Two-step Approach to Unsupervised Morpheme Segmentation

This paper describes two steps of a morpheme boundary segmentation algorithm. The task is solely to find boundaries between morphemes bar and any further analysis such as phoneme deletions, insertions or alternations that may occur between or within morphemes. The algorithm presented here was designed under the premise that it is not supposed to utilize any knowledge about the language it should analyse. Neither is it supposed to rely on any kind of human supervision. The first step is to use a high-precision, low-recall algorithm to find a relatively small number of mostly correct segmentations, see (Bordag, 2005). In the second step, these segmentations are used to train a classification, which is then applied to all words to find morpheme boundaries within them.