(from an email to a friend 2 July 2003)
Language is a recursive web of syntax and semantics. A.I. systems seem
to have faltered in the last 20 years as they try to parse seemingly
simple sentences. Given a string of words, one can interact with this
recursive web, finding the interrealtionships (and bring in past conversations,
‘knowledge’ about the world, etc), but intead of diving off the end into
recursive looping, could one use a PageRank-like calculation to help converge
on the useful nodes that are what the sentence ‘means’. (by which I mean:
prune down the web to the small set of interconnected base principle nodes
that can then be machine parsed, sorta like finding the ‘concentrator’ nodes
in a web graph)
Has anyone done this?