My Parsers
- My doctoral parser was called Plink.
- It was a one pass parser that combined syntax and semantics
in grammar rule applications.
- (We used it in the message understanding competitions.)
- We used a unification based grammar.
- Constituents were put on a stack, and I tried to keep the stack small.
- Rules were picked, based on the content of the stack, by preference
heuristics.
- More recently, I got back to parsing by implementing a neural version
of the parser.
- It also combined semantics and syntax.
- Timing was done by neural cycles of 10 ms.
- Instead of a stack, we used memory activation based on neural
activity.
- It ended up as a solid cognitive model with accurate timing,
accurate PP attachment resolution, and correct semantic output.
- I'm pretty sure it's the state of the art in neuropsychological
parsing, but it is of course incomplete.
