Dear Oberseminarler,

I think what we definitely arrived at last Monday is a set of questions 
about
- what the remaining problems are for lexicalised PCFGs
and on how to attack the language-specific parsing problems for German:
- where in the topological field structure are the elements governing 
and determined by agreement and subcategorisation?

I tried to give a short summary on the slides you find enclosed.



May I ask you now to read the following papers for next Monday?

[1] Franz Beil, Glenn Carroll, Detlef Prescher, Stefan Riezler, and Mats 
Rooth. Inside-outside estimation of a lexicalized PCFG for german. In 
Proceedings of the 37th Annual Meeting of the ACL, Maryland, 1999.

[2] Joseph Bockhorst and Mark Craven. Refining the structure of a 
stochastic context-free grammar. In Proceedings of the 17th 
International Joint Conference on Artificial Intelligence (IJCAI-2001), 
2001.

[3] Rens Bod. What is the minimal set of fragments that achieves maximal 
parse accuracy? In Proceedings of the ACL, pages 66 73, 2001.

[4] Eugene Charniak. A maximum-entropy-inspired parser. Technical Report 
CS-99-12, Brown University, 1999.




[2] and [4] (esp. sect. 2) will give alternative approaches to finding 
grammar rules not seen in the treebank.

[1] will give a feeling for what the features are that you have to take 
into account when parsing German.

And I think that [3] may be a general way to specify lexical relations 
when you do not know which relations are relevant.


I hope this will serve as a good basis for discussing our (my) parsing 
problems.  Please find the papers in the usual place (or on-line at 
http://citeseer.nj.nec.com/cs).


Best,
Tylman