This repository contains the training and evaluation notebooks for my bachelor thesis as well as the final report. The topic is "Morphology Disambiguation Using Tokens Frequencies". We tried to enhance the performance of an NN-based morphological tagger by training three models on different groups of tokens and then ensembling them. The groups were formed based on the frequencies: tokens with a higher frequency are reported to display different ambiguity properties. We hypothesized that the models would more effectively resolve ambiguities within each group, leading to an overall improvement in performance when combined through ensembling. In the result we managed to achieve a slight increase in scores.
