[Corpora-List] Metrics for corpus "parseability"

Alberto Lavelli lavelli at fbk.eu
Tue Feb 5 14:26:11 CET 2008


On Mon, Feb 04, 2008 at 11:27:56PM +0100, Sandra Kuebler wrote:


> There is related work about the ambiguity of grammars induced from
> treebanks. Anna Corazza, Alberto Lavelli, and Giorgio Satta used
> conditional cross entropy for that. This may help to at least
> abstract away from the parser :)

Sandra, thanks for mentioning our work related to the part of the original request concerning parsing. The work is described in a paper submitted to a journal and currently under revision. We plan to make the current draft available on our web site soon. Below the abstract.

best

alberto

---------------

Measuring Parsing Difficulty Across Treebanks Anna Corazza, Alberto Lavelli and Giorgio Satta

Abstract One of the main difficulties in statistical parsing is associated with the task of choosing the correct parse tree for the input sentence, among all possible parse trees allowed by the adopted grammar model. While this difficulty is usually evaluated by means of empirical performance measures, such as labeled precision and recall, several theoretical measures have also been proposed in the literature, mostly based on the notion of cross-entropy of a treebank. In this article we show how cross-entropy can be misleading to this end. We propose an alternative theoretical measure, called the expected conditional cross-entropy (ECC), which can be approximated through the inverse and normalized conditional log-likelihood of a treebank, relative to some model. We conjecture that the ECC provides a measure of the informativeness of a treebank, in such a way that more informative treebanks are easier to parse under the chosen model. We test our conjecture by comparing ECC values against standard performance measures across several treebanks for English, French, German and Italian, as well as other treebanks with different degrees of ambiguity and informativeness, obtained by means of artificial transformations of a source treebank. All of our experiments show the effectiveness of the ECC in characterizing parsing difficulty across different treebanks, making it possible treebank comparison.



More information about the Corpora mailing list