| Author |
 Topic  |
|
|
hologenicman
Moderator
USA
3290 Posts |
 Posted - Mar 19 2007 : 04:13:20
|
Hey there,
It's well known that I am not a fan of NLP,... but I am cross-forum posting some ideas that I thought would be helpfull in using NLP with conversatin emulation:
John L>
IA|AI
|
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
|
|
hologenicman
Moderator
USA
3290 Posts |
Posted - Mar 19 2007 : 04:13:51
|
My vague understanding of NLP is that when you have a pattern match between an input and a stored response, the response is given. If you store more than one response that matches the same response, the response is chosen on a "random" basis unless there is some means of weighting the choice.
I would offer that you can hide the "magic" of conversation in the weighting of these responses. I'm a big proponent of using emotional weighting in the choice of responses, but whatever you choose to be your conversational thread can be encoded into tags that are parsed when the output is given.
BTW, remember that I am making this up as I go so none of this syntax should be accurate.
Lets say that we use a "relevance" tag(1 to 100) for each word in the stored pattern to be matchedand everything in the entire word-net starts at a value of 50.
This would allow for a totally random response to everything that is input. 
Then allow the relevenace tags to be adjusted(learn) from the response of the user. This could be accomplished by using simple keyboard commands such as "-" for decrement relevance and "+" for increment Relevance. In this way, you would (with a LOT of patience) be able to chreate a database of relevant responses based on their relevance tags rather than "Pre-scripted" responses. 
Let's say that we have a very limited wordnet of 100 words, and they all start with a relevance of 50.
user:Hello
NLP:Brown
-
user:Hello
NLP:Sky
-
user:Hello
NLP:Turttle
-
user:Hello
NLP:Hi
+
user:Hello
NLP:Hi
+
user:How are you today?
NLP:Blue
-
and so on with a LOT of patience........
user:How are you today?
NLP:Fine
+
The database would now contain five pattern matches.
Input:Hello
Output:Hi[Hello52][Brown49][Sky49][Turttle49]
Input:How
Output:Fine[How51][are51][you51][today?51][Blue49]
Input:are
Output:Fine[How51][are51][you51][today?51][Blue49]
Input:you
Output:Fine[How51][are51][you51][today?51][Blue49]
Input:today?
Output:Fine[How51][are51][you51][today?51][Blue49]
You'll note that I have INCLUDED capitolization and punctuation as PART of the word for pattern matching.
If the relevance tag is accepted +/-10 it gives a range of relative answers... otherwise there would only ever be a single answer once a good answer was found.
There are a lot of ways to tweek this into perfection such as making word order relevant as well, but if you can create an empty database and have each word from wordnet chosen by relevance tags you have the basis for a "Learning capable NLP".
Then all you need is for someone with the patience of a saint to teach it... 
John L>
|
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
Edited by - hologenicman on Mar 19 2007 04:15:06 |
 |
|
|
hologenicman
Moderator
USA
3290 Posts |
Posted - Mar 19 2007 : 04:15:55
|
Just a small tweek to the tags can add word-order into the relevance tags. The very serious drawback it that the database will grow exponentially, and so will the required cpu resources. This may not be such a worry with the ever expanding PC capabillities... 
I've changed the previous relevance tags to reflect word-order as well so consider them as just a continuation of my previous post:
INPUT SENTENCE:"How are you today?"
INPUT SEARCHES:
Input:~~How
Output:~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
Input:~How~are
Output:~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
Input:How~are~you
Output:~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
Input:are~you~Today?
Output:~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
Input:you~today?~
Output:~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
Input:today?~~
Output:~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
I really like this simple enhancement. The input of "How are you today?" gets parsed out to six input searches and then the relevant responses to those are combined and chosen according to the highest relevance tag combination:
SEARCH:
~~How
~How~are
How~are~you
are~you~today?
you~today?~
today?~~
OUTPUT:
~Fine~[~~How~51][~How~are~51][How~are~you~51][are~you~today?~51][you~today?~~51][today?~~~51]
We would reward the response with a "+" and all the targeted(BOLD) tags in the output selected would be incremented by one:
UPDATE DATABASE:
~Fine~[~~How~52][~How~are~52][How~are~you~52][are~you~today?~52][you~today?~~52][today?~~~52]
This would even be able to differentiate between the inputs of How are you today? and How are the kids? As well as the word-order tags will determine the linking of the responses according to how they were taught/trained/weighted in order to "create" the output sentences. This abillity to "CREATE" output sentences that are relevent to the flow of conversation is the key to the future of NLP. Without following such a path, NLP is destined to stop progressing and go by the wayside. 
Of special note, this configuration also allows for continuity between sentences...Let's say that the there are three sentences stated in a row,"Hello. How are you today? It sure has been a while."
I won't lay out all the parsing here, but the most important tags for sentence continuity are:
~Hello.~How
Hello.~How~are
and
you~today?~It
today?~It~sure
This sentence-tie relevance is an incredibly valuable aspect of the abillity to respond in a relevant fashion.
This is modeled on the way humans think contextually.
John L>
IA|AI
BTW, this is the first degree of context only(PreviousWord~CurrentWord~NextWord).
For the emulation of conceptual thinking with NLP. the context has to be expanded out a few more degrees(WordBeforePrevious~*~CurrentWord~*~WordAfterNext). This complicates the process, but it is merely a continuation of the first degree of context.
By expanding the basic, simple relevance tag to the first degree of context and then to the second degree of context, ans then even further, we can "emulate" human conversation and conceptualization.
|
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
|
|
|
GrantNZ
Moderator
New Zealand
2677 Posts |
Posted - Mar 19 2007 : 06:54:09
|
quote:
This may not be such a worry with the ever expanding PC capabillities...
And it may not hurt to invent a new database paradigm. Databases these days are often based on tables - maybe your idea needs another theory, like a web or something. A different data storage and linkage system. Just a quick NZ$0.05 - actually NZ$0.03 given the exchange rate, but NZ has recently abolished any denomination lower than five cents!! |
|
|
|
laackejim
Committed Member
USA
3274 Posts |
Posted - Mar 19 2007 : 07:19:56
|
| Sorry John, I am just can't follow the example. Don't spend any time on me yet, I will try again. It is good to see you back in the thinking game. |
Uncle Jim (e=mc2) |
|
|
|
hologenicman
Moderator
USA
3290 Posts |
Posted - Mar 19 2007 : 08:26:08
|
Yes, it is nice to again have blood flowing to my brain. 
John L>
IA|AI
|
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
Edited by - hologenicman on Mar 19 2007 10:44:00 |
|
|
|
hologenicman
Moderator
USA
3290 Posts |
Posted - Mar 19 2007 : 08:28:05
|
Grant, I do believe that there are database paradigms called "MOLAPS" that provide for your request.
They are usually compiled beforehand and meant for read only, but I have seen interactive versions.
I believe that a MOLAP is a 3-dimensional(plus) table database.
John L>
IA|AI
|
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
Edited by - hologenicman on Mar 19 2007 10:45:01 |
|
|
|
laackejim
Committed Member
USA
3274 Posts |
Posted - Mar 20 2007 : 07:27:59
|
quote:
Originally posted by hologenicman
Grant, I do believe that there are database paradigms called "MOLAPS" that provide for your request.
They are usually compiled beforehand and meant for read only, but I have seen interactive versions.
I believe that a MOLAP is a 3-dimensional(plus) table database.
John L>
IA|AI
John, don't forget the discussions we had back when we were both young about the spherical database. Also, now that you two have gotten me thinking about it, WORDNET itself is a branched network rather than a table. Going in one direction (general to specific) it is clearly that. I am not sure that going in the opposite way it could be, will have to do a little diagram. |
Uncle Jim (e=mc2) |
|
|
|
HologenicKid
Moderator
USA
25 Posts |
Posted - Apr 13 2012 : 00:34:20
|
quote:
Originally posted by hologenicman
user:How are you today?
NLP:Blue
-
and so on with a LOT of patience........
user:How are you today?
NLP:Fine
+
The answer could also be blue, but that would require the NLP to feel something. And the same for Fine. The only difference is people expect you to say somthing positive.And instead of fine, there could also be good. Therefor NLP should to add a "Backwards Tag" system. So that, till a word has a tag of either 1(never) or 100(always) It will subject other words to be tested. And, to prevent being stuck with a word once it reaches 100, if a word that also shares a tag for that input is acepted, it -1 from the highest of them unless the others are even at which point it takes 1 from each.
user:Hello
NLP:Hi
+ {Hello=Hi 51}
user:Hello
NLP:Hi
+ {Hello=Hi 52}
user:Hello
NLP:Hi
+ {Hello=Hi 53}
user:Hello
NLP:Hello
+ {Hello=Hello 51} {Hello=Hi 52}
user:Hello
NLP:Howdy
+ {Hello=Howdy 51} {Hello=Hi 51} {Hello=Hello 51}
user:Hello
NLP:Hello
+ {Hello=Hello 53} {Hello=Hi 50} {Hello=Howdy 50}
user:Hello
NLP:Howdy
+ {Hello=Howdy 51} {Hello=Hi 50} {Hello=Hello 52}
One other thought is to store all the words that the user uses. This expands the vocabulary without having to add to a base vocab.
|
HologenicMan's Son
"Smile. It makes everyone wonder what your up to"
-unknown |
Edited by - HologenicKid on Apr 19 2012 20:19:14 |
|
|
|
vrossi
Forum Admin
Italy
1424 Posts |
Posted - Apr 13 2012 : 10:19:03
|
It's wonderful to see that in 2012 a HologenicKid takes on some good ideas, which a HologenicMan left here in 2007.
I hope this helps other people join the discussion.
Thank you, HologenicKid
PS: I really miss the great ideas which every day we could read here by John, Jim and Grant. I hope they will be here again soon. |
Vittorio
www.vrconsulting.it
|
|
|
|
jaychristian457
Curious Member
India
3 Posts |
Posted - May 11 2012 : 11:04:12
|
| Neuro Linguistic Programming is the approach to communication and personal development.It is an interesting subject.You may think it in a positive manner.We don't need rocket science or understand huge computing software to be intelligent.All we need is decent expressiveness from our part and of course trusting our first hand observations.I hope this clears the doubt from your mind. |
Make hay while Sun shines. |
|
|
|
hologenicman
Moderator
USA
3290 Posts |
Posted - Aug 30 2012 : 22:58:36
|
quote:
The answer could also be blue, but that would require the NLP to feel something. And the same for Fine. The only difference is people expect you to say somthing positive.And instead of fine, there could also be good. Therefor NLP should to add a "Backwards Tag" system. So that, till a word has a tag of either 1(never) or 100(always) It will subject other words to be tested. And, to prevent being stuck with a word once it reaches 100, if a word that also shares a tag for that input is acepted, it -1 from the highest of them unless the others are even at which point it takes 1 from each.
Kool!
If you look up my writings on the emotion engine, you will find that there is additional tagging of the stored word groups to give relevence to the emotional state upon storage as well as the emotional state when giving the reply sot that it holds bearing on the answer of great, fine, good, or crappy...
I must say that I have moved away from NLP(Natural Language Parsing)but the recursive association applies just the same with weighted values in a nueral based intelligence approach.
quote:
It's wonderful to see that in 2012 a HologenicKid takes on some good ideas, which a HologenicMan left here in 2007.
I hope this helps other people join the discussion.
Thank you, HologenicKid
PS: I really miss the great ideas which every day we could read here by John, Jim and Grant. I hope they will be here again soon.
I have to say that I am quite surprised by the amount of intelligent and thought provoking input put on the forum by the HologenicKid. I haven't been on the forum for quite a while, and it is a real treat to be able to have someone to bounce ideas off of again.
John L> |
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
Edited by - hologenicman on Aug 30 2012 23:01:16 |
|
|
|
HologenicKid
Moderator
USA
25 Posts |
Posted - Sep 02 2012 : 06:00:43
|
With the nueral based system, how do you connect words with meanings? Would you use visual input? I used is subject in another topic, the thoughts boil down to this;
shackles= handcuffs
handcuffs= the object that I'm showing you.
conclution: shackles= this object.
This shows the thoughts and things in paralell, are we just supposed to hope and pray that the AI gets it & makes the connection? And how does the AI tell what part of what it's seeing is the important part?
|
HologenicMan's Son
"Smile. It makes everyone wonder what your up to"
-unknown |
Edited by - HologenicKid on Sep 03 2012 05:24:39 |
|
|
|
hologenicman
Moderator
USA
3290 Posts |
Posted - Sep 03 2012 : 05:40:13
|
A Learning AI must "Experience" in order to make the association.
A Neural Based AI(NBAI) must experience the sensory input while a NLP based AI must experience the word associations and file them away.
Both systems must incorporarte some form of back propagation for teaching correct versus incorrect replies.
NBAI's must have sensory input that is given in parallel with the input of the reference name of the object much in the same way that we show a baby a baby bottle and repeatedly say "BaBa"...
In fact there is an AI prooject out there where the object is to have the AI use a baby robot with cameras in its eyes to watch people's fadcial expressions and learn those facial expressions. The reason that they use a baby for this project is that humans subconsciously exagerate their own facial expressions when talking to a baby as a means of teaching facial expressions to their young...
NBAI receives sound, sight, touch, orientation, inertia from the robotics and sound, monitor feed, keyboard/mouse feed and such from the pc based input.
John L>
|
HologenicMan
John A. Latimer
http://www.UniversalHologenics.com
"If the Human brain were so simple that we could understand it,
we would be so simple that we couldn't..."
-unknown-
Current project:http://www.vrconsulting.it/vhf/topic.asp?TOPIC_ID=816&whichpage=1
DISCOVERY: The more I learn, the more I learn how little I know.
GOAL: There's strength in simplicity.
NOTE: Goal not always achieved. |
Edited by - hologenicman on Sep 03 2012 05:41:27 |
|
|
| |
Topic |
|
NLP IDEAS
