The Statistical Analysis of Morphosyntactic Distributions

Proceedings of the Fourth International Conference on Language Resources and Evaluation (LREC 2004)

Abstract

This paper describes methods for the statistical analysis of quantitative data on the distribution of morphosyntactic features. A key problem is the large amount of ambiguity in automatically extracted data. In the paper, I argue for a conservative approach that treats ambiguous instances as counter-evidence. It is nonetheless possible to obtain detailed morphosyntactic information from the corpus data with the help of partial disambiguation and by exploiting systematic ambiguity classes.