Strategies for the Chatterbot: "Tricks"

There are various tricks we intend to implement to convince the user that he/she is actually conversing with a human and not a machine. They are referred to as tricks, but many humans use the same tricks everyday.

We will use the same trick that ELIZA uses. That is, to draw a conversation out of the user, with little or no actual contribution from the program. This works because most people like talking about themselves, and believe the program is actually listening to them.

Including substrings of the user�s input in the response sentence creates the illusion of listening.

�

USER: What does he like?

ARTHUR: How should I know what he likes?

�

Another way to "trick" the user is that the program should admit ignorance. Unlike ELIZA, our Chatterbot on some occasions will admit, "I don�t understand the question". This will happen quite often with the first version, as only two bit stream patterns are recognised.

We will also try to guide the user back to the subject of films. This is really just a way of changing the subject. The knowledge base holds many keyword responses; it is these responses which will trick the user into believing Arthur actually exists.

We have introduced a few humorous statements into the program to make it appear more human like.

�

USER: Do you exist?

ARTHUR: No I don�t exist, you are talking to a plantpot.

�

We would also like to simulate human typing. This is demonstrated by "Julia". By including realistic delays between characters and making errors it makes the program appear more realistic. We believe this is only possible if used through Telnet, but if possible we will implement it.

Our program will show that it is not limited to a single domain. Not only will it "talk" about the subject of films, but can also parse interrogative questions, such as "What does he like".

The fact that it is extensible means that new topics and generic answers can easily be added.

�