Semi-supervised extensions to Morfessor Baseline
Oskar Kohonen, Sámi Virpioja, Laura Leppänen, Krista Lagus · 2010
We have extended Morfessor Baseline, which is a well-known method for unsupervised morphological segmentation, to semi-supervised learning. As submission to Morpho Challenge 2010, we provide results from three methods: The first one is based on the unsupervised algorithm, but includes a weight parameter that can be used to control the amount of segmentation. The second one applies the semisupervised extension, where the labeled training data is used also during the learning. The third one is based on the second, but as an additional step we label the segments using a Hidden Markov Model trained on the labeled data.