Point the Point: Uyghur Morphological Segmentation Using PointerNetwork with GRU

Uyghur is an agglutinative language that has many morphemes. It is necessary for processing Uyghur to segment words into morphemes. This work is called morphological segmentation. Previous works treat morphological segmentation as a tagging task and classify each character as one of four classes, which are \(\{b,m,e,s\}\). However, these labels are not independent from each other, which makes the models easily overfitted. We propose a new method for the segmentation task. Instead of using these labels, we use only segmentation points for modeling. The model used in our method is more robust and easier to train than previous methods. Applying our model to Uyghur morphological segmentation, it achieves high accuracy and higher recall and f1 score than previous models.