Hi YKY,
You asked:
Do you think it's possible for me to "plug in" my learning module to
Texai? It may not be easy but certainly is very desirable.
Yes I do. Because you are developing in Lisp, the inter-process communication
to the Texai Java code will likely be a TCP socket over which remote procedure
calls are performed by your plug in. In a few more months, I hope that Ben
publishes the OpenCog specifications, and I will create a suitable interface
from Texai to OpenCog. I think that you may, in turn, be able to use that same
interface to Texai for your purposes, at some level of abstraction.
Also, there may be problems with using a COTS DB for the KB. The
learning algorithm needs a huge amount of access to the KB. Some
specialized algorithms may be better than existing DBs. For example,
it may be desirable to build a big index based on predicate hashing of
all the rules.
I recall this issue from your previous posts. I plan for the Texai KB to be
partitioned, distributed, and locally cached. Therefore, only a working set of
knowledge will be resident at any particular Texai instance (agent). But there
is no reason why a particular agent could not acquire from its peers a huge
amount of facts, to the limits of its host memory, for your learning algorithm.
Or you might be able to distribute your algorithm to multiple hosts, so as to
partition the whole of the input data.
One question I have is how do you represent complex English sentences
in first order logic, or RDF (which is similar anyway). For example:
"John eats spaghetti slowly"
"John believes Peter is crazy"
"John drinks the wine that is poisoned"
The first sentence is different from the two others in that I would represent
it as follows. Each of these English statements can be transcribed to
RDF-triples (subject, predicate, object).
* there is an eating event
* the performer is John
* the food eaten is spaghetti
* the manner of performance is slowlyRegarding the other two complex
sentences, I first refer you to Dr. Jerry Ball's slide presentation here that
illustrates the application of Double-R grammar to sentences having a Verb +
Situation Complement. Jerry does not however describe the semantics of the
diagrammed sentences. For that, I am using the Cyc KR (knowledge
representation) conventions. For example, in the sentence John believes that
Peter is crazy, Texai should, when it is taught how, create the following
semantics, which are expressible in RDF:
* there is a believing situation
* the believing situation is happening now and is ongoing
* the believer is John
* the thing believed is an attributive situation
* the attribute situation is happening now and is ongoing
* the subject of the attributive situation is Peter
* the attribute in this situation is craziness
One has to be careful when applying deductive inference to the above KR. For
example, at the time I left Cycorp, the Cyc inference engine was not then
capable of giving the correct answer to the question "Is Peter crazy?". Simple
indexing on Peter and Craziness would indicate that Peter is crazy. But it may
turn out that Peter is actually not crazy, rather that John is mistaken in his
belief. The correct answer is "unknown, but John believes so". This particular
sub-situation must not be directly accessible to the inference look-up
mechanism, rather it must be accessed via John's beliefs in order to answer
correctly.
Cheers.
-Steve
Stephen L. Reed
Artificial Intelligence Researcher
http://texai.org/blog
http://texai.org
3008 Oak Crest Ave.
Austin, Texas, USA 78704
512.791.7860
----- Original Message ----
From: YKY (Yan King Yin) <[EMAIL PROTECTED]>
To: [email protected]
Sent: Monday, May 5, 2008 12:39:37 AM
Subject: Re: [agi] about Texai
On 5/4/08, Stephen Reed <[EMAIL PROTECTED]> wrote:
> Interesting that you should ask about Texai and reasoning / learning
> algorithms. As you know, my initial approach to learning is learning by
> being taught. Therefore I do not have much yet to offer with regard to
> machine learning, learning by discovery and so forth.
Do you think it's possible for me to "plug in" my learning module to
Texai? It may not be easy but certainly is very desirable.
Also, there may be problems with using a COTS DB for the KB. The
learning algorithm needs a huge amount of access to the KB. Some
specialized algorithms may be better than existing DBs. For example,
it may be desirable to build a big index based on predicate hashing of
all the rules.
> However, I had to write a simple subsumption reasoner and I would be glad to
> license that particular code to you under a BSD or other commercial-friendly
> license as you wish. Subsumption reasoning is rather simple and you may find
> it easier to read my Java code and re-implement in Lisp. You can replace my
> RDF queries with graph traversal or SQL queries depending on how your
> knowledge base is implemented. My source code is archived in this Subversion
> repository, in the src directory.
My KB is just a DB of logic formulae, not graphical. It's close to
Cyc's KR, with a few inventions of my own.
One question I have is how do you represent complex English sentences
in first order logic, or RDF (which is similar anyway). For example:
"John eats spaghetti slowly"
"John believes Peter is crazy"
"John drinks the wine that is poisoned"
etc.
These sentences require a formula to "quote" another formula. If you
don't have this construct in your logic, the sentences may be
unrepresentable.
> Having completed the very basic dialog capability to comprehend and generate
> a single English sentence, I am moving on to coding the ability to acquire
> lexicon / grammar vocabulary and skills. To acquire skills, I have decided
> to implement an agent control language, that I call Texai behavior language,
> which compiles directly into Java for execution. I am following the design
> outlined in Gerhard Wickler's thesis whose project page is here for
> Capability Description Language. For the CDL state representation language,
> I am using RDF formulas elaborated with logical operators for implies, not,
> or, and and.
I don't know how exactly CDL is used in Texai. Do you have a blog
entry or some explanations for this?
> I need to enhance the subsumption reasoner to perform unification and
> subsumption reasoning over sets of precondition and postcondition formulas.
> This facility will not provide general deductive binding-gathering, but still
> it might be useful. And it will give you an idea of the low-level term,
> literal and formula objects that Texai will use. This work will be archived
> in this Subversion repository as I write it.
Yes, that'd be a good way for me to learn more about Texai.
Thanks =)
YKY
-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription: http://www.listbox.com/member/?&;
Powered by Listbox: http://www.listbox.com
____________________________________________________________________________________
Be a better friend, newshound, and
know-it-all with Yahoo! Mobile. Try it now.
http://mobile.yahoo.com/;_ylt=Ahu06i62sR8HDtDypao8Wcj9tAcJ
-------------------------------------------
agi
Archives: http://www.listbox.com/member/archive/303/=now
RSS Feed: http://www.listbox.com/member/archive/rss/303/
Modify Your Subscription:
http://www.listbox.com/member/?member_id=8660244&id_secret=101455710-f059c4
Powered by Listbox: http://www.listbox.com
