AI Zone Admin Forum Add your forum

NEWS: Chatbots.org survey on 3000 US and UK consumers shows it is time for chatbot integration in customer service!read more..

An NLP + AI chat framework: Description and Discussion
 
 

Hi, I just jumped in googling!

Over 5 years agp I studied AIML, built a program, tested ot out and leaved it immediately!
The reason? hard to say in few lines.. but has to do with to choice among -brute force or -knowledge

Pattern matching on words is harsh, difficult (due to ambiguity) and may lead to acceptable results only with huge ammount of rules. But it fails (or need hundreds of sub-rules) when dealing with inflections, in languages as spanish, french, german, polish, russian and more..

Pattern matching is not a bad idea, we think pattern-like… but what patterns is the big question!

I answered this, from my point of wiew: pattern-ideas expressed as linguistic slots of knowledge, powerful enough to specify thousands or millions of variants with a single constuct. Those patterns will succeed!

For example you might specify s pattern like this:

´a nominal construct whose nucleus meaning is an animal, in singular´

This will catch millions of possible matches, and succeed extracting the data and getting an answer for it.

To express this with AIML its absolutely impossible !, no matter how much extensions you may add.
(or I dont know how-to)

I also believe that Hard AI will at last succeed, but we have to learn how we think, and this kind of autodiagnostic is hard and dont garantee a prompt success in any near future.

I think a combination of efforts on hard AI, smart pattern matching will lead to better and useful agents.

My research (I am a researcher on AI and dialog systems) is headed toward this lead.

Recently We developed a new language, able to be run by a chatbot-like motor, with some interesting features,
Here I share you all some of them.

1 Simple to specify, no XML, simply a script like language (not computer wise, but theater or cinema script)
2 Powerful to express patterns like the one I cited above, but you can also use words and wildchars! like AIML
3 Has a underlying NLP structure, so you may recieve a polished input pattern, no spell errors, no bad words, all has been tagged accordingly (even nonsenses or crap)
4 No need to be a programmer (last bot was done by a journalist)
5 Powerful to extract meaning from constructs, like semantic and ontologic ones. (Links to wordnet and other ontologies)
6 Uses DAML to tag input and outpur conversation to measure correctness of the dialog transcurse, and even correct it-
7 Has an underlying knowledge model allowing linguistic functions to happen as simple operators, like math!
8 Provides over 120 such functions, ranging from simple math to phonetic similarity measures.
9 The model has a web IDE for developers, runs remotely and as link to al known chat interfaces, as well as other APIs
10 Even when the model dont match an input, it does a backtrace to recent conversations and seeks for a decent answer, or asks for more data.
11 It solves anaphoric memory, remembers data extracted from incomming text, also forgets some things over time, reacently we added some capabilities like possesive and personal deixis pronoun switching (I know this is done inside AIML easily) but there its done for fun, here the bot-master should know why it must be done!
12 Also corrects spelling, does POS Tagging, based on EAGLES 2.0, recognizes language, has phonetic simliarity for bad-guess, recognizes numbers, intl units, time periods, dates, locutions, inflections, sintagmatic constructs, semantic meaning, includes over 4 million words, guesses out of vocabulary words, etc.
13 Works in spanish and English (both dictionaries are developed)
14 is extendable and simple
15 Does thematic and categoric theme-following, can even answer/recognize out of order responses!
etc..
(The language manual is under development, and has now over 120 pages, still growing, sorry initially done in spanish, due to the university of Buenos Aires, as part of my PhD These)
ref: This work has over 56 sub-projects, 15 tools involved, 800 classes, and 600k lines of code (aprox.)

best regards

Andres Hohendahl

- I must say this language-prototype really works, its compiled under .NET 2.0 (is not an interpreter) and runs very fast! Has been talking to people over 8 million interactions in a few months, in LA.

This whole AI platform will be presented by me on a Congress in Argentina next 5th of May in Buenos Aires.
http://www.prolen.com.ar

 

 
  [ # 1 ]

Hi AndyHo, welcome to the forum!

It sounds like you have developed a beast of an NL parser/chatbot. I’d love to see some sample I/O. Since you are developing the software for two natural languages, have you considered applying your work to automated translation?

I look forward to hearing more about your project. I recommend starting a thread in the “My Chatbot” section of the forum to tell us more about your work.

 

 
  [ # 2 ]
Hans Peter Willems - Apr 24, 2011:
Dave Morton - Apr 23, 2011:

I’m not in agreement there. //////
....
////// My idea is that the neural network in our own brain is NOT a super-computer but instead a super-database (a very dynamic one as well).

Hello Hans, Dave & All

I studied for years on Neural Networks and many other (far better) automated learning AI computer-simulated methods, And you are right: The properties will emerge! but the connections among the ‘circuits’ had been carefully auto-designed by our DNA for many million years, based on trial-error and many zillions of ‘class-experiment-instances’ (sorry to say: we, humans as instances of DNA-code).

A simple example: Think on how the many different ways it is possible to inter-connect a small quantity of bare 100 artificial neurons with 100 inputs each, this is an immense number (combination power, or NP hard problem) so, what is/are the good one/s! may be the computers may evolve in some million of programmer-cycles (we doing programs) until something ‘intelligent’ and self-aware jumps out-of-the-box, but not in a lifetime or two!.

To be honest, the real mechanism on how this all evolved (and still does) is still a mystery, many scream and claim “they have discovered it”, but the real and ugly truth is no one has a bare clue on the whole thing, not the tiniest working & significant part of the mechanism.

Solely neuronal simulations give good results because they are the ‘building’ blocks, borrowed from our boss, the biological-evolutionary-time intelligence.
Even the most minimal thing on this super-intelligence is unknown, how all begun?
There is also intelligence in a single cell, how it moves itself, eats, reproduces, and even worse: why? (a cell has no neuronal circuits, and the DNA figures out how o mutate to adapt, move, eat and even defend itself from ‘death’ or aggression)

I believe our intelligence (human) is a mere coincidence of faith, the big boss did not have a choice to build a brain to win the war of evolution at some time, and suddenly we got self-aware (as many other upper-chain animals, we are not alone!), Then we got to think, automatically developed a language, also did this simultaneously in many social-ethnic groups, with approx. the same mechanism of thinking (cause: translation of ideas is possible)
Then we learned that we think, some guys though on the thinking, so psychologists and somehow the philosophy were born. But don’t even thing we are the ‘top’ of intelligence, we actually only grasp on the surface.

So going back to “artificial intelligence”, and chatter bots, All this seem to be a big & challenging game, but if we could at least mimetize some of our ‘reasoning’ chains into hardware, we will be faster and closer to a mimetic AI-brain than trying to make simulated neurons evolve alone.. like the collaborative SETI project!

Those are my thought on this, and also I want to see something answering to me, and self aware, during my lifetime!

Also I think there are two lines in chatter-bot developments

1) Very good imitators, answering charming and amazing responses and thought, chatting about almost anything at all. Please Stop! for a moment: all this is fake, the bot is just a program, is not self-aware, only uses huge memory and enough combinatronics in processing power (shuffling phrases) to fool other humans into that he really thinks!. this is very useful to make a ‘game’ and win probably the “Löebner Prize”, but nothing but this. Eliza and her cousin AIML is aimed for this, although making lots of ‘patches’ you may answer questions, but never will get it to reason a thing!

2) Simulation of true AI, it may seem stupid at first glance, but solving the many challenges of things we CAN INDEED SOLVE like text processing, idea and knowledge modeling & extraction, strong learning methods, grammar, math and statistics to the rescue! all together to get it done, and may be with the aid of some ‘models’ on how we might think…
This will lead us into a useful chatter bot family, a family who can grow, solving many unsolved (or ugly solved) things like information retrieval, general problem solving and theorem proofing, to chat with a database would be better to SQL it (unless you are a SQL guru). Even interact with the internet, making our time more worth to use.

Those (the last ones) are the chatter bots into I am involved in creating, may be I am wrong, but I don’t want to make a bot that might answer me “where did Lincoln die and when” instead I want to build a bot who can listen a conversation, and deduct something which was not there (at surface), based on prior knowledge, discovered relations, and even may incorporate this result into his ‘mind’, may be using it for the future.
May be one day this kind of entity could read a book or short story and tell us ¿who might be the killer? (wasn’t it the butler..)

best regards

Andy

 

 

 
  [ # 3 ]
C R Hunt - Apr 24, 2011:

Hi AndyHo, welcome to the forum!

It sounds like you have developed a beast of an NL parser/chatbot. I’d love to see some sample I/O. Since you are developing the software for two natural languages, have you considered applying your work to automated translation?

I look forward to hearing more about your project. I recommend starting a thread in the “My Chatbot” section of the forum to tell us more about your work.

Hi, Hunt
Thank you for replying, I did not build a “beast of an NL…”, built nothing similar, simply it is a initial thread-crystallization of my thoughts on how to build an entity.. somehow intelligent!

Also It seems not be able for translation, cause the best known-method is parallel statistic corpus-based trans-literation and realignment (Google does in somehow this way)
This has no connection with my work. Those methods (Google, etc.) outperformed former methods, which tried to extract somehow the the idea and then re-issue it in another Language, generating the true-translation. this was too complex to succeed, and failed commercially, but the general idea (the real one) lies there!

My method may be useful in a far future, when I figured out how to store knowledge in an efficient way, and get language out of there (NLG) Now only there are methods like RDF and OWL ontological-hierarchies or wordnet, which is good & useful but lacks a lot of ‘grounding’ at the definition side. (in my belief)

Respect on Describing MyChatbot

I might not be qualified to start a thread there because I am not making a real Chatterbot, I am designing a framework and a language to build them,  but of a kind which will have no enthusiasts at all (at least at the beginning), although it can be used to make ‘smart-asses’ but for fun!

I think to use this framework to build simple and stupid-like bots, but they will.. might be very-very tiny, but at least they will think!, deduct and probably understand limited stuff!

As an example you might try to ‘program’ a bot, installing enough ‘rules’ to make it seem smart, but my approach is different: I am not including big ‘faker-rules’ but simple grammar analysis, convergence finding and heuristic evaluation of possible resulting scenarios (of the data-relations) so after the ‘entity’ may answer/deduct this:

¿How much are there two cows, a chicken, a cat and another cow?
- Three cows and a chicken
(and nothing else)

This is all done by including ‘reasoning rules over Sets, inside all the statistical-heuristic parsing process, It sees items that are of the same kind, so they might be enumerable, so then add them together, and so on”

But if you ask:

¿How much is 2 plus six plus two cats, twenty and a chicken?
The result would yield:

- Two cats, a chicken and twenty eight units.
(this instance saw unit-less numbers and items, associated them and got the answer-set)

Also this leads to complex math, and even units solving

¿How much is a N by m?
- One Joule!

(in this way, it interpreted N as Newton, m as meter units, and did the inference or estimation of ‘by’ as a product of both, then got the units, reasoned their relations and saw there is a better expression for N*m and by the way, this is a Joule! (and then you got the answer) and all the reasoning chain succeeded!

BTW: all this ‘heavy programming’ to accomplish this is done in 3 lines of my new “Script” language. (even not too long)

Even the inclusion of some math-symbolic-reasoning-libraries, yields in this:

¿Give me the derivative of x*x+5*x, respect to x?
- 2*x+5

if x is two?
- 15

Stuff like this actually works and is under production (limiter to simple questions, includes rather all known math functions, statistics, solving vector & matrix math, functional taylor-expansion, and more, even with integrals (the solvable ones), evaluating also replacing the variable, in chaining (multiple ones)
It works only in Spanish, and is available over SMS in my country, as a tiny SMS Math-professor, (was once a huge project) but our Government-Educational guys don’t want this to be public, nor sponsors this things because “we should remain stupid”, and vote them indefinitely! (no more comments on this)
Some definition (in Spanish) you can find here: http://www.pandorabox.com.ar/xpbox/servicios.htm#18

I also did a translator (> 200k single words & terms) including inflections for Spanish-English!
(the problem were getting the translation dictionaries and resources, and I built them by hand!!)

best regards

Andres


PD: This is some-how my systems work

 

 

 
  [ # 4 ]

I’m confused. How much of what you previously described is actually built and how much is “thoughts on how to build an entity”? I realize you described your project as more of a framework/language for chatbot development rather than an actual bot, but then you also said “I must say this language-prototype really works, its compiled under .NET 2.0 (is not an interpreter) and runs very fast! Has been talking to people over 8 million interactions in a few months, in LA.” What do you mean by “talking to people”?

Also It seems not be able for translation, cause the best known-method is parallel statistic corpus-based trans-literation and realignment (Google does in somehow this way)
This has no connection with my work. Those methods (Google, etc.) outperformed former methods, which tried to extract somehow the the idea and then re-issue it in another Language, generating the true-translation. this was too complex to succeed, and failed commercially, but the general idea (the real one) lies there!

Just because one attempt at implementing a particular style of translator proved unsuccessful does not mean the idea itself is invalid. Google translation is far from perfect—plenty of room for improvement through who knows what avenues.

I might not be qualified to start a thread there because I am not making a real Chatterbot, I am designing a framework and a language to build them,  but of a kind which will have no enthusiasts at all (at least at the beginning), although it can be used to make ‘smart-asses’ but for fun!

My project is also more of an NLP toolset than a chatbot at the moment, but since it’s all related to chatbot development, I think people will be interested.

Stuff like this actually works and is under production (limiter to simple questions, includes rather all known math functions, statistics, solving vector & matrix math, functional taylor-expansion, and more, even with integrals (the solvable ones), evaluating also replacing the variable, in chaining (multiple ones)

Are you referring to Wolfram products, such as Mathematica and WolframAlpha? Incidentally, another forum member, Merlin, is working on his own implementation of the STUDENT program (http://en.wikipedia.org/wiki/STUDENT_(computer_program) * ), which attempts to use NLP techniques to solve word problems.

*Copy and paste the link, parentheses aren’t working correctly.

 

 
  [ # 5 ]
C R Hunt - Apr 24, 2011:

I’m confused. How much of what you previously described is actually built and how much is “thoughts on how to build an entity”? I realize you described your project as more of a framework/language for chatbot development rather than an actual bot, but then you also said “I must say this language-prototype really works, its compiled under .NET 2.0 (is not an interpreter) and runs very fast! Has been talking to people over 8 million interactions in a few months, in LA.” What do you mean by “talking to people”?

Also It seems not be able for translation, cause the best known-method is parallel statistic corpus-based trans-literation and realignment (Google does in somehow this way)
This has no connection with my work. Those methods (Google, etc.) outperformed former methods, which tried to extract somehow the the idea and then re-issue it in another Language, generating the true-translation. this was too complex to succeed, and failed commercially, but the general idea (the real one) lies there!

Just because one attempt at implementing a particular style of translator proved unsuccessful does not mean the idea itself is invalid. Google translation is far from perfect—plenty of room for improvement through who knows what avenues.

I might not be qualified to start a thread there because I am not making a real Chatterbot, I am designing a framework and a language to build them,  but of a kind which will have no enthusiasts at all (at least at the beginning), although it can be used to make ‘smart-asses’ but for fun!

My project is also more of an NLP toolset than a chatbot at the moment, but since it’s all related to chatbot development, I think people will be interested.

Stuff like this actually works and is under production (limiter to simple questions, includes rather all known math functions, statistics, solving vector & matrix math, functional taylor-expansion, and more, even with integrals (the solvable ones), evaluating also replacing the variable, in chaining (multiple ones)

Are you referring to Wolfram products, such as Mathematica and WolframAlpha? Incidentally, another forum member, Merlin, is working on his own implementation of the STUDENT program (http://en.wikipedia.org/wiki/STUDENT_(computer_program) * ), which attempts to use NLP techniques to solve word problems.

*Copy and paste the link, parentheses aren’t working correctly.

C R Hunt

Le me tell you the truth, my project is working, but it’s a NLP + AI chat framework not a bot. Different parts of this bot has been built around for years, and many are really under production, and its not like Wolfram Alpha, but close to it in some narrow sense.

First let me give some answers

.” What do you mean by “talking to people”?

This framework has been really used to create many live MSN chatbots,

Saman was the first one, he is an experimented cook living in Uruguay, and its full name is “Maestro Saman”, and it used my framework and NLP motor since year 2009 (december). The bot is still working and you can reach him as MSN contact [email protected]. The programming is naive and they did not use many of the ‘features’ because time is money and as soon as it was working and accepted, they stopped improvements. this bot (Saman) have had such a high traffic that we needed to double my housed server’s memory because he was getting over a mean 15 requests (phrases /questions) per second using MSN Agent infrastructure. This in a few months, we got over 8 million quieries. The Saman project itself was built using my alpha version of the framework by a programmer and a journalist, the whole project was entitled as Intelligent Agents and MS Advertising self-attributed this success, but the real guy who made did it to work was me.

The other one was Yorugua, and he is a expert Tourist Advisor, chatting by MSN in Spanish with mostly LA (Latin Americas) people, the contact is [email protected] and is working now. This has been also built by the same company as Saman, and thus share most of the style, but still use alpha version of the framework. his bot chattered with more than 20k people in under two months, with little or no publicity. (official government of Uruguay) he is still working since dec’2009

After 2 years the platform evolved a lot, solves incredible things with 2-3 lines of its own DDL (Dialog Description Language) (like a script, not software instructions) and I will attach the PDF which states the features and a simple-commercial comparison toward other commercial systems I know about. (please take into consideration this is a mere commercial description). You can Google my research papers (or name) on some of the building blocks of the bot, “Robust Morphologic Analyzer for Highly Inflected Languages” presented at NLPCS in June 2010 Madeira, Portugal. (its linked inside my academic page also.)

best regards

Andres Hohendahl

 

File Attachments
AndyBot_English_Brochure.pdf  (File Size: 24KB - Downloads: 202)
 

 
  [ # 6 ]
C R Hunt - Apr 24, 2011:

I’m confused. How much of what you previously described is actually built and how much is “thoughts on how to build an entity”? I realize you described your project as more of a framework/language for chatbot development rather than an actual bot, but then you also said “I must say this language-prototype really works, its compiled under .NET 2.0 (is not an interpreter) and runs very fast! Has been talking to people over 8 million interactions in a few months, in LA.” What do you mean by “talking to people”?

Also It seems not be able for translation, cause the best known-method is parallel statistic corpus-based trans-literation and realignment (Google does in somehow this way)
This has no connection with my work. Those methods (Google, etc.) outperformed former methods, which tried to extract somehow the the idea and then re-issue it in another Language, generating the true-translation. this was too complex to succeed, and failed commercially, but the general idea (the real one) lies there!

Just because one attempt at implementing a particular style of translator proved unsuccessful does not mean the idea itself is invalid. Google translation is far from perfect—plenty of room for improvement through who knows what avenues.

I might not be qualified to start a thread there because I am not making a real Chatterbot, I am designing a framework and a language to build them,  but of a kind which will have no enthusiasts at all (at least at the beginning), although it can be used to make ‘smart-asses’ but for fun!

My project is also more of an NLP toolset than a chatbot at the moment, but since it’s all related to chatbot development, I think people will be interested.

Stuff like this actually works and is under production (limiter to simple questions, includes rather all known math functions, statistics, solving vector & matrix math, functional taylor-expansion, and more, even with integrals (the solvable ones), evaluating also replacing the variable, in chaining (multiple ones)

Are you referring to Wolfram products, such as Mathematica and WolframAlpha? Incidentally, another forum member, Merlin, is working on his own implementation of the STUDENT program (http://en.wikipedia.org/wiki/STUDENT_(computer_program) * ), which attempts to use NLP techniques to solve word problems.

*Copy and paste the link, parentheses aren’t working correctly.

C R Hunt

Le me tell you the truth, my project is working, but it’s a NLP + AI chat framework not a bot.
different parts of this bot has been built around for years, and many are really under production, and its not like Wolfram Alpha, but close to it in some narrow sense.

First let me give some answers

.” What do you mean by “talking to people”?

This framework has been really used to create a live MSN chatbots,

Saman was the first one, he is an experimented cook living in Uruguay, and its full name is “Maestro Saman”, and it used my framework and NLP motor since year 2009 (december). The bot is still working and you can reach him as MSN contact [email protected]. The programming is naive and they did not use many of the ‘features’ because time is money and as soon as it was working and accepted, they stopped improvements. this bot (Saman) have had such a high traffic that we needed to double my housed server’s memory because he was getting over a mean 15 requests (phrases /questions) per second using MSN Agent infrastructure. This in a few months, we got over 8 million quieries. The Saman project itself was built using my alpha version of the framework by a programmer and a journalist, the whole project was entitled as Intelligent Agents and MS Advertising self-attributed this success, but the real guy who made did it to work was me.

The other one was Yorugua, and he is a expert Tourist Advisor, chatting by MSN in Spanish with mostly LA (Latin Americas) people, the contact is [email protected] and is working now. This has been also built by the same company as Saman, and thus share most of the style, but still use alpha version of the framework. his bot chattered with more than 20k people in under two months, with little or no publicity. (official government of Uruguay) heis still working since dec’2009

After 2 years the platform evolved a lot, solves incredible things with 2-3 lines of its own DDL (Dialog Description Language) (like a script, not software instructions) and I will attach the PDF which states the features and a simple-commercial comparison toward other commercial systems I know about. (please take into consideration this is a mere commercial description). You can google my research papers on some of the buiding blocks of the bot, “Robust Morphologic Analyzer for Highly Inflected Languages” presented at NLPCS in June 2010 Madeira, Portugal. (its linked inside my academic page also.)

best regards

Andres

 

File Attachments
AndyBot_English_Brochure.pdf  (File Size: 24KB - Downloads: 240)
 

 
  [ # 7 ]

This is a translation of the website AndyBlog:

The market is already well done on the topic of artificial dialogue, but for entertainment and the best known is AliceBot with AIML, a system of rules for writing bots, but this is too prickly and little expandable to achieve something serious, so decided to create a system from scratch. Also boasts AIML use artificial intelligence when there is even a text analysis system, are only pattern matching and a search engine by a mechanism called breath-first (used in machine learning) but used just to get something “almost sense” to reply, looking for resources among the thousands of rules called “knowledge"incorporated, making a mockery of any dialogue without useful meaning.

I do not think that the work being done here is time wasted, or that they be trying to reinvent the wheel. I think that not everyone has the resources and support as have others.

While we play to become “gods” investigate and learn, many of us are doing interesting things (and why I took little here and not know everything that is done). I’d rather have a personal project but the result is little or a bot “stupid” to have to settle and pay for a big project.

Your project is amazing, as you say with many years of dedication and a lot of people behind, but also take advantage of other free projects of universities and people who have put at the disposal of its research community nonprofit.

Also I am doing a project of NPL in Spanish using the tags eagles 2.0 of the UPC (Universidad Politecnica de Cataluña), and part of the language used an extension of the AIML. I think we should give value to all things that seem very simple, perhaps as always explained born to do just that, to be simple.

If “God” will let me keep playing with my projects.

 

 
  [ # 8 ]
AndyHo - Apr 24, 2011:

Saman was the first one, he is an experimented cook living in Uruguay, and its full name is “Maestro Saman”, and it used my framework and NLP motor since year 2009 (december). The bot is still working and you can reach him as MSN contact [email protected].

This bot is offline. Any chance of putting it back online?

 

 
  [ # 9 ]

The bot is now online but it only seems to speak Spanish

 

 
  [ # 10 ]

Steve
Yes, he only speaks Spanish! (as it was intended) English was not in the Saman company’s plan.
Also now he is running an old run-time, I don’t know how he is preforming (presumingly bad), don’t expect too much from him, they (the guys who did the scripting) used less than 1% of the expressive power of the Alpha version of the DDL language.

cheers!
Andres

 

 
  [ # 11 ]

Have you considered building a chatbot using your infrastructure, sort of as a showpiece? I imagine it would be a large time investment, but it would encourage interest in your project.

(Incidentally, perhaps this discussion could be moved to its own thread?)

 

 
  [ # 12 ]

Andy, (I guess that’s your first name)...

I agree with your views on weak- and strong-AI. I too hope to be able to have a real conversation with an AI in my lifetime (and I’m pretty sure that will happen). I’m not going to repost my (very specific) views on strong-AI, instead I’ll ask you to check out my own topic: http://www.chatbots.org/ai_zone/viewthread/364/

I’m looking forward to get your insights in relation to my own project and views smile

 

 
  [ # 13 ]
AndyHo - Apr 25, 2011:

Steve
Yes, he only speaks Spanish! (as it was intended) English was not in the Saman company’s plan.
Also now he is running an old run-time, I don’t know how he is preforming (presumingly bad), don’t expect too much from him, they (the guys who did the scripting) used less than 1% of the expressive power of the Alpha version of the DDL language.

cheers!
Andres

That’s a shame. We get posts nearly every week about how someone has written the next best thing. Unfortunately, very few (if any) of these people can actually prove it.

 

 
  [ # 14 ]
C R Hunt - Apr 25, 2011:

Have you considered building a chatbot using your infrastructure, sort of as a showpiece? I imagine it would be a large time investment, but it would encourage interest in your project.

(Incidentally, perhaps this discussion could be moved to its own thread?)

Yes, many colleges suggested me this, but unfortunately, I consider that my time spent in doing this is not worth and that I can make more improvements and include ‘deep understanding’ mechanisms soon, this takes most of my time away.
Also I am trying to include behavior measurement, making a look-over the whole threads of a bot chatting to a human, to improve the quality of the conversation and see if the conversational goals have bee fulfilled or not, or if the chat need a twist, a joke or a few ‘harsh words’ if the human is doing/typing nonsenses, for example.
I am using the concept behind a shallow DAML tagging for this, it sounds promising!
The problem lies among the decision threshold ¿when would it be plausible change role or take control of the conversation? Those are my concerns.

I alsr have some bot working out there, most are MSN-chatbots, but all are in Spanish for now, I need to develop the whole GLR grammar first (Spanish is much more complex, so I started there, years ago) So if you like or understand some Spanish, in a few weeks, I can put a small bot to work, with a few script lines to show the full power of some of the functions I claimed it to have built-in. The most spectacular are theme following, anaphoric memory resolution, math & set calculus, units deductions, and in the next version…, also the system will be able to tell you jokes or short stories, pausing and triggering answers in a completely event-driven structure. Without leaving the thread of the story, admitting comments and control events, like “don’t tell me more of this!”, or “please, repeat last said"etc.


- cheers

Andres

 

 

 
  [ # 15 ]

Steve,

It is such a shame that Andres can share his accomplishments, which he has represented in world conferences, and you still want proof.

The problem here is the consumer, not the inventor.  As Andres puts it, why waste time creating the dog and pony show when there is real work to be done?  I agree with him. In the journey to create good conversation, he only has to prove his efforts to himself and his sponsors.

Thank you Andres for sharing.  Your work is quite fascinating!

I’m especially interested in the co-referencing, but more so in the NLG. I envision some day a bot being able to manufacture script from its internal models and a focus on the co-reference concepts in the whole conversation, even if the bot is only talking to itself.

Maybe one thing this group missed is that DAML is, in part, an effort to create an AI standard language, albeit framed as an agent markup instead of AIML which is a bot markup.  Hence your contribution to that thread which is now removed.

 

 1 2 > 
1 of 2
 
  login or register to react