Question:Answer plugin???

[Cefwyn]

EDIT: Hal with fuzzy logic could be quite an interesting thing. I had thought about it while making my earlier comment about it, but I also think it would be quite difficult to do correctly as hal would either need quite a complex inference system or a lot of pre-defined rules (ie. if topic is X, measurement Y is short, measurement Z is tall; so X being YY tall is fairly short) for just about everything you can think of (height, weight, cost, etc.) that would need it but which would be quite slow to run in script.

quote:Originally posted by ll420ll

If HAL isn't going to store the Q:A pair, then why does Hal store a question at all?  Just to ask question the later and not remember that it equals the user response?  And why store the answer in User_temp and not paired with the question?

The reason hal stores a question is so that it can parrot it back to you. In many cases this is simply adding to the conversational side of hal, but if your response is written in just the right way (sometimes tricky from what I've seen) it does indeed get paired with your response as a sort of QA pair. I've not managed to look over everything in the main brain script but I'm sure that functionality could be made more reliable.

quote:
Your never going to get an answer to a question by selecting few chosen words and doing a search of an entire DB.  
If I ask HAL "How old am I? Hal never answers this correct for me.
It ignores all but the word "old". It will respond with a random response using the word "old"

The way hal uses it's database does indeed have it's issues, and while a very good A.I. it still has room for improvement. The way hal parses your messages you will find it to be more responsive if you type longer messenges. For a question like "How old am I" without putting the answer directly into the QA brain you may find it difficult getting an answer. As hal sees it the word old is simply a modifier to I, and with the word "How" int he beginning it sees it as a question, but since it doesn't strictly know what the word old means in the same way a human does it doesn't recognize that as being a question about your age. You could say "My age is 24", and then later say "What is my age?" and probably get an answer, but as I said teaching hal can be tricky from my experiences.

quote:
Cefwyn that's really cool your looking into this. Even if HAL didn't reload Q:A table at least it would know next time.  I can see where one huge Q:A table would be a problem. Could Q:A tables be sorted by topic?

As far as I can tell at the moment, hal is programmed to treat specific types of tables differently (ie. brain table, sentence table, etc.). The mainQA table is a brain table, and I'm currently unsure exactly how that affects hals scripts, but I vaguely remember seeing some lines which pass the response into the mainQA to get the reply, so I should be able to mess about with it to see what will work.

[onthecuttingedge2005]

the TopicSearch feature is the most accurate of the response functions.
I know this through years of experience with building public plug-in and ultra secret brains, I can tell you the correct methodology of appending more accuracy and understanding.
is like this:

Code: [Select]

1."","" <-- 2 deep
2."","","" <-- 3 deep
3."","","","" <-- 4 deep
4."","","","","" <-- 5 deep
and so on. this is the method I developed for pattern matches to achieve some very excellent response results.

I am at the moment writing and testing and debugging another special human parser type pattern matching plug-in using a similar code I wrote.

the Q/A Response function will probably not be needed in the next 1 or 2 versions of Ultra Hal Assistant when I get done.

when I say ultra secret, I mean stuff I don't want outside A.I programmers to know that they would employ in A.I systems outside of HAL that don't give credit for coming on this forum looking for ideas to steal for their own personal fame, I have seen this over the years and it annoys me a little. give credit where credit is due is my motto.

Jerry[8D]

[Cefwyn]

onthecuttingedge2005:
Nice to hear that at least your having fun []
I don't particularly care for people who do work for their own fame and don't share, as that and copyrights are what has caused many technologies like TTS to stagnate over the years, but I do understand your reasons for not posting stuff on this forum where just anyone can grab it and pretend it's there own. If I had the time I would go through all my old perl chatbots and see if any of it's as good as I remember it. A few years back I coded a chatbot with a parser that could understand the grammar quite well and most of the time could guess if a word was a noun, verb, or pronoun based off of a very minimal list of the more common words. I was even beginning to write code for it to understand latin roots inside of words and use those to gather the meaning of the words and sentences, but unfortunately the parser took me so long to get working that I never actually finished the bot. Something like that would work quite well inside of hal I think as the roots could be stored in the database and retrieved much quicker then my rather messy text file parsing. For some reason english grammar and how it's made up of just about every other languages has always interested me, so I could easily spend years working on a grammar parser and never finish it to my satisfaction.

EDIT: I wish I had discovered hal much sooner as it's exactly the sort of thing I did time and time again with perl, only to a much grander scale.