Abstract
Syntactic analysis of natural languages is the fundamental requirement of many applied tasks. We propose a new module between morphological and syntactic analysis that aims at determining the overall structure of a sentence prior to its complete analysis.
We exploit a concept of segments, easily automatically detectable and linguistically motivated units. The output of the module, so-called `segmentation chart', describes the relationship among segments, especially relations of coordination and apposition or relation of subordination.
In this text we present a framework that enables us to develop and test rules for automatic identification of segmentation charts. We describe two basic experiments -- an experiment with segmentation patterns obtained from the Prague Dependency Treebank and an experiment with the segmentation rules applied to plain text.
Further, we discuss the evaluation measures suitable for our task.