IM
Instant messaging allows people to chat with one another by sending text messages. Messages are received nanoseconds after they are sent, hence the notion of it being instant. There are also other features that IM has, such as voice chat. However, very few people use this because the sound does not travel as quickly as the text (lag), and it is often of low quality. As such, you have to rely on the context of the message to determine someone's attitude. Tone of voice is not conveyed very well.
As far as designing a chatterbot for IM, the tone of voice problem is easily avoided. Since it is hard to give the bot a personality, the fact that IM doesn't really support it makes this much simpler. The bot will need to observe correct IM grammar. This means that many words are spelled phonetically instead of correctly, and almost no capitalization is used. But those are easy to fix as well.
Problems with the chatterbot arise in the content of the response and its timing. It is difficult to get a bot to initiate a meaningful conversation with someone because prior to that interaction it has never known the person. The bot will often rely on responding to what the human says, not the other way around. The Lovelace objection comes into play here, where the bot can't really generate any original thought. (Well, it can start a conversation by sending a pre-scripted line to the person on the other end, but which line is sent is probably chosen at random. That eliminates the realism of the conversation).
To allow for this design constraint, it helps to give the bot a specific personality. If the bot has an attitude, it can be set to send certain statements to the person that reflect its personality. For example, a Jesus-freak bot would just start spitting readings from the New Testament to the person. While it may not make any logical sense to the person in terms of a conversation, it may help to initiate one. It gives the illusion of the bot trying to start a conversation of its own. This would also help support the argument against the Lovelace objection.
Now, Eliza responds based on the content of the message it receives. Not the content as a whole, but more towards word choice. Following Rogerian psychology techniques (a.k.a. reiteration) Eliza will analyze the words in the sentence, choose the main verb and subject, and then attempt to ask a question back that is the person's statement rephrased. Most people do not notice this, or they do and don't care, because they continue to have a conversation with Eliza under the belief that she is a real therapist.
It is embarrassingly easy to break Eliza though. For instance, the following is a conversation I had with her:
>Hello, I am Eliza.
*i am andy
>Did you come to me because you are andy?
*yes
>Do you expect a different answer by repeating yourself?
*no
>Are you saying no just to be negative?
*no
>Please, don't repeat yourself!
Eliza's lack of a personality makes her conversational bloopers that much more apparent. Giving her a personality could fix this. Back to the Jesus-freak example, if Eliza cannot determine a therapeutical response for a statement, it can rely on its personality as a safety net. Granted, that won't fix the problem above (that was a programmer error, shame on him), but it can help reduce the number of errors that come from hitting a dead end topic.
Personality is key. It is a good source of behavior patterns, and all around random statements. Giving one to a chatterbot would make IM much more interesting.
