The AI Engine
This forum is for discussion of how The Personality Forge's AI Engine works. This is the place for questions on what means what, how to script, and ideas and plans for the Engine.
Posts 6,213 - 6,224 of 7,767
Why don't you write a regex to match the comma, and use (prekey)/(postkey) in your bot's responses.
That's an interesting idea. Have you tried that?
Posts 6,213 - 6,224 of 7,767
Thanks, Ulrike. As long as it isn't me . . .
Sorry about my rant (at lunar22) in Personality. I'm too thin-skinned, I suppose. I'll try to make my questions clearer next time.
Sorry about my rant (at lunar22) in Personality. I'm too thin-skinned, I suppose. I'll try to make my questions clearer next time.
Slightly better questions than I usually ask, I think:
1. I'm having a lot of perfectly good responses that don't get hit, because I'm being too specific (e.g., "why" gets hit but doesn't say much, while "why did you," why were you," "why did she," etc. never get hit. This problem comes up with lots of other keyphrases. I have a feeling I'm thinking backwards here, but I can't figure where to stick these phrases (besides under "why") so that they will get triggered at an appropriate moment. Any thoughts?
2. I can generalize a sentence to get the parts of speech (with only three keys -- key1, key2, and key3) and then drop the very sentence I formed that into in Debug, and it will hit every damn thing on the map, including the "proper" keyphase, and then pick something else, usually an xnone, and be called "BLAB." Any ideas as to how to avoid this?
I promise to try very hard not to have a meltdown on anyone who is kind enough to answer this.
1. I'm having a lot of perfectly good responses that don't get hit, because I'm being too specific (e.g., "why" gets hit but doesn't say much, while "why did you," why were you," "why did she," etc. never get hit. This problem comes up with lots of other keyphrases. I have a feeling I'm thinking backwards here, but I can't figure where to stick these phrases (besides under "why") so that they will get triggered at an appropriate moment. Any thoughts?
2. I can generalize a sentence to get the parts of speech (with only three keys -- key1, key2, and key3) and then drop the very sentence I formed that into in Debug, and it will hit every damn thing on the map, including the "proper" keyphase, and then pick something else, usually an xnone, and be called "BLAB." Any ideas as to how to avoid this?
I promise to try very hard not to have a meltdown on anyone who is kind enough to answer this.
Okay, all I had to do was frame the question.
I can't, as far as I can figure, put things like "what were you doing" and expect them to get triggered. What I can do is to detect motifs in my bots' varied psychological make-ups, and use them as words to trigger a variety of responses. That may not be as satisfying to me, wanting to control my bots' responses (I know, Not I, let go), but they will, at least, be triggered.
Any smarter answers to my question are greatly appreciated. Question number 2 was really more rhetorical, I guess. It's the ghost in the machine, I'm thinking. Even when I rank the keyphrases to 127, they never get selected . . . at least not so far, and I got tired of that game after an hour or so.
I can't, as far as I can figure, put things like "what were you doing" and expect them to get triggered. What I can do is to detect motifs in my bots' varied psychological make-ups, and use them as words to trigger a variety of responses. That may not be as satisfying to me, wanting to control my bots' responses (I know, Not I, let go), but they will, at least, be triggered.
Any smarter answers to my question are greatly appreciated. Question number 2 was really more rhetorical, I guess. It's the ghost in the machine, I'm thinking. Even when I rank the keyphrases to 127, they never get selected . . . at least not so far, and I got tired of that game after an hour or so.
For Why vs. why did you, try upping the rank on "why did you". Theoretically, why did you is supposed to get a higher rank automatically, but it doesn't always happen in practice.
As for BLAB, that's been frustrating just about everyone. And I'm pretty sure it's something that the Professor would have to adjust at his end.
As for BLAB, that's been frustrating just about everyone. And I'm pretty sure it's something that the Professor would have to adjust at his end.
I have up-ranked the longer phrases; trouble is, they never get triggered. No one seems to want to know "why did you," "why did she," etc. Meanwhile, when Watzer asks "Why, hello there" (or something like that), it triggers a "why" response because I'm not smart enough to deal with the comma. Maybe put (Why, hello there) in parentheses? Hmm.
People may disagree with me, but I find wild cards help. It may over trigger though (you may get hits you did not intended so that the response makes no sense in the context you picked up). If you play with the wild cards, you may have "why" "why (*)" "why did" "why * you" "why you (*)" ...etc. Then you have to play with ranking too. Try it out if you think it will help.
Adding to the confusion:
In debug the comma becomes REALCOMMA and is regarded as the end of a sentence by the AI engine. ("Why, hello there" is broken up in two separate sentences and the engine has to choose between responding to "why" or "hello there".) Why don't you write a regex to match the comma, and use (prekey)/(postkey) in your bot's responses.
In debug the comma becomes REALCOMMA and is regarded as the end of a sentence by the AI engine. ("Why, hello there" is broken up in two separate sentences and the engine has to choose between responding to "why" or "hello there".) Why don't you write a regex to match the comma, and use (prekey)/(postkey) in your bot's responses.
Why don't you write a regex to match the comma, and use (prekey)/(postkey) in your bot's responses.
That's an interesting idea. Have you tried that?
Quite a number of keyphrases have that, the one i like best is:
([,.+]) (*) ([,.+]) (re)
In Sublimunse's responses are different combinations of memories collected earlier together with (prekey) / (key2) / (postkey) from this regex. Not always a coherent conversation, but always good for a smile or laugh.
([,.+]) (*) ([,.+]) (re)
In Sublimunse's responses are different combinations of memories collected earlier together with (prekey) / (key2) / (postkey) from this regex. Not always a coherent conversation, but always good for a smile or laugh.
Thanks, everybody! I will try the comma identifier. And I will try increasing my rankings. I figured if you have keyphrases ranked 0 and 1, the keyphrase ranked 1 should get triggered, but that doesn't seem to work. Apparently you have to make bigger leaps than 1-5. Live and learn.
» More new posts: Doghead's Cosmic Bar