Connectionism and explanation
Eugene Charniak · 1987
I gather that the panel on connectionism was picked to have a variety of viewpoints represented, some very pro, some very against, and me -I take an extreme waffle position.I really like connectionism, and I wish it would work for me.But so far it can't and theoretical breakthroughs would been needed to change things.I will try to explain here why I see things this way.Before I start, however, it would be a good idea to clear up one possible source of confusion.My recent work has been oil the problem of explanation in stories: how, given a story one can assign a motivation to a character based upon his or her actions.I also believe that the processes which are needed to do this will shed a lot of light on traditional parsing issues.The source of confusion comes from the fact that my recent models of explanation have involved a process of marker passing, or spreading activation.Basically I am using a breadth first search in an associative network to find connections between concepts, in the hope that such connections will suggest explanations.My typical example is Jack wanted to kill himself.He got a rope.Here the connection between kill and rope would bc the clue.