While part-of-speech tagging on small (e.g. universal) tagsets has reached human-like performance levels for many languages, full morphological tagging is still considerably more difficult, in particular for languages with rich morphology. At the same time, the "morphological features" column in the CoNLL-U data format typically includes features that do not really encode morphology (e.g. Typo).
The goal of this thesis would be to propose linguistically relevant evaluation measures for morphological tagging and validate them on a diverse subset of languages covered by the Universal Dependencies project.