Hi all,

I have the idea of keeping track of certain characteristics of a conversation of a human and my bot, in order to have it react better. As example, AIML has the concept “that” and “topic”, they could be more precise, but are already extremely powerful. So what else do you see has important/relevant information to attach to the state of a conversation?

Here are some rough ideas that you are invited to complete, precise and comment!

- Time (are we talking about the present?)
- Topic
- People (are we referencing people)
- Goal (what is the goal of the current conversation)
- Emotion (is there any emotion from the human?)
- Historic of said sentences/information
.......

Thanks

Sam

Did the bot just ask a question? Did the user respond.
Environment parameters, Bot parameters, user parameters (likes, dislikes, things owned, favorites), repetition.

Timing + full history of previous conversations with the same user.

The use of “it” in AIML is also very much underused. It is a really handy way of keeping context. For example:

Human: Do you like chess?
Bot: Yes, I play it a little
Human: Do you like it?
Bot: Yes, chess is a fun game (instead of “do I like what”)

The bot sets the value of “it” to chess in the first reply, then is answering, “do you like chess” instead of “do you like it”. This will make your bot more believable.

Keeping track of the history of said information is useful, but how would you concretely use this information?

What Steve said. Pronouns often stand in for the most important objects in a sentence/conversation and can set the context.

Keeping track of all pronouns, what they (likely) represent currently, and what they’ve represented recently in the conversation is critical. People have a tendency to pick up dropped trains of thought and expect each other to use context to understand what they are referring to.

“It” is a particularly tricky pronoun in English, because it can stand in for about any noun, and often represents nothing at all. An understanding of idiomatic usage of “it” would be beneficial in this context. Steve’s example where the bot defines “it”—leading the user to use the same definition—is quite clever.

Or when taking the initiative during a gap in the conversation. If the user told you last week he was going to Paris - either “How was Paris” or for online chats “are you still in Paris?”

I’ve thought that being able to hear the user speak would be very valuable as the stresses on words would give clues to meaning.  Sarcasm and subtle meaning could be detected more easily.

I tend to store the entire sentence, chopped up into understandable parts. Depending on how you configure things, this can be permanently or the information can fade out over time.

A lot of chatbots nowadays interprets the first lines of Wikipedia, and I understood that Siri is making use of Yelp to find local information http://www.yelp.com

Nodding the head, “no”, shaking the head “yes” and so many other facial expressions have existed as human communication for millions of years.  Facial expressions should be the primary simulation of any humanoid robot.