|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
• |
Supervised
learning
|
|
|
|
– |
No attempts (to
our knowledge) to apply
|
|
|
general
inductive methods to the full
|
|
|
problem
|
|
|
|
• |
Some very small
experiments with neural nets
|
|
|
– |
Why not?
|
|
|
|
• |
problem is hard
(Gold’s theorem)
|
|
|
|
• |
fully supervised
case seems unrealistic
|
|
|
|
– |
though small
tagged corpus is cheap
|
|
|
|
– |
production can be
parallelized
|
|