This paper describes POS tagging experiments with semi-supervised training as an extension to the (supervised) averaged perceptron algorithm, first introduced for this task by (Collins, 2002). Experiments with an iterative training on standard-sized supervised (manually annotated) dataset (10 6 tokens) combined with a relatively modest (in the order of 10 8 tokens) unsupervised (plain) data in a bagging-like fashion showed significant improvement of the POS classification task on typologically different languages, yielding better than state-of-the-art results for English and Czech (4.12 % and 4.86 % relative error reduction, respectively; absolute accuracies being 97.44 % and 95.89 %).

The aim of this paper is to explore the feasibility of solving the dependency parsing problem using sequence labeling tools. We introduce an algorithm to transform a dependency tree into a tag sequence suitable for a sequence labeling algorithm and evaluate several parameter settings on the standard treebank data. We focus mainly on Czech, as a high-inflective freeword-order language, which is not so easy to parse using traditional techniques, but we also test our approach on English for comparison.

This paper presents the results of the preliminary experiments in the automatic extraction of definitions (for semi-automatic glossary construction) from usually unstructured or only weakly structured e-learning texts in Bulgarian, Czech and Polish. The extraction is performed by regular grammars over XML-encoded morphosyntacticallyannotated documents. The results are less than satisfying and we claim that the reason for that is the intrinsic difficulty of the task, as measured by the low interannotator agreement, which calls for more sophisticated deeper linguistic processing, as well as for the use of machine learning classification techniques.

No abstract