Morphological Inflection

What happens when a neural transducer learns morphology from 100, 1,000, or 10,000 examples?

This demonstrator investigates how transition-based neural string transducers (Makarov & Clematide, 2020) generalize morphological inflection across four European languages under low-, medium-, and high-resource supervision.

Explore

Generate inflected forms across English, German, French, and Italian from lemma and feature inputs.

Compare

Examine side-by-side predictions across low (100), medium (1,000), and high (10,000) supervision regimes.

Understand

Inspect whether inputs were observed during training and how edit actions construct the output.

Data: Official CoNLL-SIGMORPHON 2017 Shared Task 1 benchmark data. Detailed documentation, UniMorph feature tag guides, and data split links are available in the Documentation tab.