importance of units

Mick O'Donnell micko at WAGSOFT.COM
Sun May 11 14:02:34 UTC 2003


Hi Kamel,

 > I am wondering if anybody on this list has used or has some references
 > on assigning wights to RST relations to capture the
 > importance of the units of the text.

    I did this in a program which pruned back RST trees to summarise.
Each relation type was assigned a particular weight and each node
of the tree was assigned an importance by multiplying weights of
relations back to the top of the discourse tree. E.g.,

1) the root of the tree is assigned a weight of 1.0.

2) The elaboration relation is assigned a multiplier of 0.7

3) An elaboration attached to the root node has an importance
     of 0.7

4) An elaboration on the elaboration has value 0.49.

5) In pruning, leaves with lowest importance are dropped first.

This approach was described in:

Michael 1997 Variable-Length On-Line Document Generation. Proceedings of
the 6th European Workshop on Natural Language Generation March 24 - 26,
1997 Gerhard-Mercator University, Duisburg, Germany.
http://www.wagsoft.com/PapersHTML/variable/variable.html
PS: http://www.wagsoft.com/Papers/EWNLG97-Varlen.ps

O'Donnell, Michael 1997 Variable-Length On-Line Document Generation,
Proceedings of the Flexible Hypertext Workshop of the Eighth ACM
International Hypertext Conference, Southhampton, UK. [PS]
http://www.wagsoft.com/PapersHTML/flexhyp/flexhyp.html
PS: http://www.wagsoft.com/Papers/flexhyp.ps

Others have applied similar techniques to RST pruning, except
all relations were assigned the same weights, e.g., Ono et al,
Daniel Marcu.

My approach was also used in the ILEX text generation system, see:

O'Donnell, Michael, Chris Mellish, Jon Oberlander and Alisair Knott 2001
"ILEX: An architecture for a dynamic hypertext generation system".
Natural Language Engineering 7, 225--250.
http://www.hcrc.ed.ac.uk/~jon/papers/jnle.pdf

Mick



More information about the Rstlist mailing list