Imagine that you take a chatbot along the lines of Cleverbot. Feed it the script of a long-running TV series or perhaps a special transcript of a set of novels. It is important that the chatbot be modified learn from a specific character and treat dialogue from others as the user input and the responses of that character as the ideal response to a statement.
When this parsing is complete, what do you have? Something almost- but not quite- totally unlike the character. But it may be close enough to fool you!
I imagine that this would require a huge database of scripts- Homer Simpson would probably be most ideal as a character, since he is predictable yet prone to non-sequitors (that is, easy for the computer to understand, and strange enough for users to disregard mistakes), has been in a huge variety of situations, is present in practically every episode of a long-running series, and is unlikely to be much more intelligent than the computer is.
Other possible characters might be from Friends or- quite possibly- Frasier Crane, since we have two whole long-running series to use as source material.
Star Trek might be too small to use, and the bizarre situations the crew gets into would mean that the set of more mundane responses would be smaller. You'll say "Hello, Captain" and it'll say "Greetings. Disarm your weapons now or we will destroy your vessel!", since when it comes down to it, the enemies greet you politely more often than your allies do when you're Picard. That said, a generalized "captain" character using all the Star Trek scripts in an anonymous fashion could work- at least if you presume that the ship was caught in a negative space wedgie the digitized it into the past, which would explain the tension.
Most movies and games are far, far to small to use, (however, very large series with whole dialogue trees have a nice, even distribution of responses) but novels, particularly long series, have potential.
The problem with these over TV scripts is that scripts are far easier to parse than prose is. If you could find a way to easily digest novels, then Vimes would be a reasonably good source, though again, he spends a disproportionately large portion of his time under stress and thus any responses the bot gives you will be distorted because of that.
We can't use Granny because she might escape. Same goes for GLaDOS.
One more possibility would be God. The bible might be easier to parse than most books, and a bot that responds to virtually everything with "I AM THAT WHICH IS THAT WHICH SMITES THAT WHICH IS THOU AND THOU'S SIN." might be interesting for a while.
Other potential sources could be someone from Stargate,if you're hell-bent on science fiction characters. I still feel that sitcoms, with such long runs and relatively mild levels of tension, would be the best sources for conversational speech.
However, there is one other thing- real historical persons with a large enough set of recorded letters would also be reasonable sources of data. I can't think of anyone with a suitably large database of conversational dialogue, though.
Author
Topic: Character Bots (Read 2129 times)