AIML is a question oriented knowledge model. It focus on separate questions. AIML is hard to connect different questions by deduction or reasoning.

I saw some abstract for paper of AIML+ontology . However, I still not found a working system which can support knowledge reasoning based on AIML.

Let’s question A.L.I.C.E. to see what she has to say about that:

@A.L.I.C.E.
► Nathan said,“AIML is a question oriented knowledge model.”

Code:

@8pla, very funny, this way of interaction design!!!!

“Anyone who knows HTML can learn AIML.”  alice bot said.

Why not we use HTML then? Because HTML can not handle the problem.

We need a reasonable knowledge representation model. Actually, AIML attempt to solve the linguistic problems and semantic problems together. All the problems need to be solved by the chat masters. I believe we need a better model that chat masters focus on the knowledge and rules.

8PLA • NET - Oct 3, 2010:

Let’s question A.L.I.C.E. to see what she has to say about that:

@A.L.I.C.E.
► Nathan said,“AIML is a question oriented knowledge model.”

Code:

<a href="http://www.newsAI.com/programe/src/reply.php?who=Nathan&said=AIML+is+a+question+oriented+knowledge+model.">
Nathan said,"AIML is a question oriented knowledge model."</a> 

Well, I think that’s why many of us are developing our own projects without relying on the AIML language. But it does provide a useful tool for those interested in chatbots and who aren’t necessarily able or willing to devote the time to programming from scratch.

What I am thinking is that we need an understanding oriented knowledge representation model. So that the chat masters need only build their knowledge base on these model and we can provide a chatbot platform to support such a standard.

This model should be simpler than AIML. It may be based on natural language text. We may define some simple tags to help the chat system to understand the knowledge in natural language.

By the improvement of the chat system, fewer and fewer tags are needed. The better system, the fewer tags are needed. Finally, the knowledge model may come back to natural language.

I believe this is a right way. I will work for it.

C R Hunt - Oct 5, 2010:

Well, I think that’s why many of us are developing our own projects without relying on the AIML language. But it does provide a useful tool for those interested in chatbots and who aren’t necessarily able or willing to devote the time to programming from scratch.

What you’re describing doesn’t sound *simpler* than AIML.

Is your meaning that the system doesn’t simpler than AIML or a tagged natural language text doesn’t simpler than AIML?

C R Hunt - Oct 5, 2010:

What you’re describing doesn’t sound *simpler* than AIML.

I mean that the interpreter you’d have to design for the tagged natural language would be a huge challenge. (Well, basically we’re talking about restricted NLP interpretation.)

As a representation standard, I don’t think it will be restricted natural language. Some of the systems may only support restricted natural language. Besides, I don’t think the tags are so hard to design. Some simple tags may help the system to understand many texts. The tags need only represent the relations between some of the words.

So, you’re proposing something like what Cyc uses? (http://cyc.com/cyc/cycrandd/areasofrandd_dir/nlu)

For those of us who wish to start from a “simple” beginning, and who don’t have the advantages of higher education in a field that would actually be of use (I only have an Associate’s in Electronics, after all), AIML interpreters provide a good base to start from, while we learn the requisite skills to create something more advanced. To be honest, Nathan, you seem to be dismissing AIML as useless, and I strongly disagree. I know that AIML isn’t the “End All, Be, All” of Artificial Intelligence, but it DOES provide a sound, solid foundation for folks to learn from. As I learn from the people here who are working on other projects, I’m gaining insights that I can use to expand upon what I’ve already learned, and none of this would have happened for me if I hadn’t learned AIML first. Your comments on AIML sound to me like someone who is saying that grade school arithmetic is useless in the face of calculus, and I say that you can’t learn the advanced concepts of calculus without a solid grounding in arithmetic. The only difference between these two analogies is that AIML is only one “starting point” among many, but I feel that the analogy still holds true.

Ok, hopping off my soap box now.

No, Cyc’s tags are still on linguistic level. What I’m talking about is semantic tags.

For example.

David<1> is a boy. He<1> is a member of a football team.

<1> shows that the word “David” and the word “He” related to a same entity.

C R Hunt - Oct 5, 2010:

So, you’re proposing something like what Cyc uses? (http://cyc.com/cyc/cycrandd/areasofrandd_dir/nlu)

Sorry Dave, I did not say AIML is useless. My meaning is that AIML is on the wrong way.
Imagine we are climbing a mountain, one of the paths may easy to start, but it can not lead the climbers to the top.

I understand some guys may go to the easy way for fun, but I’m considering how to get to the top.

Dave Morton - Oct 5, 2010:

For those of us who wish to start from a “simple” beginning, and who don’t have the advantages of higher education in a field that would actually be of use (I only have an Associate’s in Electronics, after all), AIML interpreters provide a good base to start from, while we learn the requisite skills to create something more advanced. To be honest, Nathan, you seem to be dismissing AIML as useless, and I strongly disagree. I know that AIML isn’t the “End All, Be, All” of Artificial Intelligence, but it DOES provide a sound, solid foundation for folks to learn from. As I learn from the people here who are working on other projects, I’m gaining insights that I can use to expand upon what I’ve already learned, and none of this would have happened for me if I hadn’t learned AIML first. Your comments on AIML sound to me like someone who is saying that grade school arithmetic is useless in the face of calculus, and I say that you can’t learn the advanced concepts of calculus without a solid grounding in arithmetic. The only difference between these two analogies is that AIML is only one “starting point” among many, but I feel that the analogy still holds true.

Ok, hopping off my soap box now.

Nathan Hu - Oct 5, 2010:

Sorry Dave, I did not say AIML is useless. My meaning is that AIML is on the wrong way.
Imagine we are climbing a mountain, one of the paths may easy to start, but it can not lead the climbers to the top.

I understand some guys may go to the easy way for fun, but I’m considering how to get to the top.

For “strong AI” or whatever you’d like to call it, perhaps you’re right. But for the more practically minded, an AIML bot can be a useful tool for learning more about a product/service/topic, navigating a file system or website, etc.

Nathan Hu - Oct 5, 2010:

No, Cyc’s tags are still on linguistic level. What I’m talking about is semantic tags.

For example.

David<1> is a boy. He<1> is a member of a football team.

<1> shows that the word “David” and the word “He” related to a same entity.

Interpreting the example you gave would be an example of restricted NL parsing.

I mean that the interpreter you’d have to design for the tagged natural language would be a huge challenge. (Well, basically we’re talking about restricted NLP interpretation.)

I understand the benefit of what you’re proposing. Sort of a way to ease into full NLP that allows the bot to encounter examples of natural language and build a system for deciding in the future how to interpret vague text. (For instance, the chatbot would use examples like you gave to learn how to define pronouns.) Building a large set of text with a standardized set of tags would be useful for that.

What I’m trying to point out is that the real challenge is designing an algorithm that can actually use this tagged text to arrive at workable rules. This is where the work lies, and it is an interesting topic, but one that has absolutely zero to do with AIML bashing. (I’ve never used AIML, but the existence/sphere of interest of this particular markup language seems unrelated to parsing algorithm development.)