[plt-scheme] Natural Language parsing in CS1

On Tue, Jun 2, 2009 at 4:01 PM, Todd O'Bryan <toddobryan at gmail.com> wrote:
> One of my computational linguistics professors said that the
> statistical revolution of the 1990s was incredibly important, but he
> worried that it was the result of competitive systems that encouraged
> people to create something that worked, not necessarily something that
> was based in good research. His view was that people had kind of hit a
> wall and the field needed to go back to doing some basic research to
> figure out how to get past the limitations people had hit.

My opinion:

Logic is very expressive but doesn't handle uncertainty
Stats handles uncertainty but is not expressive (the most commonly
used formalism is equivalent to propositional logic).

The stats guys are slowly building more expressive representations.
For example, various people are looking at first-order logic +
probability. There are issues here that relate to PL. Current machine
learning systems are built in matrix oriented languages like Matlab,
with speed being a motivation. These languages make it hard to build
complex systems (e.g. a type system doesn't help when, for example,
every function is matrix -> matrix). Work like the Church system at
MIT and Neil Toronto's work at BYU may provide the tools needed to
build more expressive systems that have reasonable performance.

N.