Slovene Morphological and Word Formation Segmentation: A Novel Dataset and Evaluation

Abstract

We introduce the first publicly available manually annotated dataset for morphological segmentation and word-formation analysis for Slovene, containing 1,935 words annotated by two domain experts. The dataset provides three types of linguistic information: morphological and word-formation segments with zero-morpheme and simplex annotations. We present a four-stage annotation approach achieving inter-annotator agreement of 86.80% Krippendorff’s Alpha for morphological segmentation and 85.16% for word-formation segments. Computational validation using a morphological segmentation model achieves 87.78% BPR F1 on morphological segmentation and 83.05% on word-formation segments. Despite being smaller than previous datasets derived from non-public esources, our dataset enables high performance and supports reproducible research for morphological analysis tools for Slovene.

Resources

Details