Second call for participation in the 4th SIGMORPHON shared task: Crosslinguality and Context in Morphological Inflection! This year, we present three new tasks.
All details, data, code (with pretrained models), and instructions are available on the website (https://sigmorphon.github.io/sharedtasks/2019/). We provide an overview of the tasks below:
1. Crosslingual inflection generation in 100 language pairs (including 12 held-out surprise pairs). This is a twist on the primary inflection task we have offered in the past three years: The focus is on transfer into low-resource languages.
2. Context-driven morphological analysis and lemmatization. The data here are taken from the Universal Dependencies corpora, re-annotated according to the UniMorph schema. We will feature nearly 100 languages here.
3. An open challenge to improve the open-source state of the art. You may submit a strong, open-source system to address any previous task using their splits (SIGMORPHON 2016, CoNLL-SIGMORPHON 2017/2018).
We have released very strong baselines with pretrained models for the first two tasks.
The models are state-of-the-art on related tasks and have been modified for the current datasets. We encourage you to build on these codebases.
We look forward to your participation! Also, consider submitting a paper to our host workshop SIGMORPHON 2019. The call for papers is available: sigmorphon.github.io/workshops/2019.