Jump to content

  • Log In with Google      Sign In   
  • Create Account

We're offering banner ads on our site from just $5!

1. Details HERE. 2. GDNet+ Subscriptions HERE. 3. Ad upload HERE.


Narrative interpolation


Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.

  • You cannot reply to this topic
71 replies to this topic

#41 Infinisearch   Members   -  Reputation: 268

Like
Likes
Like

Posted 26 January 2002 - 09:41 PM

Oluseyi, I thought u might find this article real interesting. Its not directly about narrative interpolation but the authors concept of a ''frame'' can be applied to both dynamic plots and making intelligent NPC''s.

http://web.media.mit.edu/~minsky/papers/Frames/frames.html

-potential energy is easily made kinetic-

Sponsor:

#42 Oluseyi   Staff Emeritus   -  Reputation: 1678

Like
Likes
Like

Posted 27 January 2002 - 05:15 AM

Thanks, Infinisearch. I''ve bookmarked it for future reference.

[ GDNet Start Here | GDNet FAQ | MS RTFM | STL | Google ]
Thanks to Kylotan for the idea!


#43 Siebharinn   Members   -  Reputation: 122

Like
Likes
Like

Posted 27 January 2002 - 06:16 PM

I''ve followed this thread with some interest, since this is similar to what I''m attempting to do with my efforts. I have a couple thoughts that I''d like to throw out.

The original impression I got from this discussion was a way to interpolate the story between fixed plot nodes. Is that accurate (at least as it applies to games)? The nodes themselves are fairly static, and built into the game. In my vision of how this would work, the game itself doesn''t generate nodes, it only seamlessly presents story elements to take the player from one node to the next. At each node, elements of the player''s "style" could be analyzed, and the next set of story elements presented in a way that appeals to the player. At each node, there would be multiple paths leading to other nodes. The game designer would build a huge web of nodes, covering most conceivable activities. If a player goes off and does something totally offbase, then the system would gently nudge them towards the nearest node. In this way, the player gets a great deal of freedom, but you don''t need a HAL9000 to run the game.

Is this even close to what you guys are thinking?

My other thought is on state. You''ve mentioned databases for relationships and other activities. The way that I''m approaching it is with a dictionary system. Each entity in the game has it''s own dictionary, and as events affect it, entries are added to it''s internal dictionary. Some of those events just get ignored (the entity doesn''t know how to handle them), while others modify the behaviour of the entity. The nice thing about a dictionary is that it frees you from defining (and thus limiting) the types of information you can hold at the beginning. Here''s an example: Say the hero slays a dragon. In addition to whatever story specific things happen as a result, a "Deed" entry is added to the dictionary of every wandering bard within a certain range. Within that entry are things like what the deed was, where it happened, if the deed doer was wounded, etc. The bard then goes on to compose his songs and poetry from the deed information that he has.
Another example is a "Gossip" entry. Whenever a certain class of events happens, a "Gossip" entry is added to anyone who witnessed it. They in turn share that entry with everyone they meet. Gossip has a time limit though, after a while the person stops sharing it. So if the player goes into a village and tries to rob someone, that will get around and affect how other people view the player.
The entire state (all of the dictionaries) can be written to or read from disk at any time.

Whew! That''s getting pretty far afield. Tying the two thoughts above together, the story nodes would actually be able to manipulate the state of other entities, to build the story around the player. The player''s activities would be reflected in the state of everything around them, which the system could then analyze to present the player with more of what they appear to want.

I''m not sure if this is really what you guys are discussing, but this is the way that I''m approaching it in my project, and thought it might be relevent.

Thanks for such a compelling topic Oluseyi!

Take care,
Bill

#44 Oluseyi   Staff Emeritus   -  Reputation: 1678

Like
Likes
Like

Posted 27 January 2002 - 08:24 PM

quote:
Original post by Siebharinn
I''ve followed this thread with some interest, since this is similar to what I''m attempting to do with my efforts. I have a couple thoughts that I''d like to throw out.

It is always in "throwing out" such thoughts that some are inspired.

quote:
The original impression I got from this discussion was a way to interpolate the story between fixed plot nodes.

This is a fairly accurate description of how the system I envisaged stands right now - at least theoretically. The key thing is, though, that I''m not trying to impose any specific intent upon it. I want this to be a melting pot, for different people to leave this discussion with different ideas and for those ideas to be implemented in different ways. Everything that adds to the interactivity, non-linearity and non-deterministic nature of electronic entertainment is a definite bonus for us all.

In my current version, I don''t have the gentle nudges to get the user back into the story, but that would be a logical extension/modification/feature for videogames (you don''t want a user to think your game sucked because he never ventured in adventure''s direction, the ass!)

quote:
Is this even close to what you guys are thinking?

Heck, i''d say it was spot on!

quote:
My other thought is on state.

I like the "deed" and "gossip" systems, and briefly discussed something similar. Such background relationships are crucial to truly bringing environments to life with people that the user can enjoy interacting with. Being able to seek hospitality at a nobleman''s court only to find that your reputation has preceded you, and that the local bards have embellished your deeds so that you''ve now killed an entire brood of dragons rather than one - with no shield and only your trusty sword!

Ah, I love this stuff!

quote:
I''m not sure if this is really what you guys are discussing, but this is the way that I''m approaching it in my project, and thought it might be relevent.

I feel you''ve given us some very practical scenarios within which to consider the applications of the techniques we''re discussing. I''ll probably refer back to your examples frequently from now on.

quote:
Thanks for such a compelling topic Oluseyi!

Thanks to all for being such a receptive and intelligent community! I''m just trying to give a little back...

[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! ]
Thanks to Kylotan for the idea!


#45 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 06:19 AM

The chief problem I see with this approach is that the resulting story/narrative will have little to no life or pizazz whenever it veers away from the pregenerated content/storyline. Yes, it will be able to nudge you back towards one or more of the author's intended story arcs, and it might even generate believable dialog, given that the B characters are all on drugs, or had lobotomies, but I think it would be challenging to get the story to be truly dynamic.

At best, this will be analogous to the expert system that listens to 100 Mozart pieces, builds itself a Hidden Markov network, and then spits out a million inferior pieces that sound like what Mozart would produce after he had a stroke. Yes, it will indeed capture his style, but it will not string it together in anything better than a probabilistic construction. The psychological heuristics that distinguish this from the real thing will be absent.

In the end, the more human-generated content you feed this thing, the longer it will be able to improv on it before the player sees the man behind the curtain.

The scary part is that I think you COULD use something like this generate the scripts for any Aaron Spelling television show or the more recent Star Trek series. I remember a madlib program from 20 years ago that didn't do a half-bad job generating plotlines for network television.



Edited by - varelse on January 28, 2002 1:23:16 PM

#46 Oluseyi   Staff Emeritus   -  Reputation: 1678

Like
Likes
Like

Posted 28 January 2002 - 06:35 AM

quote:
Original post by varelse
The chief problem I see with this approach is that the resulting story/narrative will have little to no life or pizazz whenever it veers away from the pregenerated content/storyline.

I disagree. This is not a computer science experiment; the field of interactive narrative has professionals from literature, theatre and film, liberal arts academics and computer scientists collaborating and sharing information to build reasonable models. You're thinking purely in terms of existing technology rather than trying to find the limitations of such systems and solutions to those limitations.

[Edit:] For propriety.

[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! ]
Thanks to Kylotan for the idea!


Edited by - Oluseyi on January 29, 2002 5:28:57 PM

#47 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 08:07 AM

quote:
Well aren't we cynical?


No, "we" plan to ship something in the next year or two. I looked into Author. Beyond its superior use of pronouns, it's not all that far beyond SIR from the Minsky lab.

quote:
I disagree. This is not a computer science experiment; the field of interactive narrative has professionals from literature, theatre and film, liberal arts academics and computer scientists collaborating and sharing information to build reasonable models.


Your point? Does that mean if I put a digital watch and some nuclear material into a microwave oven and turn it on, then I can travel back in time? Putting a bunch of "experts" together achieves little on its own IMO. Exactly why is this going to make a difference other than to create a lot of sound and fury?

quote:
You're thinking purely in terms of existing technology rather than trying to find the limitations of such systems and solutions to those limitations.


OK, here's the problem. Have you seen one of these contraptions tell a joke? Have you ever seen an original creation from one with more life in it that See Spot Run? I think there's quite a bit of use for this technology in bending an existing storyline to better fit the actions of a player (as in make the story come to him rather than vice versa), but call me a human-chauvinist, I've never seen an act of artistic creativity arise from a computer (yet).

So I looked into the claims of "believable dialog." Let's see what the state of the art, Author, can put out:

"Hello," greeted the wolf, who was a cunning looking creature. He asked, "Where are you going?"

"I am going to my grandmother's house," she replied.

This isn't real dialog IMO. This is an oral recitation of facts from a database. This is dialog (IMO of course):

LEONARD
Beg my forgiveness! Beg my wife's
forgiveness before I blow your brains
out!

TEDDY
Leonard, you don't have a clue what's
going on. You don't even know my name.

LEONARD
(triumphant smile)
Teddy!

TEDDY
You read it off your f*$k!n6 photo. You
don't know me, you don't even know who
you are.

Can you tell the difference? Because your audience sure can. If you use something like Author (or some sort of spinoff) to generate your dialog, you're going to end up with something with the life of an Ed Wood script, and people purposely imitating it in order to make fun of it:

LEONARD
I want you to apologize, then I will shoot you
with my gun.

TEDDY
Do you know what my name is?

LEONARD
Your name is Teddy!

TEDDY
Did you know my name is on the photograph?
You have amnesia.

Tell me the data structures and interface you'd create in order to allow a writer to imbue a script generator such that it could spit out the human-generated dialog fragment above without having every single element of each scene spelled out to it in more detail than writing the scene itself and you'll have a convert. Until then, yep, color me skeptical.

quote:
I'll stop before I say something truly unretractable.


Now this is truly puzzling. We both have the same goal here: creating a believable narrative. However, I think you're buying too hard into a dream that's just that for now. Small steps are important, and Author definitely constitutes a small step, but we are far, far away from generating a halfway interesting story by computer.














Edited by - varelse on January 28, 2002 3:31:53 PM

#48 Siebharinn   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 09:46 AM

quote:

Tell me the data structures and interface you''d create in order to allow a writer to imbue a script generator such that it could spit out the human-generated dialog fragment above without having every single element of each scene spelled out to it in more detail than writing the scene itself and you''ll have a convert. Until then, yep, color me skeptical.



Obviously we''re not there yet. But we won''t ever get there if someone isn''t willing to TRY and get there. You''re basically saying that it can''t be done, so don''t try. If the rest of the industry had that attitude, we''d still be playing pong.

In my project, I''m not attempting to auto-generate dialog. What I''m trying to do is to select the best pre-written dialog based on the actions of the player. The story is written in fragments, and those fragments are strung together in a way that''s (hopefully) more meaningful than forcing a simple linear story on the player.

Take care,
Bill

#49 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 10:24 AM

quote:
Bad analogy. The point is that those individuals have certain knowledge that they can contribute, while individuals outside those disciplines can adapt and apply that in innovative ways.


From 10 years in academia, I can sadly assert that putting together individuals from disparate fields generates sound and fury, but little more. The artists have all sorts of crazy ideas, but no clue which ones are practical, and the techies won't understand a word of what the artists are getting at. It's like herding cats. You'll see: it's a valuable learning experience.

The people that step into catifghts like this and produce magic are the rare ones that are both artistic and technical. But there just aren't that many James Camerons or Orson Welles in this world.

quote:
I can't understand your fixation on Author either. I don't use Author; in fact, I've never even evaluated it, and I'm not in the least bit interested in it. Author was an example of primitive applications in a new field. The objective here is to test the limits of imagination, tethered by the reality of existing knowledge to come up with creative new models, tools and applications. It's not to help you ship your fucking title.


Actually, you brought it up in the initial post to this thread. It's less a fixation as it was an evaluation. I'm looking for cool tricks. It didn't impress, but the creator was definitely clever, and some of his tricks might make it into my own pet project.

quote:
If you actually read the posts, you might have noticed that we're not simply interested in text generation. We're interested in representing situations and events, and the relationships between them; in modeling interaction between non-central characters in such a way as to provide a believable context for the story. It is this context that will provide the additional information necessary to generate believable narratives.


And this is where we diverge and yet agree. The marines and the Covenant characters in Halo construct quite convincing narratives for themselves based on a combination of canned content and a great state machine implementation. In particular, the aliens charge you, they run from fire, they group attacks, and they wait until you're close and then lob a sticky grenade at you. Meanwhile, they have a huge pregenerated set of insults and taunts. The illusion is quite convincing for a few hours. Then it starts repeating itself and it breaks down to just really nice combat AI.

Any advance that reduces the combinatorial explosion in creating the above canned content or extends the life of it for NPCs is indeed an advance. Setting story nodes and then gently nudging the protagonist back to the story nodes whenever he strays for too long would also be an advance as it could cut down the aimless wandering that plagues so many RPGs. It's the same sort of thing a good Game Master would do. Generating events that will do this ought to not be so hard. Getting a sense of when and how to do this is the devil in the details that IMO will require human involvement for some time yet.

quote:
Factor in advances in AI - neural networks and learning systems capable of understanding/assimilating the tenets of human emotion as well as the methods of representing them textually to an audience, and we might have something worthwhile.


Neural nets are 40+ years old! Little has happened with them in this time other than a few refinements on experiments done 30 years ago. And if you create a system capable of understanding and representing human emotion, you'll be a legend. But there's a good reason it hasn't happened yet. Chris Crawford has spent over a decade on this and Will Wright neatly trumped him comparatively overnight with _The Sims_ IMO. Note that the latter had the pressure of actually producing something rather than whittling away his days eating lotus root. It's much like the space race: the right amount of pressure can really focus the mind.

quote:
We don't have the same goal. I don't, frankly, give a flying fuck about the next narrative. I'm interested in developing a technology that may take years to mature to any reasonable level, but one that has the potential to revolutionize the interactive entertainment experience. If every time an idea popped into someone's head they said "well, this is so far away from being useful - I'm buying too hard into a dream yadda yadda", where the fuck would you be? No car. No computer. No pencil. No electricity.


If every time an idea popped into your head, you spent 5 years chasing it, you'd be dead before you reinvented the wheel. And that's where the bullshit detector distinguishes a dreamer from an inventor. If Edison had spent his life trying to perfect a warp drive, we wouldn't have the light bulb, the phonograph, or motion pictures (well, we probably would have them, but not from him). Dreamers are all sound and fury. Inventors change the world.

quote:
Comments, critiques, suggestions all welcome.


False advertising at its best it would appear...



Edited by - varelse on January 28, 2002 5:29:42 PM

#50 Oluseyi   Staff Emeritus   -  Reputation: 1678

Like
Likes
Like

Posted 28 January 2002 - 11:52 AM

quote:
Original post by varelse
And this is where we diverge and yet agree. The marines and the Covenant characters in Halo construct quite convincing narratives for themselves based on a combination of canned content and a great state machine implementation.

Consider this idea (which I have also proposed on these boards somewhere):
There exist rules-based text-to-speech (TTS) engines, but they provide rather poor voice quality. On the other hand lie concatenative speech synths, which "smush" prerecorded words together to construct sentences. I propose a middle road, where the rules-based approach is applied to prerecorded allophones (phoneme transitions) to generate coherent speech. Emotion can be applied through annotations (often implemented as waveform modulation), a practice that is already widespread in the TTS community. Coupling this speech synthesizer with a comprehensive dictionary, a "thesaurus" (to allow the AI generate similar-but-not-identical statements and a decent state machine, it would be possible to create in-simulation NPCs that converse in significantly more interesting fashion than the current crop of repetitive drones.

Fantasy? Not only is this idea very workable (one of my former employers had a rules-based engine that could switch language and voice parameters on the fly, for 13 languages at the time I was there - all within 4KB and in real-time), but it is applicable to the matter being discussed here.

quote:
[If] you create a system capable of understanding and representing human emotion, you'll be a legend. But there's a good reason it hasn't happened yet. Chris Crawford has spent over a decade on this and Will Wright neatly trumped him comparatively overnight with _The Sims_ IMO.

Perhaps the secret lies in following the Will Wright approach and approximating these ethereals rather than trying to completely represent them? Perhaps extending that approach to the topic of discussion will yield near-immediate, useful results?

quote:
If every time an idea popped into your head, you spent 5 years chasing it, you'd be dead before you reinvented the wheel. And that's where the bullshit detector distinguishes a dreamer from an inventor. If Edison had spent his life trying to perfect a warp drive, we wouldn't have the light bulb, the phonograph, or motion pictures (well, we probably would have them, but not from him). Dreamers are all sound and fury. Inventors change the world.

To be an inventor, one must dream. One must then attempt to concretize one's dreams by expressing them in logical terms, after which one attempts to implement them. Should initial attempts fail miserably, one may choose to persevere, postpone or just plain give up. Alexander Graham Bell never built the hearing device he promised his deaf wife, but in its pursuit he developed the telephone. And maybe we will not complete this, but it may inspire us (or even others) to build other tools and systems, and if that is the case I shall consider my life well spent.

Ideas are born to be shared, to be expressed and pondered and examined. We reject thousands of our ideas each year - well, at least I do. Spending time examining some is not a loss, but a gainful pastime.

[Edit:] For propriety.

[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! ]
Thanks to Kylotan for the idea!


Edited by - Oluseyi on January 29, 2002 5:31:00 PM

#51 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 12:50 PM

quote:
From 20 years living with 2 academics in a residential academic community, I can tell you that you ar asbolutely right - provided the academics in question are narrow minded and myopic.


We are all narrow-minded and myopic about some things: it''s just a question of realizing where we''re not that makes the difference. I confess upfront that the techie side of me tends to express stories in terms of objects and events, which is the first problem that needs to be solved with automated story generation. However, this outlook served me well when designing games without storylines: I have a great instinct for building worlds with a high fun-factor (something easily expressed by objects and events).

However, I''m putting much more effort into understanding character now that I''ve decided the next project will tell a great story. How that turns out is an incomplete experiment at this time, but I have plenty of artsy friends smacking me around at the moment because I''m finally stumbling around on their home turf. And I even have a backup plan should I fail: several of them offered to write it for me, but where''s the journey in that?

quote:
There exist rules-based text-to-speech (TTS) engines, but they provide rather poor voice quality. On the other hand lie concatenative speech synths, which "smush" prerecorded words together to construct sentences. I propose a middle road, where the rules-based approach is applied to prerecorded allophones (phoneme transitions) to generate coherent speech. Emotion can be applied through annotations (often implemented as waveform modulation), a practice that is already widespread in the TTS community. Coupling this speech synthesizer with a comprehensive dictionary, a "thesaurus" (to allow the AI generate similar-but-not-identical statements and a decent state machine, it would be possible to create in-simulation NPCs that converse in significantly more interesting fashion than the current crop of repetitive drones.


That''s a great idea, I''d only question what the transitions between segments sound like when one tries to express a character under duress with such a tool. Since one of your former employers has implemented this, how did it sound? The remaining problem is giving them something interesting to say, and there you run into the barrier between reciting facts like a glorified database ala Author and acting human.

quote:
Perhaps the secret lies in following the Will Wright approach and approximating these ethereals rather than trying to completely represent them? Perhaps extending that approach to the topic of discussion will yield near-immediate, useful results?


I''d insist that you need to approximate these ethereals in a way that leads to good gameplay, lest you wind up like Chris Crawford did when he didn''t see the value in the _Balance of The Planet_ review that called it a "work of art" that just wasn''t any fun. There has been a disturbing trend in the past few years to forget F U N as the primary element of games in an attempt to elevate them to art. Once the fun is gone, it''s no longer a game IMO. Interactive, sure, a game, nope.

quote:
Ideas are born to be shared, to be expressed and pondered and examined. We reject thousands of our ideas each year - well, at least I do. Spending time examining some is not a loss, but a gainful pastime.


I''m ravenously filtering ideas at the moment myself. I change my mind at a moment''s notice. Then after all the factions in my head have battled it out, I put the aluminum foil helmet back on and write something.

quote:
"Those who can''t do/don''t know shit, bitch." - Dwight Yorke, Manchester United (paraphrased).


Alas poor me, never finished anything

#52 Kylotan   Moderators   -  Reputation: 3338

Like
Likes
Like

Posted 28 January 2002 - 01:18 PM

A mud called PhysMud does remarkable event -> text translation. The resulting descriptions are detailed enough to be immersive, and don''t appear to be computer generated. So I believe this can be done satisfactorily, providing the underlying structure is available.

[ MSVC Fixes | STL | SDL | Game AI | Sockets | C++ Faq Lite | Boost ]

#53 Oluseyi   Staff Emeritus   -  Reputation: 1678

Like
Likes
Like

Posted 28 January 2002 - 02:18 PM

varelse:

Now that you seem to have capitulated to a more constructive criticism approach, fire at will!

(Oh, and don''t take the Dwight Yorke dib personally; I always have to respond to verbal challenges. You''ll get to know this about me if you hang around here a bit.)

My former employer''s TTS, while very intelligent in ways that the competition wasn''t (correctly interpreting dates, measurements, acronyms etc) sounded flat and devoid of emotion. The voice also sounded very robotic (the actual waveform generation was implemented using certain libraries whose author I can''t recall, which simulated the motion of air through human vocal and nasal cavities). One of my primary modifications is to use the pre-recorded allophones to achieve more "human" sounding. The limitation to that is that you will need to have a person record all the allophones for the given language, but that''s much better than recording an entire dictionary!

I''m looking into making it possible to have an "accent dictionary" in addition to the lexical dictionary, to allow us programmatically simulate interesting combinations like a French person speaking German. What do you think?

My pet project "showcase" for this technology is a commentary system for sports games (starting with basketball). Since this is very limited context, it will be easier to provide enough content and context for the commentary not to be repetitive (using a grammar-based text generation and a spoken words state machine, so we don''t use the same phrases several times in a row except as emphasis). It''s still in the conceptualization stage, but I expect to make significant progress over the next month.

As a final tidbit, a classmate and I are exploring the possibility of writing a paper on all of these ideas collected into a coherent whole. I''ll let y''all know how that goes.

[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! ]
Thanks to Kylotan for the idea!


#54 Infinisearch   Members   -  Reputation: 268

Like
Likes
Like

Posted 28 January 2002 - 02:24 PM

It amazes me how people who program study abstraction but don''t apply it to general life. All knowlege interrelates, and people who concentrate in only one thing can never go beyond a certain point because they lack the creativity to create. I don''t know about anyone else but varelse leave your credentials at the door. The furthering of any field can happen gradually, suddenly, or even accidentally. It can occur from a person who has never even studied that field before. Why?because that person approaches the problem with no preconceptions and will bend there mind to the fit the problem and not the other way around. If u only use what u learn in the context of what u learned it in, u will not innovate.

Lets bring up a fundamental rule taught to people of a engineering background, feedback. So far in this thread we have only covered abstract data structures, and relationships. We haven''t got to the UI abstaction yet, more specifically user feedback. IMO it is as important if not more important than the underlying data structures. Have u ever read a story to a child? The difference between a ''good'' performance and a bad one could be as simple as the way u speak, or changing the main character name to the childs name. The feedback loop between two human''s in a vocal narration is complete. It involves the intrinsic nature of our existence, our five senses. It involves the communicative nature of each of those senses. If I took a professional narrarator and have them ICQ me a story, wouldn''t that take away from the narrative.

For a computer to fully be able to evoke emotion, it must be able to both understand them, have the abililty to properly convey that emotion, and have the feedback to interpret the users emotional reaction. When I say understand them I don''t mean feel them. I know i''ll catch alot of flack for this but, emotions are logical . Given enough information you can predict what a person will feel given a certain situation.

-potential energy is easily made kinetic-


#55 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 03:05 PM

quote:
It amazes me how people who program study abstraction but don''t apply it to general life. All knowlege interrelates, and people who concentrate in only one thing can never go beyond a certain point because they lack the creativity to create. I don''t know about anyone else but varelse leave your credentials at the door. The furthering of any field can happen gradually, suddenly, or even accidentally. It can occur from a person who has never even studied that field before. Why?because that person approaches the problem with no preconceptions and will bend there mind to the fit the problem and not the other way around. If u only use what u learn in the context of what u learned it in, u will not innovate.


I rather prefer the adage "In the fields of observation, chance favors only the prepared mind", which of course came from a techie, so it must be suspect, eh?

For example, no one is going to walk off the street with zero musical background and write the successor to a Mozart concerto: that just doesn''t happen. What can happen is that a prodigy can pick up the rules an order of magnitude or more faster than the best in the field and do so within a decade wherein most of the best would need to spend a lifetime on it. I''m married to such a prodigy: I have firsthand experience watching this process and I''m in awe of it. But even she spent most of her childhood playing many musical instruments before a 10-year diversion into the hard sciences.

There are patterns and rules for any field that require a great degree of study before inspiration is of any use whatsoever. People can and do come in from the outside, look around, find an analogy to something they''ve already studied, and turn a field upside down. And that rocks when it happens because they''ve exercised our greatest advantage over computers: non-linear pattern matching. Capture that in a computer, scale it with Moore''s Law, and we''ll be on our way out of the solar system in under a decade. Vernor Vinge has termed this sudden burst in intellectual power the singularity. But that''s a tangent, just like your observation, and this reply.

I''ve made my living jumping from field to field back to the very beginning. I have no formal training in Computer Science or Engineering whatsoever. I am flattered that you have concluded that I must have majored in such a field. And you''re right, all knowledge can interrelate, but the domain-specific knowledge required for greatness is usually too esoteric to do so. Interrelated knowledge will at best make you "good" at something, not great.

And if you have the counterexample wherein a rube walked out of the country, took one look at a great unsolved problem, and spat out the answer, I''m all ears.


















#56 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 03:16 PM

quote:
(Oh, and don''t take the Dwight Yorke dib personally; I always have to respond to verbal challenges. You''ll get to know this about me if you hang around here a bit.)


No offense taken, it was such an absurd statement to apply to me that I was mostly amused, and merely presented one of my more recent counterexamples. If you find out more about me, you''ll find I''m all over the map.

quote:
My former employer''s TTS, while very intelligent in ways that the competition wasn''t (correctly interpreting dates, measurements, acronyms etc) sounded flat and devoid of emotion. The voice also sounded very robotic (the actual waveform generation was implemented using certain libraries whose author I can''t recall, which simulated the motion of air through human vocal and nasal cavities). One of my primary modifications is to use the pre-recorded allophones to achieve more "human" sounding. The limitation to that is that you will need to have a person record all the allophones for the given language, but that''s much better than recording an entire dictionary!


That seems like quite a reasonable approach, but probably a bear for the voice actor to get perfect. Do you have any samples of this thing in action out there? I worry that this approach has been around for a while (I remember the cheesy JFK speech made from samples of all his other speeches), and if there weren''t subtle (or perhaps obvious) obstacles, I would think it would be in wider use by now, so I would look deeper into its implementation before concluding it''s anything more than promising.

quote:
I''m looking into making it possible to have an "accent dictionary" in addition to the lexical dictionary, to allow us programmatically simulate interesting combinations like a French person speaking German. What do you think?


If the former approach works, the latter seems reasonable as well, though there might be some weird succession problems for phonemes not present in French, but they''re supposed to sound wrong, right?

quote:
My pet project "showcase" for this technology is a commentary system for sports games (starting with basketball). Since this is very limited context, it will be easier to provide enough content and context for the commentary not to be repetitive (using a grammar-based text generation and a spoken words state machine, so we don''t use the same phrases several times in a row except as emphasis). It''s still in the conceptualization stage, but I expect to make significant progress over the next month.


Have you ever played NFL2K ot NFL2K1? One of the two strongest features of these games was the running commentary. Top that, and you have best-seller. I believe it strings together sentences, more or less flawlessly. The effect is quite convincing: it''ll drone on for hours without seeming to repeat itself.

quote:
As a final tidbit, a classmate and I are exploring the possibility of writing a paper on all of these ideas collected into a coherent whole. I''ll let y''all know how that goes.


I would definitely be interested in a review paper that summarizes the state of the art. Good luck!



#57 Infinisearch   Members   -  Reputation: 268

Like
Likes
Like

Posted 28 January 2002 - 05:31 PM

I neither concluded or was suggesting that u had any kind of CS or engineering education, I was refering to your '10 years in academia'. There is no doubt observation and preparedness is important, but without creativity innovation is stifled. Would you agree that if you were to enter in a conversation about space-time, some people will speak on abstract levels, while others will immediately start talking about einstien's or someone else's theory as if they are fact? Others still might balance the two.

The point I was trying to make is, systems of knowlege is like playing with blocks. Every layer above the foundation is dependent on the layers below it, and if one of the lower block is found to be false or doesn't work the whole system comes tumbling down. In respect to this topic we are modeling one system to mimic another, and since as even u stated the current technologies failed in that respect don't u think it is that innovation that is necessary at the moment. 'Neural Networks are 40+ years old' and yet we are nowhere near the point of understanding or mimicing the human mind, again I ask don't u believe a shake up might be in order?

My observation was not tangent nor was yours. We are speaking of thought, creativity, and patterns, that is what this thread is essentially about. It seems to me you seperate creativity and intellectuality in your mind. Yet it is that creativity that we are trying to reproduce, and it is that creativity that will produce a system of such capability. You spoke of Edison, what about Tesla? His creative genius and ferver was that of true genius, at least IMHO. He pictured things in his head and worked from there, and he knew his invention would work. That was due to forethought and imagination, not one or the other.

quote:
There are patterns and rules for any field that require a great degree of study before inspiration is of any use whatsoever.

I must say i'm really suprised u say that and then use the word analogy in the next sentence. Patterns occur between fields more often than people care to admit or see. Domain specific knowlege can be learned at any time, the ability to see beyond and between them can be applied anywhere.

EDIT - HTML

-potential energy is easily made kinetic-

Edited by - Infinisearch on January 28, 2002 12:39:11 AM


#58 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 28 January 2002 - 05:34 PM

quote:
My former employer''s TTS, while very intelligent in ways that the competition wasn''t (correctly interpreting dates, measurements, acronyms etc) sounded flat and devoid of emotion. The voice also sounded very robotic (the actual waveform generation was implemented using certain libraries whose author I can''t recall, which simulated the motion of air through human vocal and nasal cavities). One of my primary modifications is to use the pre-recorded allophones to achieve more "human" sounding. The limitation to that is that you will need to have a person record all the allophones for the given language, but that''s much better than recording an entire dictionary!


Made curious by you enthusiasm, I went looking for the state of the art in TTS tonight. I was disappointed in what I found, but I suspect that means there are significant obstacles to getting this right. Unless you have evidence to the contrary, I''m fairly skeptical that this is practical right now for anything but a computer character in a game.

There were two main disappointments in this search: one, at least half the projects I found didn''t bother to put examples of their synthesis on their web pages, and two, many of them didn''t sound much better, if at all, than what I used to coerce out of S.A.M. on my Atari 800 20 years ago.

But there were a couple bright spots: AT&T''s Natural Voices web site has some of the best (and worst) synthesis I''ve heard. The highlight is a web page allowing you to submit 30 word sequences for immediate synthesis: instant juvenile fun. I wouldn''t use any of these voices in a game, but I''d definitely make crank calls and ransom notes with them (jk), and that''s gotta be good for something.

And here''s a nice list of speech synthesis projects so you can make your own mind up.

I needed to look into this anyway. I think you''ll need to use larger fragments than allophones for your task. But since RAM is so incredibly cheap these days, this ought not to be much of a hardship, especially if you''re limited to a specific domain such as sports commentary. Good luck!







#59 varelse   Members   -  Reputation: 122

Like
Likes
Like

Posted 29 January 2002 - 09:53 AM

quote:
I neither concluded or was suggesting that u had any kind of CS or engineering education, I was refering to your ''10 years in academia''.


Ah, the problem there being I don''t consider ''10 years in academia'' to be a credential of any sort, but more a raging sign that I didn''t know what to do with my life for the longest time. I cited it in order to convey the idea that I''ve been around a lot of academics, and besides heavy drinking and arguing over esoterica, I didn''t see them do much else than write grants to keep the band playing. I''ve heard the situation is a bit better in Europe.

quote:
There is no doubt observation and preparedness is important, but without creativity innovation is stifled.


Who''s stifling creativity? Quoting Chris Crawford:

"A fantasy universe can be fragmented or inconsistent, because human desire is often fragmented and inconsistent, but a dream universe cannot be so self-indulgent. A good dream universe is a complete image, not a partial sketch. Every detail of the ideal is clearly specified, every consequence worked out. Only when we know precisely where we''re going can we begin planning how to get there."

I think this summarizes what I''ve been trying to express in this thread nicely.

quote:
Would you agree that if you were to enter in a conversation about space-time, some people will speak on abstract levels, while others will immediately start talking about einstien''s or someone else''s theory as if they are fact? Others still might balance the two.

The point I was trying to make is, systems of knowlege is like playing with blocks. Every layer above the foundation is dependent on the layers below it, and if one of the lower block is found to be false or doesn''t work the whole system comes tumbling down.




Yes and no. Here are two simple examples: the central dogma of molecular biology for decades was DNA->RNA->Protein, a one-way street. Then along came retroviruses, and it turned out they could pull a U-turn at RNA. There was a lot of bickering, aome wounded egos, and then they called in the interior decorators to fix up the house and the field continued onward. More recently, Stanley Prusiner was dismissed as a crank for a decade for insisting prions were effectively genetic material encoded as protein. It turned out he was risht, but it took a bit of a mindshift to figure out how and why he was right. In both cases, the remainder of molecular biology kept right on chugging even though these exceptions were both staggering.

You seem to see everything as a dichotomy: it''s either status quo, or Martin Luther nailing 95 theses to the door of the Wittenburg Church. The reality is more a continuum: Legos as opposed to building blocks. You can pick at the foundation quite a bit without bringing the house down.

quote:
In respect to this topic we are modeling one system to mimic another, and since as even u stated the current technologies failed in that respect don''t u think it is that innovation that is necessary at the moment. ''Neural Networks are 40+ years old'' and yet we are nowhere near the point of understanding or mimicing the human mind, again I ask don''t u believe a shake up might be in order?


The shakeup could happen at any time, but I''m not expecting it any time soon. Why? Because a lot of smart people have spent entire careers on this without solving it. And that''s in an age where there are vastly more smart people with the opportunity to apply themselves to such a problem, as opposed to the days when only the privileged class ever had time to consider such things.

quote:
You spoke of Edison, what about Tesla? His creative genius and ferver was that of true genius, at least IMHO. He pictured things in his head and worked from there, and he knew his invention would work. That was due to forethought and imagination, not one or the other.


Why not take this to its logical extreme, and simply bring up Edward Leedskalnin as the ultimate counterexample? But even his work took decades: plenty of preparation there.

quote:
I must say i''m really suprised u say that and then use the word analogy in the next sentence. Patterns occur between fields more often than people care to admit or see. Domain specific knowlege can be learned at any time, the ability to see beyond and between them can be applied anywhere.


Sure, my wife the prodigy was once a mathematician specializing in group theory. One Summer, she did a stint in an X-Ray crystallography lab. She saw a pattern between her own work and how protein crystals form. Within a couple months, she had reduced 20 years of the research of others into a simple 3-term equation. She published, and turned the field upside down. But even she would concede her mind was prepared for it. It was the chance encounter of an analogy that allowed this to transpire. She also got lucky that literate and creative mathematicians were a rarety in crystallography.

I did something on a lesser scale a decade ago wherein I reduced an O(n^3) calculation performed by biologists down to O(n). But that was by seeing an analogy between how computers work and what the calculation was actually measuring as opposed to slavishly following the mechanics of the calculation itself. And it was in an age where computer-literate people in biology were quite rare. These days, it''s a completely different story with the rise of bioinformatics. But I got there first.

The overarching problem is stuff like this doesn''t happen every day. These are highly unlikely events and the only way to encounter them is go looking for them. Lots of really smart and creative people have gone fishing around trying to model the human mind. They got their nibbles, a few bites, and then nothing ever since. The odds are staggeringly stacked against us truly figuring this out at this point IMO.

An aside: Mathematicians as a group bemoan the difficulty in finding things to prove that haven''t already been proven. All that''s left these days are the Goldbach conjectures and the Riemann hypotheses. And that''s enough to drive someone nuts.

Blatant opinion: I suspect it will all come down to _The Halting Problem_ mentioned in other threads: we simply are not intelligent enough to understand our own intelligence. And some day, something highly unlikely will happen on a machine and it will bootstrap its own consciousness into being much like we likely did 200 million years ago or so. And the irony will be that the machine will not be able to know itself either, but it will build a bigger machine, and so on, and so on.



#60 Oluseyi   Staff Emeritus   -  Reputation: 1678

Like
Likes
Like

Posted 29 January 2002 - 10:28 AM

I intend for this thread to last at least the entire semester, being updated as ideas come and discoveries occur, so please take all tiffs somewhere else.

In fact, if things get any more counterproductive, I''ll start deleting/editing incitive posts (starting with mine). I know it''s abuse of my priviledges, which is why I''m appealing to everyone to show some restraint. Thanks.

And now, I''ll go set an example by deleting all my outbursts.

[ GDNet Start Here | GDNet Search Tool | GDNet FAQ | MS RTFM [MSDN] | SGI STL Docs | Google! ]
Thanks to Kylotan for the idea!





Old topic!
Guest, the last post of this topic is over 60 days old and at this point you may not reply in this topic. If you wish to continue this conversation start a new topic.



PARTNERS