A benchmark for morphology-aware tokenizers and language models in German.
Eight numbered tasks probe inflection, segmentation, derivation, compounding, affix
semantics, and whole-word meaning. Each task is a config; difficulty/generalization
tiers are splits.
from datasets import load_dataset
ds = load_dataset("yuanxin112/morphbench-de", "task3_derivation", split="test_main")
Tasks
task1_inflection… See the full description on the dataset page: https://huggingface.co/datasets/yuanxin112/morphbench-de.
