[Freddy]

Thank you one and both 

[one]

A note to One...

You have, in the past, been successful in finding some bug that generates an infinite log file (filling the diisk).. Hence why Suzette died on you. Unfortunately I can't manipulate an infinite log so have trouble finding the bug. I do keep trying to set up traps to find it earlier, and I may yet succeed. However... I had to erase your identiy to Suzette, hence why she is now repeating stuff to you she has said before.  So-- on one hand, sorry.  On the other--  do it again!  I want to find that bug.

bruce

No problem ( It was a loss of good Data) I contribute it to having problems with 'Bugs' before.

[wgb14]

<<Patterns are stored in topics. Topics have keywords that make them admissible. In the case of the test sentence, the only keyword available is morning. So I'd have to store the pattern in the TIME topic or the DAILY_ACTIVITY topic as the only ones currently having morning in them. Or put it in a topic that requires no keywords and is always checked if no other topic finds a match to the input (the XOPINION topic).  >>

If I am not mistaken, you are using two kind of patterns in your bot: The synonym collections (e.g., s:(Aliens AND ~movie) I have seen the movie aliens) and one to deal with information from the parser (e.g., s: ( |subject=you * |timephrase)   What else do you like to do _#1 .) . You have covered extensively the first in your paper (no problem with that) but I am still confused on what exactly have you done with the second (i hope your company allows you to share some information on that)

Please correct me if i am wrong:

As you said, topics are activated based on keywords assigned to each topic. Each topic may contain patterns of any of the two forms (keyword/parse patterns). Once a topic is activated (e.g., in our example DIALY_ACTIVITY topic) the engine then checks the topic's assigned patterns. If the parser has given any data (e.g., (I (proposition) walk (verb) home  in the morning(time phrase) it will activate the following pattern  s: ( |subject=you * |timephrase = in the morning)  What else do you like to do _#1 that will produce the output what else do you like to do in the morning. Is that correct? If yes then brilliant, your approach is absolutely brilliant. It reminds me of someone coming to England without knowing English at all. In this scenario the role of parser plays the human gestures... so for example if someone comes and says to that person "shall we go to the market?" and accompanies "go" and "we" with the proper gestures the person will understand that "he" (preposition) is being asked to go (verb) somewhere but he will not understand the destination because he doesn't know the language. It is as you said the person will roughly understand what has been said but not what actually has been said.

Again brilliant.... and thanks for sharing with us....      

[wgb14]

@ONE

<<OK as long as you are sure you don't belong in prison for stalking>>

I truly hope you don't mean what you are writing... otherwise its a very heavy insult. I never stalked you or anything and I certainly don't belong to any prison. If you check the discussion with Bruce you will see:

(Bruce): A note to One...
Myself: Not sure which one you are referring to... but i guess it would be me..
Which means that I was not sure if he referred to me as I was the one annoying his bot all the time with repeated inference questions.

Bruce: One- dont apologize for creating problems..........

I think it was only natural to assume that he was referring to me and this was just a way of expressing himself. It is the first time I see you in the forum and hence there is no chance that I would have any intentions of stealing your identity or even stoke you in any way....

I consider the matter closed... I know you now (like I know Freddy and the others) and there is no way to repeat the same mistake again....

[bruce wilcox]

I should briefly note- the syntax of the original paper is obsolete, the AND NEXT etc notation simplified into what you see above. The intent and capability of the original paper is not lost, only syntactically altered. And the original paper didnt cover all the other "keyword" types because they were not essential to expression how patterns worked similar to AIML. (Except that aiml cannot tolerate patterns except as real words).

The notion of sets is very valuable. PersonalityForge has them, but doesn't do enough with them. And the notion of ! (not) and { } optional, and the short range wildcard *~ -- these are all extremely valuable to match a pattern as precisely as you want your intent to be. AIML pattern match is way to vague, and so is personality forge.  I think that Suzette's pattern match, combined with the theory of chat that I have that is supported by CHAT-L and the runtime system are what gives her her current strengths.

[bruce wilcox]

By the way....I know that Art found Suzette worth chatting to. And he voted in the chatterbox competition for her as 10th bot. 10th bot is useful because it allows her to continue in the competition to the final round (which means I get even more debugging data).  If you (One or whomever) feel that she is more worthy than CyberTy (our nearest competitor), then a vote for Suzette would be potentially useful to her. Voting for her as "most popular bot", if 'twere done fairly would require you try out the other contenders, which is a lot of work. But 10th bot probably doesnt require as much rigor.

[wgb14]

I understand why you don't give me any clear answers but you are confusing me even further.... Once I say to myself yes I got it, a new post appears and I have to rethink. Let me clarify myself to avoid any misuderstandings: I don't want the code of your chatterbot, I am not interested for any further insights on your technology and I am not planning to steal anything. If I will use any of your work i will reference it appropriately. All I am interested to understand is the relationship between pattern matching and parse output and although you have given me some information it is still unclear to me.

Would you agree with what I have written in the previous post?

If I am not mistaken, you are using two kind of patterns in your bot: The synonym collections (e.g., s:(Aliens AND ~movie) I have seen the movie aliens) and one to deal with information from the parser (e.g., s: ( |subject=you * |timephrase)   What else do you like to do _#1 .) . You have covered extensively the first in your paper (no problem with that) but I am still confused on what exactly have you done with the second (i hope your company allows you to share some information on that)

Please correct me if i am wrong:

<<<As you said, topics are activated based on keywords assigned to each topic. Each topic may contain patterns of any of the two forms (keyword/parse patterns). Once a topic is activated (e.g., in our example DIALY_ACTIVITY topic) the engine then checks the topic's assigned patterns. If the parser has given any data (e.g., (I (proposition) walk (verb) home  in the morning(time phrase) it will activate the following pattern  s: ( |subject=you * |timephrase = in the morning)  What else do you like to do _#1 that will produce the output what else do you like to do in the morning. Is that correct? If yes then brilliant, your approach is absolutely brilliant. It reminds me of someone coming to England without knowing English at all. In this scenario the role of parser plays the human gestures... so for example if someone comes and says to that person "shall we go to the market?" and accompanies "go" and "we" with the proper gestures the person will understand that "he" (preposition) is being asked to go (verb) somewhere but he will not understand the destination because he doesn't know the language. It is as you said the person will roughly understand what has been said but not what actually has been said.>>>


BTW I don't believe in this chatterbot challenge.I think the whole contest is a huge waste of time, but I will vote for the bot since i believe you have made a difference.

[wgb14]

just voted Suzane at the most popular bot and to enter the finals. Hope our votes will make a difference....

[one]

@ONE

<<OK as long as you are sure you don't belong in prison for stalking>>

I truly hope you don't mean what you are writing... otherwise its a very heavy insult. I never stalked you or anything and I certainly don't belong to any prison. If you check the discussion with Bruce you will see:

(Bruce): A note to One...
Myself: Not sure which one you are referring to... but i guess it would be me..
Which means that I was not sure if he referred to me as I was the one annoying his bot all the time with repeated inference questions.

Bruce: One- dont apologize for creating problems..........


I think it was only natural to assume that he was referring to me and this was just a way of expressing himself. It is the first time I see you in the forum and hence there is no chance that I would have any intentions of stealing your identity or even stoke you in any way....

I consider the matter closed... I know you now (like I know Freddy and the others) and there is no way to repeat the same mistake again....

There is more than a single thread on this subject and in that thread Freddie was told things look around you'll find it.

[one]

Bruce,
GIGO.
Garbage in garbage out! (Uranus)
some spelling mistakes ect,, we went for a long walk avoiding the lake.


I think a bug might be in a trap or something, Suzette mentioned it when I got spam in my machine?

[bruce wilcox]

thanks for your votes. I'm sure they make a difference. Given the magnitude of the total votes, you could argue I am stuffing the ballot box. But you are not my friends or family, I don't know you, and you make your own decisions. I merely remind you.


As for the strange side thread going on here about identity... It just confuses me and I don't even know if I accidently started it. Could we put it aside, please?

[wgb14]

No problem Bruce... let's forget about that.... anyway.... The only reason i vote for your bot is because i think you have something different than the others. Hope you will get what you want out of the contest.

[one]

I modified my previous post but thought another post wouldn't hurt.

wgb 14 naaah don't worry I think wgb 14 might be as worked up as I was initially , so I understand and have gave a explanation, I hope we can get along without moderators having to worry.


"Peace and wheelies to you", Bruce.

[bruce wilcox]

wgb14 --- you want clarity. I am not attempting to obfuscate or hide the chat-l stuff.  Obviously this would be easier by phone alsmost, instead of back and forth trying to clarify....I'm not really even sure what you don't understand.

I understand you are not asking for code, which is good because I'm not supplying any.

I don't "believe" in this chatterbot challenge either. BUT--- it gets people to test my chatbot, giving me debugging information. Making it into the final round means I get even more testing. Being most popular is not relveant to that goal, only 10th bot status is.

you have asked: If I am not mistaken, you are using two kind of patterns in your bot: The synonym collections (e.g., s:(Aliens AND ~movie) I have seen the movie aliens) and one to deal with information from the parser (e.g., s: ( |subject=you * |timephrase)   What else do you like to do _#1 .) . You have covered extensively the first in your paper (no problem with that) but I am still confused on what exactly have you done with the second (i hope your company allows you to share some information on that)

Your expression is true as far as it goes, but it doesnt go far enough. or it doesn't express it as I think of it. I dont have two kinds of patterns. A rule has a pattern side and an output side. On the pattern side, you run a series of tests. Some tests can say "is this user variable defined, or is it defined to be equal to this value". Some tests can say "does this english word or sequence of words occur in the input, with potential restrictions on where it occurs."  The request can be made explicitly, or it can be via a set of words. Some tests can require things NOT be in the sentence. Some tests say did the system succeed in parsing the input, and if it did do we know if there is a direct object or not. Or it can say was the subject the following, or it can say what adjectives were modifitying the subject, or WHATEVER. T

Basically there are a whole lot of tests one can ask of the input and of the state of the system and of data stored previously or avaiable in recent sentences. LOTS and LOTS of possible tests. So to say it can check keywords and it can check parse data is true. It is not sufficient.  The system supports introspecction into the working of the engine via script, so some tests are even---  When you generated a response, did you change topics.  A pattern testing for that is used to generate transitional sentences.

Does this help or just confuse things more?

[bruce wilcox]

continuing.... MOST topics dont require lots of weird tests. They can do with simple sequences of english words, the use of [] and {} and ! and sets or classes.

But I support topics way outside that norm. The entire "emotion engine" -- the thing that decides whether or not it likes you or hates you, whether or not it is interested or bored with the current topics, is also written in script (changable on the fly) and it makes use of more extensive tests.  And the system topic that manages the conversation (deciding if you seem bored with the current topic or are asking for a wordnet definition of a word, or all sorts of other "speical things" uses more complex patterns.
And then the system has a "store simulation" wherein you can enter, ask the salesperson about the three items for sale, put them into your cart, go thu checkout etc.... That uses patterns that interact heavily with the fact query system.  And the system Geography topic can give road directions between landmarks on Oahu (again a test demo of capability). Again its patterns are less simple.   Fact acqutition about you is another topic. It requires parsed sentences for its patterns.

So.... simple topics use simple patterns and complex topics use complex patterns. The mechanism is all the same. The range of tests is wide.