V1.2 AMDI currently understands.....

Hundreds of People's names
Various greetings.
Various ways of signing off.
Asking what the time or date is.

Conversations about food
Conversations about travel
Conversations about politics
Conversations about yourself

Personal conversations
Various sayings and famous phrases

+ Much more 


I fully admit that when I started this project back last year, and then
released the hopelessly buggy v1.0, I was only doing it for a laugh.
The system used a simple slot and frame technique, where it would give 
a limited variety of responses depening on keywords found in your entry.

Since then, I have actually done a little research, which involved looking
at some other programs, reading guides on the web, and doing a little of
my own thinking.

Many programs attempt to make replies directly from the entries which you 
make to them. They are not attempting to understand or truly reply to
your statements or questions. I feel that this technique is wrong.

When a you reply to another person, you draw heavily on your knowledge of
the situation and past sistuations, as well as facts and figures which you
might remember. Your ability to recall such facts is heavily dependant on
what you are talking about. For instance, if you were talking about fast 
food, the number of bits in a byte would not immediately slip into your 
mind.

It is due to this heavily topicalised style of conversation that AMDI has 
been programmed to talk mainly within predefined areas, where it has 
sufficient knowledge and experience to do so. AMDI will converse outside 
these areas, but with limited accuracy.

I also believe that for many questions and statements, society has evolved 
a set of predefined answers. For example, if I was to say "Hey, how are you?",
unless there was something on your mind, you would probably reply "Fine 
thanks.". Therefore AMDI has been programmed with many predefined responses 
such as the example. However, to increase variety, certain words and 
phrases are changed.

Therefore AMDI uses a variety of techniques for actually coming up with the 
responses that it comes out with. The coding has been carefully constructed in 
order to choose the most sensible reply in most cases out of an internal list 
of many. However, I feel I have managed to obtain a balance between variety 
and interest of conversation with flow and accuracy of replies.

In the future, and I have already started work on it, I plan to include a 
truly grammatical approach to the program, which will mean that the answers 
will be more structured and conform more tightly to the original entries. I 
may also implement an element of memory and learning. I am still unsure whether 
these will form part of AMDI or will go into an all new program. I expect the 
first implementation of this to be available around June-ish 1999.

If your interested in the way AMDI works, please feel free to E-mail me at
Ian@overmill.freeserve.co.uk  I will try to answer any questions which you 
have and would appreciate all comments from people who have more expertise in 
the field. I am sorry, but I am unwilling to make the source code available.
