Victor Shulist - Aug 8, 2011:

“....they claim will change the world…” 

— I haven’t seen any statement like that in the forums.

Victor, I think his comments were directed towards some more recent threads where new members have made exactly these types of claims and in a very condescending manner. Your work with grammar parsing for chatbots is probably more advanced in that direction than anyone here on this forum, and I don’t think anyone considers your work poorly researched nor insufficiently implemented for the claims you make. I hope you’ll continue to participate here and share your progress.

Dan Hughes - Aug 7, 2011:

I’ve come across quite a few bots that use AIML, pattern matching via regex etc, but I was wondering how many of you guys active in this field have looked at, or tried using, true NLP when it comes to processing human input and generating output? [...]

I’m currently implementing this with true grammatical and semantical tree representations, and its going pretty well.  Just curious of anyone else had attempted it and how you found it?

A couple of us here are using a mix of strategies including “true NLP.” I’d say Victor and I are more on the extreme grammatical end of things while most employ a mixed strategy of pattern matching, semantics, and parsing.

Particularly, I’m interested in breaking down user input into a knowledge representation of linked parse trees, where nouns/noun phrases represent the tree “nodes” and other parts of speech form links between trees. I have a thread in the “My Chatbot Project” section (ALEX) that outlines what I’m doing, though it’s a bit neglected.

I’m definitely interested in following your progress, so please keep the forum updated!

CR, I’m also curious as to how ALEX is progressing. Have you had any breakthroughs? Setbacks? Lulls in the coding cycle? Out with it dear lady! I wanna know! :D

Hi Victor, welcome back!

Dave, is it still a “lull” if it’s two months long? That was May and June. Fortunately last month I’ve been back in the game. I should have enough significant progress to post soon. Thanks for the interest.

Dave, CR & Jan, thanks for the welcome back.

Yes, I’ve been working hard on my project.  I -do- have some updates which I believe will interest the group, the main functionality being a powerful recursive subtree matching algorithm (the system finds a stored fact parse tree that satisfies the the question parse tree), allowing for both term synonyms and entire ‘subtree synonyms’ between question and fact.  Example, ‘fact: bob went to his house’, ‘question: did bob go to his wondeful home’, the same semantic properties are added to ‘to his house’ and ‘to his wonderful home’.  This works for any level of ‘nested’ prepositional phrase objects (example, object of prep-phrase 1 is direct object 1 and direct object 1 is modified in term by another prep-phrase2 with object2, and object is modified…..).  Right now the bot will respond with ‘Yes’ but with a ‘condition’ that house=home (based on the fact they have a common property, perhaps ‘purpose=living-quarters’).  An interesting question will be (and no, this isn’t -fully- thought out just yet ), is: when can the system know to never bother mentioning ‘if, you consider home=house’.  In other cases, it perhaps shouldn’t consider two words equal, even if they have several common properties and thus, in those cases, always add the disclaimer (‘if you consider x and y equivallent).  I’m thinking it can be simply told directly, on a case by case basis.

Anyway, I’m hijacking Dan’s thread perhaps, or not, maybe the above may trigger something in his project, since we both seem to use the concept of properties (or ‘Ontologies’ if you will).

Dave - point taken about your comment - no one wants long-term ill will.

by the way, I noticed we have an edit button !!! WHOOOHOOO!!!   Having a time limit of 15 minutes, -does- make sense also.

Wow an emotional bunch here! hehehe

Quite interesting to read other peoples approach to various issues, problems and designs.

Ive read one of your threads prior to signing up on here Victor and it seems very interesting indeed, CR, not looked at yours regarding ALEX but most certainly will do.

I’ve been away on business for a few days, but planning on now cracking on over the course of the week.  Spending time in airports, on airplanes and in a hotel with TV in everylanguage but English is a good time to think

I notice that the more “lower level” projects on here seem to include elements such as ontologies, relations between databases and using past parse trees as a way to “remember” and process but have any of you actually approached this from the angle of the bot actually “understands” what a house is in some form of context other than a linked set of relations/ontology and other information?

One part of my project is to attempt to implement some kind of “reasoning” to the system so that it “understands” what a house is.

The way that I see it from looking deeply into the process of thought, is that most understanding of an object/statement etc can actually be expressed into a functional nature.  If you have elements of that function already defined elsewhere then you can “create” a function for a newly introduced object or statement.

Granted these functions can sometimes be rather complex, but they are still a function that returns a result.  As I mentioned previous about cause and effect, if that is also taken into account, then you can then further seeming focus the function of that particular meaning.

I havent got to that portion of the implementation as of yet, infact Im quite a way off and a lot of my ideas are just theory at present.

Maybe I should start an actual thread on the project where I can explain this a bit more and what Im wanting to do now that my original question here has been answered regarding NLP.

Any function can be expressed as a relationship between data items.

I notice that the more “lower level” projects on here seem to include elements such as ontologies, relations between databases and using past parse trees as a way to “remember” and process but have any of you actually approached this from the angle of the bot actually “understands” what a house is in some form of context other than a linked set of relations/ontology and other information?

Chuck began a project to relate text input to a spatial “world” to help his bot discuss its surroundings better. And I know Hans wants to incorporate senses into his bot. Not sure how these projects are progressing.

One part of my project is to attempt to implement some kind of “reasoning” to the system so that it “understands” what a house is.

The fact that those words are in quotes should tell you you’re venturing into dangerous territory. Nobody will ever agree on what constitutes “reasoning” or “understanding” for your bot—we can barely agree that humans are capable of these things (and for some people I have my doubts ). I’d keep to define-able, clear behavioral goals, which it sounds like you’re doing from your discussion of function output. Just be careful—especially around here —about what epithets you use to describe them.