Re: OWL, RDF, AIML

Hi Eric,

Many thanks for your reply, and glad someone alerted you to this thread. Also thanks for the advice on your experiences with AIML etc. Someone emailed me a copy with your article that he had. The version he sent me was 29 pages long so I have no idea if that was the full one. I spent some time trying to find it online to no avail. Dr Wallace (ALICE) has a defunct link on his website to this article too, so I'm sure he would appreciate a copy.

While things may have moved on a little in computing power since your work, your findings re AIML and DB's are very interesting and relevant. Have you come across this site: http://www.texai.org/blog/about/texai-project ?

The aim is to release this on 23rd June (Turing's birthday). It combines Wikipedia on top of Cyc with a chatbot on top of that. There are of course a lot of claims for what it is supposed to achieve, but this is somwhat near what I had in mind. In fact as it is open source, the databases are released as RDF so could be quite useful.

However, my idea was to use AIML and OpenCyc - probably as separate databases to power a chatbot. I or rather we, need all related software to be installed in one computer, but like you are doing, pull in info not stored as required from the internet.

The idea behind this approach is that it should make the chatbot more responsive, and also we are not held hostage by the vagaries of the internet if a website or server goes down etc.

Can I ask you if the conversion you undertook from OWL -RDF - AIML was in fact automatic, or did you do this manually? Do you have any advice on how to do this automatically, or indeed what software would you recommend? I looked at Protege to convert from OWL to RDF to XML, but I have no idea if it can be set up to run a script to format it as AIML. Any ideas?

Also, here is a link to Adamus 1.08 alpha which I think uses both AIML and .mdb databases: http://www.charctercreationsanddesignworks.us/ProjectDownloads.html. This is getting nearer to our concept but we don't like the idea of .mdb databases, so we are not tied to one companies products.

Regards,
Jules



I just saw a twitter message from someone alerting me to this discussion.  Sorry to join late.

I'm a little surprised that the conference papers don't seem to be available any more.  I'll have to see if I can locate a distributable version, if you're interested.  I'm not sure how much of it is still useful.  In the paper I was looking at general RDF/OWL, focusing on ontologies like FOAF and Dublin Core and creating categories based on the resources and properties within them.  With limited ontologies and data sets the concept seemed to work quite well.  As I progressed with the work, I discovered that trying to create large numbers of categories was going to bring most bots to their knees, even with a DB backend.  So I've backed away from trying to force all the information into a set of AIML.

I did look at the Cyc information as a possible source, but at the time they hadn't released everything yet and there were still too many holes.  It might be worth looking at again, if you were trying to do some reasoning.  But again there are many places where the knowledge is very shallow.

I'm now using Program O and using PHP calls to datasets like dbPedia and web services like weather and stocks to give the bot the appearance that everything is stored, when, in fact, it is being retrieved on the fly.  The initial results look promising and I'm working on adding additional categories to do more complex queries.

I hope this gives you a little bit of background.

Eric

JulesJH wrote:

I'm thinking more of using the OWL ontology. I downloaded the entire Open Cyc knowledge base in OWL, and plan to convert that (eventually) - or at least some parts of it I can use, to AIML.

Apparently Eric Freese did something along those lines, but I don't know if it was OpenCyc? Probably not.

AIMLpad+CyN will interface with the OpenCyc server, but I really need to have everything in-house, not spread over the web.

Jules