Friday, January 14, 2011
Multple sentence answer problem
1. Edit the AIML to remove full stops in the tag (except for just one at the end). This will produce some grammatical/punctuation errors. But I'm willing to live with comma splices and poorly used dashes if it simplifies the structure and creates more meaningful dialogue.
2. I could (maybe) add some script to the .swf that looks for a full stop in a string of text and stops 'reading' at that point. This would mean that in the example - I love Star Trek. My favorite series is 'Voyager' - only the first sentence would be passed to the other bot. This is technically possible but it may dumb-down the responses taking the colour out of the language. Particularly if the second sentence is an ironic statement e.g. I love Star Trek. Not!
3. A simpler fix I will try whilst contemplating the two above is to add a simple variable to count the number of interchanges between to two bot. Then, when there have been say 10 interchanges the 'stopMe' function will be fired automatically.
4. A more sophisticated version of this is have a generic 'exit' interchange fire after a fixed number of interchanges. I'm thinking here of Pinky and the Brain.
Pinky - What are we going to do tomorrow night Brain?
Brain - Same thing we do every night Pinky - try to take over the world!
More to come.
Two bots chatting - test
To start them chatting type in some text in the text field at the top of the interface and click the 'Start Talking' button.
They will keep going until you click the 'Stop Me' button. As they go they get more and more confused - this is one of the issues to be resolved. In particular, the issue of sentence splitting needs to be resolved. If one bot answers with a reply that contains multiple sentence (e.g. I love Star Trek. My favorite series is 'Voyager'.) the other bot will try to supply an answer for both sentences. This very quickly gets out of hand - more and more sentences being passed between the two. Just hit the 'Stop Me' button.
The bots are hosted on the Pandorabots site ( Pandorabots )
You can clear all the fields and all the text by clicking the 'Clear' button.
Most of the dialog is based on the standard Alice set. However, a few AIML elements have been changed - those relating to 'mother', 'science fiction' and 'Star Trek'.
Have a play and let me know what you think.
This is a link will open a new browser window displaying the Flash .swf interface
Tuesday, December 14, 2010
Bergson - agency and structure
Bergson's concept of the mechanical encrusted ('plastered' may be a better translation of the original French) on the human is analogous to the relationship between agency and structure. The individual's agency is, for want of a better term, their free will - their ability to act in the world by making choices that have real consequences. The mechanical is then all of the structures of the world. These structures are typically described by rules and laws: the laws of physics, the rules of language, the rules of etiquette, and even the rule of law. Cartoon 'violence' toys with laws of physics - the coyote lives and dies as a result of these laws. The pun, the malaprop and the double entendre are manipulations of the laws of language. Borat shatters the rules of etiquette. The Keystone Cops, Top Cat's Officer Dibble, Rumpole of the Bailey, and so many others expose the farcical nature of the rule of law.
Giddens concept of 'structuration' brings agency and structure together - each dependent of the other. A kind of co-dependent marriage. There is a deep level of incongruity between agency and structure. Not all incongruity leads to comedy but is it a necessary if not satisfactory condition.
Bergson was talking about these subterranean incongruities.
Henri Bergson, Laughter: An essay on the meaning of the comic.
Friday, October 1, 2010
A little success story
The reason for starting up the second bot... I've got the Flash interface working. The output from one becomes the input for the other and so on... I had to build a StopMe button just to get the to shut up! They are completely nonsensical at the moment. As the 'conversations' go on they tend to generate multi-sentence / multi-line replies. These in turn generate even more milti-sentence text that is nearly always bizarrely self referential. This is something I'll serious need to address in the AIML sets. I may even try to kill of 'second sentence' categories.
As these bots share a common base of AIML there is also a lot of recursion going on. This too is something I'll have to edit out.
Jobs for another day.
Thursday, September 30, 2010
More bot technology...
Thursday, August 5, 2010
Bot technology ... continued
The URL for the test bot is - Talk to Dick at Pandorabots
Have a play and feel free to add a comment.
Summer School report
Suffice to say, at the moment, this was without a doubt the valuable research related experience of my academic career.