Joined: 14 Feb 2004 Posts: 195 Location: United Kingdom, Europe, Earth, Solar system, The Milky Way, The Universe, the 3rd Dimension.
Posted: Sat Feb 26, 2005 1:57 pm Post subject:
OK, I've had an Idea for how a new form of AI that could be developed. Basically it involves the idea that most humans get up at 6-9 o'clock and go to sleep from 9-11 o'clock. Also how humans get angry, annoyed, frustrated, or happy, calm, joyful, etc. etc. etc.
OK. The core of the idea is called "Sentence Concentration". The Bot Removes all the punctuation from the sentence, and then lowercases it. Once it has that all sorted, it goes about removing words like "how", "it", "a", "what", "the", etc, the this is a long one. So, after that, from "How is the weather?" we should be looking at "weather". Much easier to parse, don't you think? So, the bot could say, "I'm just looking out of the window now..." then give us a nice, fair result like: "It is sunny, how is it out there at yours?". It would then store that we're looking at the weather, and get angry (+1 anger) if you don't tell him what the weather is.
Another idea is Annoyance, Mood, Tiredness and Anger. After all that, the bot may find it is being insulted or sworn at. It will get angry at everyone if it's anger level is more than about 850 (max 1000), just the person who insulted it if its above 650. You get the idea, it would be the same for all the other things.
Timed Online. The Bot would, in essence, be a 27/4 bot. but at about 10 o clock, the bot would switch to "HDN" (Appear Offline) and stay that way till about 8 o'clock. Then it would stay like that for about an our, then go "Busy" (for work), at about 12 they'd come back with "Lunch", for One Hour. Then, finally, they'd be back again with online at about 5:30. Then they'd be "Lunch" 6, "Back" at 6:30, and stay there till day's end. (10 o'clock).
The bot would also appear offline for bed if it's tiredness levels reached 1000.
To the idiot who hasn't read all of this topic: Die in hell, b****
I'm sure there's more I can think of but I can't be bothered right now.
It sounds like a great idea, the only problem you'll have are time zones. Maybe your bot could ask new users what time it is, and instead of actually changing its NLN and FLN statuses, it would actually block and unblock users--so, if you're on the east coast, it will get tired and "go offline" at 11, but if you're on the west coast, it will still be online because it hasn't gotten to 11 there yet.
The real reason is because some time zones are so far away from others, that some people may never see your bot online at all, because their day is your night. See what I mean? Just a suggestion though. _________________ Current Site (2008) http://www.cuvou.com/
Form Relationships? LOL, yes is could say things like "Ohh (L) I love you!".
One of my past bot programs would form relationships and had a mood simulator as well.
The mood rating was from 0 to 100, 0 would be like really sad, then there'd be anger, calm, happy, very happy.
The relationship rating was from -100 to +100, with 0 as default (like on The Sims). If you were below 0, it would like you less, if you're above 0 it would like you more (and I did base things off this. Like, on AIM, if it got a warning and couldn't chat as much, it would still chat with people with higher relationship). _________________ Current Site (2008) http://www.cuvou.com/
OK. The core of the idea is called "Sentence Concentration". The Bot Removes all the punctuation from the sentence, and then lowercases it. Once it has that all sorted, it goes about removing words like "how", "it", "a", "what", "the", etc, the this is a long one. So, after that, from "How is the weather?" we should be looking at "weather".
This is called recursion and was thought of long ago. Many AI applications use it, including AIML. _________________ Check out Botworld! A dev resource for things bot.
Downloads, articles, news, fourm and more.
http://botworld.marzopolis.com
Joined: 30 Jul 2004 Posts: 178 Location: Auckland, New Zealand
Posted: Tue Mar 01, 2005 2:53 am Post subject:
Good point Alienz, but I think all AIML Bots dont really have a clue what they are talking about. Eg, "Is this what you do in your time" is a line I really hate. Yeah, Reading the post I knew timezones would be a problem but you seemed to have beat me to it Cer, The "Out to Lunch" idea seems good, but if the bot was in GMT+12 and you were in GMT+10 It might seem alittle off...
Overall I think this is a very good suggestion, there probably would be a large B2k3 like community over this, unless of course... the source was released to the public. But bot-depot is a open source community........
OK. The core of the idea is called "Sentence Concentration". The Bot Removes all the punctuation from the sentence, and then lowercases it. Once it has that all sorted, it goes about removing words like "how", "it", "a", "what", "the", etc, the this is a long one. So, after that, from "How is the weather?" we should be looking at "weather".
This is called recursion and was thought of long ago. Many AI applications use it, including AIML.
Actually, I believe they call that 'reduction'. Reduction of the sort you describe is more like keyword searching. Once you remove the common words, you are left with keywords you can lookup to find an answer.
For sentences like "How is the weather?" or "Do you like the weather?", reducing to "weather" and then returning "Let me look outside... yes, it's a beautiful day" works great. But conversations like "I hate this weather" => "Let me look outside... yes, it's beautiful" don't work. More complicated 'weather' statements like "Did you weather the storm?" (a common English cliche) or "The weather was terrible" fail miserably.
There are many more things you can do besides simple reduction. For instance, determine if the sentence is a question or a statement, determine intent (does, the sentence contain 'what' or 'how', etc and is the person looking for some factual answer), not to mention keeping track of context.
Natural language processing is a fairly complex problem. You can get simple results with simple algorithms, but you quickly run against walls.
Good point Alienz, but I think all AIML Bots dont really have a clue what they are talking about. Eg, "Is this what you do in your time" is a line I really hate.
Those are likely default answers, and that doesn't make AIML bad. It's just like using a bot template...unless you do your own things with it, it's just a copy. The bot is only as good as it's master makes it.
For example, my bot uses AIML but it's completely custom, so you won't see that question. Btw, I don't think I've ever seen that as a default answer in any AIML templates. The most common are ALICE and ANNA.
Quote:
Actually, I believe they call that 'reduction'.
In AIML, there are different kinds of "recursion", including symbolic reduction. "Recursion" and "reduction" have been used to describe the same thing. _________________ Check out Botworld! A dev resource for things bot.
Downloads, articles, news, fourm and more.
http://botworld.marzopolis.com
In AIML, there are different kinds of "recursion", including symbolic reduction. "Recursion" and "reduction" have been used to describe the same thing.
Are you referring to the srai tag? Does the*lazy*in that stand for reduction or recursion? I'm not much of an AIML fan or expert. But as I understood that tag, it lets you "forward" a match to another pattern, which is definitely recursion.
But the original poster was talking about removing extra unneeded words, which is definitely called reduction. My reply was to that not to any AIML definitions of those words.
The ideas of reduction and recursion have been around far longer than AIML and they mean two completely different things.
The ideal language parsing would be two-fold. Unless man can program computers to handle language like the human brain does (and we still don't know exactly how it all works), we'll have to 'teach' him certain combinations of words and pass the answer to it as a response. If all fails, reductionism can be used to shrink the sentence to a more 'parseable' format.
The process would look a little like this:
1) user enters a sentence as input. 2) Unless only one word, search for direct matches (use AIML or some other way to define matches). If only one word remains go over to point 5. 3) If found, return response and quit this subroutine. Else, go to 4. 4) Remove some common small words, go back to 2 and try again. 5) Single word 'commands'. Search a seperate archive for single word responses (these could contain references to weather, traffic, google searches,... whatever) and response accordingly.
This way would allow you to 'filter' out the obscure sentences like 'i don't like the weather', or 'did you weather the storm?' before passing over the remains to the actual weather module/default response.
The process could be done in several phases, depending on the amount of 'common' words that should be deleted. Something else that could be introduced would be to use a 'fuzzy matching' system, where typo's would be able to be corrected if a near match was found. But of course, this whole process would greatly increase 'thinking' time, and it's a matter of deciding where to stop processing and give up to conserve on resources.
Lots of work has allready put into natural language parsing, both by scientists and programmers like ourselves, but as computers turn more powerful, it creates new possibilities and less boundaries.
A simple workstation would be able to match an inputted line to a database of 50000 common english words and 'correct' misspelled words easily, not even using much resources and doing it 'on-the-fly'. Once spelling mistakes are discarded, it's a matter of defining enough matches to create a near-human responding AI.
Are you referring to the srai tag? Does the*lazy*in that stand for reduction or recursion? I'm not much of an AIML fan or expert.
The ideas of reduction and recursion have been around far longer than AIML and they mean two completely different things.
The srai tag is part of it, but in their standard they have several types of reduction, and a few other things that fall under what is called "recursion". I've seen recursion and reduction used to describe the same thing, and I wasn't just referring to AIML. _________________ Check out Botworld! A dev resource for things bot.
Downloads, articles, news, fourm and more.
http://botworld.marzopolis.com
I've seen recursion and reduction used to describe the same thing, and I wasn't just referring to AIML.
Recursion and reduction are definitely two different concepts. Why they would be used to describe the same thing I don't know. With AIML, I can understand the idea they are trying to get across by merging them, but it still doesn't change the programming concept behind recursion. Here's a good definition:
"An algorithmic technique where a function, in order to accomplish a task, calls itself with some part of the task."
Reduction is the process of reducing some content down to its core meaning.
Joined: 21 Dec 2004 Posts: 49 Location: Western Australia
Posted: Wed Mar 02, 2005 11:43 am Post subject:
My old bot, Tiler, did exactly that: got a sentence down to its keywords, and looked up a response based on it. He also learnt new replies automatically based on what people told him. So, after having done considerable work on trying to make changes to improve that system I think I have some right to say that it doesn't work. You can get reasonably good results pretty quickly with it, but like Mojave said (paraphrased), simple solutions can only yield simple results. That kind of system bottlenecks after a few thousand messages if it learns, and is a pretty poor system for various reasons already talked about above if you preprogram responses in.
User Control Panel
I'm sure there's more I can think of but I can't be bothered right now.
But as I understood that tag, it lets you "forward" a match to another pattern, which is definitely recursion.