsistency. A cursory analysis of the output shows the following types of ambiguities:

1. pos tagging inconsistencies (3088, 8.5%):
((NP (JJ clerical) (NNS workers)) . 1)
((NP (NN clerical) (NNS workers)) . 1)

2. constituent label inconsistencies (1126, 3.1%):
((VP (VBZ remains) (ADJP (JJ sound))) . 1)
((VP (VBZ remains) (NP (JJ sound))) . 1)

* Note that some constituent label changes may be legitimate:
but this does not seem to be very frequent.
((NP (CD 2) (NN %)) . 29)
((ADJP (CD 2) (NN %)) . 18)

3. bracketing changes (1432, 3.9%):
((PP (IN At) (NP (DT the) (ADJP (RB very) (JJS least)))) . 4)
((PP (IN At) (NP (DT the) (RB very) (JJS least))) . 3)

* Note that about half of these constituents are NPs which
generally have incomplete bracketing in Penn Treebank. About 25%
are PP's and the rest include VP, ADJP etc.

* Here is an example demonstrating all of the above:
((ADVP (RB earlier) (NP (DT this) (NN year))) . 1)
((NP (ADVP (RBR earlier)) (DT this) (NN year)) . 4)
((ADVP (RBR earlier) (DT this) (NN year)) . 1)
((NP (RB earlier) (DT this) (NN year)) . 8)
((NP (RBR earlier) (DT this) (NN year)) . 70)
((ADVP (RBR earlier) (NP (DT this) (NN year))) . 12)
((NP (JJR earlier) (DT this) (NN year)) . 3)


Full post... Related link

Prediction, determinism, and free will

[Books_] Imagine standing around in a train station when you notice a TV monitor hanging on the wall. You recognize your own face looking back at you from the screen. You think no big deal, it must be one of those closed circuit security monitors with a camera hidden someplace. But then the you on the screen starts walking and gets on a train you know that is not due for another five minutes. The monitor is connected to a magic camera that shows what is going to happen five minutes in the future! Well, do you have the freedom not to get on that train now? Or will you somehow get compelled to get on the train against your will? If you do have the freedom to betray the monitor, does that mean a future showing monitor is impossible to build? Maybe quantum uncertainty or chaos make future prediction at that scale technically infeasible. Even for God? What if the universe was deterministic? Would that help things?

I thought about this problem on and off for a number of years until one day I ran into a paper by Gary L. Drescher. I had met Drescher once at the AI lab, enjoyed his book Made Up Minds like most fellow grad students, admired his explanation of quantum mechanics, and puzzled over his refusal to get any support from DARPA. And here he was again, with the solution to the puzzle.

To avoid the complications of the real universe Gary's trick is to think of made up universes with simple rules. This prevents confusion and eliminates prejudice. So going back to my train station, imagine a universe simulated on a computer that has completely deterministic rules. A bit like the Matrix movies. You program the simulation and let it run until sentient beings evolve and start building train stations. It is your program, and it is deterministic, so you know what is going to happen at any point in time. When in doubt, you can just load the program on a second computer and fast forward to see the future. Do YOU have the power to place a monitor that shows the future at the train station in your simulation? Well of course, you might say, you wrote the program, you own the computer, you can do anything. Not quite, it turns out...

You can certainly put a monitor in the train station and pump images of five minutes in the future to it. But trouble starts as soon as one of your simulated creatures actually looks at the monitor and decides what to do based on it. To decide what to put on the monitor at that moment, you need to fast forward. To fast forward you need to simulate what the creature is going to do. What the creature is going to do depends on what it sees on the monitor. Hmm, we have a circular impasse. Well, you might say, why not run a bunch of simulations putting various future movies on the monitor to see how the creature is going to react - one movie shows the creature getting on the train, another shows not getting on the train, for example. You fast forward in each case, then choose the movie that actually turns out to be correct. The trouble is none of the movies are going to turn out to be correct if you have a particularly stubborn creature that keeps doing the opposite of what it sees on the screen. And that brings us to the crux of the matter - the "creature" does not even have to be a sentient being. It can be a very simple automaton with the only talent of doing the opposite of whatever it observes on the monitor. And you are helpless.

So there is nothing in principle against an omniscient creator who knows the future. He just can't share the information with the inhabitants. Unless of course he forcibly compels them to obey his prophecies, but that would be lame, interfering with their free will. Determinism and (somebody else's) knowledge of the future are not inconsistent with your free will, being forced into a particular future is.

The real universe has the added problems of quantum uncertainty and chaotic behavior which may make future prediction physically impossible. But our story shows that knowledge of your own future is logically impossible - as far as impossibilities go, that one is hard to beat ;)

Any errors and misunderstandings in the discussion above belong to me, and I acknowledge Drescher for the rest. I am looking forward to his new book.

Full post... Related link