Magisterarbeit, WSD, Tesla
Yay! My Magisterarbeit (master’s thesis) on Wortsinndisambiguierung durch hierarchische Kontextabstraktion (word sense disambiguation by hierarchical context abstraction) is done and has been accepted. It deals with the task of automatic word sense disambiguation (WSD) using corpus-based machine learning techniques and hierarchical bayesian networks (i.e. hierarchical belief propagation, inspired by Jeff Hawkins’ book On Intelligence). It uses the Senseval-3 english lexical sample task data for training and evaluation (which contains parts of the British National Corpus, BNC).
The representation of training samples, the feature computation and the classification step are implemented as separate Tesla components to allow experimental exchange of corpora, different ways of feature computation and evaluation of different classification algorithms. They also allow an integration of the WSD task and the actual application, which both will hopefully lead to a bright future of these and other components in Tesla. You can check out the complete Magisterarbeit (in german language) as PDF and as zipped plain text (LaTeX and GraphViz).