I've been quietly coding away. I did some more brainstorming too.
I've been keeping a list of good ideas that Athena will need to do. Every time I see some room for improvement or I see some other program that does it better, I will add a note.
I think I might have cracked a problem I've been having with my VLL language parser. The goal is to fully express everything a person can say into terms a computer can properly parse. I came up with something that is deceptively simple. However, I have not yet coded it. It took a lot out of me to even think of it. So I had to take a break from it.
Then I worked on conversational nodes. I did some more brain storming there too. I was considering how I was working on both an NLP and a type of neural network involving conversations. While thinking about all this, I realized that I needed to make a long term conversational field. Sometimes I just make up random names to try to express some complex idea... so please excuse my lack of vocabulary. When I was thinking about how conversations flow, I realized that there are also long term events that take place in conversations that dictate future ones. We already know that over time as new knowledge is accumulated then the ai's responses change. However, when a chatbot designer creates his or her bot they try to implement this in a one to one conversation. i.e the User says a sentence, then the Ai responds. What many fail to do is think about long term conversational trends that dictate how a conversation will evolve. I'm already using my conversational node to deal with that issue. However, the conversational field idea would take into account even longer reaches of conversations.
Consider how emotional moods are programmed into a chatbot, it is similar to what I'm trying to describe here. It is a state of a chatbot that can change over time and drive future conversational responses. For instance, you can set up a conditional field in which a User would have to use curse words 5 different times before the Ai would get agitated about you having a foul mouth. Or perhaps you have to say "I love you" five times in a row before Athena will believe you. Now think about this in a real conversation construct. A User early on says "I love you" to his chatbot. Later on, He tells his chatbot "I love you very much". The Ai responds with, "Do you think we need to get married?". Its like having a long term conversation, and in the mean time, you have lots of little conversations. Instead of directly programming this into Athena, we need to get Athena to look at long term trends of conversations. The conversational nodes I'm working on is for chain type conversations. Where you are in one chain conversation and have limited choices as to what you can say next. The more conversations you feed it the more choices you have. But, in a conversational field we have to think of a chain that periodically skip through time and then predict an appropriate response.
We humans, and Art, do this in part by remembering what we said to particular people. We randomly come up with clever ideas that progress through time. We listen to others and try it for ourselves.
Ok, so how to create a conversational field automatically through mass parsing of conversation examples. Ultimately, we would have to take periodic samples of User input sentences and then compare them with a massive archive of conversations. I suppose it would be like a scattered out version of a markov chain. Like if it was a 3 to 1 chain, in which the first three didn't need to be in sequence or even next to one another. The chain would be created by taking periodic samples (maybe based by same subjects) within a single chain conversation. Then the next time the same subject is brought up in a User/Ai conversation, with the same amount of 'subjects' needed for the Markov conversational field to work, then Athena will respond with the immediate corresponding response.
I could make a quick example of this but my brain is heating up a bit... blah... this is why I take breaks.
Anyway, I have a design for a way that a User can hard-code these conversational fields directly into Athena. Oh, it can also be used to build a game with too... Like to unlock a SECRET if enough clues were unlocked.
Perhaps it can be used to set Ai moods based on trends and triggers the User has initiated. Maybe it can be used to unlock entire areas of code within Athena.
I think I smell smoke... too much thinking...
