PhD STUDENTSHIP
Maximum-Entropy Models for Broad Coverage Unification-based Grammars
Division of Informatics, University of Edinburgh.
Maximum entropy has been shown again and again to yield
state-of-the-art results in numerous natural language applications.
However, successful maximum entropy models tend to be extremely large,
and frequently contain thousands of parameters. They also tend to be
monolithic, and hard to maintain. Instead of building monolithic
maximum entropy models, this project will investigate ways
of building models that can be factored into individual 'experts'.
Each expert will be a small maximum entropy model that is tailored to
some aspect of the task. The overall model will then consist of a
combination of these experts. It is expected that the resulting
combined model will not only be compact, but will also be better than any
of the individual component models.
The domain for the project will parse selection: taking a large
unification-based grammar, parsing the Wall Street Journal, and then
building a combined model that decides which is the preferred parse.
See www.cogsci.ed.ac.uk/~osborne/coling2.ps for background details
about the task.
The student will work with Dr Miles Osborne (osborne@cogsci.ed.ac.uk,
www.cogsci.ed.ac.uk/~osborne). It is expected that the student will
collaborate with the Institute for Adaptive and Neural Computation
(http://www.anc.ed.ac.uk) at Edinburgh and the Natural Language Group
at the University of Cambridge Computer Laboratory.
Candidates should have, or expect to obtain, a good degree in a
numerate subject (such as, but not limited to, mathematics, computer
science or engineering). Knowledge of natural language processing
will be a plus.
The project is funded for three years and is open to all suitable
applicants of any nationality. It covers all fees and maintenance.
Applications will be considered until the post is filled.
The intended start is the beginning of January, but this can be
negotiated.
Interested candidates are encouraged to contact Dr. Miles Osborne
(osborne@cogsci.ed.ac.uk) or telephone +44 (0)131 650 4430 after the
1st of October.
Details about natural language processing at Edinburgh can be found
at:
http://www.informatics.ed.ac.uk/research/iccs/
For further information on graduate study in the Division of Informatics
and for an application form, see:
http://www.informatics.ed.ac.uk/prospectus/graduate/research.html
This archive was generated by hypermail 2b29 : Mon Sep 25 2000 - 21:48:18 MET DST