[Bragi]

If you import the files into my editor, it will show you a list of all the duplicate pattern in the log (just make certain that 'log duplicate patterns as warnings' is selected in the  'warnings and exception' dialog found in the debug menu.

[Freddy]

Cool, I will try that.  In fact in the end I will more than likely swap from my Aimlbot library to use your system, that's when I have figured it out.  Then I get the best of both worlds

[squarebear]

There used to be a file kicking around the net called shadow.exe which showed duplicate files and categories that would never be called. Not sure if it is still around though.

[Freddy]

Think I found it SB ?  Shadow Checker...

http://alicebot.sweb.cz/introduction.html

[squarebear]

That's the one Freddy. I am not sure if it is actively worked on or supported though, as I notice it says to contact the alice mailing list which hasn't been active for years now. It's a good tool if it still works though.

[NickyBlue]

I disagree to some extent... Not trying underming somebody efforts. And I beg my humble appologies in advance to you Mr. Square Bear. And Its surely not about whole topic but just few starting posts of this topic and  cause I am not a programer AIML or whatever. Moreover I know lot of my younger friends here making recreational bots for themshleves so it doesn't make that difference to them. So pls do ignore my input if it doesn't interest you I just wanna feel like saying something and this happens to be the page for start...smile.

for ex:
                  
Human: Learn the sun is hot
Bot: Ok I will learn that the sun is hot.

Human: Learn London is in England
Bot: Ok I will learn that London is in England.

            Doesn't this sound somewhat ridiculas. This "Learn" thing  destroying the very essence of dialog based training. Rather it should be just "The Sun is Hot" or You know "Sun is pretty hot." Then if bot doesn't have knowledge about it, it'll store it or else it just can say yeah! Thats how we improvise or add knowledge in our day to day handling. I today also visited AICore home page as well and it also uses things like that. I mean you are naturally hard coding it what if it in AIML or whatever way. So why don't we try simulate some real conversation like. When we say dialog based learning then it should be so. And thats what really counts. It'll simplify bot training. And normal web user can interact with it like they does in chat rooms with thier friends and can contribute to bots knowledge.
            
            
            Sure it need some labour on part of Bot designer to handle wrong utterances whether due to negligence or intentional on part of user. But it can be handled by implementing "level of trust" mechanisium. We also do thing like that intentionally or subconciously vasrious times, don't we? By the way all this london thing are not real text or are they? ;-)  cause things like that can be hard coded (imported) into bot database with single click. And he knows every capital or most of the cities of any country, no? So why try train it on such ...whatever matter.... ;-) just kidding guys. The point is not every knowledge of bot should be comprehensive to 100% to make it worthy. I myself don't know most of them. Not even thier just capitals but we got lots of internet text regarding such things, so? ...its certainly gonna make me look like a dumb fool at least in this regard .


            
            
another thing is like:
            
Human: Learn London is in England
Bot: Ok I will learn that London is in England.

Human: What is in England?
Bot: London.

This seems to be wrong training. If I ask you What is in England?  There might be lot of different things or possible answer. Storing data doesn't meant it'll always be used within next few sentences or context. It might be asked by some other person in some other context. So the answer would be wrong. This kind of training will really gonna make you bot intelligence really artificial (fake).

Sorry to say but for me thats the defect of most of current AIML programmers not of AIML itself.



Even I read it on some topic:

Human: I have ate hot pizza with this or that topping.
Bot: Yeah hot pizza taste good. I love hot pizza. 

 ...something like that. The first part of answer kinda reasonable but "I love hot pizza" ain't make it a sensible bot. It destroys the integrity of bot as a whole. A good bot should be well aware of things or even human feelings to whatever extend one can implement but shouldn't assert that it feels same. Even if a child and his father is climbing a hill. Father knows and take cares that his son might be tired doesn't  means he get tired at the same time, isn't it? might be not exact analogy but you got my point.
 
 So no offence should be taken but this kinda AIML log won't help make a intelligent bot I think. But still there can be many defination of inteligence for many users depending upon his intentions sans the sanctity of its being right or wrong.
 
 But for me if I ask a bot on first meeting " Well I kicked Sandy I told you about yesterday" and if it skid by saying Wow! or anything out of context then I think its rather foolish reply than being intelligent in any sense. And if we are developing NLP for any sensible use its not gonna work like that. Its like all the effort gone to drain for no reason. All the folks I see interested in AIML building could have really generated a far more genuine response which would have been a sensible dataset in itself for some future programmer if not current ones. And I think with just a little difference in basic engine or database structuring it could make a huge difference if not absolute one (Your true AI). I think some people believe that Humans kinda synonym for Intelligence but is it really? Look around you, Thier lot of difference what we speak of and what we really think of it. For me intelligence is to perform things reasonably no matter what the circumtances are (odd or otherwise). And yeah everything is relative at some point to some thing. Its no rokect science that to measure something you got to have some anchor point. whether fixed or fuzzy. Anyway there can be difference in magnitude of results but all instances will be instances of intelligence especially if seen in the light (context?) of some True AI thing. I could goon lot on this but I think I must not disturb you folks unnescessary here, will conclude it somewhere else. Just trying tell my own intrepretation of things, might not be same as you all folks or even might not be true but still as my new friend Freddy says "Just take care nobody catches you" ... Am I caught? smile.
 
 
 
 





         

[Art]

The standard AIML based programs usually do not allow the bot to learn from conversations. Any learning is done from the Botmaster actually typing the desired information into the appropriate file.


The next time the bot encounters the same question it can find the match of the newly acquired "learned" information and produce the correct response.

While AIML bots do not "self learn" through conversation, this is a big plus from Steve to offer this to the AIML / ALICE community.

I'm sure that with more proper wording, one could obtain better results than those in the examples.

[squarebear]

This "Learn" thing  destroying the very essence of dialog based training. Rather it should be just "The Sun is Hot" or You know "Sun is pretty hot." Then if bot doesn't have knowledge about it, it'll store it or else it just can say yeah!

My learn.aiml file is around 6 years old now. I amended it a few years ago to be able to handle things like "My shirt is red. What colour is my shirt?". This was necessary for me to enter the Loebner Prize as no judges would say "learn" at the start of the fact.

[Art]

Steve,

Interesting question: Since you and I both speak English, my in American English and yours is British English. We all know that there are many differences and disparities between the two. How is this anomaly handled in your AIML code? Did you breakdown all known words of each?
color - colour, etc. Then you've got words like whilst to our while and spelt to our spelled. There's a lot to be considered. Not being a real AIML/ALICE maven, please enlighten me on your method of dealing with this.

Thanks!

[paphus]

I'm not AIML expert, but two different spelling for a single meaning should not be difficult, you just associate both words to the same meaning.  What I find difficult is that most users, especially online, cannot spell... or type so most words are not spelt or spelled correctly.  You almost need to put the text through a spell checker before processing it anyone know of any good open source spell checkers?

International language is more difficult.  I was working on that this week for BOTlibre, pictorial asian languages (Japanese, Chinese, Korean) make things a lot more difficult.  You have to detect the pictorial languages and parse and print the text differently, no spaces between words, single character words, different punctuation.  Traditionally these languages go up to down and right to left, that would make things even more interesting, thankfully most online text is left to right.

Grammar is even more complex, as different languages have different grammar rules, different tenses.  I think the best solution is to just give the bot the words, and let it figure out the grammar.  But that would take lots of training, given most people spend years learning a new language, and most never get very good at it if it was not their mother tongue.

[squarebear]

We all know that there are many differences and disparities between the two. How is this anomaly handled in your AIML code? Did you breakdown all known words of each?

Art - It's even simpler than that. If someone says "I have a red shirt" or "I have a circular table". The pattern is created with "WHAT _ IS MY (object)", so it can handle "what xxxxx is my (object)" whether xxxxx is "color", "colour", "shape" or even "fkushufbvsdufbvsdkb".

paphus is correct though, you could have <srai>s that converted color to colour and so on.

[Freddy]

I'm not AIML expert, but two different spelling for a single meaning should not be difficult, you just associate both words to the same meaning.  What I find difficult is that most users, especially online, cannot spell... or type so most words are not spelt or spelled correctly.  You almost need to put the text through a spell checker before processing it anyone know of any good open source spell checkers?

You already alluded to one solution in the same paragraph. I mean isn't it possible to SRAI common spelling mistakes ?

So "helo" becomes "hello", "dispite" becomes "despite" and so on.

Something in the back of my mind tells me this has already been done. Maybe SquareBear knows of something ?

It's a lot of work though. Did you try Google for a spell checker ?

I see one problem with the spell checker and that is if you use it to replace words you might end up with some funny auto-corrections.

[Freddy]

Here's something : http://knytetrypper.proboards.com/thread/192

[Art]

Hi Freddy. Sorry, I must have missed this.

I think Dave Morton was looking at getting it to work with Program O but yes, it is designed primarily for use with Pandorabots, as they support the <learn> and <eval> tags. It makes new categories depending on what it has learned and stores them in memory, which takes priority over the update.aiml file. Each learned fact is only available for the user who taught it, in order that it doesn't learn nonsense like "Tony iz a gay n00b lol" for everyone who talks to it.

There is also the <learnf> tag on the subscription servers which writes its learned knowledge to a file called learn.aiml and is accessible to all users. However, this needs to be checked carefully to avoid it filling up with rubbish.

One of the advantages of using a Flash interface, is that everytime Mitsuku learns something new, she emails me to let me know and I can decide whether to add it to her permanent knowledge or not.

Steve, if this is so couldn't a relatively easy routine be written to allow the user to have or keep a secret with the bot?  Other users would (hopefully) never know and the bot could be asked to reveal the secret at a later date to the original (secret holder) user. Interesting.

Could this also be done with AIML based AlICE bots?

[squarebear]

This could be done with AIML quite easily.

My secret is *
could call a template like:
<think><set name="secret"></star></think>

Then you could say "what is my secret" and use <get name="secret"/>.